diff --git "a/data/prs.json" "b/data/prs.json"
--- "a/data/prs.json"
+++ "b/data/prs.json"
@@ -1,6237 +1,7353 @@
 [
   {
-    "additions": 16,
-    "author": "sayakpaul",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Cc: @bhavya01 and @entrpn. Could I get a review on this? From the [logs](https://pastebin.com/KmgxMgzm) and the generated image, seems like things are working as expected. <img width=\"1024\" height=\"1024\" alt=\"image\"\u2026",
+    "additions": 25,
+    "author": "sywangyi",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "following case fail tests/models/transformers/test_models_transformer_bria.py::BriaTransformerLoRAHotSwapTests::test_hotswapping_compiled_model_linear_2 tests/models/transformers/test_models_transformer_bria.py::BriaTransformerLoRAHotSwapT\u2026",
+    "changed_files": 1,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13535",
+    "created_at": "2026-04-21T12:22:47Z",
+    "deletions": 6,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/diffusers/pull/13535/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13535",
+    "labels": [
+      "lora",
+      "size/S"
+    ],
+    "merged": false,
+    "number": 13535,
+    "review_comments_count": 1,
+    "state": "open",
+    "title": "fix the test_hotswapping_compiled_model_xxx case failure in xpu",
+    "updated_at": "2026-04-21T15:50:14Z"
+  },
+  {
+    "additions": 1,
+    "author": "kaixuanliu",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "This PR tries to fix 2 failed test cases for hiream pipeline tests: ``` tests/pipelines/hidream_image/test_pipeline_hidream.py::HiDreamImagePipelineFastTests::test_cpu_offload_forward_pass_twice tests/pipelines/hidream_image/test_pipeline_\u2026",
+    "changed_files": 1,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13534",
+    "created_at": "2026-04-21T07:41:02Z",
+    "deletions": 1,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/diffusers/pull/13534/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13534",
+    "labels": [
+      "tests",
+      "size/S"
+    ],
+    "merged": true,
+    "number": 13534,
+    "review_comments_count": 0,
+    "state": "closed",
+    "title": "Fix non-deterministic T5 outputs in HiDream pipeline tests",
+    "updated_at": "2026-04-21T16:28:41Z"
+  },
+  {
+    "additions": 2,
+    "author": "Ricardo-M-L",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "## What does this PR do? Both `AuraFlowAttnProcessor2_0` and `FusedAuraFlowAttnProcessor2_0` in `attention_processor.py` contain a copy-paste error where `attn.norm_added_q` is called on the encoder **key** projection while guarded by a ch\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13457",
-    "created_at": "2026-04-13T13:09:47Z",
-    "deletions": 0,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13533",
+    "created_at": "2026-04-21T06:18:39Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13457/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13457",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13533/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13533",
     "labels": [
-      "size/S",
-      "examples"
+      "models",
+      "size/S"
     ],
     "merged": false,
-    "number": 13457,
+    "number": 13533,
     "review_comments_count": 0,
     "state": "open",
-    "title": "[ptxla] fix pytorch xla inference on TPUs.",
-    "updated_at": "2026-04-13T13:37:13Z"
+    "title": "Fix AuraFlow attn processors applying norm_added_q to key projection",
+    "updated_at": "2026-04-21T18:25:49Z"
   },
   {
-    "additions": 1184,
-    "author": "HsiaWinter",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Fixes # (issue) change rope dtype from torch.float64 to torch.float32 to support more gpu backends.",
-    "changed_files": 14,
+    "additions": 2,
+    "author": "Ricardo-M-L",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## What this PR does When a user passes pre-computed `prompt_embeds` (or `negative_prompt_embeds`) alongside `num_images_per_prompt > 1`, `ErnieImagePipeline.__call__` did not replicate the provided embeddings \u2014 the embeds list kept its or\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13456",
-    "created_at": "2026-04-13T08:39:20Z",
-    "deletions": 0,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13532",
+    "created_at": "2026-04-21T06:07:21Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13456/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13456",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13532/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13532",
     "labels": [
-      "documentation",
-      "models",
-      "tests",
-      "size/L",
-      "utils",
+      "size/S",
       "pipelines"
     ],
     "merged": false,
-    "number": 13456,
+    "number": 13532,
     "review_comments_count": 0,
     "state": "open",
-    "title": "fix rope dtype from torch.float64 to torch.float32",
-    "updated_at": "2026-04-13T08:39:35Z"
+    "title": "Fix ErnieImagePipeline pre-computed prompt_embeds + num_images_per_prompt shape mismatch",
+    "updated_at": "2026-04-21T06:07:37Z"
   },
   {
-    "additions": 1,
-    "author": "Anai-Guo",
+    "additions": 0,
+    "author": "Ricardo-M-L",
     "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "## Problem The `--text_encoder_out_layers` CLI argument in `train_dreambooth_lora_flux2_klein.py` defaults to `[10, 20, 30]`, but the pipeline implementation (`pipeline_flux2_klein.py`) uses `(9, 18, 27)`. This off-by-one mismatch causes t\u2026",
+    "body_excerpt": "## What this PR does `train_sana_sprint_diffusers.py` ends training (under `--push_to_hub`) with: https://github.com/huggingface/diffusers/blob/main/examples/research_projects/sana/train_sana_sprint_diffusers.py#L1758-L1766 ```python if ar\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13455",
-    "created_at": "2026-04-13T06:56:28Z",
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13531",
+    "created_at": "2026-04-21T05:59:24Z",
     "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13455/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13455",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13531/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13531",
     "labels": [
       "size/S",
       "examples"
     ],
     "merged": false,
-    "number": 13455,
+    "number": 13531,
     "review_comments_count": 0,
     "state": "open",
-    "title": "fix: align text_encoder_out_layers default with pipeline for Flux 2 Klein",
-    "updated_at": "2026-04-13T06:56:41Z"
+    "title": "Fix AttributeError/TypeError in sana_sprint model-card save",
+    "updated_at": "2026-04-21T05:59:35Z"
   },
   {
-    "additions": 18,
-    "author": "sayakpaul",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Fix https://github.com/huggingface/diffusers/actions/runs/24241815924/job/70778130709?pr=13332",
-    "changed_files": 1,
+    "additions": 64,
+    "author": "jiqing-feng",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "## What does this PR do? This PR improves the TorchAO quantization testing infrastructure with several fixes: enabling `int4wo` tests on Intel XPU, implementing `_dequantize` for TorchAO, fixing input dtype mismatches, and fixing training\u2026",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13454",
-    "created_at": "2026-04-13T05:05:33Z",
-    "deletions": 0,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13530",
+    "created_at": "2026-04-21T05:59:22Z",
+    "deletions": 28,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13454/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13454",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13530/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13530",
     "labels": [
+      "quantization",
       "tests",
-      "size/S"
+      "size/M"
     ],
     "merged": false,
-    "number": 13454,
+    "number": 13530,
     "review_comments_count": 0,
     "state": "open",
-    "title": "[tests] xfail clip related issues.",
-    "updated_at": "2026-04-13T05:05:44Z"
+    "title": "Improve TorchAO quantization test coverage and XPU support",
+    "updated_at": "2026-04-21T14:59:43Z"
   },
   {
-    "additions": 7,
-    "author": "Beinsezii",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
+    "additions": 4,
+    "author": "Ricardo-M-L",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## What this PR does Both Wuerstchen prior training scripts build their HF Hub model card with an f-string that references `{args.weight_dtype}`: **`train_text_to_image_prior.py`** (lines 104-105): ```python pipe_prior = DiffusionPipeline.\u2026",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13453",
-    "created_at": "2026-04-13T04:58:22Z",
-    "deletions": 12,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13529",
+    "created_at": "2026-04-21T05:57:33Z",
+    "deletions": 4,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13453/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13453",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13529/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13529",
     "labels": [
       "size/S",
-      "pipelines"
+      "examples"
     ],
     "merged": false,
-    "number": 13453,
-    "review_comments_count": 2,
+    "number": 13529,
+    "review_comments_count": 0,
     "state": "open",
-    "title": "Fix SDXL Refiner with Higher Order Schedulers",
-    "updated_at": "2026-04-13T05:39:29Z"
+    "title": "Fix AttributeError in Wuerstchen prior training model card",
+    "updated_at": "2026-04-21T05:57:49Z"
   },
   {
-    "additions": 8,
-    "author": "chang-zhijie",
+    "additions": 9,
+    "author": "Ricardo-M-L",
     "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "This PR resolves the unsupported atten_mask shape error when running attention with NPU (Ascend) devices. Problem: The NPU\u2019s fusion attention operator (e.g., npu_fusion_attention) does not support automatic broadcasting for attention masks\u2026",
+    "body_excerpt": "## What this PR does The Intel/BF16 `textual_inversion_bf16.py` script reads `args.report_to` at the top of `main()`: https://github.com/huggingface/diffusers/blob/main/examples/research_projects/intel_opts/textual_inversion/textual_invers\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13451",
-    "created_at": "2026-04-13T03:17:22Z",
-    "deletions": 2,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13528",
+    "created_at": "2026-04-21T05:56:11Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13451/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13451",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13528/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13528",
     "labels": [
-      "models",
-      "size/S"
+      "size/S",
+      "examples"
     ],
     "merged": false,
-    "number": 13451,
+    "number": 13528,
     "review_comments_count": 0,
     "state": "open",
-    "title": "Fix attention_mask broadcasting for NPU compatibility",
-    "updated_at": "2026-04-13T06:17:45Z"
+    "title": "Fix AttributeError in textual_inversion_bf16 by adding --report_to arg",
+    "updated_at": "2026-04-21T05:56:27Z"
   },
   {
-    "additions": 2,
-    "author": "conanna",
-    "author_association": "FIRST_TIMER",
-    "body_excerpt": "# What does this PR do? `UniPCMultistepScheduler.set_timesteps` hardcodes `self.sigmas = self.sigmas.to(\"cpu\")` at the end of the method. This causes device mismatch errors in `multistep_uni_p_bh_update` and `multistep_uni_c_bh_update` whe\u2026",
+    "additions": 4,
+    "author": "Ricardo-M-L",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## What this PR does `ZImageOmniPipeline.__call__` defines `negative_condition_siglip_embeds` only inside the CFG guard: https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/z_image/pipeline_z_image_omni.py#L581-L582\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13450",
-    "created_at": "2026-04-12T19:13:13Z",
-    "deletions": 1,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13527",
+    "created_at": "2026-04-21T05:42:57Z",
+    "deletions": 3,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13450/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13450",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13527/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13527",
     "labels": [
       "size/S",
-      "schedulers"
+      "pipelines"
     ],
     "merged": false,
-    "number": 13450,
+    "number": 13527,
     "review_comments_count": 0,
     "state": "open",
-    "title": "fix: align sigmas device in UniPCMultistepScheduler.set_timesteps",
-    "updated_at": "2026-04-12T19:13:27Z"
+    "title": "Fix NameError in ZImageOmniPipeline when guidance_scale=0",
+    "updated_at": "2026-04-21T05:43:10Z"
   },
   {
-    "additions": 9,
+    "additions": 1,
     "author": "Ricardo-M-L",
     "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "## Summary - Fix all 9 QwenImage pipeline variants where CFG is silently disabled when passing `negative_prompt_embeds` with `None` mask - `encode_prompt()` converts all-ones masks to `None` as an optimization (line 266-267), but `has_neg_\u2026",
-    "changed_files": 9,
+    "body_excerpt": "## What this PR does `LongCatImageEditPipeline._encode_prompt` calls `len()` twice on an `int` when building its truncation warning message: https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/longcat_image/pipeline_\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13449",
-    "created_at": "2026-04-12T15:32:31Z",
-    "deletions": 9,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13526",
+    "created_at": "2026-04-21T05:37:58Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13449/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13449",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13526/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13526",
     "labels": [
       "size/S",
       "pipelines"
     ],
     "merged": false,
-    "number": 13449,
+    "number": 13526,
     "review_comments_count": 0,
     "state": "open",
-    "title": "fix: QwenImage pipelines silently disable CFG with pre-computed negative_prompt_embeds",
-    "updated_at": "2026-04-12T15:32:46Z"
+    "title": "Fix TypeError in LongCatImageEditPipeline truncation warning",
+    "updated_at": "2026-04-21T05:38:14Z"
   },
   {
-    "additions": 120,
-    "author": "hlky",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? As per #13446 `trust_remote_code` fails under several circumstances: - `pretrained_model_name_or_path` as Hub repo A and `custom_pipeline` as Hub repo B, `trust_remote_code` is bypassed and remote code runs from rep\u2026",
-    "changed_files": 6,
+    "additions": 4,
+    "author": "Ricardo-M-L",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## What this PR does In `LongCatAudioDiTPipeline.__call__`, positive prompts are normalized via `_normalize_text` before being encoded: ```python normalized_prompts = [_normalize_text(text) for text in prompt] ... prompt_embeds, prompt_emb\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13448",
-    "created_at": "2026-04-12T15:05:10Z",
-    "deletions": 24,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13525",
+    "created_at": "2026-04-21T05:32:36Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13448/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13448",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13525/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13525",
     "labels": [
-      "models",
-      "tests",
-      "size/M",
-      "modular-pipelines",
-      "utils",
+      "size/S",
       "pipelines"
     ],
     "merged": false,
-    "number": 13448,
-    "review_comments_count": 13,
+    "number": 13525,
+    "review_comments_count": 0,
     "state": "open",
-    "title": "Improve `trust_remote_code`",
-    "updated_at": "2026-04-13T08:21:52Z"
+    "title": "Normalize negative_prompt in LongCatAudioDiT pipeline for CFG symmetry",
+    "updated_at": "2026-04-21T05:32:50Z"
   },
   {
-    "additions": 2226,
-    "author": "Moran232",
+    "additions": 3,
+    "author": "Ricardo-M-L",
     "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "### Description **We are the JoyAI Team,** and this is the Diffusers implementation for the JoyAI-Image-Edit model. GitHub Repository: [https://github.com/jd-opensource/JoyAI-Image] Hugging Face Model: [https://huggingface.co/jdopensource/\u2026",
-    "changed_files": 10,
+    "body_excerpt": "## What does this PR do? The ORT variant of the unconditional training script, `examples/research_projects/onnxruntime/unconditional_image_generation/train_unconditional.py`, defines the tracker flag as `--logger` on the parser: ```python\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13444",
-    "created_at": "2026-04-10T15:34:41Z",
-    "deletions": 0,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13524",
+    "created_at": "2026-04-21T04:05:52Z",
+    "deletions": 3,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13444/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13444",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13524/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13524",
     "labels": [
-      "models",
-      "size/L",
-      "pipelines"
+      "size/S",
+      "examples"
     ],
     "merged": false,
-    "number": 13444,
-    "review_comments_count": 11,
+    "number": 13524,
+    "review_comments_count": 0,
     "state": "open",
-    "title": "[feat] JoyAI-JoyImage-Edit support",
-    "updated_at": "2026-04-11T05:42:17Z"
+    "title": "Fix AttributeError in onnxruntime train_unconditional (args.report_to \u2192 args.logger)",
+    "updated_at": "2026-04-21T04:06:06Z"
   },
   {
     "additions": 1,
-    "author": "sayakpaul",
-    "author_association": "MEMBER",
-    "body_excerpt": null,
+    "author": "Ricardo-M-L",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## What does this PR do? `extract_glyph_texts` in the HunyuanVideo 1.5 modular pipeline had a copy-paste typo in its regex: ```python # src/diffusers/modular_pipelines/hunyuan_video1_5/encoders.py pattern = r\"\\\"(.*?)\\\"|\\\"(.*?)\\\"\" ``` Both\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13443",
-    "created_at": "2026-04-10T15:02:22Z",
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13523",
+    "created_at": "2026-04-21T04:02:54Z",
     "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13443/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13443",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13523/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13523",
     "labels": [
-      "models",
-      "size/S"
+      "size/S",
+      "modular-pipelines"
     ],
-    "merged": true,
-    "number": 13443,
+    "merged": false,
+    "number": 13523,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "[core] fix fa4 integration",
-    "updated_at": "2026-04-13T05:58:18Z"
+    "state": "open",
+    "title": "Fix HunyuanVideo 1.5 modular glyph regex for curly quotes",
+    "updated_at": "2026-04-21T04:03:06Z"
   },
   {
-    "additions": 111,
-    "author": "sayakpaul",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Fix https://github.com/huggingface/diffusers/actions/runs/24241815924/job/70778130703?pr=13332",
-    "changed_files": 10,
-    "cluster_id": null,
-    "cluster_ids": [],
-    "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13442",
-    "created_at": "2026-04-10T14:23:50Z",
-    "deletions": 56,
+    "additions": 1,
+    "author": "Ricardo-M-L",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## What does this PR do? Fixes an `AttributeError` in `examples/research_projects/diffusion_dpo/train_diffusion_dpo.py` that crashes the IPO loss path. The IPO branch references `args.beta`: ```python elif args.loss_type == \"ipo\": losses =\u2026",
+    "changed_files": 1,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13522",
+    "created_at": "2026-04-21T03:57:51Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13442/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13442",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13522/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13522",
     "labels": [
-      "size/M",
+      "size/S",
       "examples"
     ],
     "merged": false,
-    "number": 13442,
+    "number": 13522,
     "review_comments_count": 0,
     "state": "open",
-    "title": "[wip] [tests] fix training tests",
-    "updated_at": "2026-04-11T04:35:33Z"
+    "title": "Fix AttributeError in diffusion_dpo --loss_type ipo",
+    "updated_at": "2026-04-21T03:58:04Z"
   },
   {
-    "additions": 4,
-    "author": "azolotenkov",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Fixes the prior-preservation batching issue in the Qwen Image DreamBooth LoRA script. PR #13396 corrected the repeat count under `--with_prior_preservation`, but the script still repeats prompt embeddings with `repe\u2026",
+    "additions": 1,
+    "author": "Ricardo-M-L",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## What does this PR do? Fixes a latent `AttributeError` in `examples/t2i_adapter/train_t2i_adapter_sdxl.py` that breaks `--resume_from_checkpoint`. The accelerate `load_model_hook` references `args.control_type`, which is never declared a\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13441",
-    "created_at": "2026-04-10T11:53:31Z",
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13521",
+    "created_at": "2026-04-21T03:55:25Z",
     "deletions": 3,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13441/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13441",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13521/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13521",
     "labels": [
       "size/S",
       "examples"
     ],
     "merged": false,
-    "number": 13441,
+    "number": 13521,
     "review_comments_count": 0,
     "state": "open",
-    "title": "Fix Qwen Image DreamBooth prior-preservation batch ordering",
-    "updated_at": "2026-04-10T11:53:42Z"
+    "title": "Fix AttributeError in train_t2i_adapter_sdxl checkpoint resume",
+    "updated_at": "2026-04-21T03:55:37Z"
   },
   {
-    "additions": 2,
-    "author": "akshan-main",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "\u2026nstead of latent resolution # What does this PR do? Fixes #13439 prepare_cond_latents_and_mask shadows the pixel height/width parameters with latent dims from latents.shape (line 614): ``` batch, channels, frames, height, width = latents.\u2026",
+    "additions": 1,
+    "author": "Ricardo-M-L",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## What does this PR do? Fixes an `AttributeError` in `examples/dreambooth/train_dreambooth_lora_flux2_img2img.py` that makes the validation flow unusable. The validation block references `args.validation_image_path`: ```python if args.val\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13440",
-    "created_at": "2026-04-10T10:04:45Z",
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13520",
+    "created_at": "2026-04-21T03:50:12Z",
+    "deletions": 1,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/diffusers/pull/13520/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13520",
+    "labels": [
+      "size/S",
+      "examples"
+    ],
+    "merged": false,
+    "number": 13520,
+    "review_comments_count": 0,
+    "state": "open",
+    "title": "Fix AttributeError in train_dreambooth_lora_flux2_img2img validation",
+    "updated_at": "2026-04-21T03:50:28Z"
+  },
+  {
+    "additions": 6,
+    "author": "Ricardo-M-L",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## What does this PR do? `FluxControlNetImg2ImgPipeline` and `FluxControlNetInpaintPipeline` already pre-resize `control_image` to `batch_size * num_images_per_prompt`, but only reshape `control_mode` to `[1, 1]`: ```python if control_mode\u2026",
+    "changed_files": 2,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13519",
+    "created_at": "2026-04-21T03:18:03Z",
     "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13440/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13440",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13519/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13519",
     "labels": [
       "size/S",
       "pipelines"
     ],
-    "merged": true,
-    "number": 13440,
+    "merged": false,
+    "number": 13519,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "Fix HunyuanVideo 1.5 I2V by preprocessing image at pixel resolution i\u2026",
-    "updated_at": "2026-04-10T19:54:36Z"
+    "state": "open",
+    "title": "Fix FluxControlNet img2img/inpainting with num_images_per_prompt>1",
+    "updated_at": "2026-04-21T03:18:20Z"
   },
   {
-    "additions": 96,
-    "author": "paulinebm",
-    "author_association": "MEMBER",
-    "body_excerpt": "cc @sayakpaul",
+    "additions": 1,
+    "author": "Ricardo-M-L",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## What does this PR do? Fixes #13062. \\`diffusers.utils.import_utils\\` probes for onnxruntime by first calling \\`find_spec(\\\"onnxruntime\\\")\\` and then iterating a list of package names to resolve the installed version via \\`importlib_meta\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13438",
-    "created_at": "2026-04-09T14:22:48Z",
-    "deletions": 26,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13518",
+    "created_at": "2026-04-21T03:06:07Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13438/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13438",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13518/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13518",
     "labels": [
-      "size/M",
-      "CI"
+      "size/S",
+      "utils"
     ],
     "merged": false,
-    "number": 13438,
-    "review_comments_count": 5,
+    "number": 13518,
+    "review_comments_count": 0,
     "state": "open",
-    "title": "add PR fork workable",
-    "updated_at": "2026-04-10T06:27:38Z"
+    "title": "Recognize onnxruntime-windowsml in onnxruntime version detection",
+    "updated_at": "2026-04-21T03:06:20Z"
   },
   {
-    "additions": 3110,
-    "author": "RuixiangMa",
+    "additions": 13,
+    "author": "Ricardo-M-L",
     "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "Support JoyAIImagePipeline * https://github.com/jd-opensource/JoyAI-Image * [ jdopensource/JoyAI-Image-Edit](https://huggingface.co/jdopensource/JoyAI-Image-Edit) | Prompt | image | |----------|-------------| |NO |<img width=\"512\" height=\"\u2026",
-    "changed_files": 22,
+    "body_excerpt": "## What does this PR do? Fixes #12540. When \\`SanaTransformer2DModel\\` is built with \\`guidance_embeds=True\\` and then driven by a pipeline that does not thread \\`guidance\\` through (\\`SanaPipeline\\`, \\`SanaPAGPipeline\\`), the transformer\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13437",
-    "created_at": "2026-04-09T11:42:22Z",
-    "deletions": 0,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13517",
+    "created_at": "2026-04-21T03:02:51Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13437/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13437",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13517/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13517",
     "labels": [
-      "documentation",
       "models",
-      "tests",
-      "size/L",
-      "pipelines",
-      "schedulers"
+      "size/S"
     ],
     "merged": false,
-    "number": 13437,
-    "review_comments_count": 6,
-    "state": "closed",
-    "title": "[Feat] support JoyAIImagePipeline ",
-    "updated_at": "2026-04-11T07:11:30Z"
+    "number": 13517,
+    "review_comments_count": 0,
+    "state": "open",
+    "title": "Fix SanaTransformer2DModel crash when guidance_embeds=True with non-guidance pipelines",
+    "updated_at": "2026-04-21T03:03:06Z"
   },
   {
-    "additions": 1,
-    "author": "NIK-TIGER-BILL",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "test",
+    "additions": 13,
+    "author": "Ricardo-M-L",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## What does this PR do? Fixes #12656. \\`AuraFlowPatchEmbed.pe_selection_index_based_on_dim\\` silently produces negative / out-of-range gather indices when the input latent grid is larger than the pretrained positional-embedding grid (\\`h_\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13436",
-    "created_at": "2026-04-08T23:17:59Z",
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13516",
+    "created_at": "2026-04-21T03:00:22Z",
     "deletions": 0,
-    "draft": true,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13436/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13436",
+    "draft": false,
+    "files_url": "https://github.com/huggingface/diffusers/pull/13516/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13516",
     "labels": [
+      "models",
       "size/S"
     ],
     "merged": false,
-    "number": 13436,
+    "number": 13516,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "test",
-    "updated_at": "2026-04-08T23:18:14Z"
+    "state": "open",
+    "title": "Fix AuraFlow CUDA context crash when latent size exceeds pos_embed_max_size",
+    "updated_at": "2026-04-21T03:00:37Z"
   },
   {
-    "additions": 282,
+    "additions": 7,
     "author": "Ricardo-M-L",
     "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "## Summary - Adds `scripts/convert_cogvideox_to_original.py`, a reverse conversion script that converts CogVideoX models (transformer and VAE) from diffusers format back to the original CogVideo checkpoint format - Reverses all weight name\u2026",
+    "body_excerpt": "## What does this PR do? `CosineDPMSolverMultistepScheduler.step` constructs `BrownianTreeNoiseSampler` with the **config** bounds: ```python self.noise_sampler = BrownianTreeNoiseSampler( model_output, sigma_min=self.config.sigma_min, sig\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13435",
-    "created_at": "2026-04-08T17:38:45Z",
-    "deletions": 0,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13515",
+    "created_at": "2026-04-21T02:45:25Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13435/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13435",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13515/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13515",
     "labels": [
-      "size/L"
+      "size/S",
+      "schedulers"
     ],
     "merged": false,
-    "number": 13435,
+    "number": 13515,
     "review_comments_count": 0,
     "state": "open",
-    "title": "Add CogVideoX diffusers-to-original format conversion script",
-    "updated_at": "2026-04-08T17:39:01Z"
+    "title": "Fix RecursionError in CosineDPMSolver Brownian tree noise sampler",
+    "updated_at": "2026-04-21T02:45:41Z"
   },
   {
-    "additions": 2,
+    "additions": 8,
     "author": "Ricardo-M-L",
     "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "## Summary Fixes two functional bugs in the experimental guiders module, both introduced in #11311 (Modular Diffusers Guiders): ### Bug 1: `TangentialClassifierFreeGuidance.is_conditional` \u2014 AttributeError The `is_conditional` property ref\u2026",
-    "changed_files": 2,
+    "body_excerpt": "## What does this PR do? `gather_size_by_comm` in `src/diffusers/models/_modeling_parallel.py` unconditionally calls `torch.accelerator.current_accelerator()` whenever the backend isn't CPU: ```python gather_device = \"cpu\" if \"cpu\" in comm\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13434",
-    "created_at": "2026-04-08T15:05:39Z",
-    "deletions": 2,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13514",
+    "created_at": "2026-04-21T02:35:14Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13434/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13434",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13514/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13514",
     "labels": [
-      "size/S",
-      "guiders"
+      "models",
+      "size/S"
     ],
     "merged": false,
-    "number": 13434,
+    "number": 13514,
     "review_comments_count": 0,
     "state": "open",
-    "title": "fix: correct AttributeError in TangentialCFG and NameError in FrequencyDecoupledGuidance",
-    "updated_at": "2026-04-08T15:06:29Z"
+    "title": "Fix context-parallel gather crash on PyTorch < 2.6",
+    "updated_at": "2026-04-21T02:35:28Z"
   },
   {
-    "additions": 1,
-    "author": "DN6",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Narrow scoped token to just access inference providers when labelling issues. <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release n\u2026",
+    "additions": 6,
+    "author": "Ricardo-M-L",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## What does this PR do? Fixes #12913. \\`Lumina2Pipeline.__call__\\` passes \\`latents.shape[1]\\` as \\`image_seq_len\\` when computing \\`mu\\` for the flow-match scheduler: \\`\\`\\`python image_seq_len = latents.shape[1] mu = calculate_shift( im\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13433",
-    "created_at": "2026-04-08T08:28:55Z",
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13513",
+    "created_at": "2026-04-21T02:31:16Z",
     "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13433/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13433",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13513/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13513",
     "labels": [
       "size/S",
-      "CI"
+      "pipelines"
     ],
-    "merged": true,
-    "number": 13433,
+    "merged": false,
+    "number": 13513,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "[CI] Use finegrained token for Issue Labeler",
-    "updated_at": "2026-04-08T09:18:27Z"
+    "state": "open",
+    "title": "Fix image_seq_len computation in Lumina2Pipeline",
+    "updated_at": "2026-04-21T02:31:30Z"
   },
   {
-    "additions": 1184,
-    "author": "HsiaWinter",
+    "additions": 7,
+    "author": "Ricardo-M-L",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## What does this PR do? Fixes #13243. When \\`use_dynamic_shifting=False\\` and \\`shift != 1.0\\` (the default SD3/FLUX-style config), calling \\`scheduler.set_timesteps(num_train_timesteps)\\` produced sigmas that were shifted **twice**, so t\u2026",
+    "changed_files": 1,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13512",
+    "created_at": "2026-04-21T02:23:14Z",
+    "deletions": 2,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/diffusers/pull/13512/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13512",
+    "labels": [
+      "size/S",
+      "schedulers"
+    ],
+    "merged": false,
+    "number": 13512,
+    "review_comments_count": 0,
+    "state": "open",
+    "title": "Fix double-shift in FlowMatchEulerDiscreteScheduler.__init__",
+    "updated_at": "2026-04-21T02:23:27Z"
+  },
+  {
+    "additions": 293,
+    "author": "RuixiangMa",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? We have introduced a new text-to-image model called ERNIE-Image, which will soon be open-sourced to the community. This PR includes the model architecture definition, the pipeline, as well as the related documentati\u2026",
-    "changed_files": 14,
+    "body_excerpt": "# What does this PR do? support AutoPipelineForText2Audio for ext2Audio model ``` import torch import soundfile as sf from diffusers import AutoPipelineForText2Audio pipeline = AutoPipelineForText2Audio.from_pretrained( \"stabilityai/stable\u2026",
+    "changed_files": 6,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 11,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13432",
-    "created_at": "2026-04-08T04:14:03Z",
-    "deletions": 0,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13511",
+    "created_at": "2026-04-21T00:29:42Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13432/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13432",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13511/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13511",
     "labels": [
       "documentation",
-      "models",
-      "tests",
       "size/L",
       "utils",
       "pipelines"
     ],
-    "merged": true,
-    "number": 13432,
-    "review_comments_count": 71,
-    "state": "closed",
-    "title": "Add ernie image",
-    "updated_at": "2026-04-13T05:37:37Z"
+    "merged": false,
+    "number": 13511,
+    "review_comments_count": 0,
+    "state": "open",
+    "title": "[Feat] support AutoPipelineForText2Audio",
+    "updated_at": "2026-04-21T01:53:56Z"
   },
   {
-    "additions": 2,
-    "author": "huemin-art",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Adds an optional `decoder_block_out_channels` parameter to `AutoencoderKLFlux2` so the decoder can use different channel widths than the encoder. This is needed to support the FLUX.2 small decoder VAE (`black-forest\u2026",
+    "additions": 10,
+    "author": "Ricardo-M-L",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## What does this PR do? Fixes a regression that aborts \\`examples/dreambooth/train_dreambooth_lora.py\\` on the first training step **after** the first validation, whenever the user combines \\`--mixed_precision=fp16\\` and \\`--validation_pr\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13428",
-    "created_at": "2026-04-06T23:48:03Z",
-    "deletions": 1,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13510",
+    "created_at": "2026-04-20T12:59:16Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13428/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13428",
-    "labels": [],
-    "merged": true,
-    "number": 13428,
+    "files_url": "https://github.com/huggingface/diffusers/pull/13510/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13510",
+    "labels": [
+      "size/S",
+      "examples"
+    ],
+    "merged": false,
+    "number": 13510,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "FLUX.2 small decoder",
-    "updated_at": "2026-04-07T01:59:40Z"
+    "state": "open",
+    "title": "Fix DreamBooth LoRA fp16 training crash after validation",
+    "updated_at": "2026-04-20T12:59:33Z"
   },
   {
-    "additions": 18,
-    "author": "asomoza",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Fixes the use of `record_stream` with `torchao` and `group offload`. Without this, the resulting image is wrong. |Without this PR|With this PR| |---|---| |<img width=\"1024\" height=\"1024\" alt=\"flux_benchmark_dynamic_\u2026",
+    "additions": 1,
+    "author": "Ricardo-M-L",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## What does this PR do? Fixes the off-by-one in the \\`--text_encoder_out_layers\\` CLI default of \\`examples/dreambooth/train_dreambooth_lora_flux2_klein.py\\`: the script was defaulting to \\`[10, 20, 30]\\` (the Flux 2 **non-Klein** default\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13427",
-    "created_at": "2026-04-06T17:37:24Z",
-    "deletions": 0,
-    "draft": true,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13427/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13427",
-    "labels": [],
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13509",
+    "created_at": "2026-04-20T12:53:59Z",
+    "deletions": 1,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/diffusers/pull/13509/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13509",
+    "labels": [
+      "size/S",
+      "examples"
+    ],
     "merged": false,
-    "number": 13427,
+    "number": 13509,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "[torchao] Fix `record_stream` with torchao and group offloading",
-    "updated_at": "2026-04-06T17:59:53Z"
+    "state": "open",
+    "title": "Fix Flux 2 Klein dreambooth default text_encoder_out_layers",
+    "updated_at": "2026-04-20T12:54:15Z"
   },
   {
-    "additions": 6,
-    "author": "Akash504-ai",
+    "additions": 51,
+    "author": "Ricardo-M-L",
     "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? This PR fixes a numerical stability issue in `rescale_noise_cfg` that can lead to division by zero and produce NaN/inf values during inference. ## Problem The current implementation performs: noise_pred_rescaled = n\u2026",
-    "changed_files": 3,
+    "body_excerpt": "## What does this PR do? Fixes a latent bug in `randn_tensor` where the MPS-specific info-log suppression has been broken since PR #1902 (Jan 2023). ### The bug ```python # src/diffusers/utils/torch_utils.py if isinstance(device, str): dev\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13426",
-    "created_at": "2026-04-06T15:44:38Z",
-    "deletions": 3,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13508",
+    "created_at": "2026-04-20T12:17:18Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13426/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13426",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/diffusers/pull/13508/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13508",
+    "labels": [
+      "close-to-merge",
+      "tests",
+      "size/M",
+      "utils"
+    ],
     "merged": false,
-    "number": 13426,
-    "review_comments_count": 0,
+    "number": 13508,
+    "review_comments_count": 1,
     "state": "open",
-    "title": "Fix division by zero in rescale_noise_cfg causing NaNs during inference",
-    "updated_at": "2026-04-11T06:54:31Z"
+    "title": "fix(randn_tensor): compare device.type, not torch.device, when suppressing MPS info log",
+    "updated_at": "2026-04-21T00:50:15Z"
   },
   {
-    "additions": 10,
-    "author": "sayakpaul",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Fixes https://github.com/huggingface/diffusers/actions/runs/24034864516/job/70092362359?pr=13417. Got missed in the earlier refactor.",
+    "additions": 12,
+    "author": "songh11",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13424",
-    "created_at": "2026-04-06T14:55:04Z",
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13507",
+    "created_at": "2026-04-20T11:02:53Z",
     "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13424/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13424",
-    "labels": [],
-    "merged": true,
-    "number": 13424,
+    "files_url": "https://github.com/huggingface/diffusers/pull/13507/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13507",
+    "labels": [
+      "models",
+      "size/S"
+    ],
+    "merged": false,
+    "number": 13507,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "[tests] fix autoencoderdc tests",
-    "updated_at": "2026-04-06T19:08:03Z"
+    "state": "open",
+    "title": "fix: handle non-zero storage_offset in sage attention for ring compatibility",
+    "updated_at": "2026-04-20T11:03:10Z"
   },
   {
     "additions": 2,
-    "author": "Xyc2016",
+    "author": "Ricardo-M-L",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "## Summary - Fix `LoRA's` \u2192 `LoRAs` in two places (lines 104 and 106) - plural, not possessive - Fix `trigger it` \u2192 `trigger them` (pronoun agreement: \"Some LoRAs\" is plural) ## Changes **Line 104:** > Among adapters, [LoRA's] are the most\u2026",
+    "body_excerpt": "## What does this PR do? Fixes the misleading error raised by `GGUFQuantizer.check_quantized_param_shape` when a loaded GGUF weight doesn't match the model's expected shape. ### Before ```python inferred_shape = _quant_shape_from_byte_shap\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13423",
-    "created_at": "2026-04-06T14:42:14Z",
-    "deletions": 2,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13504",
+    "created_at": "2026-04-19T15:13:43Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13423/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13423",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13504/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13504",
     "labels": [
-      "documentation",
+      "quantization",
       "size/S"
     ],
+    "merged": false,
+    "number": 13504,
+    "review_comments_count": 0,
+    "state": "open",
+    "title": "fix(gguf): correct mismatched-shape error message in check_quantized_param_shape",
+    "updated_at": "2026-04-21T14:26:11Z"
+  },
+  {
+    "additions": 47,
+    "author": "Ricardo-M-L",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "## What does this PR do? Closes #12504. `fourier_filter` (the FFT helper used by `enable_freeu`) already upcasts `bfloat16` inputs to `float32` before calling `torch.fft.fftn`, because PyTorch's FFT does not support bf16. The same is true\u2026",
+    "changed_files": 2,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13503",
+    "created_at": "2026-04-19T08:23:39Z",
+    "deletions": 2,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/diffusers/pull/13503/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13503",
+    "labels": [
+      "tests",
+      "size/S",
+      "utils"
+    ],
     "merged": true,
-    "number": 13423,
+    "number": 13503,
     "review_comments_count": 1,
     "state": "closed",
-    "title": "Fix grammar in LoRA documentation",
-    "updated_at": "2026-04-10T16:18:30Z"
+    "title": "fix(freeu): run FFT in float32 for float16 inputs to avoid ComplexHalf",
+    "updated_at": "2026-04-21T11:45:11Z"
   },
   {
-    "additions": 16,
-    "author": "akshan-main",
+    "additions": 19,
+    "author": "Dev-next-gen",
     "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "## What does this PR do? Fixes #13421 Adds `negative_prompt` parameter to `GLMImagePipeline.__call__()` and `encode_prompt()`. Previously, the unconditional prompt was hardcoded to `\"\"` when CFG was active, with no way for users to provide\u2026",
-    "changed_files": 1,
+    "body_excerpt": "## Summary Two independent bugs that together prevent `use_stream=True` group offloading and context-parallel (ring/Ulysses) attention from working on **AMD ROCm** (tested on gfx1101 / RX 7800 XT, ROCm 7.1, PyTorch 2.7). Both fixes are bac\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13422",
-    "created_at": "2026-04-06T14:23:33Z",
-    "deletions": 4,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13502",
+    "created_at": "2026-04-19T00:36:23Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13422/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13422",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/diffusers/pull/13502/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13502",
+    "labels": [
+      "models",
+      "size/S",
+      "hooks"
+    ],
     "merged": false,
-    "number": 13422,
+    "number": 13502,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "Add negative_prompt parameter to GLMImagePipeline",
-    "updated_at": "2026-04-06T23:58:06Z"
+    "state": "open",
+    "title": "fix(ROCm): stream sync race in group_offloading + LSE shape mismatch in ring/Ulysses attention",
+    "updated_at": "2026-04-19T00:36:38Z"
   },
   {
-    "additions": 15,
-    "author": "akshan-main",
+    "additions": 28,
+    "author": "GitGlimpse895",
     "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "## What does this PR do? Fixes #13416 Adds `negative_prompt` parameter to `Flux2KleinPipeline.__call__()`. Previously, the unconditional prompt was hardcoded to `\"\"` when CFG was active, with no way for users to provide a custom negative p\u2026",
-    "changed_files": 1,
+    "body_excerpt": "# What does this PR do? `VaeImageProcessor` exposes a `resample` config parameter (defaulting to `\"lanczos\"`) and correctly applies it when resizing **PIL images** via `PIL_INTERPOLATION`. However, the two `torch.nn.functional.interpolate`\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13420",
-    "created_at": "2026-04-06T14:12:33Z",
-    "deletions": 5,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13500",
+    "created_at": "2026-04-18T11:27:36Z",
+    "deletions": 10,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13420/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13420",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/diffusers/pull/13500/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13500",
+    "labels": [
+      "size/S",
+      "utils"
+    ],
     "merged": false,
-    "number": 13420,
-    "review_comments_count": 0,
-    "state": "closed",
-    "title": "Add negative_prompt to Flux2KleinPipeline",
-    "updated_at": "2026-04-06T23:58:45Z"
+    "number": 13500,
+    "review_comments_count": 1,
+    "state": "open",
+    "title": "Fix: use resample-aware bilinear+antialias interpolation for tensor/numpy resize in VaeImageProcessor",
+    "updated_at": "2026-04-21T16:11:58Z"
   },
   {
-    "additions": 14,
-    "author": "azolotenkov",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Fixes a device mismatch in the Flux2 DreamBooth training scripts when `--offload` is used and latents are not precomputed. non-cached path: `pixel_values` were converted to `vae.dtype` but not moved onto `accelerato\u2026",
-    "changed_files": 5,
+    "additions": 2,
+    "author": "sayakpaul",
+    "author_association": "MEMBER",
+    "body_excerpt": null,
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13417",
-    "created_at": "2026-04-05T20:31:33Z",
-    "deletions": 16,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13499",
+    "created_at": "2026-04-17T11:59:11Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13417/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13417",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/diffusers/pull/13499/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13499",
+    "labels": [
+      "documentation",
+      "size/S"
+    ],
     "merged": true,
-    "number": 13417,
+    "number": 13499,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix VAE offload encode device mismatch in DreamBooth scripts",
-    "updated_at": "2026-04-08T17:19:11Z"
+    "title": "[docs] add a mention of torchao and other backends in speed memory docs.",
+    "updated_at": "2026-04-18T03:54:31Z"
   },
   {
-    "additions": 16,
-    "author": "azolotenkov",
+    "additions": 1331,
+    "author": "akshan-main",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Fixes a prior-preservation batch size mismatch in the `Flux2 DreamBooth LoRA` scripts. When custom instance prompts are not used, `prompt_embeds` already contains concatenated instance + class embeddings under `--wi\u2026",
-    "changed_files": 2,
+    "body_excerpt": "# What does this PR do? Adds the modular pipeline for ErnieImage (`ErnieImageAutoBlocks` + `ErnieImageModularPipeline`). Parity verified on A100, bf16, 50 steps, 1024x1024 with `baidu/ERNIE-Image`: - MAD (AutoBlocks vs standard): 0.000033\u2026",
+    "changed_files": 13,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13415",
-    "created_at": "2026-04-04T18:08:44Z",
-    "deletions": 8,
+    "comments_count": 18,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13498",
+    "created_at": "2026-04-17T09:11:31Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13415/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13415",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/diffusers/pull/13498/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13498",
+    "labels": [
+      "tests",
+      "size/L",
+      "modular-pipelines",
+      "utils"
+    ],
     "merged": false,
-    "number": 13415,
-    "review_comments_count": 3,
+    "number": 13498,
+    "review_comments_count": 7,
     "state": "open",
-    "title": "Fix Flux2 DreamBooth prior preservation prompt repeats",
-    "updated_at": "2026-04-05T12:01:01Z"
+    "title": "Add Ernie-Image modular pipeline",
+    "updated_at": "2026-04-21T02:31:20Z"
   },
   {
-    "additions": 11,
-    "author": "Ricardo-M-L",
+    "additions": 1,
+    "author": "GitGlimpse895",
     "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "## Summary - Fix article: `a inpainting` \u2192 `an inpainting` in auto_pipeline.py - Fix article: `a aesthetic_score` \u2192 `an aesthetic_score` in SDXL pipelines - Fix docstring: remove extraneous `pass a` in quantizers/base.py - Fix grammar: `al\u2026",
-    "changed_files": 6,
+    "body_excerpt": "## What does this PR do? Fixes two bugs in `PyramidAttentionBroadcastHook.new_forward`: **1. Redundant `iteration == 0` condition** After every `reset_state()`, both `iteration` and `cache` are reset to `0` and `None` respectively. So at i\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13414",
-    "created_at": "2026-04-04T17:07:02Z",
-    "deletions": 11,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13497",
+    "created_at": "2026-04-17T06:31:03Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13414/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13414",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/diffusers/pull/13497/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13497",
+    "labels": [
+      "size/S",
+      "hooks"
+    ],
     "merged": false,
-    "number": 13414,
+    "number": 13497,
     "review_comments_count": 0,
     "state": "open",
-    "title": "fix: correct grammar in docstrings and error messages",
-    "updated_at": "2026-04-04T17:07:02Z"
+    "title": "hooks/pyramid_attention_broadcast: remove redundant iteration==0 guard and fix stale cache VRAM leak",
+    "updated_at": "2026-04-19T02:19:52Z"
   },
   {
-    "additions": 35,
-    "author": "Ricardo-M-L",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "## Summary - Fix `it's` \u2192 `its` (possessive form) in ~35 instances across ~32 model files - The phrase \"Check the superclass documentation for it's generic methods\" should use \"its\" (possessive), not \"it's\" (contraction of \"it is\") - Also\u2026",
-    "changed_files": 32,
+    "additions": 1,
+    "author": "kaixuanliu",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "This PR fixes failed test case: `tests/models/transformers/test_models_transformer_ernie_image.py::TestErnieImageTransformerCompile::test_torch_compile_repeated_blocks` @yiyixuxu pls help review, thx!",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13413",
-    "created_at": "2026-04-04T17:05:39Z",
-    "deletions": 35,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13496",
+    "created_at": "2026-04-17T05:59:52Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13413/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13413",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/diffusers/pull/13496/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13496",
+    "labels": [
+      "models",
+      "size/S"
+    ],
     "merged": false,
-    "number": 13413,
+    "number": 13496,
     "review_comments_count": 0,
     "state": "open",
-    "title": "fix: correct possessive 'it's' to 'its' in model docstrings",
-    "updated_at": "2026-04-04T17:05:39Z"
+    "title": "add _repeated_blocks for ErnieImageTransformer2DModel",
+    "updated_at": "2026-04-20T14:30:10Z"
   },
   {
-    "additions": 307,
-    "author": "VittoriaLanzo",
-    "author_association": "NONE",
-    "body_excerpt": "## What does this PR do? Fixes #13411. `LTXEulerAncestralRFScheduler.set_timesteps(sigmas=...)` silently accepted non-monotone sigma schedules. In `step()`, a reversed schedule computes: ``` downstep_ratio = 1 + (sigma_next / sigma - 1) *\u2026",
-    "changed_files": 4,
+    "additions": 78,
+    "author": "ParamChordiya",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## Summary Fixes #13487 - `load_lora_weights` crashes with `AttributeError: 'Flux2ModularPipeline' object has no attribute 'transformer'` when called on a modular sub-pipeline that doesn't have the transformer component (e.g., a text-encod\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13412",
-    "created_at": "2026-04-04T15:49:39Z",
-    "deletions": 11,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13495",
+    "created_at": "2026-04-17T04:23:15Z",
+    "deletions": 27,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13412/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13412",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/diffusers/pull/13495/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13495",
+    "labels": [
+      "lora",
+      "tests",
+      "size/M"
+    ],
     "merged": false,
-    "number": 13412,
-    "review_comments_count": 0,
-    "state": "closed",
-    "title": "[Scheduler] Add monotonicity validation and tests for LTXEulerAncestr\u2026",
-    "updated_at": "2026-04-06T22:54:31Z"
+    "number": 13495,
+    "review_comments_count": 1,
+    "state": "open",
+    "title": "Fix load_lora_weights crash on modular sub-pipelines without transformer",
+    "updated_at": "2026-04-21T00:59:55Z"
   },
   {
-    "additions": 40,
-    "author": "yiyixuxu",
-    "author_association": "MEMBER",
-    "body_excerpt": "follow https://github.com/huggingface/diffusers/pull/13334 moved `model-integrawtion/skillls/modular-conversion.md` to `.ai/modular.md`, and link it from `AGENTS.md`, `model-integrawtion/skillls` and `review-rules.md ` added a Gotcha secti\u2026",
-    "changed_files": 4,
+    "additions": 3,
+    "author": "RuixiangMa",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "Fixes # (issue) ``` RuntimeError: Error(s) in loading state_dict for LongCatAudioDiTTransformer: size mismatch for blocks.0.ffn.ff.0.weight: copying a param with shape torch.Size([9216, 2560]) from checkpoint, the shape in current model is\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13410",
-    "created_at": "2026-04-04T08:37:28Z",
-    "deletions": 14,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13494",
+    "created_at": "2026-04-16T16:23:46Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13410/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13410",
-    "labels": [],
-    "merged": false,
-    "number": 13410,
-    "review_comments_count": 1,
-    "state": "open",
-    "title": "[agent] add modular doc",
-    "updated_at": "2026-04-06T16:26:42Z"
+    "files_url": "https://github.com/huggingface/diffusers/pull/13494/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13494",
+    "labels": [
+      "models"
+    ],
+    "merged": true,
+    "number": 13494,
+    "review_comments_count": 0,
+    "state": "closed",
+    "title": "[Bugfix] Fix shape mismatch in LongCatAudioDiTTransformer conversion",
+    "updated_at": "2026-04-16T23:49:58Z"
   },
   {
-    "additions": 32,
-    "author": "awsr",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Fixes `NameError: name 'ftfy' is not defined` This was encountered by one of our users on SDNext. I'm not entirely sure why `ftfy` didn't get installed, but if it's optionally imported then it should also be optiona\u2026",
-    "changed_files": 16,
+    "additions": 1,
+    "author": "rtrompier",
+    "author_association": "MEMBER",
+    "body_excerpt": "Bump the pinned doc-builder SHA so that main documentation builds also sync to the HF bucket (dual-write).",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13409",
-    "created_at": "2026-04-04T02:02:23Z",
-    "deletions": 16,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13409/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13409",
-    "labels": [],
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13493",
+    "created_at": "2026-04-16T12:04:22Z",
+    "deletions": 1,
+    "draft": true,
+    "files_url": "https://github.com/huggingface/diffusers/pull/13493/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13493",
+    "labels": [
+      "size/S",
+      "CI"
+    ],
     "merged": false,
-    "number": 13409,
+    "number": 13493,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "Guard against calling optional `ftfy` module if not available",
-    "updated_at": "2026-04-07T01:53:00Z"
+    "state": "closed",
+    "title": "chore: bump doc-builder SHA for main doc build workflow",
+    "updated_at": "2026-04-17T09:16:26Z"
   },
   {
-    "additions": 7,
-    "author": "Ricardo-M-L",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "## Summary - Fix duplicate words: `schedule schedule` \u2192 `schedule`, `obtain obtain` \u2192 `obtain` - Fix article usage: `a image` \u2192 `an image` (2 instances), `a image-text` \u2192 `an image-text` (2 instances), `a upsampling` \u2192 `an upsampling` ## T\u2026",
-    "changed_files": 6,
+    "additions": 4,
+    "author": "sayakpaul",
+    "author_association": "MEMBER",
+    "body_excerpt": null,
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13408",
-    "created_at": "2026-04-03T23:40:41Z",
-    "deletions": 7,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13491",
+    "created_at": "2026-04-16T08:07:04Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13408/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13408",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/diffusers/pull/13491/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13491",
+    "labels": [
+      "tests",
+      "size/S"
+    ],
     "merged": false,
-    "number": 13408,
+    "number": 13491,
     "review_comments_count": 0,
     "state": "open",
-    "title": "fix: remove duplicate words and fix grammar in docstrings and error messages",
-    "updated_at": "2026-04-03T23:40:41Z"
+    "title": "[tests] fix group offloading with disk tests",
+    "updated_at": "2026-04-16T08:08:19Z"
   },
   {
-    "additions": 54,
-    "author": "Ricardo-M-L",
+    "additions": 8,
+    "author": "chang-zhijie",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "This PR resolves the unsupported atten_mask shape error when running attention with NPU (Ascend) devices. Problem: The NPU's fusion attention operator (e.g., npu_fusion_attention) does not support automatic broadcasting for attention masks\u2026",
+    "changed_files": 1,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 7,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13490",
+    "created_at": "2026-04-16T07:26:08Z",
+    "deletions": 9,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/diffusers/pull/13490/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13490",
+    "labels": [
+      "close-to-merge",
+      "models",
+      "size/S"
+    ],
+    "merged": true,
+    "number": 13490,
+    "review_comments_count": 2,
+    "state": "closed",
+    "title": "add _native_npu_attention support mask shape like [B,1,1,S]",
+    "updated_at": "2026-04-18T22:04:15Z"
+  },
+  {
+    "additions": 6,
+    "author": "ParamChordiya",
     "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "## Summary Add missing `f` prefix to ~50 string literals where variables like `{self.__class__}`, `{self.config}`, `{unet.config.*}`, `{type(...)}` are printed as literal text instead of being interpolated. ### Categories fixed: - Safety c\u2026",
-    "changed_files": 49,
+    "body_excerpt": "# What does this PR do? Fixes a device mismatch error in `UniPCMultistepScheduler` when using any form of offloading (`model_cpu_offload`, `sequential_cpu_offload`, or `group_offloading`). When offloading is active, `self.sigmas` may resid\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13407",
-    "created_at": "2026-04-03T23:40:27Z",
-    "deletions": 54,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13489",
+    "created_at": "2026-04-16T05:37:27Z",
+    "deletions": 6,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13407/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13407",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/diffusers/pull/13489/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13489",
+    "labels": [
+      "size/S",
+      "schedulers"
+    ],
     "merged": false,
-    "number": 13407,
+    "number": 13489,
     "review_comments_count": 0,
     "state": "open",
-    "title": "fix: add missing f-string prefixes across pipelines and schedulers",
-    "updated_at": "2026-04-03T23:40:27Z"
+    "title": "Fix UniPC scheduler device mismatch when using offloading",
+    "updated_at": "2026-04-21T01:44:59Z"
   },
   {
-    "additions": 25,
-    "author": "akshan-main",
+    "additions": 16,
+    "author": "yadferhad",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "## What does this PR do? Part of #13401 `QwenEmbedRope.forward()` copies `pos_freqs` and `neg_freqs` from CPU to GPU via `.to(device)` on every transformer forward call. These tensors are fixed at init and never change, so the repeated tra\u2026",
+    "body_excerpt": "## Summary - map Flux2 `guidance_in.*` non-diffusers LoRA keys to the diffusers guidance embedder - raise a clear error when Flux2 LoRAs are loaded into the text-only modular sub-pipeline - add regression tests covering both behaviors ## R\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 24,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13406",
-    "created_at": "2026-04-03T21:17:17Z",
-    "deletions": 10,
+    "comments_count": 5,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13486",
+    "created_at": "2026-04-15T17:24:00Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13406/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13406",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13486/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13486",
     "labels": [
-      "performance",
-      "models",
+      "lora",
       "size/S"
     ],
     "merged": true,
-    "number": 13406,
-    "review_comments_count": 2,
+    "number": 13486,
+    "review_comments_count": 8,
     "state": "closed",
-    "title": "Cache RoPE freqs on device to avoid repeated CPU-GPU copy in QwenImage",
-    "updated_at": "2026-04-10T16:38:12Z"
+    "title": "Fix Flux2 non-diffusers guidance LoRA conversion",
+    "updated_at": "2026-04-20T18:36:23Z"
   },
   {
-    "additions": 23,
-    "author": "andrewor14",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "**Summary:** TorchAO recently deprecated AffineQuantizedTensor and related classes (https://github.com/pytorch/ao/issues/2752). These will be removed in the next release. We should remove references of these classes in diffusers before the\u2026",
-    "changed_files": 2,
+    "additions": 222,
+    "author": "JingyaHuang",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Following #13289 Adding torch.compile support for Neuron to diffusers. Neuron supports `torch.compile` with the following mechanism to improve performance: ``` torch.compile(model, backend=\"neuron\", fullgraph=True)\u2026",
+    "changed_files": 13,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13405",
-    "created_at": "2026-04-03T20:31:12Z",
-    "deletions": 38,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13405/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13405",
-    "labels": [],
-    "merged": true,
-    "number": 13405,
-    "review_comments_count": 5,
-    "state": "closed",
-    "title": "Remove references to torchao's AffineQuantizedTensor",
-    "updated_at": "2026-04-08T15:43:10Z"
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13485",
+    "created_at": "2026-04-15T16:57:37Z",
+    "deletions": 26,
+    "draft": true,
+    "files_url": "https://github.com/huggingface/diffusers/pull/13485/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13485",
+    "labels": [
+      "lora",
+      "models",
+      "tests",
+      "size/L",
+      "utils",
+      "pipelines",
+      "examples"
+    ],
+    "merged": false,
+    "number": 13485,
+    "review_comments_count": 0,
+    "state": "open",
+    "title": "[Neuron] Enable `torch.comppile` compatibility with Neuron device",
+    "updated_at": "2026-04-15T17:27:20Z"
   },
   {
-    "additions": 152,
-    "author": "varaprasadtarunkumar",
-    "author_association": "NONE",
-    "body_excerpt": "## What does this PR do? Fixes a per-step `torch.tensor()` reconstruction in `QwenImageTransformer2DModel.forward()` that was identified as a known performance issue in the [profiling guide](https://github.com/huggingface/diffusers/tree/ma\u2026",
+    "additions": 39,
+    "author": "RuixiangMa",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? * add diffusers-format example(repo_id: ruixiangma/LongCat-AudioDiT-1B-Diffusers) * support seed parameter ``` import soundfile as sf import torch from diffusers import LongCatAudioDiTPipeline pipeline = LongCatAudi\u2026",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 6,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13404",
-    "created_at": "2026-04-03T18:58:56Z",
-    "deletions": 4,
+    "comments_count": 8,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13483",
+    "created_at": "2026-04-15T16:20:34Z",
+    "deletions": 16,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13404/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13404",
-    "labels": [],
-    "merged": false,
-    "number": 13404,
-    "review_comments_count": 2,
+    "files_url": "https://github.com/huggingface/diffusers/pull/13483/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13483",
+    "labels": [
+      "documentation",
+      "size/M",
+      "pipelines"
+    ],
+    "merged": true,
+    "number": 13483,
+    "review_comments_count": 4,
     "state": "closed",
-    "title": "fix: cache modulate_index tensor to eliminate per-step DtoH sync",
-    "updated_at": "2026-04-03T19:26:05Z"
+    "title": "[chore] Add diffusers-format example to LongCatAudioDiTPipeline",
+    "updated_at": "2026-04-16T04:52:15Z"
   },
   {
-    "additions": 4,
-    "author": "Ricardo-M-L",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "## Summary - Add missing `f` prefix to 4 log messages in `bnb_quantizer.py` where `{target_dtype}` and `{current_device}` are printed as literal text instead of being interpolated - Affects both 4-bit and 8-bit BnB quantizer classes ## Tes\u2026",
+    "additions": 6,
+    "author": "christopher5106",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "In standard pipeline, without negative prompt, the pipeline is run without CFG. In the modular pipeline, the absence of a negative_prompt is by default replaced by an empty string \"\", outputing negative_prompt_embeds. Better align for 1/ m\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13403",
-    "created_at": "2026-04-03T18:27:20Z",
-    "deletions": 6,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13482",
+    "created_at": "2026-04-15T13:43:08Z",
+    "deletions": 9,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13403/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13403",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/diffusers/pull/13482/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13482",
+    "labels": [
+      "size/S",
+      "modular-pipelines"
+    ],
     "merged": false,
-    "number": 13403,
+    "number": 13482,
     "review_comments_count": 0,
     "state": "open",
-    "title": "fix: add missing f-string prefixes in BnB quantizer log messages",
-    "updated_at": "2026-04-03T18:27:20Z"
+    "title": "Fix encoder difference between modular and standard pipelines for Qwen Image and consistency with other models",
+    "updated_at": "2026-04-16T07:51:14Z"
   },
   {
-    "additions": 8,
-    "author": "Ricardo-M-L",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "## Summary - Fix 6 instances of `have have` \u2192 `have` in modular pipeline error messages - Fix `to to` \u2192 `to` in UNet docstring - Fix `be be` \u2192 `be` in Flux Kontext inpaint pipeline docstring ## Test plan - [ ] Verify error messages read co\u2026",
+    "additions": 64,
+    "author": "DN6",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Fix a few failures in the BnB test suite. <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make\u2026",
     "changed_files": 8,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13402",
-    "created_at": "2026-04-03T18:23:45Z",
-    "deletions": 8,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13481",
+    "created_at": "2026-04-15T10:22:08Z",
+    "deletions": 39,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13402/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13402",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/diffusers/pull/13481/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13481",
+    "labels": [
+      "models",
+      "tests",
+      "size/M"
+    ],
     "merged": false,
-    "number": 13402,
-    "review_comments_count": 0,
+    "number": 13481,
+    "review_comments_count": 16,
     "state": "open",
-    "title": "fix: remove duplicate words in error messages and docstrings",
-    "updated_at": "2026-04-03T18:23:45Z"
+    "title": "[CI] Fix BnB tests",
+    "updated_at": "2026-04-21T17:34:17Z"
   },
   {
-    "additions": 5,
+    "additions": 7,
     "author": "sayakpaul",
     "author_association": "MEMBER",
-    "body_excerpt": null,
+    "body_excerpt": "# What does this PR do? Verify with the following: <details> <summary>Code</summary> ```py \"\"\"Minimal reproducer: XLA rejects out-of-bounds negative index that CPU/CUDA silently clamps.\"\"\" import torch import torch_xla.core.xla_model as xm\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13400",
-    "created_at": "2026-04-03T16:51:50Z",
-    "deletions": 1,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13480",
+    "created_at": "2026-04-15T10:07:06Z",
+    "deletions": 7,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13400/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13400",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/diffusers/pull/13480/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13480",
+    "labels": [
+      "models",
+      "size/S"
+    ],
     "merged": true,
-    "number": 13400,
+    "number": 13480,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Update README.md of the profiling guide",
-    "updated_at": "2026-04-03T17:00:13Z"
+    "title": "[core] fix autoencoderkl qwenimage for xla",
+    "updated_at": "2026-04-16T05:59:27Z"
   },
   {
-    "additions": 4,
-    "author": "Ricardo-M-L",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "Two types of exception handling bugs: **1. `raise \"string\"` in attention_processor.py (lines 328-332)** \\`raise \"string\"\\` causes \\`TypeError: exceptions must derive from BaseException\\` at runtime instead of the intended error message. Ch\u2026",
-    "changed_files": 2,
+    "additions": 237,
+    "author": "zhtmike",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? This PR adds support for attention mask input when using the attention backend with `set_attention_backend(\"flash\")`. With this change, `QwenImagePipeline` can run with the flash backend w/ or w/o `Ulysses SP`. For\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13398",
-    "created_at": "2026-04-03T05:41:14Z",
-    "deletions": 4,
+    "comments_count": 5,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13479",
+    "created_at": "2026-04-15T09:51:35Z",
+    "deletions": 18,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13398/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13398",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/diffusers/pull/13479/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13479",
+    "labels": [
+      "models",
+      "size/L"
+    ],
     "merged": false,
-    "number": 13398,
-    "review_comments_count": 0,
+    "number": 13479,
+    "review_comments_count": 14,
     "state": "open",
-    "title": "fix: use proper exception objects instead of bare raise/assert strings",
-    "updated_at": "2026-04-03T05:41:14Z"
+    "title": "Add attention mask input support for flash backend",
+    "updated_at": "2026-04-17T01:47:05Z"
   },
   {
     "additions": 3,
-    "author": "yiyixuxu",
-    "author_association": "MEMBER",
-    "body_excerpt": null,
-    "changed_files": 2,
-    "cluster_id": null,
-    "cluster_ids": [],
-    "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13397",
-    "created_at": "2026-04-03T05:40:53Z",
-    "deletions": 2,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13397/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13397",
-    "labels": [],
-    "merged": true,
-    "number": 13397,
-    "review_comments_count": 7,
-    "state": "closed",
-    "title": "improve Claude CI",
-    "updated_at": "2026-04-06T20:43:14Z"
-  },
-  {
-    "additions": 4,
-    "author": "chenyangzhu1",
+    "author": "songh11",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? I found that the same problem in https://github.com/huggingface/diffusers/issues/13292 also appears in Qwen-Image's dreambooth lora script. https://github.com/huggingface/diffusers/blob/8070f6ec54a7699d5ee285090d973\u2026",
+    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 8,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13396",
-    "created_at": "2026-04-03T04:50:53Z",
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13478",
+    "created_at": "2026-04-15T09:14:25Z",
     "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13396/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13396",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13478/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13478",
     "labels": [
       "size/S",
-      "examples"
+      "pipelines"
     ],
     "merged": true,
-    "number": 13396,
+    "number": 13478,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix(qwen-image dreambooth): correct prompt embed repeats when using `--with_prior_preservation`",
-    "updated_at": "2026-04-10T04:47:06Z"
+    "title": "fix(ernie-image): avoid locals() comprehension scope issue in callback kwargs",
+    "updated_at": "2026-04-17T02:10:09Z"
   },
   {
-    "additions": 19,
-    "author": "Akash504-ai",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? This PR fixes a missing validation in `DDPMScheduler.set_timesteps`. Currently, passing `num_inference_steps=0` does not raise an error and can lead to invalid internal state or unexpected behavior. This behavior is\u2026",
-    "changed_files": 4,
+    "additions": 1,
+    "author": "rtrompier",
+    "author_association": "MEMBER",
+    "body_excerpt": "Switch the PR doc upload flow from the legacy dataset push to the new HF bucket.",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13395",
-    "created_at": "2026-04-03T04:38:48Z",
-    "deletions": 0,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13395/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13395",
-    "labels": [],
-    "merged": false,
-    "number": 13395,
-    "review_comments_count": 0,
-    "state": "open",
-    "title": "Fix missing validation for num_inference_steps in DDPMScheduler and add tests",
-    "updated_at": "2026-04-03T04:38:48Z"
-  },
-  {
-    "additions": 4,
-    "author": "GalacticAvenger",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "## Summary Fixes several typos and grammar errors across the documentation: 1. **`controlling_generation.md`**: \"allows to generate\" \u2192 \"allows you to generate\" (missing object after \"allows\") 2. **`sdxl.md`**: \"it's refiner model\" \u2192 \"its r\u2026",
-    "changed_files": 3,
-    "cluster_id": null,
-    "cluster_ids": [],
-    "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13391",
-    "created_at": "2026-04-02T20:15:48Z",
-    "deletions": 4,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13476",
+    "created_at": "2026-04-15T08:25:11Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13391/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13391",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/diffusers/pull/13476/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13476",
+    "labels": [
+      "size/S",
+      "CI"
+    ],
     "merged": true,
-    "number": 13391,
+    "number": 13476,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix typos and grammar errors in documentation",
-    "updated_at": "2026-04-02T20:42:32Z"
+    "title": "chore: bump doc-builder SHA for PR upload workflow",
+    "updated_at": "2026-04-15T10:16:27Z"
   },
   {
-    "additions": 2144,
-    "author": "RuixiangMa",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Adds [LongCat-AudioDiT](https://github.com/meituan-longcat/LongCat-AudioDiT) model support to diffusers. Although LongCat-AudioDiT can be used for TTS-like generation, it is fundamentally a diffusion-based audio gen\u2026",
-    "changed_files": 17,
+    "additions": 229,
+    "author": "sayakpaul",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Add an example of model parallelism for Flux using PyTorch XLA. Tested on v5e-8. Cc: @entrpn if you could review.",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 6,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13390",
-    "created_at": "2026-04-02T17:19:18Z",
-    "deletions": 0,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13474",
+    "created_at": "2026-04-15T03:29:08Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13390/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13390",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13474/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13474",
     "labels": [
-      "documentation",
-      "models",
-      "tests",
       "size/L",
-      "utils",
-      "pipelines"
+      "examples"
     ],
-    "merged": false,
-    "number": 13390,
-    "review_comments_count": 120,
-    "state": "open",
-    "title": "[Feat] Adds LongCat-AudioDiT pipeline ",
-    "updated_at": "2026-04-11T13:00:07Z"
+    "merged": true,
+    "number": 13474,
+    "review_comments_count": 14,
+    "state": "closed",
+    "title": "add an example of spmd for flux on v5e-8",
+    "updated_at": "2026-04-17T09:16:29Z"
   },
   {
-    "additions": 2027,
-    "author": "akshan-main",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "## What does this PR do? Adds modular pipeline blocks for HunyuanVideo 1.5 with both text-to-video (`HunyuanVideo15Blocks`) and image-to-video (`HunyuanVideo15Image2VideoBlocks`). **Parity verified on Colab G4 GPU:** - T2V: MAD 0.000000 vs\u2026",
-    "changed_files": 13,
+    "additions": 124,
+    "author": "aws-zhenguo",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? This PR adds SDPA forward and backward calls for neuron backend to support ring attention. User would need [torch_neuronx](https://awsdocs-neuron.readthedocs-hosted.com/en/latest/frameworks/torch/pytorch-native-over\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 5,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13389",
-    "created_at": "2026-04-02T16:09:27Z",
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13473",
+    "created_at": "2026-04-14T22:28:18Z",
     "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13389/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13389",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13473/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13473",
     "labels": [
-      "tests",
-      "size/L",
-      "modular-pipelines",
-      "utils"
+      "performance",
+      "models",
+      "size/M"
     ],
     "merged": false,
-    "number": 13389,
+    "number": 13473,
     "review_comments_count": 0,
     "state": "open",
-    "title": "Add modular pipeline for HunyuanVideo 1.5",
-    "updated_at": "2026-04-11T05:00:53Z"
+    "title": "[attention backends] Add Neuron backend for context parallel",
+    "updated_at": "2026-04-15T01:35:26Z"
   },
   {
-    "additions": 160,
-    "author": "Ricardo-M-L",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "160 occurrences across 158 pipeline files use the incorrect phrase \"should be the same type to\" instead of \"should be the same type as\" in the negative_prompt type check error message. Example (before): ```python f\"`negative_prompt` should\u2026",
-    "changed_files": 158,
+    "additions": 11,
+    "author": "yiyixuxu",
+    "author_association": "MEMBER",
+    "body_excerpt": "based on https://github.com/huggingface/diffusers/pull/13432#discussion_r3078646179",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13388",
-    "created_at": "2026-04-02T16:08:54Z",
-    "deletions": 160,
+    "comments_count": 4,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13472",
+    "created_at": "2026-04-14T20:28:05Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13388/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13388",
-    "labels": [],
-    "merged": false,
-    "number": 13388,
-    "review_comments_count": 0,
-    "state": "open",
-    "title": "fix: correct grammar 'same type to' \u2192 'same type as' in error messages",
-    "updated_at": "2026-04-02T16:08:54Z"
+    "files_url": "https://github.com/huggingface/diffusers/pull/13472/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13472",
+    "labels": [
+      "size/S"
+    ],
+    "merged": true,
+    "number": 13472,
+    "review_comments_count": 11,
+    "state": "closed",
+    "title": "[agents docs] add float64 gotcha",
+    "updated_at": "2026-04-16T23:41:52Z"
   },
   {
-    "additions": 66,
-    "author": "chenyangzhu1",
+    "additions": 10,
+    "author": "Akash504-ai",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? When running the `train_dreambooth_lora_qwen_image.py `script with the `--with_prior_preservation` flag, the training crashes during the text embedding extraction phase. Fixing two distinct bugs related to tensor co\u2026",
+    "body_excerpt": "## What does this PR do? Fixes #13462 This PR resolves an issue where profiling decorators wrap bound methods, capturing the original instance and breaking instance isolation when components are deep-copied. Previously, decorating a bound\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 6,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13387",
-    "created_at": "2026-04-02T15:33:42Z",
-    "deletions": 2,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13471",
+    "created_at": "2026-04-14T18:39:45Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13387/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13387",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13471/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13471",
     "labels": [
-      "size/M",
+      "size/S",
       "examples"
     ],
     "merged": true,
-    "number": 13387,
+    "number": 13471,
     "review_comments_count": 2,
     "state": "closed",
-    "title": "Handle prompt embedding concat in Qwen dreambooth example",
-    "updated_at": "2026-04-10T03:43:33Z"
+    "title": "fix(profiling): preserve instance isolation when decorating methods",
+    "updated_at": "2026-04-15T02:16:21Z"
   },
   {
-    "additions": 46,
-    "author": "paulinebm",
+    "additions": 2,
+    "author": "McPatate",
     "author_association": "MEMBER",
-    "body_excerpt": "## \ud83d\udd12 Pin GitHub Actions to commit SHAs This PR pins all GitHub Actions to their exact commit SHA instead of mutable tags or branch names. **Why?** Pinning to a SHA prevents supply chain attacks where a tag (e.g. `v4`) could be moved to poi\u2026",
-    "changed_files": 11,
+    "body_excerpt": "Bumping the minimum safetensors version to its new (pre)release!",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13385",
-    "created_at": "2026-04-02T09:17:16Z",
-    "deletions": 46,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13470",
+    "created_at": "2026-04-14T17:27:04Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13385/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13385",
-    "labels": [],
-    "merged": true,
-    "number": 13385,
-    "review_comments_count": 0,
-    "state": "closed",
-    "title": "\ud83d\udd12 Pin GitHub Actions to commit SHAs",
-    "updated_at": "2026-04-02T15:34:46Z"
-  },
-  {
-    "additions": 14,
-    "author": "jiqing-feng",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "## What does this PR do? Fixes Dynamo `lru_cache` warnings when using `torch.compile` on diffusion pipelines. Two changes: 1. **`attention_dispatch.py`**: `dispatch_attention_fn` calls `is_torch_version(\">=\", \"2.5.0\")` at runtime, which is\u2026",
-    "changed_files": 2,
-    "cluster_id": null,
-    "cluster_ids": [],
-    "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13384",
-    "created_at": "2026-04-02T06:35:02Z",
-    "deletions": 2,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13384/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13384",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/diffusers/pull/13470/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13470",
+    "labels": [
+      "size/S"
+    ],
     "merged": true,
-    "number": 13384,
-    "review_comments_count": 3,
+    "number": 13470,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix Dynamo `lru_cache` warnings during `torch.compile`",
-    "updated_at": "2026-04-03T05:12:55Z"
+    "title": "feat: bump safetensors to `0.8.0-rc.0`",
+    "updated_at": "2026-04-17T02:12:41Z"
   },
   {
-    "additions": 160,
-    "author": "yiyixuxu",
+    "additions": 6,
+    "author": "sayakpaul",
     "author_association": "MEMBER",
-    "body_excerpt": null,
-    "changed_files": 3,
+    "body_excerpt": "# What does this PR do? https://github.com/huggingface/diffusers/actions/runs/24397001944/job/71257467343 Safest is to xfail.",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13382",
-    "created_at": "2026-04-01T20:08:19Z",
-    "deletions": 0,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13469",
+    "created_at": "2026-04-14T13:07:30Z",
+    "deletions": 3,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13382/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13382",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/diffusers/pull/13469/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13469",
+    "labels": [
+      "tests",
+      "size/S"
+    ],
     "merged": true,
-    "number": 13382,
-    "review_comments_count": 11,
+    "number": 13469,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "[docs] add auto docstring and parameter templates documentation for m\u2026",
-    "updated_at": "2026-04-02T20:34:46Z"
+    "title": "[tests] fix deprecated attention processor testing.",
+    "updated_at": "2026-04-14T16:35:45Z"
   },
   {
-    "additions": 984,
-    "author": "asomoza",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Add a skill for all the optimizations that can be done with diffusers. At the moment it has a lot of information and still with this the LLM (Opus 4.6) is still a hit or miss with the optimizations, specially with w\u2026",
-    "changed_files": 7,
+    "additions": 18,
+    "author": "Anai-Guo",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "Fixes #13462 ## Root cause `annotate_pipeline` called `setattr(component, method_name, annotate(method, label))` where `method` is a **bound method** (retrieved via `getattr(component, method_name)`). The resulting wrapper is a plain funct\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13381",
-    "created_at": "2026-04-01T14:42:10Z",
-    "deletions": 0,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13468",
+    "created_at": "2026-04-14T06:44:25Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13381/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13381",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/diffusers/pull/13468/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13468",
+    "labels": [
+      "size/S",
+      "examples"
+    ],
     "merged": false,
-    "number": 13381,
-    "review_comments_count": 0,
+    "number": 13468,
+    "review_comments_count": 1,
     "state": "open",
-    "title": "[Agents] Optimizations skill",
-    "updated_at": "2026-04-01T14:42:10Z"
+    "title": "fix(profiling): patch sub-component methods at class level to prevent instance-isolation bug",
+    "updated_at": "2026-04-14T12:23:59Z"
   },
   {
-    "additions": 405,
-    "author": "DN6",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Addresses suggestions brought up in #13343 by adding scripts/workflows to auto labels issues and PRs . For Issues: Use `Qwen/Qwen3.5-35B-A3B` via Inference Providers to parse issues titles and content and assign app\u2026",
-    "changed_files": 5,
+    "additions": 11,
+    "author": "GitGlimpse895",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## What does this PR do? Fixes two bugs in `PyramidAttentionBroadcastHook.new_forward`: 1. **Redundant `iteration == 0` condition** \u2014 `self.state.cache is None` already covers the first-call case after every `reset_state`, making the extra\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13380",
-    "created_at": "2026-04-01T13:23:14Z",
-    "deletions": 0,
+    "comments_count": 6,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13467",
+    "created_at": "2026-04-14T04:29:39Z",
+    "deletions": 12,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13380/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13380",
-    "labels": [],
-    "merged": true,
-    "number": 13380,
-    "review_comments_count": 11,
+    "files_url": "https://github.com/huggingface/diffusers/pull/13467/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13467",
+    "labels": [
+      "size/S",
+      "hooks"
+    ],
+    "merged": false,
+    "number": 13467,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "[CI] Add PR/Issue Auto Labeler",
-    "updated_at": "2026-04-07T04:32:19Z"
+    "title": "hooks/pyramid_attention_broadcast: fix redundant recompute at iteration 0 and free stale cache when outside timestep range",
+    "updated_at": "2026-04-17T06:32:56Z"
   },
   {
-    "additions": 216,
-    "author": "Sunhill666",
+    "additions": 8,
+    "author": "Anai-Guo",
     "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
-    "changed_files": 16,
+    "body_excerpt": "## What does this PR do? Adds a monotonicity check to `LTXEulerAncestralRFScheduler.set_timesteps()` when an explicit sigma schedule is provided. **Bug:** When `sigmas` are supplied explicitly (ComfyUI-style), `set_timesteps()` accepts non\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 5,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13379",
-    "created_at": "2026-04-01T13:12:53Z",
-    "deletions": 104,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13466",
+    "created_at": "2026-04-14T04:16:25Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13379/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13379",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13466/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13466",
     "labels": [
-      "tests",
       "size/S",
-      "size/L",
-      "modular-pipelines",
-      "pipelines"
+      "schedulers"
     ],
     "merged": false,
-    "number": 13379,
+    "number": 13466,
     "review_comments_count": 0,
     "state": "open",
-    "title": "fix(qwen): fix CFG failing when passing neg prompt embeds with none mask",
-    "updated_at": "2026-04-13T13:22:38Z"
+    "title": "fix(LTXEulerAncestralRFScheduler): validate monotonicity of explicit sigma schedules",
+    "updated_at": "2026-04-14T04:16:38Z"
   },
   {
-    "additions": 1996,
-    "author": "akshan-main",
+    "additions": 8,
+    "author": "HsiaWinter",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Adds modular pipeline support for LTX Video, covering both text-to-video and image-to-video. The implementation follows the same structure as the existing Wan modular pipeline. ### Text-to-video ``` LTXBlocks (Seque\u2026",
-    "changed_files": 13,
+    "body_excerpt": "# What does this PR do? Fix attention_mask broadcasting for NPU compatibility",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 22,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13378",
-    "created_at": "2026-04-01T10:57:41Z",
-    "deletions": 0,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13465",
+    "created_at": "2026-04-14T03:34:39Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13378/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13378",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13465/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13465",
     "labels": [
-      "tests",
-      "size/L",
-      "modular-pipelines",
-      "utils"
+      "models",
+      "size/S"
     ],
-    "merged": true,
-    "number": 13378,
-    "review_comments_count": 42,
-    "state": "closed",
-    "title": "[modular] Add LTX Video modular pipeline",
-    "updated_at": "2026-04-11T03:05:39Z"
+    "merged": false,
+    "number": 13465,
+    "review_comments_count": 4,
+    "state": "open",
+    "title": "fix npu compatibility",
+    "updated_at": "2026-04-16T07:30:08Z"
   },
   {
-    "additions": 11,
-    "author": "hf-security-analysis[bot]",
+    "additions": 3,
+    "author": "HsiaWinter",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "Update `.github/workflows/claude_review.yml` workflow configuration. cc @sayakpaul Closes huggingface/tracking-issues#42",
+    "body_excerpt": "# What does this PR do? Fixes # (issue) Fix rope dtype to float32 to support some old gpu backends and opt the time_proj dtype to support gguf",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13374",
-    "created_at": "2026-03-31T09:26:39Z",
-    "deletions": 4,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13464",
+    "created_at": "2026-04-14T01:48:39Z",
+    "deletions": 3,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13374/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13374",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/diffusers/pull/13464/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13464",
+    "labels": [
+      "models",
+      "size/S"
+    ],
     "merged": true,
-    "number": 13374,
+    "number": 13464,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "chore: update claude_review.yml",
-    "updated_at": "2026-04-01T04:48:30Z"
+    "title": "fix some dtype issue for gguf / some gpu backends",
+    "updated_at": "2026-04-14T04:41:01Z"
   },
   {
-    "additions": 63,
-    "author": "christopher5106",
+    "additions": 1,
+    "author": "entrpn",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "This PR adds local flash attention 4. Package required: [flash-attn-4](https://pypi.org/project/flash-attn-4/) (flash_attn.cute) Note: flash_4_hub already in the code but points to kernels-staging/flash-attn4 not publicly available. When i\u2026",
-    "changed_files": 3,
+    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13373",
-    "created_at": "2026-03-31T08:51:42Z",
-    "deletions": 0,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13463",
+    "created_at": "2026-04-14T00:19:13Z",
+    "deletions": 4,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13373/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13373",
-    "labels": [],
-    "merged": false,
-    "number": 13373,
+    "files_url": "https://github.com/huggingface/diffusers/pull/13463/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13463",
+    "labels": [
+      "size/S",
+      "pipelines"
+    ],
+    "merged": true,
+    "number": 13463,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "FA4 local (_flash_4 mode)",
-    "updated_at": "2026-03-31T09:04:04Z"
+    "state": "closed",
+    "title": "[ptxla] fix pytorch xla inference on TPUs.",
+    "updated_at": "2026-04-14T03:51:26Z"
   },
   {
-    "additions": 173,
-    "author": "sayakpaul",
-    "author_association": "MEMBER",
-    "body_excerpt": null,
+    "additions": 13,
+    "author": "hitchhiker3010",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? Fixes performance issues identified by profiling `ZImagePipeline` with `torch.profiler` as part of #13401 . ### What does this PR do? Profiled `ZImagePipeline` (using `Tongyi-MAI/Z-Image-Turbo`) in both eager and `t\u2026",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13371",
-    "created_at": "2026-03-31T05:30:09Z",
-    "deletions": 42,
+    "comments_count": 6,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13461",
+    "created_at": "2026-04-13T20:40:44Z",
+    "deletions": 11,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13371/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13371",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/diffusers/pull/13461/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13461",
+    "labels": [
+      "performance",
+      "models",
+      "size/S",
+      "pipelines"
+    ],
     "merged": true,
-    "number": 13371,
-    "review_comments_count": 3,
+    "number": 13461,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "[tests] refactor wan autoencoder tests",
-    "updated_at": "2026-04-03T05:36:46Z"
+    "title": "Remove compile bottlenecks from ZImage pipeline",
+    "updated_at": "2026-04-16T15:30:26Z"
   },
   {
-    "additions": 133,
-    "author": "HMAKT99",
+    "additions": 9,
+    "author": "neo",
     "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "## What this does Adds a community example for embedding AI provenance metadata into diffusers-generated images. When you generate an image with diffusers, the output carries no metadata indicating it's AI-generated. This utility embeds pr\u2026",
+    "body_excerpt": "# What does this PR do? This is to replicate what transformers already has: https://github.com/huggingface/transformers/blob/357f414cae88d5f592105bd97ec4a24ef7735f9f/src/transformers/dynamic_module_utils.py#L296-L310 from: https://github.c\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13370",
-    "created_at": "2026-03-30T15:12:13Z",
-    "deletions": 0,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13460",
+    "created_at": "2026-04-13T14:46:02Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13370/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13370",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/diffusers/pull/13460/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13460",
+    "labels": [
+      "size/S",
+      "utils"
+    ],
     "merged": false,
-    "number": 13370,
+    "number": 13460,
     "review_comments_count": 0,
     "state": "open",
-    "title": "Add AI image provenance community example",
-    "updated_at": "2026-04-13T06:27:45Z"
+    "title": "Add hash check for dynamic modules cache",
+    "updated_at": "2026-04-13T15:23:58Z"
   },
   {
-    "additions": 33,
-    "author": "sayakpaul",
-    "author_association": "MEMBER",
-    "body_excerpt": null,
+    "additions": 16,
+    "author": "neo",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? Currently the behaviour/final module path of the are different in the following way: - if it's a standalone custom `AutoModel`, the dynamic module is being loaded directly from the hub into `diffusers_modules.local.\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13369",
-    "created_at": "2026-03-30T10:00:40Z",
-    "deletions": 27,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13459",
+    "created_at": "2026-04-13T14:45:54Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13369/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13369",
-    "labels": [],
-    "merged": true,
-    "number": 13369,
+    "files_url": "https://github.com/huggingface/diffusers/pull/13459/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13459",
+    "labels": [
+      "size/S",
+      "utils"
+    ],
+    "merged": false,
+    "number": 13459,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "[tests] refactor autoencoderdc tests",
-    "updated_at": "2026-04-06T09:10:25Z"
+    "state": "open",
+    "title": "Ensure consistent dynamic module path for custom components",
+    "updated_at": "2026-04-13T14:46:07Z"
   },
   {
-    "additions": 37,
-    "author": "sayakpaul",
-    "author_association": "MEMBER",
-    "body_excerpt": null,
-    "changed_files": 1,
+    "additions": 20,
+    "author": "neo",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? by using `get_class_obj_and_candidates`, this now can also check against custom components being passed in and not error out as not found. ##### side note I also noticed the second argument `library` from `maybe_rai\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13368",
-    "created_at": "2026-03-30T08:03:51Z",
-    "deletions": 43,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13458",
+    "created_at": "2026-04-13T14:45:40Z",
+    "deletions": 10,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13368/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13368",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13458/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13458",
     "labels": [
-      "tests",
-      "size/M"
+      "size/S",
+      "pipelines"
     ],
     "merged": false,
-    "number": 13368,
+    "number": 13458,
     "review_comments_count": 1,
     "state": "open",
-    "title": "[tests] refactor autoencoderkl tests",
-    "updated_at": "2026-04-12T06:54:07Z"
+    "title": "Accommodate custom components in `maybe_raise_or_warn`",
+    "updated_at": "2026-04-13T17:33:27Z"
   },
   {
-    "additions": 12,
-    "author": "NIK-TIGER-BILL",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "## What does this PR do? Adds input validation for the `eta` parameter in `DDIMPipeline.__call__`. Fixes #13362 ### Context The DDIM paper defines \u03b7 (eta) as a value that lies strictly in [0, 1]: - \u03b7 = 0 \u2192 deterministic DDIM - \u03b7 = 1 \u2192 DDPM\u2026",
+    "additions": 16,
+    "author": "sayakpaul",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Cc: @bhavya01 and @entrpn. Could I get a review on this? From the [logs](https://pastebin.com/KmgxMgzm) and the generated image, seems like things are working as expected. <img width=\"1024\" height=\"1024\" alt=\"image\"\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13367",
-    "created_at": "2026-03-30T07:15:17Z",
+    "comments_count": 5,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13457",
+    "created_at": "2026-04-13T13:09:47Z",
     "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13367/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13367",
-    "labels": [],
-    "merged": true,
-    "number": 13367,
-    "review_comments_count": 3,
+    "files_url": "https://github.com/huggingface/diffusers/pull/13457/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13457",
+    "labels": [
+      "size/S",
+      "examples"
+    ],
+    "merged": false,
+    "number": 13457,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "fix(ddim): validate eta is in [0, 1] in DDIMPipeline",
-    "updated_at": "2026-04-03T02:07:29Z"
+    "title": "[ptxla] fix pytorch xla inference on TPUs.",
+    "updated_at": "2026-04-14T03:25:00Z"
   },
   {
-    "additions": 2,
-    "author": "sayakpaul",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? To unblock the refactor of existing autoencoder tests to use the latest tester mixins, I will merge this PR once the CI is green.",
-    "changed_files": 1,
+    "additions": 1190,
+    "author": "HsiaWinter",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? Fixes # (issue) change rope dtype from torch.float64 to torch.float32 to support more gpu backends.",
+    "changed_files": 14,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13366",
-    "created_at": "2026-03-30T05:18:53Z",
-    "deletions": 2,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13456",
+    "created_at": "2026-04-13T08:39:20Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13366/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13366",
-    "labels": [],
-    "merged": true,
-    "number": 13366,
+    "files_url": "https://github.com/huggingface/diffusers/pull/13456/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13456",
+    "labels": [
+      "documentation",
+      "models",
+      "tests",
+      "size/L",
+      "utils",
+      "pipelines"
+    ],
+    "merged": false,
+    "number": 13456,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "[ci] move to assert instead of self.Assert*",
-    "updated_at": "2026-03-30T05:39:18Z"
+    "state": "open",
+    "title": "fix rope dtype from torch.float64 to torch.float32",
+    "updated_at": "2026-04-14T03:28:21Z"
   },
   {
-    "additions": 32,
-    "author": "sayakpaul",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? https://github.com/huggingface/diffusers/actions/runs/23678409392/job/68986000951",
+    "additions": 1,
+    "author": "Anai-Guo",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## Problem The `--text_encoder_out_layers` CLI argument in `train_dreambooth_lora_flux2_klein.py` defaults to `[10, 20, 30]`, but the pipeline implementation (`pipeline_flux2_klein.py`) uses `(9, 18, 27)`. This off-by-one mismatch causes t\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13365",
-    "created_at": "2026-03-30T05:06:45Z",
-    "deletions": 3,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13455",
+    "created_at": "2026-04-13T06:56:28Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13365/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13365",
-    "labels": [],
-    "merged": true,
-    "number": 13365,
-    "review_comments_count": 6,
-    "state": "closed",
-    "title": "[ci] support claude reviewing on forks.",
-    "updated_at": "2026-03-31T09:26:12Z"
+    "files_url": "https://github.com/huggingface/diffusers/pull/13455/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13455",
+    "labels": [
+      "size/S",
+      "examples"
+    ],
+    "merged": false,
+    "number": 13455,
+    "review_comments_count": 0,
+    "state": "open",
+    "title": "fix: align text_encoder_out_layers default with pipeline for Flux 2 Klein",
+    "updated_at": "2026-04-13T06:56:41Z"
   },
   {
-    "additions": 1,
-    "author": "YangKai0616",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Per the title. Ref PR https://github.com/huggingface/diffusers/pull/13339 .",
+    "additions": 18,
+    "author": "sayakpaul",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Fix https://github.com/huggingface/diffusers/actions/runs/24241815924/job/70778130709?pr=13332",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13364",
-    "created_at": "2026-03-30T04:20:27Z",
-    "deletions": 2,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13454",
+    "created_at": "2026-04-13T05:05:33Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13364/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13364",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/diffusers/pull/13454/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13454",
+    "labels": [
+      "tests",
+      "size/S"
+    ],
     "merged": true,
-    "number": 13364,
+    "number": 13454,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix MotionConv2d to cast blur_kernel to input dtype instead of reverse",
-    "updated_at": "2026-03-31T09:53:12Z"
+    "title": "[tests] xfail clip related issues.",
+    "updated_at": "2026-04-14T17:03:25Z"
   },
   {
-    "additions": 25,
-    "author": "andrew-w-ross",
+    "additions": 7,
+    "author": "Beinsezii",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Fix's loading `ModularPipelines` loading component specs where `pretrained_model_name_or_path` is a path. Didn't bother opening an issue considering the tiny size of the fix. <!-- Congratulations! You've made it thi\u2026",
+    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13363",
-    "created_at": "2026-03-30T01:44:18Z",
-    "deletions": 2,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13453",
+    "created_at": "2026-04-13T04:58:22Z",
+    "deletions": 12,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13363/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13363",
-    "labels": [],
-    "merged": true,
-    "number": 13363,
-    "review_comments_count": 1,
-    "state": "closed",
-    "title": "corrects single file path validation logic",
-    "updated_at": "2026-04-08T00:28:00Z"
+    "files_url": "https://github.com/huggingface/diffusers/pull/13453/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13453",
+    "labels": [
+      "size/S",
+      "pipelines"
+    ],
+    "merged": false,
+    "number": 13453,
+    "review_comments_count": 2,
+    "state": "open",
+    "title": "Fix SDXL Refiner with Higher Order Schedulers",
+    "updated_at": "2026-04-13T05:39:29Z"
   },
   {
-    "additions": 3244,
-    "author": "yiyixuxu",
-    "author_association": "MEMBER",
-    "body_excerpt": null,
-    "changed_files": 23,
+    "additions": 19,
+    "author": "chang-zhijie",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "This PR resolves the unsupported atten_mask shape error when running attention with NPU (Ascend) devices. Problem: The NPU\u2019s fusion attention operator (e.g., npu_fusion_attention) does not support automatic broadcasting for attention masks\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13360",
-    "created_at": "2026-03-28T16:27:47Z",
-    "deletions": 46,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13451",
+    "created_at": "2026-04-13T03:17:22Z",
+    "deletions": 10,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13360/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13360",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/diffusers/pull/13451/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13451",
+    "labels": [
+      "models",
+      "size/S"
+    ],
     "merged": false,
-    "number": 13360,
+    "number": 13451,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "[WIP]LTX modular + 1:1 match + improve agent debugging skills ",
-    "updated_at": "2026-03-31T08:36:31Z"
+    "state": "closed",
+    "title": "Fix attention_mask broadcasting for NPU compatibility",
+    "updated_at": "2026-04-19T01:14:32Z"
   },
   {
-    "additions": 841,
-    "author": "sayakpaul",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? TL;DR: Adds a guide on how to profile a pipeline and fix issues like CPU overhead, CPU<->GPU syncs, etc. ## Motivation Since we provide first-class `torch.compile` support, it's important that our pipelines are set\u2026",
-    "changed_files": 8,
+    "additions": 2,
+    "author": "conanna",
+    "author_association": "FIRST_TIMER",
+    "body_excerpt": "# What does this PR do? `UniPCMultistepScheduler.set_timesteps` hardcodes `self.sigmas = self.sigmas.to(\"cpu\")` at the end of the method. This causes device mismatch errors in `multistep_uni_p_bh_update` and `multistep_uni_c_bh_update` whe\u2026",
+    "changed_files": 1,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13450",
+    "created_at": "2026-04-12T19:13:13Z",
+    "deletions": 1,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/diffusers/pull/13450/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13450",
+    "labels": [
+      "size/S",
+      "schedulers"
+    ],
+    "merged": false,
+    "number": 13450,
+    "review_comments_count": 0,
+    "state": "open",
+    "title": "fix: align sigmas device in UniPCMultistepScheduler.set_timesteps",
+    "updated_at": "2026-04-12T19:13:27Z"
+  },
+  {
+    "additions": 9,
+    "author": "Ricardo-M-L",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## Summary - Fix all 9 QwenImage pipeline variants where CFG is silently disabled when passing `negative_prompt_embeds` with `None` mask - `encode_prompt()` converts all-ones masks to `None` as an optimization (line 266-267), but `has_neg_\u2026",
+    "changed_files": 9,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13449",
+    "created_at": "2026-04-12T15:32:31Z",
+    "deletions": 9,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/diffusers/pull/13449/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13449",
+    "labels": [
+      "size/S",
+      "pipelines"
+    ],
+    "merged": false,
+    "number": 13449,
+    "review_comments_count": 0,
+    "state": "closed",
+    "title": "fix: QwenImage pipelines silently disable CFG with pre-computed negative_prompt_embeds",
+    "updated_at": "2026-04-18T17:22:56Z"
+  },
+  {
+    "additions": 178,
+    "author": "hlky",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? As per #13446 `trust_remote_code` fails under several circumstances: - `pretrained_model_name_or_path` as Hub repo A and `custom_pipeline` as Hub repo B, `trust_remote_code` is bypassed and remote code runs from rep\u2026",
+    "changed_files": 7,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 7,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13448",
+    "created_at": "2026-04-12T15:05:10Z",
+    "deletions": 27,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/diffusers/pull/13448/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13448",
+    "labels": [
+      "models",
+      "tests",
+      "size/L",
+      "modular-pipelines",
+      "utils",
+      "pipelines"
+    ],
+    "merged": false,
+    "number": 13448,
+    "review_comments_count": 20,
+    "state": "open",
+    "title": "Improve `trust_remote_code`",
+    "updated_at": "2026-04-21T09:16:37Z"
+  },
+  {
+    "additions": 2201,
+    "author": "Moran232",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "### Description **We are the JoyAI Team,** and this is the Diffusers implementation for the JoyAI-Image-Edit model. GitHub Repository: [https://github.com/jd-opensource/JoyAI-Image] Hugging Face Model: [https://huggingface.co/jdopensource/\u2026",
+    "changed_files": 10,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13444",
+    "created_at": "2026-04-10T15:34:41Z",
+    "deletions": 0,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/diffusers/pull/13444/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13444",
+    "labels": [
+      "models",
+      "size/L",
+      "pipelines"
+    ],
+    "merged": false,
+    "number": 13444,
+    "review_comments_count": 33,
+    "state": "open",
+    "title": "[feat] JoyAI-JoyImage-Edit support",
+    "updated_at": "2026-04-20T09:02:33Z"
+  },
+  {
+    "additions": 1,
+    "author": "sayakpaul",
+    "author_association": "MEMBER",
+    "body_excerpt": null,
+    "changed_files": 1,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13443",
+    "created_at": "2026-04-10T15:02:22Z",
+    "deletions": 1,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/diffusers/pull/13443/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13443",
+    "labels": [
+      "models",
+      "size/S"
+    ],
+    "merged": true,
+    "number": 13443,
+    "review_comments_count": 0,
+    "state": "closed",
+    "title": "[core] fix fa4 integration",
+    "updated_at": "2026-04-13T05:58:18Z"
+  },
+  {
+    "additions": 111,
+    "author": "sayakpaul",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Fix https://github.com/huggingface/diffusers/actions/runs/24241815924/job/70778130703?pr=13332",
+    "changed_files": 10,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13442",
+    "created_at": "2026-04-10T14:23:50Z",
+    "deletions": 56,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/diffusers/pull/13442/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13442",
+    "labels": [
+      "size/M",
+      "examples"
+    ],
+    "merged": true,
+    "number": 13442,
+    "review_comments_count": 0,
+    "state": "closed",
+    "title": "[tests] fix training tests",
+    "updated_at": "2026-04-15T01:26:34Z"
+  },
+  {
+    "additions": 4,
+    "author": "azolotenkov",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? Fixes the prior-preservation batching issue in the Qwen Image DreamBooth LoRA script. PR #13396 corrected the repeat count under `--with_prior_preservation`, but the script still repeats prompt embeddings with `repe\u2026",
+    "changed_files": 1,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13441",
+    "created_at": "2026-04-10T11:53:31Z",
+    "deletions": 3,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/diffusers/pull/13441/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13441",
+    "labels": [
+      "size/S",
+      "examples"
+    ],
+    "merged": true,
+    "number": 13441,
+    "review_comments_count": 0,
+    "state": "closed",
+    "title": "Fix Qwen Image DreamBooth prior-preservation batch ordering",
+    "updated_at": "2026-04-15T07:54:48Z"
+  },
+  {
+    "additions": 2,
+    "author": "akshan-main",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "\u2026nstead of latent resolution # What does this PR do? Fixes #13439 prepare_cond_latents_and_mask shadows the pixel height/width parameters with latent dims from latents.shape (line 614): ``` batch, channels, frames, height, width = latents.\u2026",
+    "changed_files": 1,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13440",
+    "created_at": "2026-04-10T10:04:45Z",
+    "deletions": 2,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/diffusers/pull/13440/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13440",
+    "labels": [
+      "size/S",
+      "pipelines"
+    ],
+    "merged": true,
+    "number": 13440,
+    "review_comments_count": 0,
+    "state": "closed",
+    "title": "Fix HunyuanVideo 1.5 I2V by preprocessing image at pixel resolution i\u2026",
+    "updated_at": "2026-04-10T19:54:36Z"
+  },
+  {
+    "additions": 96,
+    "author": "paulinebm",
+    "author_association": "MEMBER",
+    "body_excerpt": "cc @sayakpaul",
+    "changed_files": 1,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13438",
+    "created_at": "2026-04-09T14:22:48Z",
+    "deletions": 26,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/diffusers/pull/13438/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13438",
+    "labels": [
+      "size/M",
+      "CI"
+    ],
+    "merged": true,
+    "number": 13438,
+    "review_comments_count": 10,
+    "state": "closed",
+    "title": "add PR fork workable",
+    "updated_at": "2026-04-16T09:49:18Z"
+  },
+  {
+    "additions": 3110,
+    "author": "RuixiangMa",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "Support JoyAIImagePipeline * https://github.com/jd-opensource/JoyAI-Image * [ jdopensource/JoyAI-Image-Edit](https://huggingface.co/jdopensource/JoyAI-Image-Edit) | Prompt | image | |----------|-------------| |NO |<img width=\"512\" height=\"\u2026",
+    "changed_files": 22,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13437",
+    "created_at": "2026-04-09T11:42:22Z",
+    "deletions": 0,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/diffusers/pull/13437/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13437",
+    "labels": [
+      "documentation",
+      "models",
+      "tests",
+      "size/L",
+      "pipelines",
+      "schedulers"
+    ],
+    "merged": false,
+    "number": 13437,
+    "review_comments_count": 6,
+    "state": "closed",
+    "title": "[Feat] support JoyAIImagePipeline ",
+    "updated_at": "2026-04-11T07:11:30Z"
+  },
+  {
+    "additions": 1,
+    "author": "NIK-TIGER-BILL",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "test",
+    "changed_files": 1,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13436",
+    "created_at": "2026-04-08T23:17:59Z",
+    "deletions": 0,
+    "draft": true,
+    "files_url": "https://github.com/huggingface/diffusers/pull/13436/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13436",
+    "labels": [
+      "size/S"
+    ],
+    "merged": false,
+    "number": 13436,
+    "review_comments_count": 0,
+    "state": "closed",
+    "title": "test",
+    "updated_at": "2026-04-08T23:18:14Z"
+  },
+  {
+    "additions": 282,
+    "author": "Ricardo-M-L",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## Summary - Adds `scripts/convert_cogvideox_to_original.py`, a reverse conversion script that converts CogVideoX models (transformer and VAE) from diffusers format back to the original CogVideo checkpoint format - Reverses all weight name\u2026",
+    "changed_files": 1,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13435",
+    "created_at": "2026-04-08T17:38:45Z",
+    "deletions": 0,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/diffusers/pull/13435/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13435",
+    "labels": [
+      "size/L"
+    ],
+    "merged": false,
+    "number": 13435,
+    "review_comments_count": 0,
+    "state": "open",
+    "title": "Add CogVideoX diffusers-to-original format conversion script",
+    "updated_at": "2026-04-08T17:39:01Z"
+  },
+  {
+    "additions": 2,
+    "author": "Ricardo-M-L",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## Summary Fixes two functional bugs in the experimental guiders module, both introduced in #11311 (Modular Diffusers Guiders): ### Bug 1: `TangentialClassifierFreeGuidance.is_conditional` \u2014 AttributeError The `is_conditional` property ref\u2026",
+    "changed_files": 2,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13434",
+    "created_at": "2026-04-08T15:05:39Z",
+    "deletions": 2,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/diffusers/pull/13434/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13434",
+    "labels": [
+      "size/S",
+      "guiders"
+    ],
+    "merged": false,
+    "number": 13434,
+    "review_comments_count": 0,
+    "state": "open",
+    "title": "fix: correct AttributeError in TangentialCFG and NameError in FrequencyDecoupledGuidance",
+    "updated_at": "2026-04-18T17:26:02Z"
+  },
+  {
+    "additions": 1,
+    "author": "DN6",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Narrow scoped token to just access inference providers when labelling issues. <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release n\u2026",
+    "changed_files": 1,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13433",
+    "created_at": "2026-04-08T08:28:55Z",
+    "deletions": 1,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/diffusers/pull/13433/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13433",
+    "labels": [
+      "size/S",
+      "CI"
+    ],
+    "merged": true,
+    "number": 13433,
+    "review_comments_count": 0,
+    "state": "closed",
+    "title": "[CI] Use finegrained token for Issue Labeler",
+    "updated_at": "2026-04-08T09:18:27Z"
+  },
+  {
+    "additions": 1184,
+    "author": "HsiaWinter",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? We have introduced a new text-to-image model called ERNIE-Image, which will soon be open-sourced to the community. This PR includes the model architecture definition, the pipeline, as well as the related documentati\u2026",
+    "changed_files": 14,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 11,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13432",
+    "created_at": "2026-04-08T04:14:03Z",
+    "deletions": 0,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/diffusers/pull/13432/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13432",
+    "labels": [
+      "documentation",
+      "models",
+      "tests",
+      "size/L",
+      "utils",
+      "pipelines"
+    ],
+    "merged": true,
+    "number": 13432,
+    "review_comments_count": 72,
+    "state": "closed",
+    "title": "Add ernie image",
+    "updated_at": "2026-04-14T10:01:10Z"
+  },
+  {
+    "additions": 2,
+    "author": "huemin-art",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? Adds an optional `decoder_block_out_channels` parameter to `AutoencoderKLFlux2` so the decoder can use different channel widths than the encoder. This is needed to support the FLUX.2 small decoder VAE (`black-forest\u2026",
+    "changed_files": 1,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13428",
+    "created_at": "2026-04-06T23:48:03Z",
+    "deletions": 1,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/diffusers/pull/13428/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13428",
+    "labels": [],
+    "merged": true,
+    "number": 13428,
+    "review_comments_count": 0,
+    "state": "closed",
+    "title": "FLUX.2 small decoder",
+    "updated_at": "2026-04-07T01:59:40Z"
+  },
+  {
+    "additions": 18,
+    "author": "asomoza",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Fixes the use of `record_stream` with `torchao` and `group offload`. Without this, the resulting image is wrong. |Without this PR|With this PR| |---|---| |<img width=\"1024\" height=\"1024\" alt=\"flux_benchmark_dynamic_\u2026",
+    "changed_files": 1,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13427",
+    "created_at": "2026-04-06T17:37:24Z",
+    "deletions": 0,
+    "draft": true,
+    "files_url": "https://github.com/huggingface/diffusers/pull/13427/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13427",
+    "labels": [],
+    "merged": false,
+    "number": 13427,
+    "review_comments_count": 0,
+    "state": "closed",
+    "title": "[torchao] Fix `record_stream` with torchao and group offloading",
+    "updated_at": "2026-04-06T17:59:53Z"
+  },
+  {
+    "additions": 6,
+    "author": "Akash504-ai",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? This PR fixes a numerical stability issue in `rescale_noise_cfg` that can lead to division by zero and produce NaN/inf values during inference. ## Problem The current implementation performs: noise_pred_rescaled = n\u2026",
+    "changed_files": 3,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13426",
+    "created_at": "2026-04-06T15:44:38Z",
+    "deletions": 3,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/diffusers/pull/13426/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13426",
+    "labels": [
+      "size/S",
+      "pipelines",
+      "guiders"
+    ],
+    "merged": false,
+    "number": 13426,
+    "review_comments_count": 0,
+    "state": "open",
+    "title": "Fix division by zero in rescale_noise_cfg causing NaNs during inference",
+    "updated_at": "2026-04-14T18:26:24Z"
+  },
+  {
+    "additions": 10,
+    "author": "sayakpaul",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Fixes https://github.com/huggingface/diffusers/actions/runs/24034864516/job/70092362359?pr=13417. Got missed in the earlier refactor.",
+    "changed_files": 1,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13424",
+    "created_at": "2026-04-06T14:55:04Z",
+    "deletions": 0,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/diffusers/pull/13424/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13424",
+    "labels": [],
+    "merged": true,
+    "number": 13424,
+    "review_comments_count": 0,
+    "state": "closed",
+    "title": "[tests] fix autoencoderdc tests",
+    "updated_at": "2026-04-06T19:08:03Z"
+  },
+  {
+    "additions": 2,
+    "author": "Xyc2016",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "## Summary - Fix `LoRA's` \u2192 `LoRAs` in two places (lines 104 and 106) - plural, not possessive - Fix `trigger it` \u2192 `trigger them` (pronoun agreement: \"Some LoRAs\" is plural) ## Changes **Line 104:** > Among adapters, [LoRA's] are the most\u2026",
+    "changed_files": 1,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13423",
+    "created_at": "2026-04-06T14:42:14Z",
+    "deletions": 2,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/diffusers/pull/13423/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13423",
+    "labels": [
+      "documentation",
+      "size/S"
+    ],
+    "merged": true,
+    "number": 13423,
+    "review_comments_count": 1,
+    "state": "closed",
+    "title": "Fix grammar in LoRA documentation",
+    "updated_at": "2026-04-10T16:18:30Z"
+  },
+  {
+    "additions": 16,
+    "author": "akshan-main",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## What does this PR do? Fixes #13421 Adds `negative_prompt` parameter to `GLMImagePipeline.__call__()` and `encode_prompt()`. Previously, the unconditional prompt was hardcoded to `\"\"` when CFG was active, with no way for users to provide\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13356",
-    "created_at": "2026-03-28T04:35:05Z",
-    "deletions": 14,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13422",
+    "created_at": "2026-04-06T14:23:33Z",
+    "deletions": 4,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/diffusers/pull/13422/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13422",
+    "labels": [],
+    "merged": false,
+    "number": 13422,
+    "review_comments_count": 0,
+    "state": "closed",
+    "title": "Add negative_prompt parameter to GLMImagePipeline",
+    "updated_at": "2026-04-06T23:58:06Z"
+  },
+  {
+    "additions": 15,
+    "author": "akshan-main",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## What does this PR do? Fixes #13416 Adds `negative_prompt` parameter to `Flux2KleinPipeline.__call__()`. Previously, the unconditional prompt was hardcoded to `\"\"` when CFG was active, with no way for users to provide a custom negative p\u2026",
+    "changed_files": 1,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13420",
+    "created_at": "2026-04-06T14:12:33Z",
+    "deletions": 5,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/diffusers/pull/13420/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13420",
+    "labels": [],
+    "merged": false,
+    "number": 13420,
+    "review_comments_count": 0,
+    "state": "closed",
+    "title": "Add negative_prompt to Flux2KleinPipeline",
+    "updated_at": "2026-04-06T23:58:45Z"
+  },
+  {
+    "additions": 14,
+    "author": "azolotenkov",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? Fixes a device mismatch in the Flux2 DreamBooth training scripts when `--offload` is used and latents are not precomputed. non-cached path: `pixel_values` were converted to `vae.dtype` but not moved onto `accelerato\u2026",
+    "changed_files": 5,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13417",
+    "created_at": "2026-04-05T20:31:33Z",
+    "deletions": 16,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/diffusers/pull/13417/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13417",
+    "labels": [],
+    "merged": true,
+    "number": 13417,
+    "review_comments_count": 0,
+    "state": "closed",
+    "title": "Fix VAE offload encode device mismatch in DreamBooth scripts",
+    "updated_at": "2026-04-08T17:19:11Z"
+  },
+  {
+    "additions": 16,
+    "author": "azolotenkov",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? Fixes a prior-preservation batch size mismatch in the `Flux2 DreamBooth LoRA` scripts. When custom instance prompts are not used, `prompt_embeds` already contains concatenated instance + class embeddings under `--wi\u2026",
+    "changed_files": 2,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 4,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13415",
+    "created_at": "2026-04-04T18:08:44Z",
+    "deletions": 8,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/diffusers/pull/13415/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13415",
+    "labels": [
+      "size/S",
+      "examples"
+    ],
+    "merged": true,
+    "number": 13415,
+    "review_comments_count": 3,
+    "state": "closed",
+    "title": "Fix Flux2 DreamBooth prior preservation prompt repeats",
+    "updated_at": "2026-04-15T09:31:58Z"
+  },
+  {
+    "additions": 11,
+    "author": "Ricardo-M-L",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## Summary - Fix article: `a inpainting` \u2192 `an inpainting` in auto_pipeline.py - Fix article: `a aesthetic_score` \u2192 `an aesthetic_score` in SDXL pipelines - Fix docstring: remove extraneous `pass a` in quantizers/base.py - Fix grammar: `al\u2026",
+    "changed_files": 6,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13414",
+    "created_at": "2026-04-04T17:07:02Z",
+    "deletions": 11,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/diffusers/pull/13414/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13414",
+    "labels": [],
+    "merged": false,
+    "number": 13414,
+    "review_comments_count": 0,
+    "state": "closed",
+    "title": "fix: correct grammar in docstrings and error messages",
+    "updated_at": "2026-04-18T17:25:44Z"
+  },
+  {
+    "additions": 35,
+    "author": "Ricardo-M-L",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## Summary - Fix `it's` \u2192 `its` (possessive form) in ~35 instances across ~32 model files - The phrase \"Check the superclass documentation for it's generic methods\" should use \"its\" (possessive), not \"it's\" (contraction of \"it is\") - Also\u2026",
+    "changed_files": 32,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13413",
+    "created_at": "2026-04-04T17:05:39Z",
+    "deletions": 35,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/diffusers/pull/13413/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13413",
+    "labels": [],
+    "merged": false,
+    "number": 13413,
+    "review_comments_count": 0,
+    "state": "closed",
+    "title": "fix: correct possessive 'it's' to 'its' in model docstrings",
+    "updated_at": "2026-04-18T17:25:40Z"
+  },
+  {
+    "additions": 307,
+    "author": "VittoriaLanzo",
+    "author_association": "NONE",
+    "body_excerpt": "## What does this PR do? Fixes #13411. `LTXEulerAncestralRFScheduler.set_timesteps(sigmas=...)` silently accepted non-monotone sigma schedules. In `step()`, a reversed schedule computes: ``` downstep_ratio = 1 + (sigma_next / sigma - 1) *\u2026",
+    "changed_files": 4,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13412",
+    "created_at": "2026-04-04T15:49:39Z",
+    "deletions": 11,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/diffusers/pull/13412/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13412",
+    "labels": [],
+    "merged": false,
+    "number": 13412,
+    "review_comments_count": 0,
+    "state": "closed",
+    "title": "[Scheduler] Add monotonicity validation and tests for LTXEulerAncestr\u2026",
+    "updated_at": "2026-04-06T22:54:31Z"
+  },
+  {
+    "additions": 46,
+    "author": "yiyixuxu",
+    "author_association": "MEMBER",
+    "body_excerpt": "follow https://github.com/huggingface/diffusers/pull/13334 moved `model-integrawtion/skillls/modular-conversion.md` to `.ai/modular.md`, and link it from `AGENTS.md`, `model-integrawtion/skillls` and `review-rules.md ` added a Gotcha secti\u2026",
+    "changed_files": 4,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13410",
+    "created_at": "2026-04-04T08:37:28Z",
+    "deletions": 15,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/diffusers/pull/13410/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13410",
+    "labels": [
+      "size/M"
+    ],
+    "merged": true,
+    "number": 13410,
+    "review_comments_count": 1,
+    "state": "closed",
+    "title": "[agent] add modular doc",
+    "updated_at": "2026-04-14T17:43:30Z"
+  },
+  {
+    "additions": 32,
+    "author": "awsr",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? Fixes `NameError: name 'ftfy' is not defined` This was encountered by one of our users on SDNext. I'm not entirely sure why `ftfy` didn't get installed, but if it's optionally imported then it should also be optiona\u2026",
+    "changed_files": 16,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13409",
+    "created_at": "2026-04-04T02:02:23Z",
+    "deletions": 16,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/diffusers/pull/13409/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13409",
+    "labels": [],
+    "merged": false,
+    "number": 13409,
+    "review_comments_count": 0,
+    "state": "open",
+    "title": "Guard against calling optional `ftfy` module if not available",
+    "updated_at": "2026-04-07T01:53:00Z"
+  },
+  {
+    "additions": 7,
+    "author": "Ricardo-M-L",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## Summary - Fix duplicate words: `schedule schedule` \u2192 `schedule`, `obtain obtain` \u2192 `obtain` - Fix article usage: `a image` \u2192 `an image` (2 instances), `a image-text` \u2192 `an image-text` (2 instances), `a upsampling` \u2192 `an upsampling` ## T\u2026",
+    "changed_files": 6,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13408",
+    "created_at": "2026-04-03T23:40:41Z",
+    "deletions": 7,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/diffusers/pull/13408/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13408",
+    "labels": [],
+    "merged": false,
+    "number": 13408,
+    "review_comments_count": 0,
+    "state": "closed",
+    "title": "fix: remove duplicate words and fix grammar in docstrings and error messages",
+    "updated_at": "2026-04-18T17:25:36Z"
+  },
+  {
+    "additions": 54,
+    "author": "Ricardo-M-L",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## Summary Add missing `f` prefix to ~50 string literals where variables like `{self.__class__}`, `{self.config}`, `{unet.config.*}`, `{type(...)}` are printed as literal text instead of being interpolated. ### Categories fixed: - Safety c\u2026",
+    "changed_files": 49,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13407",
+    "created_at": "2026-04-03T23:40:27Z",
+    "deletions": 54,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13356/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13356",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13407/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13407",
+    "labels": [],
+    "merged": false,
+    "number": 13407,
+    "review_comments_count": 0,
+    "state": "open",
+    "title": "fix: add missing f-string prefixes across pipelines and schedulers",
+    "updated_at": "2026-04-03T23:40:27Z"
+  },
+  {
+    "additions": 25,
+    "author": "akshan-main",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "## What does this PR do? Part of #13401 `QwenEmbedRope.forward()` copies `pos_freqs` and `neg_freqs` from CPU to GPU via `.to(device)` on every transformer forward call. These tensors are fixed at init and never change, so the repeated tra\u2026",
+    "changed_files": 1,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 24,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13406",
+    "created_at": "2026-04-03T21:17:17Z",
+    "deletions": 10,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/diffusers/pull/13406/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13406",
     "labels": [
-      "performance"
+      "performance",
+      "models",
+      "size/S"
     ],
     "merged": true,
-    "number": 13356,
-    "review_comments_count": 50,
+    "number": 13406,
+    "review_comments_count": 2,
     "state": "closed",
-    "title": "Add examples on how to profile a pipeline",
-    "updated_at": "2026-04-03T14:13:33Z"
+    "title": "Cache RoPE freqs on device to avoid repeated CPU-GPU copy in QwenImage",
+    "updated_at": "2026-04-10T16:38:12Z"
   },
   {
-    "additions": 17,
-    "author": "howardzhang-cv",
+    "additions": 23,
+    "author": "andrewor14",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? ## Summary Update the minimum torchao version guarding to 0.15.0, as part of a larger motivation to update torchao quantizers to newer versions. Comment initially made in https://github.com/huggingface/diffusers/pul\u2026",
-    "changed_files": 3,
+    "body_excerpt": "**Summary:** TorchAO recently deprecated AffineQuantizedTensor and related classes (https://github.com/pytorch/ao/issues/2752). These will be removed in the next release. We should remove references of these classes in diffusers before the\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13355",
-    "created_at": "2026-03-27T22:12:14Z",
-    "deletions": 20,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13405",
+    "created_at": "2026-04-03T20:31:12Z",
+    "deletions": 38,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13355/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13355",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13405/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13405",
     "labels": [],
     "merged": true,
-    "number": 13355,
-    "review_comments_count": 0,
+    "number": 13405,
+    "review_comments_count": 5,
     "state": "closed",
-    "title": "change minimum version guard for torchao to 0.15.0",
-    "updated_at": "2026-03-28T03:41:51Z"
+    "title": "Remove references to torchao's AffineQuantizedTensor",
+    "updated_at": "2026-04-08T15:43:10Z"
   },
   {
-    "additions": 3,
-    "author": "hf-security-analysis[bot]",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "Update `.github/workflows/claude_review.yml` workflow configuration. cc @sayakpaul Closes huggingface/tracking-issues#34",
-    "changed_files": 1,
+    "additions": 152,
+    "author": "varaprasadtarunkumar",
+    "author_association": "NONE",
+    "body_excerpt": "## What does this PR do? Fixes a per-step `torch.tensor()` reconstruction in `QwenImageTransformer2DModel.forward()` that was identified as a known performance issue in the [profiling guide](https://github.com/huggingface/diffusers/tree/ma\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13354",
-    "created_at": "2026-03-27T11:59:04Z",
-    "deletions": 3,
+    "comments_count": 6,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13404",
+    "created_at": "2026-04-03T18:58:56Z",
+    "deletions": 4,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13354/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13354",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13404/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13404",
     "labels": [],
     "merged": false,
-    "number": 13354,
-    "review_comments_count": 1,
+    "number": 13404,
+    "review_comments_count": 2,
     "state": "closed",
-    "title": "chore: update claude_review.yml",
-    "updated_at": "2026-04-02T08:43:45Z"
+    "title": "fix: cache modulate_index tensor to eliminate per-step DtoH sync",
+    "updated_at": "2026-04-03T19:26:05Z"
   },
   {
-    "additions": 115,
-    "author": "azolotenkov",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "## What does this PR do? This PR fixes a bug where `BucketBatchSampler` reshuffled precomputed batches on each `__iter__()` call. DreamBooth training scripts precompute latents and/or prompt embeddings and later consume them by dataloader\u2026",
-    "changed_files": 5,
+    "additions": 4,
+    "author": "Ricardo-M-L",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## Summary - Add missing `f` prefix to 4 log messages in `bnb_quantizer.py` where `{target_dtype}` and `{current_device}` are printed as literal text instead of being interpolated - Affects both 4-bit and 8-bit BnB quantizer classes ## Tes\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 4,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13353",
-    "created_at": "2026-03-27T11:21:35Z",
-    "deletions": 30,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13403",
+    "created_at": "2026-04-03T18:27:20Z",
+    "deletions": 6,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13353/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13353",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13403/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13403",
     "labels": [],
     "merged": false,
-    "number": 13353,
-    "review_comments_count": 6,
+    "number": 13403,
+    "review_comments_count": 0,
     "state": "open",
-    "title": "Fix BucketBatchSampler cache alignment in DreamBooth scripts",
-    "updated_at": "2026-03-30T11:56:46Z"
+    "title": "fix: add missing f-string prefixes in BnB quantizer log messages",
+    "updated_at": "2026-04-03T18:27:20Z"
   },
   {
-    "additions": 3,
+    "additions": 8,
+    "author": "Ricardo-M-L",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## Summary - Fix 6 instances of `have have` \u2192 `have` in modular pipeline error messages - Fix `to to` \u2192 `to` in UNet docstring - Fix `be be` \u2192 `be` in Flux Kontext inpaint pipeline docstring ## Test plan - [ ] Verify error messages read co\u2026",
+    "changed_files": 8,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13402",
+    "created_at": "2026-04-03T18:23:45Z",
+    "deletions": 8,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/diffusers/pull/13402/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13402",
+    "labels": [],
+    "merged": false,
+    "number": 13402,
+    "review_comments_count": 0,
+    "state": "closed",
+    "title": "fix: remove duplicate words in error messages and docstrings",
+    "updated_at": "2026-04-18T17:25:31Z"
+  },
+  {
+    "additions": 5,
     "author": "sayakpaul",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Fixes https://github.com/huggingface/diffusers/actions/runs/23641475244/job/68862961669",
+    "body_excerpt": null,
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13352",
-    "created_at": "2026-03-27T10:29:09Z",
-    "deletions": 0,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13400",
+    "created_at": "2026-04-03T16:51:50Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13352/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13352",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13400/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13400",
     "labels": [],
     "merged": true,
-    "number": 13352,
+    "number": 13400,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "[ci] include checkout step in claude review workflow",
-    "updated_at": "2026-03-27T11:58:34Z"
+    "title": "Update README.md of the profiling guide",
+    "updated_at": "2026-04-03T17:00:13Z"
   },
   {
-    "additions": 262,
-    "author": "DN6",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
-    "changed_files": 5,
+    "additions": 4,
+    "author": "Ricardo-M-L",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "Two types of exception handling bugs: **1. `raise \"string\"` in attention_processor.py (lines 328-332)** \\`raise \"string\"\\` causes \\`TypeError: exceptions must derive from BaseException\\` at runtime instead of the intended error message. Ch\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13348",
-    "created_at": "2026-03-26T11:31:31Z",
-    "deletions": 204,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13398",
+    "created_at": "2026-04-03T05:41:14Z",
+    "deletions": 4,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13348/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13348",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13398/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13398",
     "labels": [],
     "merged": false,
-    "number": 13348,
-    "review_comments_count": 5,
+    "number": 13398,
+    "review_comments_count": 0,
     "state": "open",
-    "title": "[CI] Refactor Skyreels, Lumina, Ominigen, Mochi transformer tests",
-    "updated_at": "2026-03-30T09:49:38Z"
+    "title": "fix: use proper exception objects instead of bare raise/assert strings",
+    "updated_at": "2026-04-03T05:41:14Z"
   },
   {
-    "additions": 305,
-    "author": "DN6",
+    "additions": 3,
+    "author": "yiyixuxu",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
-    "changed_files": 4,
+    "body_excerpt": null,
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13347",
-    "created_at": "2026-03-26T11:22:21Z",
-    "deletions": 67,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13397",
+    "created_at": "2026-04-03T05:40:53Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13347/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13347",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13397/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13397",
     "labels": [],
-    "merged": false,
-    "number": 13347,
-    "review_comments_count": 4,
-    "state": "open",
-    "title": "[CI] Refactor Chronoedit, PRX, EasyAnimate, Ovis transformer tests",
-    "updated_at": "2026-03-30T03:45:29Z"
+    "merged": true,
+    "number": 13397,
+    "review_comments_count": 7,
+    "state": "closed",
+    "title": "improve Claude CI",
+    "updated_at": "2026-04-06T20:43:14Z"
+  },
+  {
+    "additions": 4,
+    "author": "chenyangzhu1",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? I found that the same problem in https://github.com/huggingface/diffusers/issues/13292 also appears in Qwen-Image's dreambooth lora script. https://github.com/huggingface/diffusers/blob/8070f6ec54a7699d5ee285090d973\u2026",
+    "changed_files": 1,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 8,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13396",
+    "created_at": "2026-04-03T04:50:53Z",
+    "deletions": 1,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/diffusers/pull/13396/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13396",
+    "labels": [
+      "size/S",
+      "examples"
+    ],
+    "merged": true,
+    "number": 13396,
+    "review_comments_count": 0,
+    "state": "closed",
+    "title": "fix(qwen-image dreambooth): correct prompt embed repeats when using `--with_prior_preservation`",
+    "updated_at": "2026-04-10T04:47:06Z"
   },
   {
-    "additions": 201,
-    "author": "DN6",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
-    "changed_files": 3,
+    "additions": 19,
+    "author": "Akash504-ai",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? This PR fixes a missing validation in `DDPMScheduler.set_timesteps`. Currently, passing `num_inference_steps=0` does not raise an error and can lead to invalid internal state or unexpected behavior. This behavior is\u2026",
+    "changed_files": 4,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13346",
-    "created_at": "2026-03-26T11:11:59Z",
-    "deletions": 141,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13395",
+    "created_at": "2026-04-03T04:38:48Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13346/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13346",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/diffusers/pull/13395/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13395",
+    "labels": [
+      "tests",
+      "size/S",
+      "schedulers"
+    ],
     "merged": false,
-    "number": 13346,
-    "review_comments_count": 3,
+    "number": 13395,
+    "review_comments_count": 0,
     "state": "open",
-    "title": "[CI] Refactor Cog Transformer Model Tests",
-    "updated_at": "2026-03-30T03:42:26Z"
+    "title": "Fix missing validation for num_inference_steps in DDPMScheduler and add tests",
+    "updated_at": "2026-04-15T05:30:31Z"
   },
   {
-    "additions": 266,
-    "author": "DN6",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
+    "additions": 4,
+    "author": "GalacticAvenger",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "## Summary Fixes several typos and grammar errors across the documentation: 1. **`controlling_generation.md`**: \"allows to generate\" \u2192 \"allows you to generate\" (missing object after \"allows\") 2. **`sdxl.md`**: \"it's refiner model\" \u2192 \"its r\u2026",
     "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13345",
-    "created_at": "2026-03-26T11:05:29Z",
-    "deletions": 102,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13391",
+    "created_at": "2026-04-02T20:15:48Z",
+    "deletions": 4,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13345/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13345",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13391/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13391",
     "labels": [],
-    "merged": false,
-    "number": 13345,
+    "merged": true,
+    "number": 13391,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "[CI] Refactor Chroma , LongCat and HiDream Transformer Tests",
-    "updated_at": "2026-03-26T11:05:29Z"
+    "state": "closed",
+    "title": "Fix typos and grammar errors in documentation",
+    "updated_at": "2026-04-02T20:42:32Z"
   },
   {
-    "additions": 96,
-    "author": "DN6",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
-    "changed_files": 2,
+    "additions": 2070,
+    "author": "RuixiangMa",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? Adds [LongCat-AudioDiT](https://github.com/meituan-longcat/LongCat-AudioDiT) model support to diffusers. Although LongCat-AudioDiT can be used for TTS-like generation, it is fundamentally a diffusion-based audio gen\u2026",
+    "changed_files": 18,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13344",
-    "created_at": "2026-03-26T11:03:48Z",
-    "deletions": 1,
+    "comments_count": 14,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13390",
+    "created_at": "2026-04-02T17:19:18Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13344/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13344",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13390/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13390",
     "labels": [
+      "documentation",
       "models",
       "tests",
-      "size/M"
+      "size/L",
+      "utils",
+      "pipelines"
     ],
     "merged": true,
-    "number": 13344,
-    "review_comments_count": 1,
+    "number": 13390,
+    "review_comments_count": 126,
     "state": "closed",
-    "title": "[CI] Add GLM Image Transformer Model Tests",
-    "updated_at": "2026-04-07T10:58:06Z"
+    "title": "[Feat] Adds LongCat-AudioDiT pipeline ",
+    "updated_at": "2026-04-16T15:25:36Z"
   },
   {
-    "additions": 379,
-    "author": "DN6",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
-    "changed_files": 5,
+    "additions": 2036,
+    "author": "akshan-main",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "## What does this PR do? Adds modular pipeline blocks for HunyuanVideo 1.5 with both text-to-video (`HunyuanVideo15Blocks`) and image-to-video (`HunyuanVideo15Image2VideoBlocks`). **Parity verified on Colab G4 GPU:** - T2V: MAD 0.000000 vs\u2026",
+    "changed_files": 13,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13342",
-    "created_at": "2026-03-26T10:43:26Z",
-    "deletions": 346,
+    "comments_count": 15,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13389",
+    "created_at": "2026-04-02T16:09:27Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13342/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13342",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/diffusers/pull/13389/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13389",
+    "labels": [
+      "close-to-merge",
+      "tests",
+      "size/L",
+      "modular-pipelines",
+      "utils"
+    ],
     "merged": true,
-    "number": 13342,
-    "review_comments_count": 3,
+    "number": 13389,
+    "review_comments_count": 5,
     "state": "closed",
-    "title": "[CI] Hunyuan Transformer Tests Refactor",
-    "updated_at": "2026-04-06T14:46:20Z"
+    "title": "Add modular pipeline for HunyuanVideo 1.5",
+    "updated_at": "2026-04-16T20:47:52Z"
   },
   {
     "additions": 160,
-    "author": "DN6",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
-    "changed_files": 2,
-    "cluster_id": null,
-    "cluster_ids": [],
-    "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13341",
-    "created_at": "2026-03-26T10:07:29Z",
-    "deletions": 96,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13341/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13341",
-    "labels": [],
-    "merged": false,
-    "number": 13341,
-    "review_comments_count": 2,
-    "state": "open",
-    "title": "[CI] Refactor Bria Transformer Tests",
-    "updated_at": "2026-03-27T12:02:51Z"
-  },
-  {
-    "additions": 128,
-    "author": "DN6",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
-    "changed_files": 1,
+    "author": "Ricardo-M-L",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "160 occurrences across 158 pipeline files use the incorrect phrase \"should be the same type to\" instead of \"should be the same type as\" in the negative_prompt type check error message. Example (before): ```python f\"`negative_prompt` should\u2026",
+    "changed_files": 158,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13340",
-    "created_at": "2026-03-26T09:56:24Z",
-    "deletions": 102,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13388",
+    "created_at": "2026-04-02T16:08:54Z",
+    "deletions": 160,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13340/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13340",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13388/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13388",
     "labels": [],
     "merged": false,
-    "number": 13340,
-    "review_comments_count": 1,
-    "state": "open",
-    "title": "[CI] Refactor SD3 Transformer Test",
-    "updated_at": "2026-03-27T11:19:12Z"
-  },
-  {
-    "additions": 15,
-    "author": "YangKai0616",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# Issue: The `test_save_load_float16` test has a hidden non-determinism problem: 1. When the pipeline is constructed directly, its components default to training mode (e.g., `text_encoder` contains `dropout` layers). However, after loading\u2026",
-    "changed_files": 2,
-    "cluster_id": null,
-    "cluster_ids": [],
-    "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13339",
-    "created_at": "2026-03-26T09:19:39Z",
-    "deletions": 0,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13339/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13339",
-    "labels": [],
-    "merged": true,
-    "number": 13339,
-    "review_comments_count": 12,
+    "number": 13388,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix: ensure consistent dtype and eval mode in pipeline save/load tests",
-    "updated_at": "2026-03-31T08:53:23Z"
+    "title": "fix: correct grammar 'same type to' \u2192 'same type as' in error messages",
+    "updated_at": "2026-04-18T17:25:27Z"
   },
   {
-    "additions": 1,
-    "author": "sayakpaul",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Fix https://github.com/huggingface/diffusers/actions/runs/23581004473",
+    "additions": 66,
+    "author": "chenyangzhu1",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? When running the `train_dreambooth_lora_qwen_image.py `script with the `--with_prior_preservation` flag, the training crashes during the text embedding extraction phase. Fixing two distinct bugs related to tensor co\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13338",
-    "created_at": "2026-03-26T07:22:48Z",
-    "deletions": 0,
+    "comments_count": 6,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13387",
+    "created_at": "2026-04-02T15:33:42Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13338/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13338",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/diffusers/pull/13387/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13387",
+    "labels": [
+      "size/M",
+      "examples"
+    ],
     "merged": true,
-    "number": 13338,
-    "review_comments_count": 0,
+    "number": 13387,
+    "review_comments_count": 2,
     "state": "closed",
-    "title": "fix claude workflow to include id-token with write.",
-    "updated_at": "2026-03-26T10:09:15Z"
+    "title": "Handle prompt embedding concat in Qwen dreambooth example",
+    "updated_at": "2026-04-10T03:43:33Z"
   },
   {
-    "additions": 300,
-    "author": "dg845",
+    "additions": 46,
+    "author": "paulinebm",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? This PR updates the LTX-2 docs to cover multimodal guidance and prompt enhancement, which were added with LTX-2.3 model support in #13217. Additionally, the LTX-2.X official default negative prompt, T2V system promp\u2026",
-    "changed_files": 2,
+    "body_excerpt": "## \ud83d\udd12 Pin GitHub Actions to commit SHAs This PR pins all GitHub Actions to their exact commit SHA instead of mutable tags or branch names. **Why?** Pinning to a SHA prevents supply chain attacks where a tag (e.g. `v4`) could be moved to poi\u2026",
+    "changed_files": 11,
     "cluster_id": null,
     "cluster_ids": [],
-    "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13337",
-    "created_at": "2026-03-26T06:15:28Z",
-    "deletions": 15,
+    "cluster_role": null,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13385",
+    "created_at": "2026-04-02T09:17:16Z",
+    "deletions": 46,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13337/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13337",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13385/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13385",
     "labels": [],
     "merged": true,
-    "number": 13337,
-    "review_comments_count": 10,
+    "number": 13385,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "Update LTX-2 Docs to Cover LTX-2.3 Models",
-    "updated_at": "2026-03-27T00:51:33Z"
+    "title": "\ud83d\udd12 Pin GitHub Actions to commit SHAs",
+    "updated_at": "2026-04-02T15:34:46Z"
   },
   {
-    "additions": 2,
-    "author": "kaixuanliu",
+    "additions": 14,
+    "author": "jiqing-feng",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "@sayakpaul pls help review, thx!",
+    "body_excerpt": "## What does this PR do? Fixes Dynamo `lru_cache` warnings when using `torch.compile` on diffusion pipelines. Two changes: 1. **`attention_dispatch.py`**: `dispatch_attention_fn` calls `is_torch_version(\">=\", \"2.5.0\")` at runtime, which is\u2026",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13336",
-    "created_at": "2026-03-26T06:00:04Z",
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13384",
+    "created_at": "2026-04-02T06:35:02Z",
     "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13336/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13336",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13384/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13384",
     "labels": [],
     "merged": true,
-    "number": 13336,
-    "review_comments_count": 0,
+    "number": 13384,
+    "review_comments_count": 3,
     "state": "closed",
-    "title": "avoid hardcode device in flux-control example",
-    "updated_at": "2026-03-26T07:10:54Z"
+    "title": "Fix Dynamo `lru_cache` warnings during `torch.compile`",
+    "updated_at": "2026-04-03T05:12:55Z"
   },
   {
-    "additions": 101,
-    "author": "DN6",
+    "additions": 160,
+    "author": "yiyixuxu",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
-    "changed_files": 1,
+    "body_excerpt": null,
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13335",
-    "created_at": "2026-03-26T04:07:43Z",
-    "deletions": 72,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13382",
+    "created_at": "2026-04-01T20:08:19Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13335/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13335",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13382/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13382",
     "labels": [],
     "merged": true,
-    "number": 13335,
-    "review_comments_count": 0,
+    "number": 13382,
+    "review_comments_count": 11,
     "state": "closed",
-    "title": "[CI] Refactor Cosmos Transformer Tests",
-    "updated_at": "2026-04-06T04:35:37Z"
+    "title": "[docs] add auto docstring and parameter templates documentation for m\u2026",
+    "updated_at": "2026-04-02T20:34:46Z"
   },
   {
-    "additions": 101,
-    "author": "stevhliu",
+    "additions": 984,
+    "author": "asomoza",
     "author_association": "MEMBER",
-    "body_excerpt": "follow up to #13297: - created `models.md`, a shared reference for holding model coding style, dependencies, common conventions, attention patterns, gotchas, etc. - updated `AGENTS.md`, `model-integration/SKILL.md`, and `review-rules.md` t\u2026",
-    "changed_files": 4,
+    "body_excerpt": "# What does this PR do? Add a skill for all the optimizations that can be done with diffusers. At the moment it has a lot of information and still with this the LLM (Opus 4.6) is still a hit or miss with the optimizations, specially with w\u2026",
+    "changed_files": 7,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13334",
-    "created_at": "2026-03-25T16:28:29Z",
-    "deletions": 85,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13381",
+    "created_at": "2026-04-01T14:42:10Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13334/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13334",
-    "labels": [],
-    "merged": true,
-    "number": 13334,
-    "review_comments_count": 12,
-    "state": "closed",
-    "title": "[docs] refactor model skill",
-    "updated_at": "2026-03-30T06:13:57Z"
+    "files_url": "https://github.com/huggingface/diffusers/pull/13381/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13381",
+    "labels": [
+      "size/L"
+    ],
+    "merged": false,
+    "number": 13381,
+    "review_comments_count": 30,
+    "state": "open",
+    "title": "[Agents] Optimizations skill",
+    "updated_at": "2026-04-14T15:15:59Z"
   },
   {
-    "additions": 29,
-    "author": "kashif",
+    "additions": 405,
+    "author": "DN6",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? move the documentation into its own Text section and clarify some docstring. <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release no\u2026",
-    "changed_files": 4,
+    "body_excerpt": "# What does this PR do? Addresses suggestions brought up in #13343 by adding scripts/workflows to auto labels issues and PRs . For Issues: Use `Qwen/Qwen3.5-35B-A3B` via Inference Providers to parse issues titles and content and assign app\u2026",
+    "changed_files": 5,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13333",
-    "created_at": "2026-03-25T11:53:03Z",
-    "deletions": 19,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13380",
+    "created_at": "2026-04-01T13:23:14Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13333/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13333",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13380/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13380",
     "labels": [],
     "merged": true,
-    "number": 13333,
-    "review_comments_count": 0,
+    "number": 13380,
+    "review_comments_count": 11,
     "state": "closed",
-    "title": "[LLADA2] documentation fixes",
-    "updated_at": "2026-03-25T19:02:06Z"
+    "title": "[CI] Add PR/Issue Auto Labeler",
+    "updated_at": "2026-04-07T04:32:19Z"
   },
   {
-    "additions": 46,
-    "author": "sayakpaul",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? We should be able to catch stuff like https://github.com/huggingface/diffusers/pull/13321. Additionally, I ported the existing dependency test suite to use `pytest`.",
-    "changed_files": 4,
+    "additions": 352,
+    "author": "Sunhill666",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
+    "changed_files": 16,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13332",
-    "created_at": "2026-03-25T08:42:00Z",
-    "deletions": 10,
+    "comments_count": 11,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13379",
+    "created_at": "2026-04-01T13:12:53Z",
+    "deletions": 93,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13332/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13332",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13379/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13379",
     "labels": [
       "tests",
-      "size/M",
-      "pipelines",
-      "CI"
+      "size/L",
+      "modular-pipelines",
+      "pipelines"
     ],
     "merged": true,
-    "number": 13332,
-    "review_comments_count": 1,
+    "number": 13379,
+    "review_comments_count": 14,
     "state": "closed",
-    "title": "[tests] tighten dependency testing.",
-    "updated_at": "2026-04-10T14:31:05Z"
+    "title": "fix(qwen): fix CFG failing when passing neg prompt embeds with none mask",
+    "updated_at": "2026-04-17T08:03:59Z"
   },
   {
-    "additions": 1417,
-    "author": "satani99",
+    "additions": 1996,
+    "author": "akshan-main",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "Added CogvideoxFunInpaintPipeline <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fu\u2026",
-    "changed_files": 7,
+    "body_excerpt": "# What does this PR do? Adds modular pipeline support for LTX Video, covering both text-to-video and image-to-video. The implementation follows the same structure as the existing Wan modular pipeline. ### Text-to-video ``` LTXBlocks (Seque\u2026",
+    "changed_files": 13,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13331",
-    "created_at": "2026-03-25T05:05:58Z",
+    "comments_count": 22,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13378",
+    "created_at": "2026-04-01T10:57:41Z",
     "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13331/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13331",
-    "labels": [],
-    "merged": false,
-    "number": 13331,
-    "review_comments_count": 11,
-    "state": "open",
-    "title": "Cogvideox_fun_inp pipeline",
-    "updated_at": "2026-03-26T02:09:17Z"
+    "files_url": "https://github.com/huggingface/diffusers/pull/13378/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13378",
+    "labels": [
+      "tests",
+      "size/L",
+      "modular-pipelines",
+      "utils"
+    ],
+    "merged": true,
+    "number": 13378,
+    "review_comments_count": 42,
+    "state": "closed",
+    "title": "[modular] Add LTX Video modular pipeline",
+    "updated_at": "2026-04-11T03:05:39Z"
   },
   {
-    "additions": 7,
-    "author": "sayakpaul",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Surfaced in https://github.com/huggingface/diffusers/pull/13291#issuecomment-4122867870. Cc: @howardzhang-cv",
-    "changed_files": 2,
+    "additions": 11,
+    "author": "hf-security-analysis[bot]",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "Update `.github/workflows/claude_review.yml` workflow configuration. cc @sayakpaul Closes huggingface/tracking-issues#42",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13330",
-    "created_at": "2026-03-25T04:09:32Z",
-    "deletions": 1,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13374",
+    "created_at": "2026-03-31T09:26:39Z",
+    "deletions": 4,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13330/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13330",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13374/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13374",
     "labels": [],
-    "merged": false,
-    "number": 13330,
-    "review_comments_count": 2,
-    "state": "open",
-    "title": "[tests] fix torchao tests",
-    "updated_at": "2026-03-25T05:19:46Z"
+    "merged": true,
+    "number": 13374,
+    "review_comments_count": 0,
+    "state": "closed",
+    "title": "chore: update claude_review.yml",
+    "updated_at": "2026-04-15T02:03:12Z"
   },
   {
-    "additions": 34,
-    "author": "sayakpaul",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Simplify release workflow like https://github.com/huggingface/transformers/blob/main/.github/workflows/release.yml. - Removed Test PyPI \u2014 no longer publish to Test PyPI and install from there; instead, install direc\u2026",
-    "changed_files": 1,
+    "additions": 63,
+    "author": "christopher5106",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "This PR adds local flash attention 4. Package required: [flash-attn-4](https://pypi.org/project/flash-attn-4/) (flash_attn.cute) Note: flash_4_hub already in the code but points to kernels-staging/flash-attn4 not publicly available. When i\u2026",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13329",
-    "created_at": "2026-03-25T04:04:57Z",
-    "deletions": 44,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13373",
+    "created_at": "2026-03-31T08:51:42Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13329/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13329",
-    "labels": [
-      "size/M",
-      "CI"
-    ],
+    "files_url": "https://github.com/huggingface/diffusers/pull/13373/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13373",
+    "labels": [],
     "merged": false,
-    "number": 13329,
-    "review_comments_count": 2,
+    "number": 13373,
+    "review_comments_count": 0,
     "state": "open",
-    "title": "[ci] simplify release workflow.",
-    "updated_at": "2026-04-12T06:50:48Z"
+    "title": "FA4 local (_flash_4 mode)",
+    "updated_at": "2026-03-31T09:04:04Z"
   },
   {
-    "additions": 119,
-    "author": "zhtmike",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Solve the issue #13319. There are two bugs: - `grad_out` is already in `BSHD` format, which matches the shape of `out`. Therefore, we do not need to permute `grad_out` again. - `autograd.Function.backward()` does no\u2026",
+    "additions": 173,
+    "author": "sayakpaul",
+    "author_association": "MEMBER",
+    "body_excerpt": null,
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13328",
-    "created_at": "2026-03-25T03:35:26Z",
-    "deletions": 16,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13371",
+    "created_at": "2026-03-31T05:30:09Z",
+    "deletions": 42,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13328/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13328",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13371/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13371",
     "labels": [],
     "merged": true,
-    "number": 13328,
-    "review_comments_count": 0,
+    "number": 13371,
+    "review_comments_count": 3,
     "state": "closed",
-    "title": "Fix Ulysses SP backward with SDPA",
-    "updated_at": "2026-03-30T09:45:28Z"
+    "title": "[tests] refactor wan autoencoder tests",
+    "updated_at": "2026-04-03T05:36:46Z"
   },
   {
-    "additions": 106,
-    "author": "LiuYinfeng01",
+    "additions": 133,
+    "author": "HMAKT99",
     "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "Add diffusers.utils.aiter_swizzle_gemm: can_shuffle (layout k multiple of 2*IK), apply_swizzle for eligible large nn.Linear layers using shuffle_weight + hipb_mm with bpreshuffle. Requires AMD MI-series (ROCm) with the aiter package instal\u2026",
-    "changed_files": 2,
+    "body_excerpt": "## What this does Adds a community example for embedding AI provenance metadata into diffusers-generated images. When you generate an image with diffusers, the output carries no metadata indicating it's AI-generated. This utility embeds pr\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13327",
-    "created_at": "2026-03-25T03:29:24Z",
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13370",
+    "created_at": "2026-03-30T15:12:13Z",
     "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13327/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13327",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13370/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13370",
     "labels": [],
     "merged": false,
-    "number": 13327,
+    "number": 13370,
     "review_comments_count": 0,
     "state": "open",
-    "title": "feat(utils): optional aiter swizzle GEMM for ROCm MI + aiter",
-    "updated_at": "2026-03-30T09:50:48Z"
+    "title": "Add AI image provenance community example",
+    "updated_at": "2026-04-14T00:30:25Z"
   },
   {
-    "additions": 999,
-    "author": "CalamitousFelicitousness",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "Adds support for Flux2 LoKR, with dual path to benchmark implementations. - Custom lossless path: BFL LoKR keys \u2192 peft LoKrConfig (fuse-first QKV) - Generic lossy path: optional SVD conversion via peft.convert_to_lora - Fix alpha handling\u2026",
-    "changed_files": 5,
+    "additions": 33,
+    "author": "sayakpaul",
+    "author_association": "MEMBER",
+    "body_excerpt": null,
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 29,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13326",
-    "created_at": "2026-03-25T01:26:53Z",
-    "deletions": 64,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13369",
+    "created_at": "2026-03-30T10:00:40Z",
+    "deletions": 27,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13326/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13326",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13369/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13369",
     "labels": [],
-    "merged": false,
-    "number": 13326,
-    "review_comments_count": 17,
-    "state": "open",
-    "title": "Add Flux2 LoKR adapter support prototype with dual conversion paths",
-    "updated_at": "2026-04-08T09:26:07Z"
+    "merged": true,
+    "number": 13369,
+    "review_comments_count": 0,
+    "state": "closed",
+    "title": "[tests] refactor autoencoderdc tests",
+    "updated_at": "2026-04-06T09:10:25Z"
   },
   {
-    "additions": 2048,
-    "author": "linoytsaban",
+    "additions": 37,
+    "author": "sayakpaul",
     "author_association": "MEMBER",
-    "body_excerpt": "adds a flux.2 klein kv training script, with additional general changes that can be propagated to other lora scripts as well: - [x] **Default aspect ratio buckets** - uses preset aspect ratio buckets to avoid needing to pass them manually.\u2026",
-    "changed_files": 2,
+    "body_excerpt": null,
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13325",
-    "created_at": "2026-03-24T18:41:22Z",
-    "deletions": 1,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13368",
+    "created_at": "2026-03-30T08:03:51Z",
+    "deletions": 43,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13325/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13325",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13368/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13368",
     "labels": [
-      "size/L",
-      "pipelines",
-      "examples"
+      "tests",
+      "size/M"
     ],
     "merged": false,
-    "number": 13325,
-    "review_comments_count": 10,
+    "number": 13368,
+    "review_comments_count": 2,
     "state": "open",
-    "title": "[flux.2 LoRA] make lora training compatible with flux.2 klein kv",
-    "updated_at": "2026-04-09T10:50:37Z"
+    "title": "[tests] refactor autoencoderkl tests",
+    "updated_at": "2026-04-14T10:12:17Z"
   },
   {
-    "additions": 2549,
-    "author": "AlanPonnachan",
+    "additions": 12,
+    "author": "NIK-TIGER-BILL",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? This PR introduces the modular architecture for Stable Diffusion 3 (SD3), implementing both Text-to-Image (T2I) and Image-to-Image (I2I) pipelines. **Key additions:** - Added `SD3ModularPipeline` and `SD3AutoBlocks`\u2026",
-    "changed_files": 14,
+    "body_excerpt": "## What does this PR do? Adds input validation for the `eta` parameter in `DDIMPipeline.__call__`. Fixes #13362 ### Context The DDIM paper defines \u03b7 (eta) as a value that lies strictly in [0, 1]: - \u03b7 = 0 \u2192 deterministic DDIM - \u03b7 = 1 \u2192 DDPM\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 12,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13324",
-    "created_at": "2026-03-24T17:33:51Z",
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13367",
+    "created_at": "2026-03-30T07:15:17Z",
     "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13324/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13324",
-    "labels": [
-      "tests",
-      "size/L",
-      "modular-pipelines",
-      "utils"
-    ],
-    "merged": false,
-    "number": 13324,
-    "review_comments_count": 23,
-    "state": "open",
-    "title": "feat: Add Modular Pipeline for Stable Diffusion 3 (SD3)",
-    "updated_at": "2026-04-09T14:06:57Z"
+    "files_url": "https://github.com/huggingface/diffusers/pull/13367/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13367",
+    "labels": [],
+    "merged": true,
+    "number": 13367,
+    "review_comments_count": 3,
+    "state": "closed",
+    "title": "fix(ddim): validate eta is in [0, 1] in DDIMPipeline",
+    "updated_at": "2026-04-03T02:07:29Z"
   },
   {
-    "additions": 8,
+    "additions": 2,
     "author": "sayakpaul",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Fix https://github.com/huggingface/diffusers/actions/runs/23486641913/job/68344105498?pr=13313. This is because from Transformers v5, we have the device attribute on `CLIPImageProcessor`.",
+    "body_excerpt": "# What does this PR do? To unblock the refactor of existing autoencoder tests to use the latest tester mixins, I will merge this PR once the CI is green.",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13323",
-    "created_at": "2026-03-24T11:32:09Z",
-    "deletions": 4,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13366",
+    "created_at": "2026-03-30T05:18:53Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13323/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13323",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13366/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13366",
     "labels": [],
     "merged": true,
-    "number": 13323,
+    "number": 13366,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix to device and to dtype tests.",
-    "updated_at": "2026-03-25T06:17:14Z"
+    "title": "[ci] move to assert instead of self.Assert*",
+    "updated_at": "2026-03-30T05:39:18Z"
   },
   {
-    "additions": 1,
-    "author": "DN6",
+    "additions": 32,
+    "author": "sayakpaul",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Latest version of the the Hub API removes the library argument in favour of filter. <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the rel\u2026",
+    "body_excerpt": "# What does this PR do? https://github.com/huggingface/diffusers/actions/runs/23678409392/job/68986000951",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13322",
-    "created_at": "2026-03-24T10:44:09Z",
-    "deletions": 1,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13365",
+    "created_at": "2026-03-30T05:06:45Z",
+    "deletions": 3,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13322/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13322",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13365/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13365",
     "labels": [],
     "merged": true,
-    "number": 13322,
-    "review_comments_count": 0,
+    "number": 13365,
+    "review_comments_count": 6,
     "state": "closed",
-    "title": "[CI] Update fetching pipelines for latest HF Hub Version",
-    "updated_at": "2026-03-24T11:12:32Z"
+    "title": "[ci] support claude reviewing on forks.",
+    "updated_at": "2026-03-31T09:26:12Z"
   },
   {
-    "additions": 11,
-    "author": "DN6",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Noticed this is resulting in a failure on our PyPi release test because the `torchvision` import is unguarded https://github.com/huggingface/diffusers/actions/runs/22723437757/job/65891766459#step:7:408 This PR remo\u2026",
+    "additions": 1,
+    "author": "YangKai0616",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? Per the title. Ref PR https://github.com/huggingface/diffusers/pull/13339 .",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13321",
-    "created_at": "2026-03-24T09:47:57Z",
-    "deletions": 4,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13364",
+    "created_at": "2026-03-30T04:20:27Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13321/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13321",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13364/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13364",
     "labels": [],
     "merged": true,
-    "number": 13321,
+    "number": 13364,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix unguarded `torchvision` import in Cosmos",
-    "updated_at": "2026-03-24T10:30:25Z"
+    "title": "Fix MotionConv2d to cast blur_kernel to input dtype instead of reverse",
+    "updated_at": "2026-03-31T09:53:12Z"
   },
   {
-    "additions": 8,
-    "author": "Alexkkir",
+    "additions": 25,
+    "author": "andrew-w-ross",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Currently everyone patches `_SET_ADAPTER_SCALE_FN_MAPPING` when training their models with peft https://github.com/search?q=_SET_ADAPTER_SCALE_FN_MAPPING&type=code Common patching practice looks like: ```python _SET\u2026",
-    "changed_files": 1,
+    "body_excerpt": "# What does this PR do? Fix's loading `ModularPipelines` loading component specs where `pretrained_model_name_or_path` is a path. Didn't bother opening an issue considering the tiny size of the fix. <!-- Congratulations! You've made it thi\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 5,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13320",
-    "created_at": "2026-03-24T08:31:05Z",
-    "deletions": 27,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13363",
+    "created_at": "2026-03-30T01:44:18Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13320/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13320",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13363/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13363",
     "labels": [],
     "merged": true,
-    "number": 13320,
-    "review_comments_count": 0,
+    "number": 13363,
+    "review_comments_count": 1,
     "state": "closed",
-    "title": "Use defaultdict for _SET_ADAPTER_SCALE_FN_MAPPING",
-    "updated_at": "2026-03-24T12:19:57Z"
+    "title": "corrects single file path validation logic",
+    "updated_at": "2026-04-08T00:28:00Z"
   },
   {
-    "additions": 43,
-    "author": "sayakpaul",
+    "additions": 3244,
+    "author": "yiyixuxu",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? * Pytest's `caplog` fixture isn't enough to capture warnings and the errors raised from `src/diffusers/loaders/peft.py`. * Some QwenImage hotswapping tests are failing because of recompilation issues (which is known\u2026",
-    "changed_files": 2,
+    "body_excerpt": null,
+    "changed_files": 23,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13318",
-    "created_at": "2026-03-24T05:59:50Z",
-    "deletions": 17,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13360",
+    "created_at": "2026-03-28T16:27:47Z",
+    "deletions": 46,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13318/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13318",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13360/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13360",
     "labels": [],
-    "merged": true,
-    "number": 13318,
+    "merged": false,
+    "number": 13360,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "[tests] fix lora logging tests for models.",
-    "updated_at": "2026-03-24T10:18:10Z"
+    "state": "open",
+    "title": "[WIP]LTX modular + 1:1 match + improve agent debugging skills ",
+    "updated_at": "2026-03-31T08:36:31Z"
   },
   {
-    "additions": 2445,
-    "author": "sippycoder",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? This PR introduces NucleusMoE-Image series into the diffusers library. NucleusMoE-Image is a 2B active 17B parameter model trained with efficiency at its core. Our novel architecture highlights the scalability of sp\u2026",
-    "changed_files": 17,
+    "additions": 841,
+    "author": "sayakpaul",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? TL;DR: Adds a guide on how to profile a pipeline and fix issues like CPU overhead, CPU<->GPU syncs, etc. ## Motivation Since we provide first-class `torch.compile` support, it's important that our pipelines are set\u2026",
+    "changed_files": 8,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 13,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13317",
-    "created_at": "2026-03-24T03:16:39Z",
-    "deletions": 3,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13356",
+    "created_at": "2026-03-28T04:35:05Z",
+    "deletions": 14,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13317/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13317",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/diffusers/pull/13356/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13356",
+    "labels": [
+      "performance"
+    ],
     "merged": true,
-    "number": 13317,
-    "review_comments_count": 53,
+    "number": 13356,
+    "review_comments_count": 50,
     "state": "closed",
-    "title": "NucleusMoE-Image",
-    "updated_at": "2026-04-03T09:01:13Z"
+    "title": "Add examples on how to profile a pipeline",
+    "updated_at": "2026-04-03T14:13:33Z"
   },
   {
-    "additions": 32,
-    "author": "YangKai0616",
+    "additions": 17,
+    "author": "howardzhang-cv",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? This PR improves numerical stability for four custom autoencoder RMS normalization layers. For `fp16/bf16` inputs, normalization is computed in `fp32` and then cast back to the original dtype. These layers are imple\u2026",
-    "changed_files": 4,
+    "body_excerpt": "# What does this PR do? ## Summary Update the minimum torchao version guarding to 0.15.0, as part of a larger motivation to update torchao quantizers to newer versions. Comment initially made in https://github.com/huggingface/diffusers/pul\u2026",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 7,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13316",
-    "created_at": "2026-03-23T11:52:46Z",
-    "deletions": 4,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13355",
+    "created_at": "2026-03-27T22:12:14Z",
+    "deletions": 20,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13316/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13316",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13355/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13355",
     "labels": [],
     "merged": true,
-    "number": 13316,
-    "review_comments_count": 2,
+    "number": 13355,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "Stabilize low-precision custom autoencoder RMS normalization",
-    "updated_at": "2026-03-24T09:00:06Z"
+    "title": "change minimum version guard for torchao to 0.15.0",
+    "updated_at": "2026-03-28T03:41:51Z"
   },
   {
-    "additions": 157,
-    "author": "DN6",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
-    "changed_files": 2,
+    "additions": 3,
+    "author": "hf-security-analysis[bot]",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "Update `.github/workflows/claude_review.yml` workflow configuration. cc @sayakpaul Closes huggingface/tracking-issues#34",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13315",
-    "created_at": "2026-03-23T11:47:51Z",
-    "deletions": 73,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13354",
+    "created_at": "2026-03-27T11:59:04Z",
+    "deletions": 3,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13315/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13315",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13354/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13354",
     "labels": [],
     "merged": false,
-    "number": 13315,
-    "review_comments_count": 0,
-    "state": "open",
-    "title": "[CI] Refactor Sana Transformer Tests",
-    "updated_at": "2026-03-23T11:47:51Z"
+    "number": 13354,
+    "review_comments_count": 1,
+    "state": "closed",
+    "title": "chore: update claude_review.yml",
+    "updated_at": "2026-04-02T08:43:45Z"
   },
   {
-    "additions": 27,
-    "author": "sayakpaul",
-    "author_association": "MEMBER",
-    "body_excerpt": null,
-    "changed_files": 2,
+    "additions": 115,
+    "author": "azolotenkov",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "## What does this PR do? This PR fixes a bug where `BucketBatchSampler` reshuffled precomputed batches on each `__iter__()` call. DreamBooth training scripts precompute latents and/or prompt embeddings and later consume them by dataloader\u2026",
+    "changed_files": 5,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13314",
-    "created_at": "2026-03-23T10:49:59Z",
-    "deletions": 3,
+    "comments_count": 4,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13353",
+    "created_at": "2026-03-27T11:21:35Z",
+    "deletions": 30,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13314/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13314",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13353/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13353",
     "labels": [],
-    "merged": true,
-    "number": 13314,
-    "review_comments_count": 0,
-    "state": "closed",
-    "title": "[chore] properly deprecate src.diffusers.utils.testing_utils.",
-    "updated_at": "2026-03-24T05:24:38Z"
+    "merged": false,
+    "number": 13353,
+    "review_comments_count": 6,
+    "state": "open",
+    "title": "Fix BucketBatchSampler cache alignment in DreamBooth scripts",
+    "updated_at": "2026-03-30T11:56:46Z"
   },
   {
-    "additions": 193,
+    "additions": 3,
     "author": "sayakpaul",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Fix https://github.com/huggingface/diffusers/issues/13311",
-    "changed_files": 2,
+    "body_excerpt": "# What does this PR do? Fixes https://github.com/huggingface/diffusers/actions/runs/23641475244/job/68862961669",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13313",
-    "created_at": "2026-03-23T06:53:19Z",
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13352",
+    "created_at": "2026-03-27T10:29:09Z",
     "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13313/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13313",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13352/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13352",
     "labels": [],
     "merged": true,
-    "number": 13313,
+    "number": 13352,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix klein lora loading.",
-    "updated_at": "2026-03-25T02:21:43Z"
+    "title": "[ci] include checkout step in claude review workflow",
+    "updated_at": "2026-03-27T11:58:34Z"
   },
   {
-    "additions": 30,
-    "author": "zhtmike",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? This PR expands the inputs of `QwenImage` UT from a single prompt to multiple prompts. Since `QwenImagePipeline` already supports multi-prompt inputs, we added corresponding test coverage here. - Tested with followi\u2026",
-    "changed_files": 4,
+    "additions": 236,
+    "author": "DN6",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
+    "changed_files": 6,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13312",
-    "created_at": "2026-03-23T05:45:44Z",
-    "deletions": 14,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13348",
+    "created_at": "2026-03-26T11:31:31Z",
+    "deletions": 231,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13312/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13312",
-    "labels": [],
-    "merged": true,
-    "number": 13312,
-    "review_comments_count": 11,
-    "state": "closed",
-    "title": "change QwenImageTransformer UT to batch inputs",
-    "updated_at": "2026-03-24T03:34:14Z"
+    "files_url": "https://github.com/huggingface/diffusers/pull/13348/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13348",
+    "labels": [
+      "tests",
+      "size/L"
+    ],
+    "merged": false,
+    "number": 13348,
+    "review_comments_count": 5,
+    "state": "open",
+    "title": "[CI] Refactor Skyreels, Lumina, Ominigen, Mochi transformer tests",
+    "updated_at": "2026-04-14T10:16:33Z"
   },
   {
-    "additions": 6,
-    "author": "r266-tech",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "## What does this PR do? Fixes #13298 When users pass a CUDA `torch.Generator` to diffusers pipelines, they may expect the same results as with a CPU generator given the same seed. However, PyTorch uses different random number generation a\u2026",
-    "changed_files": 1,
+    "additions": 288,
+    "author": "DN6",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
+    "changed_files": 5,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13309",
-    "created_at": "2026-03-22T18:03:37Z",
-    "deletions": 0,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13347",
+    "created_at": "2026-03-26T11:22:21Z",
+    "deletions": 98,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13309/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13309",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/diffusers/pull/13347/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13347",
+    "labels": [
+      "tests",
+      "size/L"
+    ],
     "merged": false,
-    "number": 13309,
-    "review_comments_count": 0,
+    "number": 13347,
+    "review_comments_count": 4,
     "state": "open",
-    "title": "Add warning when using CUDA random generator (#13298)",
-    "updated_at": "2026-03-31T06:25:05Z"
+    "title": "[CI] Refactor Chronoedit, PRX, EasyAnimate, Ovis transformer tests",
+    "updated_at": "2026-04-14T10:21:27Z"
   },
   {
-    "additions": 11,
-    "author": "agarwalprakhar2511",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "## Summary Fixes silent weight corruption when loading models with `device_map=\"mps\"`. ### Root cause `load_model_dict_into_meta` unconditionally sets `non_blocking=True` for `set_module_tensor_to_device` (added for accelerate > 1.8.1). Wi\u2026",
-    "changed_files": 1,
+    "additions": 181,
+    "author": "DN6",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
-    "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13308",
-    "created_at": "2026-03-22T09:21:13Z",
-    "deletions": 4,
+    "cluster_role": null,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13346",
+    "created_at": "2026-03-26T11:11:59Z",
+    "deletions": 140,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13308/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13308",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/diffusers/pull/13346/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13346",
+    "labels": [
+      "tests",
+      "size/L"
+    ],
     "merged": false,
-    "number": 13308,
-    "review_comments_count": 0,
+    "number": 13346,
+    "review_comments_count": 3,
     "state": "open",
-    "title": "fix: disable non-blocking tensor copies to MPS during model loading",
-    "updated_at": "2026-03-22T09:27:02Z"
+    "title": "[CI] Refactor Cog Transformer Model Tests",
+    "updated_at": "2026-04-14T10:42:16Z"
   },
   {
-    "additions": 12,
-    "author": "agarwalprakhar2511",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "## Summary When `--with_prior_preservation` is enabled in the Flux2 dreambooth LoRA training scripts, the prompt embedding repeat logic double-counts the batch size, producing a shape mismatch against the latent tensor. **Root cause:** `co\u2026",
+    "additions": 250,
+    "author": "DN6",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
     "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13307",
-    "created_at": "2026-03-22T09:13:05Z",
-    "deletions": 3,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13345",
+    "created_at": "2026-03-26T11:05:29Z",
+    "deletions": 101,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13307/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13307",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/diffusers/pull/13345/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13345",
+    "labels": [
+      "tests",
+      "size/L"
+    ],
     "merged": false,
-    "number": 13307,
+    "number": 13345,
     "review_comments_count": 0,
     "state": "open",
-    "title": "fix(dreambooth): batch size mismatch with --with_prior_preservation in flux2 scripts",
-    "updated_at": "2026-03-22T09:13:32Z"
+    "title": "[CI] Refactor Chroma , LongCat and HiDream Transformer Tests",
+    "updated_at": "2026-04-14T10:52:50Z"
   },
   {
-    "additions": 380,
-    "author": "pthombre",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "--- What does this PR do? Adds a new documentation page for NeMo Automodel, NVIDIA's PyTorch DTensor-native training library for fine-tuning and pretraining diffusion models at scale. NeMo Automodel integrates directly with Diffusers \u2014 it\u2026",
+    "additions": 96,
+    "author": "DN6",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13306",
-    "created_at": "2026-03-22T00:51:33Z",
-    "deletions": 0,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13344",
+    "created_at": "2026-03-26T11:03:48Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13306/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13306",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/diffusers/pull/13344/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13344",
+    "labels": [
+      "models",
+      "tests",
+      "size/M"
+    ],
     "merged": true,
-    "number": 13306,
-    "review_comments_count": 24,
+    "number": 13344,
+    "review_comments_count": 1,
     "state": "closed",
-    "title": "[docs] Add NeMo Automodel training guide",
-    "updated_at": "2026-03-30T17:21:58Z"
+    "title": "[CI] Add GLM Image Transformer Model Tests",
+    "updated_at": "2026-04-07T10:58:06Z"
   },
   {
-    "additions": 18,
-    "author": "s-zx",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "## What does this PR do? Fixes #13281. When combining torchao quantization (e.g. `Float8WeightOnlyConfig`) with group offloading (`use_stream=True`), inference fails with a device mismatch: the quantized weight remains on CPU while the inp\u2026",
-    "changed_files": 1,
+    "additions": 379,
+    "author": "DN6",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
+    "changed_files": 5,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13305",
-    "created_at": "2026-03-22T00:34:28Z",
-    "deletions": 9,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13342",
+    "created_at": "2026-03-26T10:43:26Z",
+    "deletions": 346,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13305/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13305",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13342/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13342",
     "labels": [],
+    "merged": true,
+    "number": 13342,
+    "review_comments_count": 3,
+    "state": "closed",
+    "title": "[CI] Hunyuan Transformer Tests Refactor",
+    "updated_at": "2026-04-06T14:46:20Z"
+  },
+  {
+    "additions": 158,
+    "author": "DN6",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
+    "changed_files": 2,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13341",
+    "created_at": "2026-03-26T10:07:29Z",
+    "deletions": 96,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/diffusers/pull/13341/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13341",
+    "labels": [
+      "tests",
+      "size/L"
+    ],
     "merged": false,
-    "number": 13305,
-    "review_comments_count": 0,
+    "number": 13341,
+    "review_comments_count": 2,
     "state": "open",
-    "title": "fix: skip pin_memory and non_blocking transfer for tensor subclasses in group offloading",
-    "updated_at": "2026-03-22T00:34:28Z"
+    "title": "[CI] Refactor Bria Transformer Tests",
+    "updated_at": "2026-04-14T10:39:26Z"
   },
   {
-    "additions": 24,
-    "author": "s-zx",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "## What does this PR do? Fixes `ImportError` when using `TorchAoConfig` with torchao >= 0.16.0. torchao 0.15.0 deprecated the snake_case quantization functions (`int4_weight_only`, `float8_weight_only`, etc.) with deprecation warnings, and\u2026",
+    "additions": 128,
+    "author": "DN6",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13304",
-    "created_at": "2026-03-22T00:04:23Z",
-    "deletions": 10,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13340",
+    "created_at": "2026-03-26T09:56:24Z",
+    "deletions": 102,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13304/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13304",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13340/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13340",
     "labels": [],
     "merged": false,
-    "number": 13304,
-    "review_comments_count": 0,
+    "number": 13340,
+    "review_comments_count": 2,
     "state": "open",
-    "title": "fix: support torchao >= 0.16.0 by importing renamed CamelCase Config classes",
-    "updated_at": "2026-03-22T00:04:23Z"
+    "title": "[CI] Refactor SD3 Transformer Test",
+    "updated_at": "2026-04-14T19:41:09Z"
   },
   {
-    "additions": 2,
-    "author": "s-zx",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "## What does this PR do? Fixes dtype mismatch errors when running SD3 or FLUX models on CPU with `torch_dtype=torch.float16`. When the model's projection layers (Conv2d in `PatchEmbed`, Linear in `FluxTransformer2DModel`) are in float32 bu\u2026",
+    "additions": 15,
+    "author": "YangKai0616",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# Issue: The `test_save_load_float16` test has a hidden non-determinism problem: 1. When the pipeline is constructed directly, its components default to training mode (e.g., `text_encoder` contains `dropout` layers). However, after loading\u2026",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13303",
-    "created_at": "2026-03-21T23:55:48Z",
-    "deletions": 2,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13339",
+    "created_at": "2026-03-26T09:19:39Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13303/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13303",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13339/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13339",
     "labels": [],
-    "merged": false,
-    "number": 13303,
-    "review_comments_count": 0,
-    "state": "open",
-    "title": "fix: cast input to weight dtype in PatchEmbed and FluxTransformer2DModel to prevent dtype mismatch",
-    "updated_at": "2026-03-21T23:55:48Z"
+    "merged": true,
+    "number": 13339,
+    "review_comments_count": 12,
+    "state": "closed",
+    "title": "Fix: ensure consistent dtype and eval mode in pipeline save/load tests",
+    "updated_at": "2026-03-31T08:53:23Z"
   },
   {
-    "additions": 7,
-    "author": "s-zx",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "## Summary When `set_attention_backend()` is called directly on an individual attention submodule (e.g. `pipe.transformer.attn1.set_attention_backend(\"sage_hub\")`), hub-based kernels are never downloaded and the backend silently falls back\u2026",
+    "additions": 1,
+    "author": "sayakpaul",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Fix https://github.com/huggingface/diffusers/actions/runs/23581004473",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13302",
-    "created_at": "2026-03-21T23:33:44Z",
-    "deletions": 1,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13338",
+    "created_at": "2026-03-26T07:22:48Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13302/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13302",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13338/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13338",
     "labels": [],
-    "merged": false,
-    "number": 13302,
+    "merged": true,
+    "number": 13338,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "fix(attention): download hub kernel in AttentionModuleMixin.set_attention_backend",
-    "updated_at": "2026-03-21T23:33:44Z"
+    "state": "closed",
+    "title": "fix claude workflow to include id-token with write.",
+    "updated_at": "2026-03-26T10:09:15Z"
   },
   {
-    "additions": 9,
-    "author": "adi776borate",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
-    "changed_files": 3,
+    "additions": 300,
+    "author": "dg845",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? This PR updates the LTX-2 docs to cover multimodal guidance and prompt enhancement, which were added with LTX-2.3 model support in #13217. Additionally, the LTX-2.X official default negative prompt, T2V system promp\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13299",
-    "created_at": "2026-03-21T09:54:54Z",
-    "deletions": 3,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13337",
+    "created_at": "2026-03-26T06:15:28Z",
+    "deletions": 15,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13299/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13299",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13337/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13337",
     "labels": [],
-    "merged": false,
-    "number": 13299,
-    "review_comments_count": 0,
-    "state": "open",
-    "title": "Fix missing latents_bn_std dtype cast in VAE normalization",
-    "updated_at": "2026-03-21T09:54:54Z"
-  },
-  {
-    "additions": 49,
-    "author": "sayakpaul",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Use Claude for PR reviews. Some choices / comments: - A GitHub Actions workflow (using `anthropics/claude-code-action`) that auto-reviews PRs touching `src/diffusers/**` and responds to `@claude` mentions from folks\u2026",
+    "merged": true,
+    "number": 13337,
+    "review_comments_count": 10,
+    "state": "closed",
+    "title": "Update LTX-2 Docs to Cover LTX-2.3 Models",
+    "updated_at": "2026-03-27T00:51:33Z"
+  },
+  {
+    "additions": 2,
+    "author": "kaixuanliu",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "@sayakpaul pls help review, thx!",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13297",
-    "created_at": "2026-03-20T06:43:17Z",
-    "deletions": 0,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13336",
+    "created_at": "2026-03-26T06:00:04Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13297/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13297",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13336/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13336",
     "labels": [],
     "merged": true,
-    "number": 13297,
-    "review_comments_count": 14,
+    "number": 13336,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "[ci] claude in ci.",
-    "updated_at": "2026-03-25T16:00:18Z"
+    "title": "avoid hardcode device in flux-control example",
+    "updated_at": "2026-03-26T07:10:54Z"
   },
   {
-    "additions": 8,
+    "additions": 101,
     "author": "DN6",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Release tests were not updated to use the transformers `main` branch. This is resulting in unrelated failures on the release branch https://github.com/huggingface/diffusers/actions/runs/23330168017/job/67859952145#s\u2026",
+    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13296",
-    "created_at": "2026-03-20T05:46:04Z",
-    "deletions": 0,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13335",
+    "created_at": "2026-03-26T04:07:43Z",
+    "deletions": 72,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13296/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13296",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13335/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13335",
     "labels": [],
     "merged": true,
-    "number": 13296,
+    "number": 13335,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "[CI] Update transformer version in release tests ",
-    "updated_at": "2026-03-20T06:10:06Z"
+    "title": "[CI] Refactor Cosmos Transformer Tests",
+    "updated_at": "2026-04-06T04:35:37Z"
   },
   {
-    "additions": 2,
-    "author": "sayakpaul",
+    "additions": 101,
+    "author": "stevhliu",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? The WanVACE LoRA test failures are the flakiest in nature. They have been making our CI red for a while now. This is another attempt getting rid of that: https://github.com/huggingface/diffusers/actions/runs/2330825\u2026",
-    "changed_files": 1,
+    "body_excerpt": "follow up to #13297: - created `models.md`, a shared reference for holding model coding style, dependencies, common conventions, attention patterns, gotchas, etc. - updated `AGENTS.md`, `model-integration/SKILL.md`, and `review-rules.md` t\u2026",
+    "changed_files": 4,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13294",
-    "created_at": "2026-03-20T04:01:35Z",
-    "deletions": 4,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13334",
+    "created_at": "2026-03-25T16:28:29Z",
+    "deletions": 85,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13294/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13294",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13334/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13334",
     "labels": [],
     "merged": true,
-    "number": 13294,
-    "review_comments_count": 2,
+    "number": 13334,
+    "review_comments_count": 12,
     "state": "closed",
-    "title": "[ci] hoping to fix is_flaky with wanvace.",
-    "updated_at": "2026-03-20T10:34:39Z"
+    "title": "[docs] refactor model skill",
+    "updated_at": "2026-03-30T06:13:57Z"
   },
   {
-    "additions": 11,
-    "author": "sayakpaul",
+    "additions": 29,
+    "author": "kashif",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Fix https://github.com/huggingface/diffusers/actions/runs/23308255022/job/67788477405?pr=13290",
-    "changed_files": 1,
+    "body_excerpt": "# What does this PR do? move the documentation into its own Text section and clarify some docstring. <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release no\u2026",
+    "changed_files": 4,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13293",
-    "created_at": "2026-03-20T03:31:58Z",
-    "deletions": 10,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13333",
+    "created_at": "2026-03-25T11:53:03Z",
+    "deletions": 19,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13293/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13293",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13333/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13333",
     "labels": [],
     "merged": true,
-    "number": 13293,
+    "number": 13333,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "[tests] fix audioldm2 tests.",
-    "updated_at": "2026-03-20T15:23:24Z"
+    "title": "[LLADA2] documentation fixes",
+    "updated_at": "2026-03-25T19:02:06Z"
   },
   {
-    "additions": 174,
-    "author": "howardzhang-cv",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Remove the deprecated string-based quant_type path from TorchAoConfig, requiring AOBaseConfig instances instead. - TorchAoConfig.__init__ now only accepts AOBaseConfig subclass instances (e.g. Int8WeightOnlyConfig()\u2026",
-    "changed_files": 6,
+    "additions": 46,
+    "author": "sayakpaul",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? We should be able to catch stuff like https://github.com/huggingface/diffusers/pull/13321. Additionally, I ported the existing dependency test suite to use `pytest`.",
+    "changed_files": 4,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 12,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13291",
-    "created_at": "2026-03-19T20:31:22Z",
-    "deletions": 580,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13332",
+    "created_at": "2026-03-25T08:42:00Z",
+    "deletions": 10,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13291/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13291",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/diffusers/pull/13332/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13332",
+    "labels": [
+      "tests",
+      "size/M",
+      "pipelines",
+      "CI"
+    ],
     "merged": true,
-    "number": 13291,
-    "review_comments_count": 37,
+    "number": 13332,
+    "review_comments_count": 1,
     "state": "closed",
-    "title": "remove str option for quantization config in torchao",
-    "updated_at": "2026-03-27T22:12:52Z"
+    "title": "[tests] tighten dependency testing.",
+    "updated_at": "2026-04-10T14:31:05Z"
   },
   {
-    "additions": 31,
-    "author": "cbensimon",
-    "author_association": "MEMBER",
-    "body_excerpt": null,
-    "changed_files": 4,
+    "additions": 1417,
+    "author": "satani99",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "Added CogvideoxFunInpaintPipeline <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fu\u2026",
+    "changed_files": 7,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13290",
-    "created_at": "2026-03-19T17:34:08Z",
-    "deletions": 10,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13331",
+    "created_at": "2026-03-25T05:05:58Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13290/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13290",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13331/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13331",
     "labels": [],
-    "merged": true,
-    "number": 13290,
-    "review_comments_count": 6,
-    "state": "closed",
-    "title": "[export] Add export-safe LRU cache helper",
-    "updated_at": "2026-03-23T12:40:21Z"
+    "merged": false,
+    "number": 13331,
+    "review_comments_count": 11,
+    "state": "open",
+    "title": "Cogvideox_fun_inp pipeline",
+    "updated_at": "2026-03-26T02:09:17Z"
   },
   {
-    "additions": 172,
-    "author": "JingyaHuang",
+    "additions": 7,
+    "author": "sayakpaul",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? This PR adds AWS Neuron (Trainium/Inferentia) as an officially supported compute backend in Diffusers, on par with existing backends like CUDA, MPS, XPU, and MLU. **Changes** - import_utils.py \u2014 adds is_torch_neuron\u2026",
-    "changed_files": 13,
+    "body_excerpt": "# What does this PR do? Surfaced in https://github.com/huggingface/diffusers/pull/13291#issuecomment-4122867870. Cc: @howardzhang-cv",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13289",
-    "created_at": "2026-03-19T16:01:08Z",
-    "deletions": 24,
-    "draft": true,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13289/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13289",
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13330",
+    "created_at": "2026-03-25T04:09:32Z",
+    "deletions": 1,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/diffusers/pull/13330/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13330",
     "labels": [
-      "lora",
-      "models",
       "tests",
-      "size/M",
-      "utils",
-      "pipelines",
-      "examples"
+      "size/S",
+      "CI"
     ],
     "merged": false,
-    "number": 13289,
-    "review_comments_count": 0,
+    "number": 13330,
+    "review_comments_count": 2,
     "state": "open",
-    "title": "[Neuron] Add AWS Neuron (Trainium/Inferentia) as an officially supported device",
-    "updated_at": "2026-04-10T15:36:12Z"
+    "title": "[tests] fix torchao tests",
+    "updated_at": "2026-04-16T04:57:53Z"
   },
   {
-    "additions": 110,
-    "author": "DN6",
+    "additions": 34,
+    "author": "sayakpaul",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Adding a test to catch cases such as #13271 <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so ma\u2026",
+    "body_excerpt": "# What does this PR do? Simplify release workflow like https://github.com/huggingface/transformers/blob/main/.github/workflows/release.yml. - Removed Test PyPI \u2014 no longer publish to Test PyPI and install from there; instead, install direc\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13287",
-    "created_at": "2026-03-19T07:44:01Z",
-    "deletions": 1,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13329",
+    "created_at": "2026-03-25T04:04:57Z",
+    "deletions": 44,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13287/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13287",
-    "labels": [],
-    "merged": true,
-    "number": 13287,
-    "review_comments_count": 0,
-    "state": "closed",
-    "title": "[Modular] Test for catching dtype and device issues with AutoModel type hints",
-    "updated_at": "2026-03-20T05:06:03Z"
+    "files_url": "https://github.com/huggingface/diffusers/pull/13329/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13329",
+    "labels": [
+      "size/M",
+      "CI"
+    ],
+    "merged": false,
+    "number": 13329,
+    "review_comments_count": 2,
+    "state": "open",
+    "title": "[ci] simplify release workflow.",
+    "updated_at": "2026-04-12T06:50:48Z"
   },
   {
-    "additions": 159,
-    "author": "howardzhang-cv",
+    "additions": 119,
+    "author": "zhtmike",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Adds low precision attention API from TorchAO to diffusers by updating TorchAoConfig with attn_backend option. Note: this will require torchao 0.17.0 ## Todo: - [ ] Need to add new necessary tests - [ ] Need to upda\u2026",
+    "body_excerpt": "# What does this PR do? Solve the issue #13319. There are two bugs: - `grad_out` is already in `BSHD` format, which matches the shape of `out`. Therefore, we do not need to permute `grad_out` again. - `autograd.Function.backward()` does no\u2026",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13285",
-    "created_at": "2026-03-19T00:41:59Z",
-    "deletions": 9,
-    "draft": true,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13285/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13285",
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13328",
+    "created_at": "2026-03-25T03:35:26Z",
+    "deletions": 16,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/diffusers/pull/13328/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13328",
     "labels": [],
-    "merged": false,
-    "number": 13285,
+    "merged": true,
+    "number": 13328,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "Add low precision attention API from torchao to TorchAoConfig",
-    "updated_at": "2026-03-30T18:59:40Z"
+    "state": "closed",
+    "title": "Fix Ulysses SP backward with SDPA",
+    "updated_at": "2026-04-16T03:44:44Z"
   },
   {
-    "additions": 1,
-    "author": "sayakpaul",
-    "author_association": "MEMBER",
-    "body_excerpt": null,
-    "changed_files": 1,
+    "additions": 106,
+    "author": "LiuYinfeng01",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "Add diffusers.utils.aiter_swizzle_gemm: can_shuffle (layout k multiple of 2*IK), apply_swizzle for eligible large nn.Linear layers using shuffle_weight + hipb_mm with bpreshuffle. Requires AMD MI-series (ROCm) with the aiter package instal\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13283",
-    "created_at": "2026-03-18T10:14:14Z",
-    "deletions": 1,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13327",
+    "created_at": "2026-03-25T03:29:24Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13283/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13283",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13327/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13327",
     "labels": [],
     "merged": false,
-    "number": 13283,
-    "review_comments_count": 1,
+    "number": 13327,
+    "review_comments_count": 0,
     "state": "open",
-    "title": "Update attention_backends.md to update FA3 minimum support to Ampere",
-    "updated_at": "2026-03-18T18:01:41Z"
+    "title": "feat(utils): optional aiter swizzle GEMM for ROCm MI + aiter",
+    "updated_at": "2026-03-30T09:50:48Z"
   },
   {
-    "additions": 0,
-    "author": "SHYuanBest",
+    "additions": 999,
+    "author": "CalamitousFelicitousness",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "cc @sayakpaul related to: https://gist.github.com/sayakpaul/aba9d8925a24944f183d67f1ef4e009d",
-    "changed_files": 1,
+    "body_excerpt": "Adds support for Flux2 LoKR, with dual path to benchmark implementations. - Custom lossless path: BFL LoKR keys \u2192 peft LoKrConfig (fuse-first QKV) - Generic lossy path: optional SVD conversion via peft.convert_to_lora - Fix alpha handling\u2026",
+    "changed_files": 5,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13282",
-    "created_at": "2026-03-18T07:25:19Z",
-    "deletions": 2,
+    "comments_count": 29,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13326",
+    "created_at": "2026-03-25T01:26:53Z",
+    "deletions": 64,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13282/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13282",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13326/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13326",
     "labels": [],
-    "merged": true,
-    "number": 13282,
-    "review_comments_count": 0,
-    "state": "closed",
-    "title": "[Helios] Remove lru_cache for better AoTI compatibility and cleaner code",
-    "updated_at": "2026-03-18T18:11:58Z"
+    "merged": false,
+    "number": 13326,
+    "review_comments_count": 17,
+    "state": "open",
+    "title": "Add Flux2 LoKR adapter support prototype with dual conversion paths",
+    "updated_at": "2026-04-08T09:26:07Z"
   },
   {
-    "additions": 44,
-    "author": "sayakpaul",
+    "additions": 2048,
+    "author": "linoytsaban",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Needs `kernels` to be installed from source for testing. Users might have to update `nvidia-cutlass-dsl` (or might even do a `--force-reinstall`). In this PR, we're just supporting the bare minimum inference support\u2026",
+    "body_excerpt": "adds a flux.2 klein kv training script, with additional general changes that can be propagated to other lora scripts as well: - [x] **Default aspect ratio buckets** - uses preset aspect ratio buckets to avoid needing to pass them manually.\u2026",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13280",
-    "created_at": "2026-03-18T03:23:30Z",
-    "deletions": 0,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13325",
+    "created_at": "2026-03-24T18:41:22Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13280/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13280",
-    "labels": [],
-    "merged": true,
-    "number": 13280,
-    "review_comments_count": 0,
+    "files_url": "https://github.com/huggingface/diffusers/pull/13325/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13325",
+    "labels": [
+      "size/L",
+      "pipelines",
+      "examples"
+    ],
+    "merged": false,
+    "number": 13325,
+    "review_comments_count": 10,
     "state": "closed",
-    "title": "[core] fa4 support.",
-    "updated_at": "2026-03-20T11:59:28Z"
+    "title": "[flux.2 LoRA] make lora training compatible with flux.2 klein kv",
+    "updated_at": "2026-04-16T14:22:00Z"
   },
   {
-    "additions": 1,
-    "author": "zhtmike",
+    "additions": 2380,
+    "author": "AlanPonnachan",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Fix issue #13277. QwenImagePipeline cannot run with Ulysses SP together with batch prompt inputs. It is related to the mask is not correctly broadcasted. We need to broadcast the attention mask from [B, S] to [B, H,\u2026",
-    "changed_files": 2,
+    "body_excerpt": "# What does this PR do? This PR introduces the modular architecture for Stable Diffusion 3 (SD3), implementing both Text-to-Image (T2I) and Image-to-Image (I2I) pipelines. **Key additions:** - Added `SD3ModularPipeline` and `SD3AutoBlocks`\u2026",
+    "changed_files": 14,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 5,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13278",
-    "created_at": "2026-03-17T08:25:19Z",
-    "deletions": 1,
+    "comments_count": 22,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13324",
+    "created_at": "2026-03-24T17:33:51Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13278/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13278",
-    "labels": [],
-    "merged": true,
-    "number": 13278,
-    "review_comments_count": 18,
-    "state": "closed",
-    "title": "Fix the attention mask in ulysses SP for QwenImage",
-    "updated_at": "2026-03-24T09:33:56Z"
+    "files_url": "https://github.com/huggingface/diffusers/pull/13324/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13324",
+    "labels": [
+      "close-to-merge",
+      "tests",
+      "size/L",
+      "modular-pipelines",
+      "utils"
+    ],
+    "merged": false,
+    "number": 13324,
+    "review_comments_count": 23,
+    "state": "open",
+    "title": "feat: Add Modular Pipeline for Stable Diffusion 3 (SD3)",
+    "updated_at": "2026-04-21T14:49:49Z"
   },
   {
-    "additions": 105,
+    "additions": 8,
     "author": "sayakpaul",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Fix offloading when using TorchAO. This assumes that the underlying quantization tensor class implements pinning properly. But that's not something we can do in TorchAO, anyway. The benefit of this is that many new\u2026",
+    "body_excerpt": "# What does this PR do? Fix https://github.com/huggingface/diffusers/actions/runs/23486641913/job/68344105498?pr=13313. This is because from Transformers v5, we have the device attribute on `CLIPImageProcessor`.",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 7,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13276",
-    "created_at": "2026-03-17T05:16:05Z",
-    "deletions": 14,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13323",
+    "created_at": "2026-03-24T11:32:09Z",
+    "deletions": 4,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13276/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13276",
-    "labels": [
-      "performance",
-      "roadmap"
-    ],
+    "files_url": "https://github.com/huggingface/diffusers/pull/13323/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13323",
+    "labels": [],
     "merged": true,
-    "number": 13276,
-    "review_comments_count": 13,
+    "number": 13323,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "[core] fix group offloading when using torchao",
-    "updated_at": "2026-04-06T20:22:57Z"
+    "title": "fix to device and to dtype tests.",
+    "updated_at": "2026-03-25T06:17:14Z"
   },
   {
-    "additions": 3,
-    "author": "sayakpaul",
+    "additions": 1,
+    "author": "DN6",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Fixes https://github.com/huggingface/diffusers/actions/runs/23176322293/job/67339276234?pr=13270",
+    "body_excerpt": "# What does this PR do? Latest version of the the Hub API removes the library argument in favour of filter. <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the rel\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13275",
-    "created_at": "2026-03-17T03:59:44Z",
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13322",
+    "created_at": "2026-03-24T10:44:09Z",
     "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13275/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13275",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13322/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13322",
     "labels": [],
     "merged": true,
-    "number": 13275,
+    "number": 13322,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "[tests] fix llava kwargs in the hunyuan tests",
-    "updated_at": "2026-03-17T04:41:51Z"
+    "title": "[CI] Update fetching pipelines for latest HF Hub Version",
+    "updated_at": "2026-03-24T11:12:32Z"
   },
   {
-    "additions": 112,
-    "author": "gambletan",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "## Summary Fixes #13124 `log_validation()` in dreambooth training scripts runs pipeline inference using the same UNet/transformer that is being trained, without `torch.no_grad()`. With `--mixed_precision=\"fp16\"`, gradients are computed and\u2026",
-    "changed_files": 17,
-    "cluster_id": "cluster-1246-22",
-    "cluster_ids": [
-      "cluster-1246-22"
-    ],
-    "cluster_role": "member",
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13273",
-    "created_at": "2026-03-16T14:13:58Z",
-    "deletions": 99,
+    "additions": 11,
+    "author": "DN6",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Noticed this is resulting in a failure on our PyPi release test because the `torchvision` import is unguarded https://github.com/huggingface/diffusers/actions/runs/22723437757/job/65891766459#step:7:408 This PR remo\u2026",
+    "changed_files": 1,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13321",
+    "created_at": "2026-03-24T09:47:57Z",
+    "deletions": 4,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13273/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13273",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13321/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13321",
     "labels": [],
-    "merged": false,
-    "number": 13273,
+    "merged": true,
+    "number": 13321,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "fix: wrap validation inference with torch.no_grad() in dreambooth examples",
-    "updated_at": "2026-03-16T14:13:58Z"
+    "state": "closed",
+    "title": "Fix unguarded `torchvision` import in Cosmos",
+    "updated_at": "2026-03-24T10:30:25Z"
   },
   {
-    "additions": 64,
-    "author": "gambletan",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "## Summary Fixes #12913 - **Bug**: `image_seq_len = latents.shape[1]` takes the **channel dimension** (e.g. 16) instead of the spatial sequence length. Lumina2 latents have shape `(batch, channels, height, width)` and are NOT packed before\u2026",
-    "changed_files": 2,
+    "additions": 8,
+    "author": "Alexkkir",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? Currently everyone patches `_SET_ADAPTER_SCALE_FN_MAPPING` when training their models with peft https://github.com/search?q=_SET_ADAPTER_SCALE_FN_MAPPING&type=code Common patching practice looks like: ```python _SET\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13272",
-    "created_at": "2026-03-16T14:10:56Z",
-    "deletions": 1,
+    "comments_count": 5,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13320",
+    "created_at": "2026-03-24T08:31:05Z",
+    "deletions": 27,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13272/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13272",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13320/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13320",
     "labels": [],
-    "merged": false,
-    "number": 13272,
-    "review_comments_count": 0,
-    "state": "open",
-    "title": "fix: compute image_seq_len from spatial dims in Lumina2 pipeline",
-    "updated_at": "2026-03-16T14:10:56Z"
+    "merged": true,
+    "number": 13320,
+    "review_comments_count": 0,
+    "state": "closed",
+    "title": "Use defaultdict for _SET_ADAPTER_SCALE_FN_MAPPING",
+    "updated_at": "2026-03-24T12:19:57Z"
   },
   {
-    "additions": 3,
-    "author": "DN6",
+    "additions": 43,
+    "author": "sayakpaul",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Oversight on my part when adding #13168. We avoid passing `torch_dtype` to non `nn.Modules` such as Processors and Tokenizers because these objects do not accept that anymore and instead store it in init_kwargs whic\u2026",
-    "changed_files": 1,
+    "body_excerpt": "# What does this PR do? * Pytest's `caplog` fixture isn't enough to capture warnings and the errors raised from `src/diffusers/loaders/peft.py`. * Some QwenImage hotswapping tests are failing because of recompilation issues (which is known\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13271",
-    "created_at": "2026-03-16T11:44:49Z",
-    "deletions": 9,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13318",
+    "created_at": "2026-03-24T05:59:50Z",
+    "deletions": 17,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13271/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13271",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13318/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13318",
     "labels": [],
     "merged": true,
-    "number": 13271,
-    "review_comments_count": 1,
+    "number": 13318,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "[Modular] Fix dtype assignment when type hint is AutoModel",
-    "updated_at": "2026-03-17T04:17:53Z"
+    "title": "[tests] fix lora logging tests for models.",
+    "updated_at": "2026-03-24T10:18:10Z"
   },
   {
-    "additions": 26,
-    "author": "sywangyi",
+    "additions": 2445,
+    "author": "sippycoder",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "fix tests/models/transformers/test_models_transformer_flux.py::TestFluxTransformerContextParallel::test_context_parallel_inference[ulysses] tests/models/transformers/test_models_transformer_flux.py::TestFluxTransformerContextParallel::test\u2026",
-    "changed_files": 1,
+    "body_excerpt": "# What does this PR do? This PR introduces NucleusMoE-Image series into the diffusers library. NucleusMoE-Image is a 2B active 17B parameter model trained with efficiency at its core. Our novel architecture highlights the scalability of sp\u2026",
+    "changed_files": 17,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13270",
-    "created_at": "2026-03-16T07:47:18Z",
-    "deletions": 7,
+    "comments_count": 14,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13317",
+    "created_at": "2026-03-24T03:16:39Z",
+    "deletions": 3,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13270/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13270",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13317/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13317",
     "labels": [],
     "merged": true,
-    "number": 13270,
-    "review_comments_count": 1,
+    "number": 13317,
+    "review_comments_count": 53,
     "state": "closed",
-    "title": "fix parallelism case failure in xpu",
-    "updated_at": "2026-03-17T03:22:16Z"
+    "title": "NucleusMoE-Image",
+    "updated_at": "2026-04-15T00:34:07Z"
   },
   {
-    "additions": 728,
-    "author": "yiyixuxu",
-    "author_association": "MEMBER",
-    "body_excerpt": null,
-    "changed_files": 9,
+    "additions": 32,
+    "author": "YangKai0616",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? This PR improves numerical stability for four custom autoencoder RMS normalization layers. For `fp16/bf16` inputs, normalization is computed in `fp32` and then cast back to the original dtype. These layers are imple\u2026",
+    "changed_files": 4,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13269",
-    "created_at": "2026-03-15T20:47:20Z",
-    "deletions": 55,
+    "comments_count": 7,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13316",
+    "created_at": "2026-03-23T11:52:46Z",
+    "deletions": 4,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13269/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13269",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13316/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13316",
     "labels": [],
     "merged": true,
-    "number": 13269,
-    "review_comments_count": 31,
+    "number": 13316,
+    "review_comments_count": 2,
     "state": "closed",
-    "title": "[agents]support skills",
-    "updated_at": "2026-03-20T04:07:46Z"
+    "title": "Stabilize low-precision custom autoencoder RMS normalization",
+    "updated_at": "2026-03-24T09:00:06Z"
   },
   {
-    "additions": 217,
-    "author": "aryabyte21",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "## Summary Adds support for the [DiffSynth-Studio blockwise ControlNet](https://github.com/modelscope/DiffSynth-Studio) architecture to the existing `QwenImageControlNetModel`, addressing [the maintainer's feedback on PR #12317](https://gi\u2026",
-    "changed_files": 3,
+    "additions": 157,
+    "author": "DN6",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13268",
-    "created_at": "2026-03-15T05:17:57Z",
-    "deletions": 7,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13315",
+    "created_at": "2026-03-23T11:47:51Z",
+    "deletions": 73,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13268/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13268",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13315/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13315",
     "labels": [],
     "merged": false,
-    "number": 13268,
+    "number": 13315,
     "review_comments_count": 0,
     "state": "open",
-    "title": "Add DiffSynth blockwise ControlNet support to QwenImageControlNetModel",
-    "updated_at": "2026-03-15T05:17:57Z"
+    "title": "[CI] Refactor Sana Transformer Tests",
+    "updated_at": "2026-03-23T11:47:51Z"
   },
   {
-    "additions": 1,
-    "author": "yunjoochoi",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? One-line fix \u2014 missing self.gradient_checkpointing = False init, consistent with all other ControlNet implementations <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, you\u2026",
-    "changed_files": 1,
+    "additions": 27,
+    "author": "sayakpaul",
+    "author_association": "MEMBER",
+    "body_excerpt": null,
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13267",
-    "created_at": "2026-03-15T03:43:58Z",
-    "deletions": 0,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13314",
+    "created_at": "2026-03-23T10:49:59Z",
+    "deletions": 3,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13267/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13267",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13314/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13314",
     "labels": [],
-    "merged": false,
-    "number": 13267,
+    "merged": true,
+    "number": 13314,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "Fix missing gradient_checkpointing init in ZImageControlNetModel",
-    "updated_at": "2026-03-16T21:55:13Z"
+    "state": "closed",
+    "title": "[chore] properly deprecate src.diffusers.utils.testing_utils.",
+    "updated_at": "2026-03-24T05:24:38Z"
   },
   {
-    "additions": 71,
-    "author": "abhixb",
-    "author_association": "NONE",
-    "body_excerpt": "Fixed issue #13177 ## The problem If you pass a tensor or numpy array as the image to the Klein pipeline, it errors even though the docstring explicitly says these types are supported. Three things cause this: - The image processor's valid\u2026",
-    "changed_files": 3,
+    "additions": 193,
+    "author": "sayakpaul",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Fix https://github.com/huggingface/diffusers/issues/13311",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13265",
-    "created_at": "2026-03-14T13:12:44Z",
-    "deletions": 8,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13313",
+    "created_at": "2026-03-23T06:53:19Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13265/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13265",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13313/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13313",
     "labels": [],
-    "merged": false,
-    "number": 13265,
+    "merged": true,
+    "number": 13313,
     "review_comments_count": 0,
     "state": "closed",
-    "title": " add tensor and numpy array support for image input in Flux2KleinPipeline",
-    "updated_at": "2026-04-03T09:32:22Z"
+    "title": "fix klein lora loading.",
+    "updated_at": "2026-03-25T02:21:43Z"
   },
   {
-    "additions": 207,
-    "author": "sayakpaul",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? * Introduces `Flux2Transformer2DModelOutput` class as its outputs differ from the rest of the lot. * Test suite for Klein KV. * Docs.",
-    "changed_files": 5,
+    "additions": 30,
+    "author": "zhtmike",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? This PR expands the inputs of `QwenImage` UT from a single prompt to multiple prompts. Since `QwenImagePipeline` already supports multi-prompt inputs, we added corresponding test coverage here. - Tested with followi\u2026",
+    "changed_files": 4,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13264",
-    "created_at": "2026-03-13T03:55:38Z",
-    "deletions": 8,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13312",
+    "created_at": "2026-03-23T05:45:44Z",
+    "deletions": 14,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13264/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13264",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13312/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13312",
     "labels": [],
     "merged": true,
-    "number": 13264,
-    "review_comments_count": 0,
+    "number": 13312,
+    "review_comments_count": 11,
     "state": "closed",
-    "title": "[core] Flux2 klein kv followups",
-    "updated_at": "2026-03-13T04:35:11Z"
+    "title": "change QwenImageTransformer UT to batch inputs",
+    "updated_at": "2026-03-24T03:34:14Z"
   },
   {
-    "additions": 2,
-    "author": "vkuzo",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "Summary: Very brief writeup as I'm about to head out for the day: 1. we want to enable cuda graphs for qwen-image + nvfp4 at small batch sizes, because without cuda graphs are we bottlenecked on cpu ops 2. to make cuda graphs work, we need\u2026",
+    "additions": 6,
+    "author": "r266-tech",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## What does this PR do? Fixes #13298 When users pass a CUDA `torch.Generator` to diffusers pipelines, they may expect the same results as with a CPU generator given the same seed. However, PyTorch uses different random number generation a\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13263",
-    "created_at": "2026-03-12T20:15:42Z",
-    "deletions": 2,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13309",
+    "created_at": "2026-03-22T18:03:37Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13263/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13263",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13309/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13309",
     "labels": [],
     "merged": false,
-    "number": 13263,
+    "number": 13309,
     "review_comments_count": 0,
     "state": "open",
-    "title": "[not for land yet]: improve cuda graph support for Qwen-Image",
-    "updated_at": "2026-03-23T10:25:18Z"
-  },
-  {
-    "additions": 1386,
-    "author": "huemin-art",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Adds `Flux2KleinKVPipeline` \u2014 a KV-cached reference image conditioning pipeline for Flux2 Klein 9B. On the first denoising step, reference image tokens are included in the full transformer forward pass and their pos\u2026",
-    "changed_files": 6,
-    "cluster_id": null,
-    "cluster_ids": [],
-    "cluster_role": null,
-    "comments_count": 4,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13262",
-    "created_at": "2026-03-12T14:56:59Z",
-    "deletions": 22,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13262/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13262",
-    "labels": [],
-    "merged": true,
-    "number": 13262,
-    "review_comments_count": 10,
-    "state": "closed",
-    "title": "klein 9b kv",
-    "updated_at": "2026-03-13T22:39:07Z"
+    "title": "Add warning when using CUDA random generator (#13298)",
+    "updated_at": "2026-03-31T06:25:05Z"
   },
   {
-    "additions": 6,
-    "author": "s-zx",
+    "additions": 11,
+    "author": "agarwalprakhar2511",
     "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "## Summary Use safe dict access for alpha keys when converting non-diffusers Z-Image LoRA to diffusers format. Some LoRA formats use different alpha key naming (e.g. `layers.0.adaLN_modulation.0.alpha`), causing KeyError. ## Root Cause `st\u2026",
+    "body_excerpt": "## Summary Fixes silent weight corruption when loading models with `device_map=\"mps\"`. ### Root cause `load_model_dict_into_meta` unconditionally sets `non_blocking=True` for `set_module_tensor_to_device` (added for accelerate > 1.8.1). Wi\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13260",
-    "created_at": "2026-03-11T23:00:33Z",
-    "deletions": 2,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13308",
+    "created_at": "2026-03-22T09:21:13Z",
+    "deletions": 4,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13260/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13260",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13308/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13308",
     "labels": [],
     "merged": false,
-    "number": 13260,
+    "number": 13308,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "fix: use .get() for alpha keys in lora_conversion_utils.py",
-    "updated_at": "2026-03-12T16:06:04Z"
+    "state": "open",
+    "title": "fix: disable non-blocking tensor copies to MPS during model loading",
+    "updated_at": "2026-03-22T09:27:02Z"
   },
   {
-    "additions": 107,
-    "author": "yiyixuxu",
-    "author_association": "MEMBER",
-    "body_excerpt": null,
-    "changed_files": 4,
+    "additions": 12,
+    "author": "agarwalprakhar2511",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## Summary When `--with_prior_preservation` is enabled in the Flux2 dreambooth LoRA training scripts, the prompt embedding repeat logic double-counts the batch size, producing a shape mismatch against the latent tensor. **Root cause:** `co\u2026",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13259",
-    "created_at": "2026-03-11T22:34:21Z",
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13307",
+    "created_at": "2026-03-22T09:13:05Z",
     "deletions": 3,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13259/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13259",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13307/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13307",
     "labels": [],
-    "merged": true,
-    "number": 13259,
-    "review_comments_count": 9,
-    "state": "closed",
-    "title": "Add AGENTS.md",
-    "updated_at": "2026-03-14T18:35:17Z"
+    "merged": false,
+    "number": 13307,
+    "review_comments_count": 0,
+    "state": "open",
+    "title": "fix(dreambooth): batch size mismatch with --with_prior_preservation in flux2 scripts",
+    "updated_at": "2026-03-22T09:13:32Z"
   },
   {
-    "additions": 2,
-    "author": "alvarobartt",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? This PR adds the `PRXPipeline` in the `AUTO_TEXT2IMAGE_PIPELINES_MAPPING`, so that PRX models e.g. https://huggingface.co/Photoroom/prx-1024-t2i-beta, can be loaded with the `AutoPipelineForText2Image` as it's curre\u2026",
-    "changed_files": 1,
+    "additions": 380,
+    "author": "pthombre",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "--- What does this PR do? Adds a new documentation page for NeMo Automodel, NVIDIA's PyTorch DTensor-native training library for fine-tuning and pretraining diffusion models at scale. NeMo Automodel integrates directly with Diffusers \u2014 it\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13257",
-    "created_at": "2026-03-11T17:14:51Z",
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13306",
+    "created_at": "2026-03-22T00:51:33Z",
     "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13257/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13257",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13306/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13306",
     "labels": [],
     "merged": true,
-    "number": 13257,
-    "review_comments_count": 0,
+    "number": 13306,
+    "review_comments_count": 24,
     "state": "closed",
-    "title": "Add `PRXPipeline` in `AUTO_TEXT2IMAGE_PIPELINES_MAPPING`",
-    "updated_at": "2026-03-11T17:39:25Z"
+    "title": "[docs] Add NeMo Automodel training guide",
+    "updated_at": "2026-03-30T17:21:58Z"
   },
   {
-    "additions": 6,
-    "author": "sayakpaul",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Fix https://github.com/huggingface/diffusers/issues/13249. @AlpineVibrations could you test if this is working?",
+    "additions": 18,
+    "author": "s-zx",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## What does this PR do? Fixes #13281. When combining torchao quantization (e.g. `Float8WeightOnlyConfig`) with group offloading (`use_stream=True`), inference fails with a device mismatch: the quantized weight remains on CPU while the inp\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 5,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13255",
-    "created_at": "2026-03-11T11:10:12Z",
-    "deletions": 2,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13255/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13255",
-    "labels": [],
-    "merged": true,
-    "number": 13255,
-    "review_comments_count": 0,
-    "state": "closed",
-    "title": "[lora] fix z-image non-diffusers lora loading.",
-    "updated_at": "2026-03-13T01:29:01Z"
-  },
-  {
-    "additions": 171,
-    "author": "DN6",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
-    "changed_files": 2,
-    "cluster_id": null,
-    "cluster_ids": [],
-    "cluster_role": null,
     "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13254",
-    "created_at": "2026-03-11T10:25:06Z",
-    "deletions": 214,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13305",
+    "created_at": "2026-03-22T00:34:28Z",
+    "deletions": 9,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13254/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13254",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13305/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13305",
     "labels": [],
     "merged": false,
-    "number": 13254,
+    "number": 13305,
     "review_comments_count": 0,
     "state": "open",
-    "title": "[CI] Refactor LTX Transformer Tests",
-    "updated_at": "2026-03-11T10:25:06Z"
+    "title": "fix: skip pin_memory and non_blocking transfer for tensor subclasses in group offloading",
+    "updated_at": "2026-03-22T00:34:28Z"
   },
   {
-    "additions": 232,
-    "author": "DN6",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
+    "additions": 24,
+    "author": "s-zx",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## What does this PR do? Fixes `ImportError` when using `TorchAoConfig` with torchao >= 0.16.0. torchao 0.15.0 deprecated the snake_case quantization functions (`int4_weight_only`, `float8_weight_only`, etc.) with deprecation warnings, and\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13253",
-    "created_at": "2026-03-11T06:04:17Z",
-    "deletions": 92,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13304",
+    "created_at": "2026-03-22T00:04:23Z",
+    "deletions": 10,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13253/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13253",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13304/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13304",
     "labels": [],
     "merged": false,
-    "number": 13253,
+    "number": 13304,
     "review_comments_count": 0,
     "state": "open",
-    "title": "[CI] Refactor Z Image Transformer Tests ",
-    "updated_at": "2026-03-11T10:24:42Z"
+    "title": "fix: support torchao >= 0.16.0 by importing renamed CamelCase Config classes",
+    "updated_at": "2026-03-22T00:04:23Z"
   },
   {
-    "additions": 3,
-    "author": "kaixuanliu",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "@yiyixuxu and @asomoza, pls help review, thx! This PR fixes bug for following sample code: ``` import torch from diffusers import PaintByExamplePipeline from PIL import Image, ImageDraw # Load the pipeline pipe = PaintByExamplePipeline.fro\u2026",
+    "additions": 2,
+    "author": "s-zx",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## What does this PR do? Fixes dtype mismatch errors when running SD3 or FLUX models on CPU with `torch_dtype=torch.float16`. When the model's projection layers (Conv2d in `PatchEmbed`, Linear in `FluxTransformer2DModel`) are in float32 bu\u2026",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 6,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13252",
-    "created_at": "2026-03-11T01:52:38Z",
-    "deletions": 0,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13303",
+    "created_at": "2026-03-21T23:55:48Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13252/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13252",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13303/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13303",
     "labels": [],
-    "merged": true,
-    "number": 13252,
+    "merged": false,
+    "number": 13303,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "fix: 'PaintByExampleImageEncoder' object has no attribute 'all_tied_w\u2026",
-    "updated_at": "2026-04-07T21:22:09Z"
+    "state": "open",
+    "title": "fix: cast input to weight dtype in PatchEmbed and FluxTransformer2DModel to prevent dtype mismatch",
+    "updated_at": "2026-03-21T23:55:48Z"
   },
   {
-    "additions": 24,
-    "author": "miguelmartin75",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "- Moves cosmos to the \"video\" model section - Updates the main example - Minor updates to docstrings in cosmos transfer2.5 pipeline Review: - Docs: @stevhliu and @sayakpaul",
-    "changed_files": 4,
+    "additions": 7,
+    "author": "s-zx",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## Summary When `set_attention_backend()` is called directly on an individual attention submodule (e.g. `pipe.transformer.attn1.set_attention_backend(\"sage_hub\")`), hub-based kernels are never downloaded and the backend silently falls back\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13251",
-    "created_at": "2026-03-11T01:21:26Z",
-    "deletions": 18,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13302",
+    "created_at": "2026-03-21T23:33:44Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13251/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13251",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13302/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13302",
     "labels": [],
-    "merged": true,
-    "number": 13251,
+    "merged": false,
+    "number": 13302,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "Update Documentation for NVIDIA Cosmos",
-    "updated_at": "2026-03-11T16:14:57Z"
+    "state": "open",
+    "title": "fix(attention): download hub kernel in AttentionModuleMixin.set_attention_backend",
+    "updated_at": "2026-03-21T23:33:44Z"
   },
   {
-    "additions": 32,
-    "author": "s-zx",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "## Summary The `_convert_non_diffusers_z_image_lora_to_diffusers` function did not consume LoKr-format keys (`.lokr_w1`, `.lokr_w2`, `.alpha`) from external Z-image LoRA checkpoints (e.g. Kohya/LyCORIS), causing a `ValueError` when the sta\u2026",
-    "changed_files": 1,
+    "additions": 12,
+    "author": "adi776borate",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
+    "changed_files": 4,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 4,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13250",
-    "created_at": "2026-03-10T22:11:42Z",
-    "deletions": 0,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13299",
+    "created_at": "2026-03-21T09:54:54Z",
+    "deletions": 4,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13250/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13250",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/diffusers/pull/13299/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13299",
+    "labels": [
+      "size/S",
+      "pipelines"
+    ],
     "merged": false,
-    "number": 13250,
+    "number": 13299,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "fix: handle LoKr format keys in Z-image LoRA conversion",
-    "updated_at": "2026-03-23T23:57:09Z"
+    "state": "open",
+    "title": "Fix missing latents_bn_std dtype cast in VAE normalization",
+    "updated_at": "2026-04-18T06:15:56Z"
   },
   {
-    "additions": 10,
-    "author": "stevhliu",
+    "additions": 49,
+    "author": "sayakpaul",
     "author_association": "MEMBER",
-    "body_excerpt": "a few maintenance items: - move \"Guiders\" out of Modular Diffusers (because its not exclusive to it). i think after \"Schedulers\" is the most logical place for guiders as they're both swappable components and it has a much higher visibility\u2026",
-    "changed_files": 10,
+    "body_excerpt": "# What does this PR do? Use Claude for PR reviews. Some choices / comments: - A GitHub Actions workflow (using `anthropics/claude-code-action`) that auto-reviews PRs touching `src/diffusers/**` and responds to `@claude` mentions from folks\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13248",
-    "created_at": "2026-03-10T16:39:07Z",
-    "deletions": 148,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13297",
+    "created_at": "2026-03-20T06:43:17Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13248/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13248",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13297/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13297",
     "labels": [],
     "merged": true,
-    "number": 13248,
-    "review_comments_count": 2,
+    "number": 13297,
+    "review_comments_count": 14,
     "state": "closed",
-    "title": "[docs] updates",
-    "updated_at": "2026-03-16T20:25:01Z"
+    "title": "[ci] claude in ci.",
+    "updated_at": "2026-03-25T16:00:18Z"
   },
   {
-    "additions": 358,
-    "author": "sayakpaul",
+    "additions": 8,
+    "author": "DN6",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Implements a test suite for the conditional blocks for the following functionalities: * `get_workflow()` works as expected. * `select_blocks()` works as expected. * Automatic workflow selection works as expected. Th\u2026",
-    "changed_files": 3,
+    "body_excerpt": "# What does this PR do? Release tests were not updated to use the transformers `main` branch. This is resulting in unrelated failures on the release branch https://github.com/huggingface/diffusers/actions/runs/23330168017/job/67859952145#s\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13247",
-    "created_at": "2026-03-10T13:00:25Z",
-    "deletions": 118,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13296",
+    "created_at": "2026-03-20T05:46:04Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13247/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13247",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13296/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13296",
     "labels": [],
     "merged": true,
-    "number": 13247,
+    "number": 13296,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "[tests] Tests for conditional pipeline blocks",
-    "updated_at": "2026-03-26T03:18:19Z"
+    "title": "[CI] Update transformer version in release tests ",
+    "updated_at": "2026-03-20T06:10:06Z"
   },
   {
-    "additions": 51,
-    "author": "kvr06-ai",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Fixes a bug where `sigma_min` and `sigma_max` were computed from already-shifted sigmas in `__init__`, causing `set_timesteps()` to apply the shift formula twice. Fixes #13243 ## Root cause In `__init__`, the shift\u2026",
-    "changed_files": 2,
+    "additions": 2,
+    "author": "sayakpaul",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? The WanVACE LoRA test failures are the flakiest in nature. They have been making our CI red for a while now. This is another attempt getting rid of that: https://github.com/huggingface/diffusers/actions/runs/2330825\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13246",
-    "created_at": "2026-03-10T10:45:14Z",
-    "deletions": 2,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13294",
+    "created_at": "2026-03-20T04:01:35Z",
+    "deletions": 4,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13246/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13246",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13294/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13294",
     "labels": [],
-    "merged": false,
-    "number": 13246,
-    "review_comments_count": 0,
-    "state": "open",
-    "title": "Fix double-shifted sigma_min/sigma_max in FlowMatchEulerDiscreteScheduler",
-    "updated_at": "2026-03-11T04:13:52Z"
+    "merged": true,
+    "number": 13294,
+    "review_comments_count": 2,
+    "state": "closed",
+    "title": "[ci] hoping to fix is_flaky with wanvace.",
+    "updated_at": "2026-03-20T10:34:39Z"
   },
   {
-    "additions": 61,
+    "additions": 11,
     "author": "sayakpaul",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Add tests (two tests: loading from a modular repo and loading from a local path) to ensure `load_components()` always loads the expected components. Here, I took the liberty of defining `expected_components` as the\u2026",
+    "body_excerpt": "# What does this PR do? Fix https://github.com/huggingface/diffusers/actions/runs/23308255022/job/67788477405?pr=13290",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13245",
-    "created_at": "2026-03-10T10:42:21Z",
-    "deletions": 0,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13293",
+    "created_at": "2026-03-20T03:31:58Z",
+    "deletions": 10,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13245/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13245",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13293/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13293",
     "labels": [],
     "merged": true,
-    "number": 13245,
+    "number": 13293,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "[tests] test load_components in modular",
-    "updated_at": "2026-03-21T04:12:20Z"
+    "title": "[tests] fix audioldm2 tests.",
+    "updated_at": "2026-03-20T15:23:24Z"
   },
   {
-    "additions": 19,
-    "author": "kaixuanliu",
+    "additions": 174,
+    "author": "howardzhang-cv",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "@yiyixuxu @asomoza, pls help review, thx!",
-    "changed_files": 2,
+    "body_excerpt": "# What does this PR do? Remove the deprecated string-based quant_type path from TorchAoConfig, requiring AOBaseConfig instances instead. - TorchAoConfig.__init__ now only accepts AOBaseConfig subclass instances (e.g. Int8WeightOnlyConfig()\u2026",
+    "changed_files": 6,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 9,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13244",
-    "created_at": "2026-03-10T10:35:36Z",
-    "deletions": 15,
+    "comments_count": 12,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13291",
+    "created_at": "2026-03-19T20:31:22Z",
+    "deletions": 580,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13244/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13244",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13291/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13291",
     "labels": [],
     "merged": true,
-    "number": 13244,
-    "review_comments_count": 4,
+    "number": 13291,
+    "review_comments_count": 37,
     "state": "closed",
-    "title": "Fix IndexError in HunyuanVideo I2V pipeline",
-    "updated_at": "2026-04-08T01:25:21Z"
+    "title": "remove str option for quantization config in torchao",
+    "updated_at": "2026-03-27T22:12:52Z"
   },
   {
-    "additions": 28,
-    "author": "sayakpaul",
+    "additions": 31,
+    "author": "cbensimon",
     "author_association": "MEMBER",
     "body_excerpt": null,
-    "changed_files": 1,
-    "cluster_id": null,
-    "cluster_ids": [],
-    "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13242",
-    "created_at": "2026-03-10T06:33:46Z",
-    "deletions": 31,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13242/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13242",
-    "labels": [],
-    "merged": false,
-    "number": 13242,
-    "review_comments_count": 0,
-    "state": "open",
-    "title": "[tests] migrate test_hooks.py to pytest",
-    "updated_at": "2026-03-10T06:33:46Z"
-  },
-  {
-    "additions": 12,
-    "author": "Chase-Xuu",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "## Summary Fixes #12771 \u2014 `DPMSolverMultistepScheduler` crashes with `IndexError: index 21 is out of bounds for dimension 0 with size 21` when using `beta_schedule='squaredcos_cap_v2'` with `use_karras_sigmas=True`. ## Root Cause When usin\u2026",
-    "changed_files": 1,
+    "changed_files": 4,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13240",
-    "created_at": "2026-03-10T02:02:34Z",
-    "deletions": 4,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13290",
+    "created_at": "2026-03-19T17:34:08Z",
+    "deletions": 10,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13240/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13240",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13290/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13290",
     "labels": [],
-    "merged": false,
-    "number": 13240,
-    "review_comments_count": 0,
-    "state": "open",
-    "title": "fix: deduplicate timesteps in DPMSolverMultistep for squaredcos_cap_v2 + karras/lu sigmas",
-    "updated_at": "2026-03-10T02:02:34Z"
+    "merged": true,
+    "number": 13290,
+    "review_comments_count": 6,
+    "state": "closed",
+    "title": "[export] Add export-safe LRU cache helper",
+    "updated_at": "2026-03-23T12:40:21Z"
   },
   {
-    "additions": 11,
-    "author": "Chase-Xuu",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "## What does this PR do? Fixes #12771 \u2014 `DPMSolverMultistepScheduler` crashes with `IndexError: index N is out of bounds` when using `beta_schedule='squaredcos_cap_v2'` with `use_karras_sigmas=True` (or `use_lu_lambdas=True`). ### Root cau\u2026",
-    "changed_files": 1,
+    "additions": 172,
+    "author": "JingyaHuang",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? This PR adds AWS Neuron (Trainium/Inferentia) as an officially supported compute backend in Diffusers, on par with existing backends like CUDA, MPS, XPU, and MLU. **Changes** - import_utils.py \u2014 adds is_torch_neuron\u2026",
+    "changed_files": 13,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13239",
-    "created_at": "2026-03-10T01:56:07Z",
-    "deletions": 4,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13239/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13239",
-    "labels": [],
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13289",
+    "created_at": "2026-03-19T16:01:08Z",
+    "deletions": 24,
+    "draft": true,
+    "files_url": "https://github.com/huggingface/diffusers/pull/13289/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13289",
+    "labels": [
+      "lora",
+      "models",
+      "tests",
+      "size/M",
+      "utils",
+      "pipelines",
+      "examples"
+    ],
     "merged": false,
-    "number": 13239,
+    "number": 13289,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "fix: deduplicate timesteps in DPMSolverMultistepScheduler to prevent IndexError",
-    "updated_at": "2026-03-10T15:07:13Z"
+    "state": "open",
+    "title": "[Neuron] Add AWS Neuron (Trainium/Inferentia) as an officially supported device",
+    "updated_at": "2026-04-13T16:35:16Z"
   },
   {
-    "additions": 2,
-    "author": "Chase-Xuu",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "## Description In `_offload_to_memory()`, when using CUDA streams (`self.stream is not None`), module buffers from `self.modules` were not being restored to their CPU tensor copies during offload. This created an asymmetry with both `_buil\u2026",
+    "additions": 110,
+    "author": "DN6",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Adding a test to catch cases such as #13271 <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so ma\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13238",
-    "created_at": "2026-03-10T01:43:26Z",
-    "deletions": 0,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13287",
+    "created_at": "2026-03-19T07:44:01Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13238/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13238",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13287/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13287",
     "labels": [],
-    "merged": false,
-    "number": 13238,
+    "merged": true,
+    "number": 13287,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix: restore module buffers during stream-based group offload",
-    "updated_at": "2026-03-10T15:07:31Z"
+    "title": "[Modular] Test for catching dtype and device issues with AutoModel type hints",
+    "updated_at": "2026-03-20T05:06:03Z"
   },
-  {
-    "additions": 13,
-    "author": "eliemichel",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? It enables detection of FLUX.2 Klein models when loading transformers from single file, like in the following example: ```python import torch from diffusers import Flux2KleinPipeline, Flux2Transformer2DModel, GGUFQu\u2026",
-    "changed_files": 1,
+  {
+    "additions": 163,
+    "author": "howardzhang-cv",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? Adds low precision attention API from TorchAO to diffusers by updating TorchAoConfig with attn_backend option. Note: this will require torchao 0.17.0 ## Todo: - [ ] Need to add new necessary tests - [ ] Need to upda\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13237",
-    "created_at": "2026-03-09T22:34:15Z",
-    "deletions": 1,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13285",
+    "created_at": "2026-03-19T00:41:59Z",
+    "deletions": 8,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13237/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13237",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13285/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13285",
     "labels": [
-      "size/S",
-      "single-file"
+      "quantization",
+      "size/M"
     ],
     "merged": false,
-    "number": 13237,
+    "number": 13285,
     "review_comments_count": 0,
     "state": "open",
-    "title": "Add support for FLUX.2 Klein 9B and 4B in single_file_utils",
-    "updated_at": "2026-04-08T11:45:03Z"
+    "title": "Add low precision attention API from torchao to TorchAoConfig",
+    "updated_at": "2026-04-20T23:54:25Z"
   },
   {
     "additions": 1,
-    "author": "eliemichel",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? I believe that the error message was wrong, since the shapes mentionned did not match the test trigering the error. I think that there is also an ambiguity on what is meant by \"quantized\" shape: the value returned b\u2026",
+    "author": "sayakpaul",
+    "author_association": "MEMBER",
+    "body_excerpt": null,
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13236",
-    "created_at": "2026-03-09T21:49:18Z",
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13283",
+    "created_at": "2026-03-18T10:14:14Z",
     "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13236/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13236",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13283/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13283",
     "labels": [],
     "merged": false,
-    "number": 13236,
-    "review_comments_count": 0,
+    "number": 13283,
+    "review_comments_count": 1,
     "state": "open",
-    "title": "Fix error message in gguf_quantizer.py",
-    "updated_at": "2026-03-10T02:22:19Z"
+    "title": "Update attention_backends.md to update FA3 minimum support to Ampere",
+    "updated_at": "2026-03-18T18:01:41Z"
   },
   {
-    "additions": 216,
+    "additions": 0,
+    "author": "SHYuanBest",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "cc @sayakpaul related to: https://gist.github.com/sayakpaul/aba9d8925a24944f183d67f1ef4e009d",
+    "changed_files": 1,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13282",
+    "created_at": "2026-03-18T07:25:19Z",
+    "deletions": 2,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/diffusers/pull/13282/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13282",
+    "labels": [],
+    "merged": true,
+    "number": 13282,
+    "review_comments_count": 0,
+    "state": "closed",
+    "title": "[Helios] Remove lru_cache for better AoTI compatibility and cleaner code",
+    "updated_at": "2026-03-18T18:11:58Z"
+  },
+  {
+    "additions": 44,
     "author": "sayakpaul",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? - [x] Refactor MagCache tests. - [x] Include TaylorSeer in our model-level caching test mixin. - [ ] Consider removing caching-related mixins from `test_pipelines_common.py`.",
-    "changed_files": 4,
+    "body_excerpt": "# What does this PR do? Needs `kernels` to be installed from source for testing. Users might have to update `nvidia-cutlass-dsl` (or might even do a `--force-reinstall`). In this PR, we're just supporting the bare minimum inference support\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13235",
-    "created_at": "2026-03-09T13:57:58Z",
-    "deletions": 245,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13280",
+    "created_at": "2026-03-18T03:23:30Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13235/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13235",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13280/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13280",
     "labels": [],
-    "merged": false,
-    "number": 13235,
+    "merged": true,
+    "number": 13280,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "[tests] refactor caching tests.",
-    "updated_at": "2026-03-10T03:55:52Z"
+    "state": "closed",
+    "title": "[core] fa4 support.",
+    "updated_at": "2026-03-20T11:59:28Z"
   },
   {
-    "additions": 55,
+    "additions": 1,
+    "author": "zhtmike",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? Fix issue #13277. QwenImagePipeline cannot run with Ulysses SP together with batch prompt inputs. It is related to the mask is not correctly broadcasted. We need to broadcast the attention mask from [B, S] to [B, H,\u2026",
+    "changed_files": 2,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 5,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13278",
+    "created_at": "2026-03-17T08:25:19Z",
+    "deletions": 1,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/diffusers/pull/13278/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13278",
+    "labels": [],
+    "merged": true,
+    "number": 13278,
+    "review_comments_count": 18,
+    "state": "closed",
+    "title": "Fix the attention mask in ulysses SP for QwenImage",
+    "updated_at": "2026-03-24T09:33:56Z"
+  },
+  {
+    "additions": 105,
     "author": "sayakpaul",
     "author_association": "MEMBER",
-    "body_excerpt": null,
+    "body_excerpt": "# What does this PR do? Fix offloading when using TorchAO. This assumes that the underlying quantization tensor class implements pinning properly. But that's not something we can always guarantee. So, we need coordination with TorchAO. Exa\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13234",
-    "created_at": "2026-03-09T13:21:38Z",
-    "deletions": 70,
+    "comments_count": 7,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13276",
+    "created_at": "2026-03-17T05:16:05Z",
+    "deletions": 14,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13234/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13234",
-    "labels": [],
-    "merged": false,
-    "number": 13234,
-    "review_comments_count": 0,
-    "state": "open",
-    "title": "[tests] migrate group offloading tests to pytest",
-    "updated_at": "2026-03-09T13:21:38Z"
+    "files_url": "https://github.com/huggingface/diffusers/pull/13276/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13276",
+    "labels": [
+      "performance",
+      "roadmap"
+    ],
+    "merged": true,
+    "number": 13276,
+    "review_comments_count": 13,
+    "state": "closed",
+    "title": "[core] fix group offloading when using torchao",
+    "updated_at": "2026-04-17T12:13:10Z"
   },
   {
     "additions": 3,
-    "author": "DN6",
+    "author": "sayakpaul",
     "author_association": "MEMBER",
-    "body_excerpt": "Potential fix for [https://github.com/huggingface/diffusers/security/code-scanning/2147](https://github.com/huggingface/diffusers/security/code-scanning/2147) In general, the fix is to explicitly set a `permissions` block in the workflow (\u2026",
+    "body_excerpt": "# What does this PR do? Fixes https://github.com/huggingface/diffusers/actions/runs/23176322293/job/67339276234?pr=13270",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13233",
-    "created_at": "2026-03-09T11:39:34Z",
-    "deletions": 0,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13275",
+    "created_at": "2026-03-17T03:59:44Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13233/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13233",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13275/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13275",
     "labels": [],
     "merged": true,
-    "number": 13233,
+    "number": 13275,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "[CI] Add Workflow permissions to PR tests",
-    "updated_at": "2026-03-09T12:23:02Z"
+    "title": "[tests] fix llava kwargs in the hunyuan tests",
+    "updated_at": "2026-03-17T04:41:51Z"
   },
   {
-    "additions": 3589,
-    "author": "plugyawn",
+    "additions": 112,
+    "author": "gambletan",
     "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? This PR adds support for [Diffusion Transformers with Representation Autoencoders](https://rae-dit.github.io) in Diffusers. It implements the Stage-2 side of the RAE setup: - `RAEDiT2DModel` - `RAEDiTPipeline` - che\u2026",
-    "changed_files": 22,
+    "body_excerpt": "## Summary Fixes #13124 `log_validation()` in dreambooth training scripts runs pipeline inference using the same UNet/transformer that is being trained, without `torch.no_grad()`. With `--mixed_precision=\"fp16\"`, gradients are computed and\u2026",
+    "changed_files": 17,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 17,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13231",
-    "created_at": "2026-03-09T04:49:54Z",
-    "deletions": 4,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13273",
+    "created_at": "2026-03-16T14:13:58Z",
+    "deletions": 99,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13231/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13231",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13273/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13273",
     "labels": [],
     "merged": false,
-    "number": 13231,
-    "review_comments_count": 17,
+    "number": 13273,
+    "review_comments_count": 0,
     "state": "open",
-    "title": "Add RAE Diffusion Transformer inference/preliminary training pipelines",
-    "updated_at": "2026-04-12T18:02:36Z"
+    "title": "fix: wrap validation inference with torch.no_grad() in dreambooth examples",
+    "updated_at": "2026-03-16T14:13:58Z"
   },
   {
-    "additions": 3,
-    "author": "DN6",
-    "author_association": "MEMBER",
-    "body_excerpt": "Potential fix for [https://github.com/huggingface/diffusers/security/code-scanning/2150](https://github.com/huggingface/diffusers/security/code-scanning/2150) In general, this issue is fixed by adding an explicit `permissions` block to the\u2026",
-    "changed_files": 1,
+    "additions": 64,
+    "author": "gambletan",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## Summary Fixes #12913 - **Bug**: `image_seq_len = latents.shape[1]` takes the **channel dimension** (e.g. 16) instead of the spatial sequence length. Lumina2 latents have shape `(batch, channels, height, width)` and are NOT packed before\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13230",
-    "created_at": "2026-03-09T04:38:14Z",
-    "deletions": 0,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13272",
+    "created_at": "2026-03-16T14:10:56Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13230/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13230",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13272/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13272",
     "labels": [],
-    "merged": true,
-    "number": 13230,
+    "merged": false,
+    "number": 13272,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "[CI] Potential fix for code scanning alert no. 2150: Workflow does not contain permissions",
-    "updated_at": "2026-03-10T03:37:16Z"
+    "state": "open",
+    "title": "fix: compute image_seq_len from spatial dims in Lumina2 pipeline",
+    "updated_at": "2026-03-16T14:10:56Z"
   },
   {
-    "additions": 115,
-    "author": "lawrence-cj",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "This PR adds LTX-VAE support for SANA-Video. Cc: @dg845 @sayakpaul GPU memory needed: 47GB for LTX refiner ### SANA-Video with LTX2-Refiner: ```python \"\"\"Sana Video + LTX2 Refiner: Stage 1 generate latent \u2192 Stage 2 refine (3 steps).\"\"\" imp\u2026",
-    "changed_files": 3,
+    "additions": 3,
+    "author": "DN6",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Oversight on my part when adding #13168. We avoid passing `torch_dtype` to non `nn.Modules` such as Processors and Tokenizers because these objects do not accept that anymore and instead store it in init_kwargs whic\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 6,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13229",
-    "created_at": "2026-03-09T04:16:06Z",
-    "deletions": 42,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13271",
+    "created_at": "2026-03-16T11:44:49Z",
+    "deletions": 9,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13229/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13229",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13271/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13271",
     "labels": [],
     "merged": true,
-    "number": 13229,
-    "review_comments_count": 4,
+    "number": 13271,
+    "review_comments_count": 1,
     "state": "closed",
-    "title": "add ltx2 vae in sana-video;",
-    "updated_at": "2026-03-18T04:09:52Z"
+    "title": "[Modular] Fix dtype assignment when type hint is AutoModel",
+    "updated_at": "2026-03-17T04:17:53Z"
   },
   {
-    "additions": 4,
-    "author": "DefTruth",
+    "additions": 26,
+    "author": "sywangyi",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "Allow pass cpu generator for helios. w/o this fix, helios distilled pipeline will raise an device mismatch error if we use cpu generator. we should use the device from generator for torch.randn. ```bash File \"/workspace/deftruth/dev/minico\u2026",
+    "body_excerpt": "fix tests/models/transformers/test_models_transformer_flux.py::TestFluxTransformerContextParallel::test_context_parallel_inference[ulysses] tests/models/transformers/test_models_transformer_flux.py::TestFluxTransformerContextParallel::test\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13228",
-    "created_at": "2026-03-09T02:40:17Z",
-    "deletions": 1,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13270",
+    "created_at": "2026-03-16T07:47:18Z",
+    "deletions": 7,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13228/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13228",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13270/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13270",
     "labels": [],
     "merged": true,
-    "number": 13228,
-    "review_comments_count": 0,
+    "number": 13270,
+    "review_comments_count": 1,
     "state": "closed",
-    "title": "fix: allow pass cpu generator for helios",
-    "updated_at": "2026-03-10T02:46:04Z"
+    "title": "fix parallelism case failure in xpu",
+    "updated_at": "2026-03-17T03:22:16Z"
   },
   {
-    "additions": 2663,
-    "author": "kashif",
+    "additions": 728,
+    "author": "yiyixuxu",
     "author_association": "MEMBER",
-    "body_excerpt": "Add support for LLaDA2/LLaDA2.1 discrete diffusion text generation: - BlockRefinementPipeline: block-wise iterative refinement with confidence-based token commitment, supporting editing threshold for LLaDA2.1 models - LLaDA2Pipeline: conve\u2026",
-    "changed_files": 20,
+    "body_excerpt": null,
+    "changed_files": 9,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 4,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13226",
-    "created_at": "2026-03-08T17:44:13Z",
-    "deletions": 1,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13269",
+    "created_at": "2026-03-15T20:47:20Z",
+    "deletions": 55,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13226/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13226",
-    "labels": [
-      "roadmap"
-    ],
+    "files_url": "https://github.com/huggingface/diffusers/pull/13269/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13269",
+    "labels": [],
     "merged": true,
-    "number": 13226,
-    "review_comments_count": 50,
+    "number": 13269,
+    "review_comments_count": 31,
     "state": "closed",
-    "title": "[Discrete Diffusion] Add LLaDA2 pipeline",
-    "updated_at": "2026-03-25T11:15:43Z"
+    "title": "[agents]support skills",
+    "updated_at": "2026-03-20T04:07:46Z"
   },
   {
-    "additions": 34,
-    "author": "SHYuanBest",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "cc @dg845 @sayakpaul @yiyixuxu ### Helios-Distilled ``` # CUDA_VISIBLE_DEVICES=0,1,2,3 torchrun --master_port 12311 --nproc_per_node 2 infer_distilled.py import os import argparse import torch import torch.distributed as dist from diffuser\u2026",
-    "changed_files": 2,
+    "additions": 217,
+    "author": "aryabyte21",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## Summary Adds support for the [DiffSynth-Studio blockwise ControlNet](https://github.com/modelscope/DiffSynth-Studio) architecture to the existing `QwenImageControlNetModel`, addressing [the maintainer's feedback on PR #12317](https://gi\u2026",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13223",
-    "created_at": "2026-03-08T07:57:07Z",
-    "deletions": 9,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13268",
+    "created_at": "2026-03-15T05:17:57Z",
+    "deletions": 7,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13223/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13223",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13268/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13268",
     "labels": [],
-    "merged": true,
-    "number": 13223,
+    "merged": false,
+    "number": 13268,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "Fix Helios Context Parallelism",
-    "updated_at": "2026-03-08T22:27:10Z"
+    "state": "open",
+    "title": "Add DiffSynth blockwise ControlNet support to QwenImageControlNetModel",
+    "updated_at": "2026-03-15T05:17:57Z"
   },
   {
-    "additions": 3,
-    "author": "SHYuanBest",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "cc @dg845 @sayakpaul @yiyixuxu",
+    "additions": 1,
+    "author": "yunjoochoi",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? One-line fix \u2014 missing self.gradient_checkpointing = False init, consistent with all other ControlNet implementations <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, you\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13222",
-    "created_at": "2026-03-08T07:44:12Z",
-    "deletions": 4,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13267",
+    "created_at": "2026-03-15T03:43:58Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13222/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13222",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13267/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13267",
     "labels": [],
-    "merged": true,
-    "number": 13222,
+    "merged": false,
+    "number": 13267,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "Optimize Helios docs",
-    "updated_at": "2026-03-08T23:54:16Z"
+    "state": "open",
+    "title": "Fix missing gradient_checkpointing init in ZImageControlNetModel",
+    "updated_at": "2026-03-16T21:55:13Z"
   },
   {
-    "additions": 26,
-    "author": "dg845",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? This PR refactors `AttentionProcessorSkipHook` to support attention processors with a `_skip_attn_scores` flag, which is meant to allow attention processors to specify custom [spatio-temporal guidance (STG)](https:/\u2026",
-    "changed_files": 1,
+    "additions": 71,
+    "author": "abhixb",
+    "author_association": "NONE",
+    "body_excerpt": "Fixed issue #13177 ## The problem If you pass a tensor or numpy array as the image to the Klein pipeline, it errors even though the docstring explicitly says these types are supported. Three things cause this: - The image processor's valid\u2026",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13220",
-    "created_at": "2026-03-07T00:36:14Z",
-    "deletions": 3,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13265",
+    "created_at": "2026-03-14T13:12:44Z",
+    "deletions": 8,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13220/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13220",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13265/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13265",
     "labels": [],
     "merged": false,
-    "number": 13220,
-    "review_comments_count": 7,
-    "state": "open",
-    "title": "Refactor `AttentionProcessorSkipHook` to Support Custom STG Logic",
-    "updated_at": "2026-03-08T02:01:06Z"
+    "number": 13265,
+    "review_comments_count": 0,
+    "state": "closed",
+    "title": " add tensor and numpy array support for image input in Flux2KleinPipeline",
+    "updated_at": "2026-04-03T09:32:22Z"
   },
   {
-    "additions": 2,
-    "author": "tcaimm",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? This PR fixes wrapped transformer config access in the Flux2 Klein DreamBooth LoRA training scripts. When the transformer is wrapped by Accelerate, DDP, or FSDP, accessing `transformer.config.guidance_embeds` direct\u2026",
-    "changed_files": 2,
+    "additions": 207,
+    "author": "sayakpaul",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? * Introduces `Flux2Transformer2DModelOutput` class as its outputs differ from the rest of the lot. * Test suite for Klein KV. * Docs.",
+    "changed_files": 5,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13219",
-    "created_at": "2026-03-06T12:53:58Z",
-    "deletions": 2,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13264",
+    "created_at": "2026-03-13T03:55:38Z",
+    "deletions": 8,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13219/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13219",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13264/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13264",
     "labels": [],
     "merged": true,
-    "number": 13219,
+    "number": 13264,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix wrapped transformer config access in Flux2 Klein training",
-    "updated_at": "2026-03-06T14:17:51Z"
+    "title": "[core] Flux2 klein kv followups",
+    "updated_at": "2026-03-13T04:35:11Z"
   },
   {
     "additions": 2,
-    "author": "kaixuanliu",
+    "author": "vkuzo",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "In helios pipeline, the latents are always in FP32 format: [L798](https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/helios/pipeline_helios.py#L798), hence we should skip the test_save_load_float16 test case, rather\u2026",
+    "body_excerpt": "Summary: Very brief writeup as I'm about to head out for the day: 1. we want to enable cuda graphs for qwen-image + nvfp4 at small batch sizes, because without cuda graphs are we bottlenecked on cpu ops 2. to make cuda graphs work, we need\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 4,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13218",
-    "created_at": "2026-03-06T10:49:10Z",
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13263",
+    "created_at": "2026-03-12T20:15:42Z",
     "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13218/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13218",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13263/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13263",
     "labels": [],
-    "merged": true,
-    "number": 13218,
+    "merged": false,
+    "number": 13263,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "skip invalid test case for helios pipeline",
-    "updated_at": "2026-03-18T06:58:36Z"
+    "state": "open",
+    "title": "[not for land yet]: improve cuda graph support for Qwen-Image",
+    "updated_at": "2026-03-23T10:25:18Z"
   },
   {
-    "additions": 2494,
-    "author": "dg845",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? This PR adds support for LTX-2.3 ([official code](https://github.com/Lightricks/LTX-2), [model weights](https://huggingface.co/Lightricks/LTX-2.3)), a new model in the LTX-2.X family of audio-video models. LTX-2.3 h\u2026",
-    "changed_files": 13,
+    "additions": 1386,
+    "author": "huemin-art",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? Adds `Flux2KleinKVPipeline` \u2014 a KV-cached reference image conditioning pipeline for Flux2 Klein 9B. On the first denoising step, reference image tokens are included in the full transformer forward pass and their pos\u2026",
+    "changed_files": 6,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 10,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13217",
-    "created_at": "2026-03-06T04:10:04Z",
-    "deletions": 566,
+    "comments_count": 4,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13262",
+    "created_at": "2026-03-12T14:56:59Z",
+    "deletions": 22,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13217/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13217",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13262/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13262",
     "labels": [],
     "merged": true,
-    "number": 13217,
-    "review_comments_count": 52,
+    "number": 13262,
+    "review_comments_count": 10,
     "state": "closed",
-    "title": "Add Support for LTX-2.3 Models",
-    "updated_at": "2026-03-19T22:13:11Z"
+    "title": "klein 9b kv",
+    "updated_at": "2026-03-13T22:39:07Z"
+  },
+  {
+    "additions": 6,
+    "author": "s-zx",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## Summary Use safe dict access for alpha keys when converting non-diffusers Z-Image LoRA to diffusers format. Some LoRA formats use different alpha key naming (e.g. `layers.0.adaLN_modulation.0.alpha`), causing KeyError. ## Root Cause `st\u2026",
+    "changed_files": 1,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13260",
+    "created_at": "2026-03-11T23:00:33Z",
+    "deletions": 2,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/diffusers/pull/13260/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13260",
+    "labels": [],
+    "merged": false,
+    "number": 13260,
+    "review_comments_count": 0,
+    "state": "closed",
+    "title": "fix: use .get() for alpha keys in lora_conversion_utils.py",
+    "updated_at": "2026-03-12T16:06:04Z"
   },
   {
-    "additions": 4441,
+    "additions": 107,
     "author": "yiyixuxu",
     "author_association": "MEMBER",
-    "body_excerpt": "```py import torch from diffusers import ModularPipeline, ClassifierFreeGuidance, ClassifierFreeZeroStarGuidance from diffusers.utils import export_to_video, load_image, load_video import gc branch = \"\" ############################# # heli\u2026",
-    "changed_files": 15,
+    "body_excerpt": null,
+    "changed_files": 4,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 4,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13216",
-    "created_at": "2026-03-05T22:24:45Z",
-    "deletions": 0,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13259",
+    "created_at": "2026-03-11T22:34:21Z",
+    "deletions": 3,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13216/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13216",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13259/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13259",
     "labels": [],
     "merged": true,
-    "number": 13216,
-    "review_comments_count": 1,
+    "number": 13259,
+    "review_comments_count": 9,
     "state": "closed",
-    "title": "[modular] helios",
-    "updated_at": "2026-03-09T20:38:00Z"
+    "title": "Add AGENTS.md",
+    "updated_at": "2026-03-14T18:35:17Z"
   },
   {
-    "additions": 57,
-    "author": "sayakpaul",
+    "additions": 2,
+    "author": "alvarobartt",
     "author_association": "MEMBER",
-    "body_excerpt": null,
-    "changed_files": 57,
+    "body_excerpt": "# What does this PR do? This PR adds the `PRXPipeline` in the `AUTO_TEXT2IMAGE_PIPELINES_MAPPING`, so that PRX models e.g. https://huggingface.co/Photoroom/prx-1024-t2i-beta, can be loaded with the `AutoPipelineForText2Image` as it's curre\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13215",
-    "created_at": "2026-03-05T15:50:25Z",
-    "deletions": 57,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13257",
+    "created_at": "2026-03-11T17:14:51Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13215/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13215",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13257/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13257",
     "labels": [],
     "merged": true,
-    "number": 13215,
+    "number": 13257,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "post release 0.37.0",
-    "updated_at": "2026-03-07T05:33:51Z"
+    "title": "Add `PRXPipeline` in `AUTO_TEXT2IMAGE_PIPELINES_MAPPING`",
+    "updated_at": "2026-03-11T17:39:25Z"
   },
   {
-    "additions": 2,
-    "author": "SHYuanBest",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "cc @dg845 @sayakpaul @yiyixuxu",
+    "additions": 6,
+    "author": "sayakpaul",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Fix https://github.com/huggingface/diffusers/issues/13249. @AlpineVibrations could you test if this is working?",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13214",
-    "created_at": "2026-03-05T14:15:09Z",
-    "deletions": 0,
+    "comments_count": 5,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13255",
+    "created_at": "2026-03-11T11:10:12Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13214/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13214",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13255/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13255",
     "labels": [],
     "merged": true,
-    "number": 13214,
+    "number": 13255,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Convert tensors to float in Helios's optimized_scale function",
-    "updated_at": "2026-03-06T03:08:45Z"
+    "title": "[lora] fix z-image non-diffusers lora loading.",
+    "updated_at": "2026-03-13T01:29:01Z"
   },
   {
-    "additions": 4,
-    "author": "SHYuanBest",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "cc @dg845 @sayakpaul",
-    "changed_files": 4,
+    "additions": 171,
+    "author": "DN6",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13213",
-    "created_at": "2026-03-05T13:15:11Z",
-    "deletions": 4,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13254",
+    "created_at": "2026-03-11T10:25:06Z",
+    "deletions": 214,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13213/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13213",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13254/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13254",
     "labels": [],
-    "merged": true,
-    "number": 13213,
+    "merged": false,
+    "number": 13254,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "Fix Helios paper link in documentation",
-    "updated_at": "2026-03-05T13:28:14Z"
+    "state": "open",
+    "title": "[CI] Refactor LTX Transformer Tests",
+    "updated_at": "2026-03-11T10:25:06Z"
   },
   {
-    "additions": 5,
-    "author": "asomoza",
+    "additions": 232,
+    "author": "DN6",
     "author_association": "MEMBER",
-    "body_excerpt": "Just an extension of #13183 to the other Z-Image pipleines @yiyixuxu @sayakpaul",
-    "changed_files": 5,
+    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13212",
-    "created_at": "2026-03-04T16:49:24Z",
-    "deletions": 5,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13253",
+    "created_at": "2026-03-11T06:04:17Z",
+    "deletions": 92,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13212/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13212",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13253/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13253",
     "labels": [],
-    "merged": true,
-    "number": 13212,
+    "merged": false,
+    "number": 13253,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "[Z-Image] Fix more `do_classifier_free_guidance` thresholds",
-    "updated_at": "2026-03-04T20:12:03Z"
+    "state": "open",
+    "title": "[CI] Refactor Z Image Transformer Tests ",
+    "updated_at": "2026-03-11T10:24:42Z"
   },
   {
-    "additions": 135,
-    "author": "SHYuanBest",
+    "additions": 3,
+    "author": "kaixuanliu",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "Related to https://github.com/huggingface/diffusers/pull/13208 and https://github.com/huggingface/diffusers/pull/13210 cc @yiyixuxu cc @sayakpaul cc @dg845",
+    "body_excerpt": "@yiyixuxu and @asomoza, pls help review, thx! This PR fixes bug for following sample code: ``` import torch from diffusers import PaintByExamplePipeline from PIL import Image, ImageDraw # Load the pipeline pipe = PaintByExamplePipeline.fro\u2026",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13211",
-    "created_at": "2026-03-04T13:22:26Z",
+    "comments_count": 6,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13252",
+    "created_at": "2026-03-11T01:52:38Z",
     "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13211/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13211",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13252/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13252",
     "labels": [],
     "merged": true,
-    "number": 13211,
-    "review_comments_count": 5,
+    "number": 13252,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix group-offloading bug",
-    "updated_at": "2026-03-04T15:04:14Z"
+    "title": "fix: 'PaintByExampleImageEncoder' object has no attribute 'all_tied_w\u2026",
+    "updated_at": "2026-04-07T21:22:09Z"
   },
   {
-    "additions": 23,
-    "author": "SHYuanBest",
+    "additions": 24,
+    "author": "miguelmartin75",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "Related to https://github.com/huggingface/diffusers/pull/13208 cc @yiyixuxu cc @sayakpaul cc @dg845",
-    "changed_files": 3,
+    "body_excerpt": "- Moves cosmos to the \"video\" model section - Updates the main example - Minor updates to docstrings in cosmos transfer2.5 pipeline Review: - Docs: @stevhliu and @sayakpaul",
+    "changed_files": 4,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13210",
-    "created_at": "2026-03-04T13:13:33Z",
-    "deletions": 17,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13251",
+    "created_at": "2026-03-11T01:21:26Z",
+    "deletions": 18,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13210/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13210",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13251/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13251",
     "labels": [],
     "merged": true,
-    "number": 13210,
+    "number": 13251,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix Helios related (optimize docs & remove redudant)",
-    "updated_at": "2026-03-04T15:14:44Z"
+    "title": "Update Documentation for NVIDIA Cosmos",
+    "updated_at": "2026-03-11T16:14:57Z"
   },
   {
-    "additions": 78,
-    "author": "sayakpaul",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Fixes https://github.com/huggingface/diffusers/issues/13203 Additionally, fixes how \"alpha\"s are handled in the diffusers format path.",
+    "additions": 32,
+    "author": "s-zx",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## Summary The `_convert_non_diffusers_z_image_lora_to_diffusers` function did not consume LoKr-format keys (`.lokr_w1`, `.lokr_w2`, `.alpha`) from external Z-image LoRA checkpoints (e.g. Kohya/LyCORIS), causing a `ValueError` when the sta\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 7,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13209",
-    "created_at": "2026-03-04T10:58:07Z",
-    "deletions": 16,
+    "comments_count": 4,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13250",
+    "created_at": "2026-03-10T22:11:42Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13209/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13209",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13250/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13250",
     "labels": [],
-    "merged": true,
-    "number": 13209,
-    "review_comments_count": 2,
+    "merged": false,
+    "number": 13250,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "[lora] fix zimage lora conversion to support for more lora.",
-    "updated_at": "2026-03-05T02:54:46Z"
+    "title": "fix: handle LoKr format keys in Z-image LoRA conversion",
+    "updated_at": "2026-03-23T23:57:09Z"
   },
   {
-    "additions": 5655,
-    "author": "dg845",
+    "additions": 10,
+    "author": "stevhliu",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? This PR adds pipelines for the Helios-14B video generation model ([project page](https://pku-yuangroup.github.io/Helios-Page/), [official code](https://github.com/PKU-YuanGroup/Helios), [model weights](https://huggi\u2026",
-    "changed_files": 33,
+    "body_excerpt": "a few maintenance items: - move \"Guiders\" out of Modular Diffusers (because its not exclusive to it). i think after \"Schedulers\" is the most logical place for guiders as they're both swappable components and it has a much higher visibility\u2026",
+    "changed_files": 10,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13208",
-    "created_at": "2026-03-04T10:49:11Z",
-    "deletions": 2,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13248",
+    "created_at": "2026-03-10T16:39:07Z",
+    "deletions": 148,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13208/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13208",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13248/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13248",
     "labels": [],
     "merged": true,
-    "number": 13208,
-    "review_comments_count": 0,
+    "number": 13248,
+    "review_comments_count": 2,
     "state": "closed",
-    "title": "Add Helios-14B Video Generation Pipelines",
-    "updated_at": "2026-03-04T16:01:50Z"
+    "title": "[docs] updates",
+    "updated_at": "2026-03-16T20:25:01Z"
   },
   {
-    "additions": 34,
-    "author": "jiqing-feng",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "Currently, every process encodes the entire dataset through the VAE during preprocessing. This is redundant in multi-GPU training \u2014 each process does the same full encoding work independently. This PR shards the VAE encoding across process\u2026",
-    "changed_files": 1,
+    "additions": 358,
+    "author": "sayakpaul",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Implements a test suite for the conditional blocks for the following functionalities: * `get_workflow()` works as expected. * `select_blocks()` works as expected. * Automatic workflow selection works as expected. Th\u2026",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13207",
-    "created_at": "2026-03-04T08:02:24Z",
-    "deletions": 7,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13247",
+    "created_at": "2026-03-10T13:00:25Z",
+    "deletions": 118,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13207/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13207",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13247/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13247",
     "labels": [],
     "merged": true,
-    "number": 13207,
+    "number": 13247,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "cogvideo example: Distribute VAE video encoding across processes in CogVideoX LoRA training",
-    "updated_at": "2026-03-04T09:39:05Z"
+    "title": "[tests] Tests for conditional pipeline blocks",
+    "updated_at": "2026-03-26T03:18:19Z"
   },
   {
-    "additions": 61,
-    "author": "kirillsst",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "## Summary - Add `scheduler.set_begin_index(0)` before the denoising loop in all flow-matching pipelines that were missing it - For img2img pipelines with strength-based schedule truncation (`wan_video2video`, `ltx_condition`), set the beg\u2026",
-    "changed_files": 56,
+    "additions": 51,
+    "author": "kvr06-ai",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? Fixes a bug where `sigma_min` and `sigma_max` were computed from already-shifted sigmas in `__init__`, causing `set_timesteps()` to apply the shift formula twice. Fixes #13243 ## Root cause In `__init__`, the shift\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13206",
-    "created_at": "2026-03-03T13:30:17Z",
-    "deletions": 0,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13246",
+    "created_at": "2026-03-10T10:45:14Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13206/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13206",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13246/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13246",
     "labels": [],
     "merged": false,
-    "number": 13206,
+    "number": 13246,
     "review_comments_count": 0,
     "state": "open",
-    "title": "Set `begin_index` for flow-matching pipelines to avoid slow `index_for_timestep`",
-    "updated_at": "2026-03-03T13:30:17Z"
+    "title": "Fix double-shifted sigma_min/sigma_max in FlowMatchEulerDiscreteScheduler",
+    "updated_at": "2026-03-11T04:13:52Z"
   },
   {
-    "additions": 1,
-    "author": "teith",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Fixes invalid type annotation for the `image` parameter in `Flux2Pipeline.__call__`. `list[PIL.Image.Image, PIL.Image.Image]` passes two type arguments to `list`, which only accepts one. Fixed to `PIL.Image.Image |\u2026",
+    "additions": 61,
+    "author": "sayakpaul",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Add tests (two tests: loading from a modular repo and loading from a local path) to ensure `load_components()` always loads the expected components. Here, I took the liberty of defining `expected_components` as the\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13205",
-    "created_at": "2026-03-03T01:07:16Z",
-    "deletions": 1,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13245",
+    "created_at": "2026-03-10T10:42:21Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13205/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13205",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13245/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13245",
     "labels": [],
     "merged": true,
-    "number": 13205,
+    "number": 13245,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix: correct invalid type annotation for `image` in `Flux2Pipeline.__call__`",
-    "updated_at": "2026-03-13T18:56:39Z"
+    "title": "[tests] test load_components in modular",
+    "updated_at": "2026-03-21T04:12:20Z"
   },
   {
-    "additions": 239,
-    "author": "yiyixuxu",
-    "author_association": "MEMBER",
-    "body_excerpt": "based on this discussion https://github.com/huggingface/blog/pull/3278#discussion_r2873972272",
-    "changed_files": 3,
+    "additions": 19,
+    "author": "kaixuanliu",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "@yiyixuxu @asomoza, pls help review, thx!",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13204",
-    "created_at": "2026-03-02T23:03:06Z",
-    "deletions": 8,
+    "comments_count": 9,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13244",
+    "created_at": "2026-03-10T10:35:36Z",
+    "deletions": 15,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13204/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13204",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13244/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13244",
     "labels": [],
     "merged": true,
-    "number": 13204,
-    "review_comments_count": 10,
+    "number": 13244,
+    "review_comments_count": 4,
     "state": "closed",
-    "title": "[modular] not pass trust_remote_code to external repos ",
-    "updated_at": "2026-03-03T12:36:37Z"
+    "title": "Fix IndexError in HunyuanVideo I2V pipeline",
+    "updated_at": "2026-04-08T01:25:21Z"
   },
   {
-    "additions": 0,
-    "author": "DN6",
+    "additions": 28,
+    "author": "sayakpaul",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Accidentally committed a bunch of extra Claude files. Removing them here. <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes\u2026",
-    "changed_files": 17,
+    "body_excerpt": null,
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13202",
-    "created_at": "2026-03-02T19:04:38Z",
-    "deletions": 1907,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13242",
+    "created_at": "2026-03-10T06:33:46Z",
+    "deletions": 31,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13202/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13202",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13242/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13242",
     "labels": [],
-    "merged": true,
-    "number": 13202,
+    "merged": false,
+    "number": 13242,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "Clean up accidental files",
-    "updated_at": "2026-03-02T19:05:59Z"
+    "state": "open",
+    "title": "[tests] migrate test_hooks.py to pytest",
+    "updated_at": "2026-03-10T06:33:46Z"
   },
   {
-    "additions": 27,
-    "author": "DN6",
-    "author_association": "MEMBER",
-    "body_excerpt": "Reverts huggingface/diffusers#13168",
-    "changed_files": 20,
+    "additions": 12,
+    "author": "Chase-Xuu",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## Summary Fixes #12771 \u2014 `DPMSolverMultistepScheduler` crashes with `IndexError: index 21 is out of bounds for dimension 0 with size 21` when using `beta_schedule='squaredcos_cap_v2'` with `use_karras_sigmas=True`. ## Root Cause When usin\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13201",
-    "created_at": "2026-03-02T18:35:59Z",
-    "deletions": 2112,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13240",
+    "created_at": "2026-03-10T02:02:34Z",
+    "deletions": 4,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13201/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13201",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13240/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13240",
     "labels": [],
     "merged": false,
-    "number": 13201,
+    "number": 13240,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "Revert \"[Modular] Save Modular Pipeline weights to Hub\"",
-    "updated_at": "2026-03-02T18:53:47Z"
+    "state": "open",
+    "title": "fix: deduplicate timesteps in DPMSolverMultistep for squaredcos_cap_v2 + karras/lu sigmas",
+    "updated_at": "2026-03-10T02:02:34Z"
   },
   {
-    "additions": 4,
-    "author": "christopher5106",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "If you don't see any problem having Lora support for Controlnet pipelines for Z-Image, here is the PR. @sayakpaul",
-    "changed_files": 2,
+    "additions": 11,
+    "author": "Chase-Xuu",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## What does this PR do? Fixes #12771 \u2014 `DPMSolverMultistepScheduler` crashes with `IndexError: index N is out of bounds` when using `beta_schedule='squaredcos_cap_v2'` with `use_karras_sigmas=True` (or `use_lu_lambdas=True`). ### Root cau\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13200",
-    "created_at": "2026-03-02T14:51:52Z",
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13239",
+    "created_at": "2026-03-10T01:56:07Z",
     "deletions": 4,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13200/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13200",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13239/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13239",
     "labels": [],
-    "merged": true,
-    "number": 13200,
+    "merged": false,
+    "number": 13239,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "adding lora support to z-image controlnet pipelines",
-    "updated_at": "2026-03-05T18:11:07Z"
+    "title": "fix: deduplicate timesteps in DPMSolverMultistepScheduler to prevent IndexError",
+    "updated_at": "2026-03-10T15:07:13Z"
   },
   {
-    "additions": 0,
-    "author": "sayakpaul",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Remove DB utils from the benchmarking workflow. Internal thread: https://huggingface.slack.com/archives/CTKK32GE8/p1772345661280549 Cc: @McPatate",
-    "changed_files": 2,
+    "additions": 2,
+    "author": "Chase-Xuu",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## Description In `_offload_to_memory()`, when using CUDA streams (`self.stream is not None`), module buffers from `self.modules` were not being restored to their CPU tensor copies during offload. This created an asymmetry with both `_buil\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13199",
-    "created_at": "2026-03-02T10:46:19Z",
-    "deletions": 180,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13238",
+    "created_at": "2026-03-10T01:43:26Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13199/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13199",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13238/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13238",
     "labels": [],
-    "merged": true,
-    "number": 13199,
+    "merged": false,
+    "number": 13238,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "remove db utils from benchmarking",
-    "updated_at": "2026-03-02T11:15:33Z"
+    "title": "fix: restore module buffers during stream-based group offload",
+    "updated_at": "2026-03-10T15:07:31Z"
   },
   {
-    "additions": 43,
-    "author": "delmalih",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? This PR improves the docstrings and type hints in `src/diffusers/schedulers/scheduling_ipndm.py` to follow the project conventions as described in #9567. ## Context This is part of a broader effort to improve docume\u2026",
+    "additions": 13,
+    "author": "eliemichel",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? It enables detection of FLUX.2 Klein models when loading transformers from single file, like in the following example: ```python import torch from diffusers import Flux2KleinPipeline, Flux2Transformer2DModel, GGUFQu\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13198",
-    "created_at": "2026-02-28T10:30:41Z",
-    "deletions": 8,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13237",
+    "created_at": "2026-03-09T22:34:15Z",
+    "deletions": 1,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/diffusers/pull/13237/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13237",
+    "labels": [
+      "size/S",
+      "single-file"
+    ],
+    "merged": false,
+    "number": 13237,
+    "review_comments_count": 0,
+    "state": "open",
+    "title": "Add support for FLUX.2 Klein 9B and 4B in single_file_utils",
+    "updated_at": "2026-04-08T11:45:03Z"
+  },
+  {
+    "additions": 1,
+    "author": "eliemichel",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? I believe that the error message was wrong, since the shapes mentionned did not match the test trigering the error. I think that there is also an ambiguity on what is meant by \"quantized\" shape: the value returned b\u2026",
+    "changed_files": 1,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13236",
+    "created_at": "2026-03-09T21:49:18Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13198/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13198",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13236/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13236",
     "labels": [],
-    "merged": true,
-    "number": 13198,
+    "merged": false,
+    "number": 13236,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "docs: improve docstring scheduling_ipndm.py",
-    "updated_at": "2026-03-02T17:42:55Z"
+    "state": "open",
+    "title": "Fix error message in gguf_quantizer.py",
+    "updated_at": "2026-03-10T02:22:19Z"
   },
   {
-    "additions": 52,
-    "author": "DN6",
+    "additions": 216,
+    "author": "sayakpaul",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? When loading custom code from a local path with AutoModel, the subfolder argument is not properly handled in `get_cached_module_file`. This results in this check for a valid file path failing and a local path being\u2026",
-    "changed_files": 2,
+    "body_excerpt": "# What does this PR do? - [x] Refactor MagCache tests. - [x] Include TaylorSeer in our model-level caching test mixin. - [ ] Consider removing caching-related mixins from `test_pipelines_common.py`.",
+    "changed_files": 4,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13197",
-    "created_at": "2026-02-28T06:33:35Z",
-    "deletions": 2,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13235",
+    "created_at": "2026-03-09T13:57:58Z",
+    "deletions": 245,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13197/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13197",
-    "labels": [],
-    "merged": true,
-    "number": 13197,
+    "files_url": "https://github.com/huggingface/diffusers/pull/13235/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13235",
+    "labels": [
+      "tests",
+      "size/L"
+    ],
+    "merged": false,
+    "number": 13235,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "[AutoModel] Fix bug with subfolders and local model paths when loading custom code",
-    "updated_at": "2026-03-02T12:14:25Z"
+    "state": "open",
+    "title": "[tests] refactor caching tests.",
+    "updated_at": "2026-04-17T04:06:37Z"
   },
   {
-    "additions": 150,
-    "author": "yiyixuxu",
+    "additions": 55,
+    "author": "sayakpaul",
     "author_association": "MEMBER",
-    "body_excerpt": "before https://huggingface.co/YiYiXu/modular-loader-t2i-0704 after https://huggingface.co/YiYiXu/modular-loader-t2i-20260227 the input/output isn't too meaningful for conditional/auto pipeline so it is better we show them for each workflow\u2026",
-    "changed_files": 4,
+    "body_excerpt": null,
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13195",
-    "created_at": "2026-02-27T11:11:16Z",
-    "deletions": 115,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13234",
+    "created_at": "2026-03-09T13:21:38Z",
+    "deletions": 70,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13195/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13195",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13234/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13234",
     "labels": [],
-    "merged": true,
-    "number": 13195,
-    "review_comments_count": 3,
-    "state": "closed",
-    "title": "[modular]Update model card to include workflow",
-    "updated_at": "2026-03-03T06:50:11Z"
+    "merged": false,
+    "number": 13234,
+    "review_comments_count": 0,
+    "state": "open",
+    "title": "[tests] migrate group offloading tests to pytest",
+    "updated_at": "2026-03-09T13:21:38Z"
   },
   {
-    "additions": 52,
-    "author": "sayakpaul",
+    "additions": 3,
+    "author": "DN6",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Why not use `pytest` features haha.",
-    "changed_files": 3,
+    "body_excerpt": "Potential fix for [https://github.com/huggingface/diffusers/security/code-scanning/2147](https://github.com/huggingface/diffusers/security/code-scanning/2147) In general, the fix is to explicitly set a `permissions` block in the workflow (\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13194",
-    "created_at": "2026-02-27T10:00:53Z",
-    "deletions": 61,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13233",
+    "created_at": "2026-03-09T11:39:34Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13194/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13194",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13233/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13233",
     "labels": [],
     "merged": true,
-    "number": 13194,
+    "number": 13233,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "[tests] Use `tmp_path` fixture modular tests",
-    "updated_at": "2026-03-09T09:58:51Z"
+    "title": "[CI] Add Workflow permissions to PR tests",
+    "updated_at": "2026-03-09T12:23:02Z"
   },
   {
-    "additions": 32,
-    "author": "yiyixuxu",
-    "author_association": "MEMBER",
-    "body_excerpt": "so that we are able to correctly load repos like https://huggingface.co/diffusers/flux2-modular When creating a pipeline using `SequentialPipelineBlocks.from_blocks_dict()` and saving it, the `modular_model_index.json` stores `_blocks_clas\u2026",
-    "changed_files": 2,
+    "additions": 3881,
+    "author": "plugyawn",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? This PR adds support for [Diffusion Transformers with Representation Autoencoders](https://rae-dit.github.io) in Diffusers. It implements the Stage-2 side of the RAE setup: - `RAEDiT2DModel` - `RAEDiTPipeline` - che\u2026",
+    "changed_files": 22,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13193",
-    "created_at": "2026-02-27T09:40:45Z",
-    "deletions": 1,
+    "comments_count": 21,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13231",
+    "created_at": "2026-03-09T04:49:54Z",
+    "deletions": 4,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13193/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13193",
-    "labels": [],
-    "merged": true,
-    "number": 13193,
-    "review_comments_count": 1,
-    "state": "closed",
-    "title": "[modular] fallback to default_blocks_name when loading base block classes in ModularPipeline",
-    "updated_at": "2026-02-28T04:58:06Z"
+    "files_url": "https://github.com/huggingface/diffusers/pull/13231/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13231",
+    "labels": [
+      "documentation",
+      "models",
+      "tests",
+      "size/L",
+      "utils",
+      "pipelines",
+      "examples"
+    ],
+    "merged": false,
+    "number": 13231,
+    "review_comments_count": 17,
+    "state": "open",
+    "title": "Add RAE Diffusion Transformer inference/preliminary training pipelines",
+    "updated_at": "2026-04-20T14:27:54Z"
   },
   {
-    "additions": 29,
-    "author": "sayakpaul",
+    "additions": 3,
+    "author": "DN6",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? `modular_model_index.json` is a fundamental block for sharing and loading modular blocks and pipelines. This PR adds a test to ensure the integrity and consistency of a `modular_model_index.json`. I think this will\u2026",
+    "body_excerpt": "Potential fix for [https://github.com/huggingface/diffusers/security/code-scanning/2150](https://github.com/huggingface/diffusers/security/code-scanning/2150) In general, this issue is fixed by adding an explicit `permissions` block to the\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13192",
-    "created_at": "2026-02-27T09:40:30Z",
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13230",
+    "created_at": "2026-03-09T04:38:14Z",
     "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13192/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13192",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13230/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13230",
     "labels": [],
     "merged": true,
-    "number": 13192,
+    "number": 13230,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "[tests] consistency tests for modular index",
-    "updated_at": "2026-02-28T03:17:25Z"
+    "title": "[CI] Potential fix for code scanning alert no. 2150: Workflow does not contain permissions",
+    "updated_at": "2026-03-10T03:37:16Z"
   },
   {
-    "additions": 33,
-    "author": "jashshah999",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "## What does this PR do? Fixes None dereference crashes and a mutable default argument bug: **1. `models/model_loading_utils.py`** - `_fetch_remapped_cls_from_config` used chained `.get().get()` without a fallback on the first call. If `pr\u2026",
+    "additions": 115,
+    "author": "lawrence-cj",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "This PR adds LTX-VAE support for SANA-Video. Cc: @dg845 @sayakpaul GPU memory needed: 47GB for LTX refiner ### SANA-Video with LTX2-Refiner: ```python \"\"\"Sana Video + LTX2 Refiner: Stage 1 generate latent \u2192 Stage 2 refine (3 steps).\"\"\" imp\u2026",
     "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13190",
-    "created_at": "2026-02-26T19:59:04Z",
-    "deletions": 11,
+    "comments_count": 6,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13229",
+    "created_at": "2026-03-09T04:16:06Z",
+    "deletions": 42,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13190/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13190",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13229/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13229",
     "labels": [],
-    "merged": false,
-    "number": 13190,
-    "review_comments_count": 0,
-    "state": "open",
-    "title": "Fix None dereference crashes and mutable default feat_idx in Wan/Cosmos autoencoders",
-    "updated_at": "2026-03-12T15:16:37Z"
+    "merged": true,
+    "number": 13229,
+    "review_comments_count": 4,
+    "state": "closed",
+    "title": "add ltx2 vae in sana-video;",
+    "updated_at": "2026-03-18T04:09:52Z"
   },
   {
-    "additions": 69,
-    "author": "christopher5106",
+    "additions": 4,
+    "author": "DefTruth",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "I added support for Kohya's Flux.2 loras to the PR as well. Reproducer code is: FLUX.2 Dev ```python import torch from diffusers import Flux2Pipeline pipe = Flux2Pipeline.from_pretrained(\"black-forest-labs/FLUX.2-dev\", torch_dtype=torch.bf\u2026",
-    "changed_files": 2,
+    "body_excerpt": "Allow pass cpu generator for helios. w/o this fix, helios distilled pipeline will raise an device mismatch error if we use cpu generator. we should use the device from generator for torch.randn. ```bash File \"/workspace/deftruth/dev/minico\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13189",
-    "created_at": "2026-02-26T18:24:54Z",
-    "deletions": 20,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13228",
+    "created_at": "2026-03-09T02:40:17Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13189/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13189",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13228/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13228",
     "labels": [],
-    "merged": false,
-    "number": 13189,
+    "merged": true,
+    "number": 13228,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "Fix for loading of Kohya's Musubi tuner's Flux.2 dev lora",
-    "updated_at": "2026-03-31T06:43:45Z"
+    "state": "closed",
+    "title": "fix: allow pass cpu generator for helios",
+    "updated_at": "2026-03-10T02:46:04Z"
   },
   {
-    "additions": 2,
-    "author": "christopher5106",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "Text encoder lora layers are dropped for some loras such as [this one](https://huggingface.co/scenario-labs/big-head-kontext-lora/blob/main/flux_kontext_lora.safetensors) A log message confirms it: `No LoRA keys associated to CLIPTextModel\u2026",
-    "changed_files": 1,
+    "additions": 2663,
+    "author": "kashif",
+    "author_association": "MEMBER",
+    "body_excerpt": "Add support for LLaDA2/LLaDA2.1 discrete diffusion text generation: - BlockRefinementPipeline: block-wise iterative refinement with confidence-based token commitment, supporting editing threshold for LLaDA2.1 models - LLaDA2Pipeline: conve\u2026",
+    "changed_files": 20,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13188",
-    "created_at": "2026-02-26T18:19:43Z",
-    "deletions": 2,
+    "comments_count": 4,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13226",
+    "created_at": "2026-03-08T17:44:13Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13188/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13188",
-    "labels": [],
-    "merged": true,
-    "number": 13188,
-    "review_comments_count": 0,
+    "files_url": "https://github.com/huggingface/diffusers/pull/13226/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13226",
+    "labels": [
+      "roadmap"
+    ],
+    "merged": true,
+    "number": 13226,
+    "review_comments_count": 50,
     "state": "closed",
-    "title": "Fixing Kohya loras loading: Flux.1-dev loras with TE (\"lora_te1_\" prefix)",
-    "updated_at": "2026-03-05T18:13:18Z"
+    "title": "[Discrete Diffusion] Add LLaDA2 pipeline",
+    "updated_at": "2026-03-25T11:15:43Z"
   },
   {
-    "additions": 75,
-    "author": "Songrui625",
+    "additions": 34,
+    "author": "SHYuanBest",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
+    "body_excerpt": "cc @dg845 @sayakpaul @yiyixuxu ### Helios-Distilled ``` # CUDA_VISIBLE_DEVICES=0,1,2,3 torchrun --master_port 12311 --nproc_per_node 2 infer_distilled.py import os import argparse import torch import torch.distributed as dist from diffuser\u2026",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 6,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13187",
-    "created_at": "2026-02-26T12:37:36Z",
-    "deletions": 3,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13223",
+    "created_at": "2026-03-08T07:57:07Z",
+    "deletions": 9,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13187/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13187",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13223/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13223",
     "labels": [],
     "merged": true,
-    "number": 13187,
-    "review_comments_count": 2,
+    "number": 13223,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix LTX-2 image-to-video generation failure in two stages generation",
-    "updated_at": "2026-02-27T08:55:02Z"
+    "title": "Fix Helios Context Parallelism",
+    "updated_at": "2026-03-08T22:27:10Z"
   },
   {
-    "additions": 115,
-    "author": "DN6",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Add `regisiter_for_auto_class` method to `ConfigMixin` so that we don't have to manually edit the `config.json` file to add the `auto_map` field that we need to load custom code models. <!-- Congratulations! You've\u2026",
-    "changed_files": 3,
+    "additions": 3,
+    "author": "SHYuanBest",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "cc @dg845 @sayakpaul @yiyixuxu",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13186",
-    "created_at": "2026-02-26T11:59:09Z",
-    "deletions": 0,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13222",
+    "created_at": "2026-03-08T07:44:12Z",
+    "deletions": 4,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13186/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13186",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13222/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13222",
     "labels": [],
     "merged": true,
-    "number": 13186,
+    "number": 13222,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "[AutoModel] Allow registering `auto_map` to model config",
-    "updated_at": "2026-03-02T16:43:25Z"
+    "title": "Optimize Helios docs",
+    "updated_at": "2026-03-08T23:54:16Z"
   },
   {
-    "additions": 13,
-    "author": "jashshah999",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "## What does this PR do? Fixes several latent bugs across the codebase: **Mutable default arguments** (shared state across calls): - `_replace_with_gguf_linear`: `modules_to_not_convert=[]` -> `None` with guard - `_dequantize_gguf_and_rest\u2026",
-    "changed_files": 5,
+    "additions": 26,
+    "author": "dg845",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? This PR refactors `AttentionProcessorSkipHook` to support attention processors with a `_skip_attn_scores` flag, which is meant to allow attention processors to specify custom [spatio-temporal guidance (STG)](https:/\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13185",
-    "created_at": "2026-02-25T22:19:12Z",
-    "deletions": 8,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13220",
+    "created_at": "2026-03-07T00:36:14Z",
+    "deletions": 3,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13185/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13185",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13220/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13220",
     "labels": [],
     "merged": false,
-    "number": 13185,
-    "review_comments_count": 0,
+    "number": 13220,
+    "review_comments_count": 7,
     "state": "open",
-    "title": "Fix mutable default arguments, resource leak, and redundant assignment",
-    "updated_at": "2026-03-12T15:16:34Z"
+    "title": "Refactor `AttentionProcessorSkipHook` to Support Custom STG Logic",
+    "updated_at": "2026-03-08T02:01:06Z"
   },
   {
-    "additions": 6,
-    "author": "dxqb",
+    "additions": 2,
+    "author": "tcaimm",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "please see https://github.com/huggingface/diffusers/issues/12893#issuecomment-3961152727 this is a PR against release v.0.36.0, therefore currently not mergable. I can change that if and when this PR is agreed upon. - Pipelines and pipelin\u2026",
-    "changed_files": 1,
+    "body_excerpt": "# What does this PR do? This PR fixes wrapped transformer config access in the Flux2 Klein DreamBooth LoRA training scripts. When the transformer is wrapped by Accelerate, DDP, or FSDP, accessing `transformer.config.guidance_embeds` direct\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 4,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13184",
-    "created_at": "2026-02-25T18:16:26Z",
-    "deletions": 0,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13219",
+    "created_at": "2026-03-06T12:53:58Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13184/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13184",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13219/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13219",
     "labels": [],
-    "merged": false,
-    "number": 13184,
+    "merged": true,
+    "number": 13219,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Z-Image: no attention mask if attention mask is a no-op",
-    "updated_at": "2026-03-24T18:01:33Z"
+    "title": "Fix wrapped transformer config access in Flux2 Klein training",
+    "updated_at": "2026-03-06T14:17:51Z"
   },
   {
-    "additions": 1,
-    "author": "kirillsst",
+    "additions": 2,
+    "author": "kaixuanliu",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "## Summary - Z-Image uses a different CFG formula: `pred = pos + guidance_scale * (pos - neg)`, where `guidance_scale = 0` corresponds to no guidance - The `do_classifier_free_guidance` property was checking `> 1` (standard diffusers conve\u2026",
+    "body_excerpt": "In helios pipeline, the latents are always in FP32 format: [L798](https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/helios/pipeline_helios.py#L798), hence we should skip the test_save_load_float16 test case, rather\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13183",
-    "created_at": "2026-02-24T12:11:43Z",
-    "deletions": 1,
+    "comments_count": 4,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13218",
+    "created_at": "2026-03-06T10:49:10Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13183/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13183",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13218/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13218",
     "labels": [],
     "merged": true,
-    "number": 13183,
+    "number": 13218,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix wrong `do_classifier_free_guidance` threshold in ZImagePipeline",
-    "updated_at": "2026-02-26T01:08:12Z"
+    "title": "skip invalid test case for helios pipeline",
+    "updated_at": "2026-03-18T06:58:36Z"
   },
   {
-    "additions": 140,
-    "author": "sayakpaul",
+    "additions": 2494,
+    "author": "dg845",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? * Fixes support for ring attention for Flash and Flash 3 * Adds a test suite for CP backends x attention backends <img width=\"1024\" height=\"1024\" alt=\"image\" src=\"https://github.com/user-attachments/assets/2600632f-\u2026",
-    "changed_files": 6,
+    "body_excerpt": "# What does this PR do? This PR adds support for LTX-2.3 ([official code](https://github.com/Lightricks/LTX-2), [model weights](https://huggingface.co/Lightricks/LTX-2.3)), a new model in the LTX-2.X family of audio-video models. LTX-2.3 h\u2026",
+    "changed_files": 13,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13182",
-    "created_at": "2026-02-24T11:39:24Z",
-    "deletions": 14,
+    "comments_count": 10,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13217",
+    "created_at": "2026-03-06T04:10:04Z",
+    "deletions": 566,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13182/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13182",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13217/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13217",
     "labels": [],
-    "merged": false,
-    "number": 13182,
-    "review_comments_count": 0,
-    "state": "open",
-    "title": "[attention backends] fix ring CP for flash and flash 3",
-    "updated_at": "2026-02-24T11:53:27Z"
+    "merged": true,
+    "number": 13217,
+    "review_comments_count": 52,
+    "state": "closed",
+    "title": "Add Support for LTX-2.3 Models",
+    "updated_at": "2026-03-19T22:13:11Z"
   },
   {
-    "additions": 1,
-    "author": "sayakpaul",
+    "additions": 4441,
+    "author": "yiyixuxu",
     "author_association": "MEMBER",
-    "body_excerpt": null,
-    "changed_files": 1,
+    "body_excerpt": "```py import torch from diffusers import ModularPipeline, ClassifierFreeGuidance, ClassifierFreeZeroStarGuidance from diffusers.utils import export_to_video, load_image, load_video import gc branch = \"\" ############################# # heli\u2026",
+    "changed_files": 15,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13181",
-    "created_at": "2026-02-24T10:40:49Z",
-    "deletions": 1,
+    "comments_count": 4,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13216",
+    "created_at": "2026-03-05T22:24:45Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13181/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13181",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13216/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13216",
     "labels": [],
     "merged": true,
-    "number": 13181,
-    "review_comments_count": 0,
+    "number": 13216,
+    "review_comments_count": 1,
     "state": "closed",
-    "title": "[docs] Fix torchrun command argument order in docs",
-    "updated_at": "2026-02-24T16:31:40Z"
+    "title": "[modular] helios",
+    "updated_at": "2026-03-09T20:38:00Z"
   },
   {
-    "additions": 7,
-    "author": "DN6",
+    "additions": 57,
+    "author": "sayakpaul",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Since Quanto is now in maintenance mode. I think we can deprecate it and remove by 1.0.0 <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in th\u2026",
-    "changed_files": 2,
+    "body_excerpt": null,
+    "changed_files": 57,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13180",
-    "created_at": "2026-02-24T09:14:48Z",
-    "deletions": 1,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13215",
+    "created_at": "2026-03-05T15:50:25Z",
+    "deletions": 57,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13180/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13180",
-    "labels": [
-      "roadmap"
-    ],
+    "files_url": "https://github.com/huggingface/diffusers/pull/13215/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13215",
+    "labels": [],
     "merged": true,
-    "number": 13180,
+    "number": 13215,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "[Quantization] Deprecate Quanto",
-    "updated_at": "2026-03-11T03:56:47Z"
+    "title": "post release 0.37.0",
+    "updated_at": "2026-03-07T05:33:51Z"
   },
   {
-    "additions": 1,
-    "author": "dg845",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? #13123 introduced an error where we use a Python `<3.9` style type hint with `Optional[...]` and `Union[...]` without importing them from `typing`. This PR fixes this by converting the type hint to a Python `>=3.9`\u2026",
+    "additions": 2,
+    "author": "SHYuanBest",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "cc @dg845 @sayakpaul @yiyixuxu",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13178",
-    "created_at": "2026-02-24T00:04:08Z",
-    "deletions": 3,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13214",
+    "created_at": "2026-03-05T14:15:09Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13178/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13178",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13214/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13214",
     "labels": [],
     "merged": true,
-    "number": 13178,
+    "number": 13214,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix AutoModel `typing` Import Error",
-    "updated_at": "2026-02-24T02:47:41Z"
+    "title": "Convert tensors to float in Helios's optimized_scale function",
+    "updated_at": "2026-03-06T03:08:45Z"
   },
   {
-    "additions": 41,
-    "author": "dhruvildarji",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "## Summary Closes #13104 Issue #13104 reports a `NameError: name 'logger' is not defined` that occurs at import time when `_update_torch_safe_globals()` in `torchao_quantizer.py` raises an `ImportError` (e.g. when `torchao.dtypes.uintx.uin\u2026",
-    "changed_files": 1,
+    "additions": 4,
+    "author": "SHYuanBest",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "cc @dg845 @sayakpaul",
+    "changed_files": 4,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13175",
-    "created_at": "2026-02-23T10:03:12Z",
-    "deletions": 0,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13213",
+    "created_at": "2026-03-05T13:15:11Z",
+    "deletions": 4,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13175/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13175",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13213/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13213",
     "labels": [],
-    "merged": false,
-    "number": 13175,
+    "merged": true,
+    "number": 13213,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "tests: add regression test for logger NameError in torchao_quantizer",
-    "updated_at": "2026-02-23T10:03:12Z"
+    "state": "closed",
+    "title": "Fix Helios paper link in documentation",
+    "updated_at": "2026-03-05T13:28:14Z"
   },
   {
-    "additions": 304,
-    "author": "sayakpaul",
+    "additions": 5,
+    "author": "asomoza",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Add attention backend tests.",
+    "body_excerpt": "Just an extension of #13183 to the other Z-Image pipleines @yiyixuxu @sayakpaul",
     "changed_files": 5,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13174",
-    "created_at": "2026-02-23T06:01:10Z",
-    "deletions": 172,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13212",
+    "created_at": "2026-03-04T16:49:24Z",
+    "deletions": 5,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13174/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13174",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13212/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13212",
     "labels": [],
-    "merged": false,
-    "number": 13174,
+    "merged": true,
+    "number": 13212,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "[tests] add attention backend tests.",
-    "updated_at": "2026-02-24T07:19:17Z"
+    "state": "closed",
+    "title": "[Z-Image] Fix more `do_classifier_free_guidance` thresholds",
+    "updated_at": "2026-03-04T20:12:03Z"
   },
   {
-    "additions": 10,
-    "author": "veeceey",
+    "additions": 135,
+    "author": "SHYuanBest",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "After [Dao-AILab/flash-attention@ed20940](https://github.com/Dao-AILab/flash-attention/commit/ed209409acedbb2379f870bbd03abce31a7a51b7), `flash_attn_3_func` no longer returns `(out, lse, ...)` by default -- it just returns `out`. This brea\u2026",
-    "changed_files": 1,
+    "body_excerpt": "Related to https://github.com/huggingface/diffusers/pull/13208 and https://github.com/huggingface/diffusers/pull/13210 cc @yiyixuxu cc @sayakpaul cc @dg845",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13173",
-    "created_at": "2026-02-23T04:19:21Z",
-    "deletions": 2,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13211",
+    "created_at": "2026-03-04T13:22:26Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13173/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13173",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13211/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13211",
     "labels": [],
     "merged": true,
-    "number": 13173,
-    "review_comments_count": 1,
+    "number": 13211,
+    "review_comments_count": 5,
     "state": "closed",
-    "title": "Fix Flash Attention 3 interface for new FA3 return format",
-    "updated_at": "2026-02-26T12:04:37Z"
+    "title": "Fix group-offloading bug",
+    "updated_at": "2026-03-04T15:04:14Z"
   },
   {
-    "additions": 4,
-    "author": "veeceey",
+    "additions": 23,
+    "author": "SHYuanBest",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "The dynamic CFG calculation in CogVideoX pipelines was using `t.item()` (the raw timestep value, which ranges from ~1 to ~999) instead of the loop index `i` (which ranges from 0 to `num_inference_steps - 1`). This made `((num_inference_ste\u2026",
-    "changed_files": 4,
+    "body_excerpt": "Related to https://github.com/huggingface/diffusers/pull/13208 cc @yiyixuxu cc @sayakpaul cc @dg845",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13172",
-    "created_at": "2026-02-23T04:17:38Z",
-    "deletions": 4,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13210",
+    "created_at": "2026-03-04T13:13:33Z",
+    "deletions": 17,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13172/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13172",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13210/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13210",
     "labels": [],
-    "merged": false,
-    "number": 13172,
+    "merged": true,
+    "number": 13210,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "Fix CogVideoX dynamic CFG to use loop index instead of timestep value",
-    "updated_at": "2026-02-23T04:17:50Z"
+    "state": "closed",
+    "title": "Fix Helios related (optimize docs & remove redudant)",
+    "updated_at": "2026-03-04T15:14:44Z"
   },
   {
-    "additions": 34,
+    "additions": 78,
     "author": "sayakpaul",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Now that the Hub-based attention backends (Flash, Flash 3, SAGE) have been fully feature-complete (`torch.compile`, CP, etc.), we can confidently remove the non-Hub variants. This has several advantages: * More Hub-\u2026",
+    "body_excerpt": "# What does this PR do? Fixes https://github.com/huggingface/diffusers/issues/13203 Additionally, fixes how \"alpha\"s are handled in the diffusers format path.",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13171",
-    "created_at": "2026-02-22T06:58:31Z",
-    "deletions": 708,
+    "comments_count": 7,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13209",
+    "created_at": "2026-03-04T10:58:07Z",
+    "deletions": 16,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13171/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13171",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13209/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13209",
     "labels": [],
-    "merged": false,
-    "number": 13171,
-    "review_comments_count": 0,
-    "state": "open",
-    "title": "[attention backends] remove non-hub attention backends.",
-    "updated_at": "2026-02-23T23:08:31Z"
+    "merged": true,
+    "number": 13209,
+    "review_comments_count": 2,
+    "state": "closed",
+    "title": "[lora] fix zimage lora conversion to support for more lora.",
+    "updated_at": "2026-03-05T02:54:46Z"
   },
   {
-    "additions": 4,
-    "author": "asomoza",
+    "additions": 5655,
+    "author": "dg845",
     "author_association": "MEMBER",
-    "body_excerpt": "FAL trained loras use the default peft key naming, this PR just rename them to the diffusers naming. Fixes #13162 Tested with [this lora](https://huggingface.co/fal/flux-2-klein-4b-spritesheet-lora) ### code ```python import torch from dif\u2026",
-    "changed_files": 1,
+    "body_excerpt": "# What does this PR do? This PR adds pipelines for the Helios-14B video generation model ([project page](https://pku-yuangroup.github.io/Helios-Page/), [official code](https://github.com/PKU-YuanGroup/Helios), [model weights](https://huggi\u2026",
+    "changed_files": 33,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13169",
-    "created_at": "2026-02-20T17:52:43Z",
-    "deletions": 0,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13208",
+    "created_at": "2026-03-04T10:49:11Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13169/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13169",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13208/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13208",
     "labels": [],
     "merged": true,
-    "number": 13169,
+    "number": 13208,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Support Flux Klein peft (fal) lora format",
-    "updated_at": "2026-02-21T05:01:40Z"
+    "title": "Add Helios-14B Video Generation Pipelines",
+    "updated_at": "2026-03-04T16:01:50Z"
   },
   {
-    "additions": 2112,
-    "author": "DN6",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Enable modular pipelines to save model weights to the Hub. Previously, only configs were saved. Note that if individual components are loaded from external repos, their model configs will continue to reference those\u2026",
-    "changed_files": 20,
+    "additions": 34,
+    "author": "jiqing-feng",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "Currently, every process encodes the entire dataset through the VAE during preprocessing. This is redundant in multi-GPU training \u2014 each process does the same full encoding work independently. This PR shards the VAE encoding across process\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13168",
-    "created_at": "2026-02-20T14:01:44Z",
-    "deletions": 27,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13207",
+    "created_at": "2026-03-04T08:02:24Z",
+    "deletions": 7,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13168/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13168",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13207/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13207",
     "labels": [],
     "merged": true,
-    "number": 13168,
-    "review_comments_count": 16,
+    "number": 13207,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "[Modular] Save Modular Pipeline weights to Hub",
-    "updated_at": "2026-03-02T16:50:42Z"
+    "title": "cogvideo example: Distribute VAE video encoding across processes in CogVideoX LoRA training",
+    "updated_at": "2026-03-04T09:39:05Z"
   },
   {
-    "additions": 4,
-    "author": "anijain2305",
+    "additions": 61,
+    "author": "kirillsst",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "Once `dequantize_gguf_tensor` fetches the quant_type attributes from the GGUFParamter tensor subclass, there is no further need of running the actual dequantize operations on the Tensor subclass, we can just convert to plain tensor right a\u2026",
-    "changed_files": 1,
+    "body_excerpt": "## Summary - Add `scheduler.set_begin_index(0)` before the denoising loop in all flow-matching pipelines that were missing it - For img2img pipelines with strength-based schedule truncation (`wan_video2video`, `ltx_condition`), set the beg\u2026",
+    "changed_files": 56,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13166",
-    "created_at": "2026-02-19T23:02:10Z",
-    "deletions": 1,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13206",
+    "created_at": "2026-03-03T13:30:17Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13166/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13166",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13206/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13206",
     "labels": [],
-    "merged": true,
-    "number": 13166,
-    "review_comments_count": 1,
-    "state": "closed",
-    "title": "[gguf][torch.compile time] Convert to plain tensor earlier in dequantize_gguf_tensor",
-    "updated_at": "2026-02-26T18:17:12Z"
+    "merged": false,
+    "number": 13206,
+    "review_comments_count": 0,
+    "state": "open",
+    "title": "Set `begin_index` for flow-matching pipelines to avoid slow `index_for_timestep`",
+    "updated_at": "2026-03-03T13:30:17Z"
   },
   {
-    "additions": 91,
-    "author": "sayakpaul",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Great pointer here https://github.com/huggingface/diffusers/pull/12812#discussion_r2827759302 from @vasqu.",
+    "additions": 1,
+    "author": "teith",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? Fixes invalid type annotation for the `image` parameter in `Flux2Pipeline.__call__`. `list[PIL.Image.Image, PIL.Image.Image]` passes two type arguments to `list`, which only accepts one. Fixed to `PIL.Image.Image |\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13165",
-    "created_at": "2026-02-19T13:28:04Z",
-    "deletions": 66,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13205",
+    "created_at": "2026-03-03T01:07:16Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13165/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13165",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13205/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13205",
     "labels": [],
     "merged": true,
-    "number": 13165,
-    "review_comments_count": 4,
+    "number": 13205,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "[attention backends] use dedicated wrappers from fa3 for cp.",
-    "updated_at": "2026-02-25T19:07:03Z"
+    "title": "fix: correct invalid type annotation for `image` in `Flux2Pipeline.__call__`",
+    "updated_at": "2026-03-13T18:56:39Z"
   },
   {
-    "additions": 5,
-    "author": "sayakpaul",
+    "additions": 239,
+    "author": "yiyixuxu",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Docker containers have been pushed: https://github.com/huggingface/diffusers/actions/runs/22182840803 Cc: @dg845 as well",
+    "body_excerpt": "based on this discussion https://github.com/huggingface/blog/pull/3278#discussion_r2873972272",
     "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13164",
-    "created_at": "2026-02-19T13:14:08Z",
-    "deletions": 5,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13204",
+    "created_at": "2026-03-02T23:03:06Z",
+    "deletions": 8,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13164/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13164",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13204/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13204",
     "labels": [],
     "merged": true,
-    "number": 13164,
-    "review_comments_count": 0,
+    "number": 13204,
+    "review_comments_count": 10,
     "state": "closed",
-    "title": "remove deps related to test from ci",
-    "updated_at": "2026-02-20T03:05:40Z"
+    "title": "[modular] not pass trust_remote_code to external repos ",
+    "updated_at": "2026-03-03T12:36:37Z"
   },
   {
-    "additions": 2,
+    "additions": 0,
     "author": "DN6",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? New LoRA Hotswap tests are failing because the method used to create dummy_inputs is still the one used in the old tests. Update the tests to use the correct method name. <!-- Congratulations! You've made it this fa\u2026",
-    "changed_files": 1,
-    "cluster_id": null,
-    "cluster_ids": [],
-    "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13163",
-    "created_at": "2026-02-19T07:22:56Z",
-    "deletions": 2,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13163/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13163",
-    "labels": [],
-    "merged": true,
-    "number": 13163,
-    "review_comments_count": 0,
-    "state": "closed",
-    "title": "[CI] Fix new LoRAHotswap tests",
-    "updated_at": "2026-02-20T03:31:20Z"
-  },
-  {
-    "additions": 33,
-    "author": "sayakpaul",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? For standardization purposes, this PR updates the `sage_attention` repo id to `sage-attention` as the former is now deprecated. The PR also updates the usage to incorporate `version` instead of `revision` which is a\u2026",
-    "changed_files": 3,
+    "body_excerpt": "# What does this PR do? Accidentally committed a bunch of extra Claude files. Removing them here. <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes\u2026",
+    "changed_files": 17,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13161",
-    "created_at": "2026-02-18T18:17:19Z",
-    "deletions": 7,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13202",
+    "created_at": "2026-03-02T19:04:38Z",
+    "deletions": 1907,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13161/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13161",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13202/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13202",
     "labels": [],
     "merged": true,
-    "number": 13161,
-    "review_comments_count": 4,
+    "number": 13202,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "[attention backends] change to updated repo and version.",
-    "updated_at": "2026-03-05T13:53:12Z"
+    "title": "Clean up accidental files",
+    "updated_at": "2026-03-02T19:05:59Z"
   },
   {
-    "additions": 86,
-    "author": "delmalih",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? This PR improves the docstrings and type hints in `src/diffusers/schedulers/scheduling_flow_match_lcm.py` to follow the project conventions as described in #9567. ## Context This is part of a broader effort to impro\u2026",
-    "changed_files": 1,
+    "additions": 27,
+    "author": "DN6",
+    "author_association": "MEMBER",
+    "body_excerpt": "Reverts huggingface/diffusers#13168",
+    "changed_files": 20,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13160",
-    "created_at": "2026-02-18T14:47:36Z",
-    "deletions": 50,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13201",
+    "created_at": "2026-03-02T18:35:59Z",
+    "deletions": 2112,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13160/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13160",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13201/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13201",
     "labels": [],
-    "merged": true,
-    "number": 13160,
+    "merged": false,
+    "number": 13201,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "docs: improve docstring scheduling_flow_match_lcm.py",
-    "updated_at": "2026-02-18T18:52:03Z"
+    "title": "Revert \"[Modular] Save Modular Pipeline weights to Hub\"",
+    "updated_at": "2026-03-02T18:53:47Z"
   },
   {
-    "additions": 7,
-    "author": "stevhliu",
-    "author_association": "MEMBER",
-    "body_excerpt": "adds some docs for `QwenImageLayeredPipeline`",
-    "changed_files": 1,
+    "additions": 4,
+    "author": "christopher5106",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "If you don't see any problem having Lora support for Controlnet pipelines for Z-Image, here is the PR. @sayakpaul",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13158",
-    "created_at": "2026-02-17T21:48:00Z",
-    "deletions": 1,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13200",
+    "created_at": "2026-03-02T14:51:52Z",
+    "deletions": 4,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13158/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13158",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13200/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13200",
     "labels": [],
     "merged": true,
-    "number": 13158,
-    "review_comments_count": 2,
+    "number": 13200,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "[docs] add docs for qwenimagelayered",
-    "updated_at": "2026-02-18T19:02:18Z"
+    "title": "adding lora support to z-image controlnet pipelines",
+    "updated_at": "2026-03-05T18:11:07Z"
   },
   {
-    "additions": 710,
-    "author": "stevhliu",
+    "additions": 0,
+    "author": "sayakpaul",
     "author_association": "MEMBER",
-    "body_excerpt": "deprecates pipelines and also: - fixes relative imports inside the moved packages - updates `deprecate/__init__.py` for each deprecated pipeline - update `pipelines/__init__.py` to remove deprecated pipeline",
-    "changed_files": 122,
+    "body_excerpt": "# What does this PR do? Remove DB utils from the benchmarking workflow. Internal thread: https://huggingface.slack.com/archives/CTKK32GE8/p1772345661280549 Cc: @McPatate",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13157",
-    "created_at": "2026-02-17T20:01:51Z",
-    "deletions": 2493,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13199",
+    "created_at": "2026-03-02T10:46:19Z",
+    "deletions": 180,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13157/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13157",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13199/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13199",
     "labels": [],
     "merged": true,
-    "number": 13157,
+    "number": 13199,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "[docs] deprecate pipelines",
-    "updated_at": "2026-04-01T17:16:27Z"
+    "title": "remove db utils from benchmarking",
+    "updated_at": "2026-03-02T11:15:33Z"
   },
   {
-    "additions": 46,
-    "author": "stevhliu",
-    "author_association": "MEMBER",
-    "body_excerpt": "docs for #12196",
+    "additions": 43,
+    "author": "delmalih",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? This PR improves the docstrings and type hints in `src/diffusers/schedulers/scheduling_ipndm.py` to follow the project conventions as described in #9567. ## Context This is part of a broader effort to improve docume\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13156",
-    "created_at": "2026-02-17T19:33:43Z",
-    "deletions": 1,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13198",
+    "created_at": "2026-02-28T10:30:41Z",
+    "deletions": 8,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13156/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13156",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13198/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13198",
     "labels": [],
     "merged": true,
-    "number": 13156,
+    "number": 13198,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "[docs] validation for custom blocks",
-    "updated_at": "2026-02-18T18:54:25Z"
+    "title": "docs: improve docstring scheduling_ipndm.py",
+    "updated_at": "2026-03-02T17:42:55Z"
   },
   {
-    "additions": 9,
+    "additions": 52,
     "author": "DN6",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? We had `k-diffusion` as a test dependency, which in turn installed `clip-anytorch` which had `ftfy` as a dependency. This is why all the tests with pipelines using `ftfy` were passing even though we never had it in\u2026",
-    "changed_files": 4,
+    "body_excerpt": "# What does this PR do? When loading custom code from a local path with AutoModel, the subfolder argument is not properly handled in `get_cached_module_file`. This results in this check for a valid file path failing and a local path being\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13155",
-    "created_at": "2026-02-17T11:38:43Z",
-    "deletions": 4,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13197",
+    "created_at": "2026-02-28T06:33:35Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13155/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13155",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13197/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13197",
     "labels": [],
     "merged": true,
-    "number": 13155,
-    "review_comments_count": 5,
+    "number": 13197,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "[CI] Add ftfy as a test dependency",
-    "updated_at": "2026-02-18T17:21:22Z"
+    "title": "[AutoModel] Fix bug with subfolders and local model paths when loading custom code",
+    "updated_at": "2026-03-02T12:14:25Z"
   },
   {
-    "additions": 4,
-    "author": "dg845",
+    "additions": 150,
+    "author": "yiyixuxu",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? This PR fixes the `ftfy` import for the PRX pipeline by guarding the import with `is_ftfy_available`. It also removes the `xfail` mark for the PRX pipeline tests as the tests appear to work now for `transformers>4.5\u2026",
-    "changed_files": 2,
+    "body_excerpt": "before https://huggingface.co/YiYiXu/modular-loader-t2i-0704 after https://huggingface.co/YiYiXu/modular-loader-t2i-20260227 the input/output isn't too meaningful for conditional/auto pipeline so it is better we show them for each workflow\u2026",
+    "changed_files": 4,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13154",
-    "created_at": "2026-02-17T02:02:38Z",
-    "deletions": 12,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13195",
+    "created_at": "2026-02-27T11:11:16Z",
+    "deletions": 115,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13154/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13154",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13195/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13195",
     "labels": [],
     "merged": true,
-    "number": 13154,
-    "review_comments_count": 2,
+    "number": 13195,
+    "review_comments_count": 3,
     "state": "closed",
-    "title": "Fix `ftfy` import for PRX Pipeline",
-    "updated_at": "2026-02-19T04:44:37Z"
+    "title": "[modular]Update model card to include workflow",
+    "updated_at": "2026-03-03T06:50:11Z"
   },
   {
-    "additions": 884,
+    "additions": 52,
     "author": "sayakpaul",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Some comments in-line. I have run the tests locally and all of them pass.",
-    "changed_files": 9,
-    "cluster_id": null,
-    "cluster_ids": [],
-    "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13153",
-    "created_at": "2026-02-16T10:16:59Z",
-    "deletions": 864,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13153/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13153",
-    "labels": [
-      "tests",
-      "size/L"
-    ],
-    "merged": false,
-    "number": 13153,
-    "review_comments_count": 3,
-    "state": "open",
-    "title": "[tests] refactor UNet model tests to align with the new pattern",
-    "updated_at": "2026-04-12T06:54:04Z"
-  },
-  {
-    "additions": 2,
-    "author": "DN6",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? k-diffusion package is no longer maintained (last release was 3 years ago). We've already deprecated the pipeline, but I think we can remove it completely since it's introducing dependency issues in the CI. <!-- Con\u2026",
-    "changed_files": 18,
+    "body_excerpt": "# What does this PR do? Why not use `pytest` features haha.",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13152",
-    "created_at": "2026-02-16T07:23:22Z",
-    "deletions": 1794,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13194",
+    "created_at": "2026-02-27T10:00:53Z",
+    "deletions": 61,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13152/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13152",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13194/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13194",
     "labels": [],
     "merged": true,
-    "number": 13152,
+    "number": 13194,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "[Pipelines] Remove k-diffusion",
-    "updated_at": "2026-02-16T08:24:25Z"
+    "title": "[tests] Use `tmp_path` fixture modular tests",
+    "updated_at": "2026-03-09T09:58:51Z"
   },
   {
-    "additions": 2,
-    "author": "sayakpaul",
+    "additions": 32,
+    "author": "yiyixuxu",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? As discussed in https://github.com/huggingface/diffusers/pull/13082/changes#r2778648922",
+    "body_excerpt": "so that we are able to correctly load repos like https://huggingface.co/diffusers/flux2-modular When creating a pipeline using `SequentialPipelineBlocks.from_blocks_dict()` and saving it, the `modular_model_index.json` stores `_blocks_clas\u2026",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13150",
-    "created_at": "2026-02-16T04:46:01Z",
-    "deletions": 18,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13193",
+    "created_at": "2026-02-27T09:40:45Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13150/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13150",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13193/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13193",
     "labels": [],
     "merged": true,
-    "number": 13150,
-    "review_comments_count": 0,
+    "number": 13193,
+    "review_comments_count": 1,
     "state": "closed",
-    "title": "[tests] accept recompile_limit from the user in tests",
-    "updated_at": "2026-02-16T09:18:24Z"
+    "title": "[modular] fallback to default_blocks_name when loading base block classes in ModularPipeline",
+    "updated_at": "2026-02-28T04:58:06Z"
   },
   {
-    "additions": 0,
-    "author": "dg845",
+    "additions": 29,
+    "author": "sayakpaul",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Follow-up to #13129 and #13132 which maybe finally fixes the `ModuleNotFoundError: No module named 'pkg_resources'` CI error. I think the previous fix didn't work because I assumed the error was triggered when insta\u2026",
-    "changed_files": 2,
+    "body_excerpt": "# What does this PR do? `modular_model_index.json` is a fundamental block for sharing and loading modular blocks and pipelines. This PR adds a test to ensure the integrity and consistency of a `modular_model_index.json`. I think this will\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13149",
-    "created_at": "2026-02-16T01:53:39Z",
-    "deletions": 10,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13192",
+    "created_at": "2026-02-27T09:40:30Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13149/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13149",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13192/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13192",
     "labels": [],
     "merged": true,
-    "number": 13149,
+    "number": 13192,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "[CI] Revert `setuptools` CI Fix as the Failing Pipelines are Deprecated",
-    "updated_at": "2026-02-19T04:34:12Z"
+    "title": "[tests] consistency tests for modular index",
+    "updated_at": "2026-02-28T03:17:25Z"
   },
   {
-    "additions": 78,
-    "author": "yiyixuxu",
-    "author_association": "MEMBER",
-    "body_excerpt": null,
-    "changed_files": 2,
+    "additions": 33,
+    "author": "jashshah999",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## What does this PR do? Fixes None dereference crashes and a mutable default argument bug: **1. `models/model_loading_utils.py`** - `_fetch_remapped_cls_from_config` used chained `.get().get()` without a fallback on the first call. If `pr\u2026",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13148",
-    "created_at": "2026-02-15T02:41:35Z",
-    "deletions": 5,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13190",
+    "created_at": "2026-02-26T19:59:04Z",
+    "deletions": 11,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13148/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13148",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13190/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13190",
     "labels": [],
-    "merged": true,
-    "number": 13148,
-    "review_comments_count": 9,
-    "state": "closed",
-    "title": "[Modular] update the auto pipeline blocks doc",
-    "updated_at": "2026-02-27T20:50:38Z"
+    "merged": false,
+    "number": 13190,
+    "review_comments_count": 0,
+    "state": "open",
+    "title": "Fix None dereference crashes and mutable default feat_idx in Wan/Cosmos autoencoders",
+    "updated_at": "2026-04-17T19:10:51Z"
   },
   {
-    "additions": 1,
-    "author": "yiyixuxu",
-    "author_association": "MEMBER",
-    "body_excerpt": "was reverted in https://github.com/huggingface/diffusers/pull/12524/changes#r2807900601 z-image generate blury image otherwise ``` import torch from diffusers import ModularPipeline device = \"cuda\" # 1. Load the pipeline # Use bfloat16 for\u2026",
-    "changed_files": 1,
+    "additions": 69,
+    "author": "christopher5106",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "I added support for Kohya's Flux.2 loras to the PR as well. Reproducer code is: FLUX.2 Dev ```python import torch from diffusers import Flux2Pipeline pipe = Flux2Pipeline.from_pretrained(\"black-forest-labs/FLUX.2-dev\", torch_dtype=torch.bf\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13147",
-    "created_at": "2026-02-14T20:47:49Z",
-    "deletions": 1,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13189",
+    "created_at": "2026-02-26T18:24:54Z",
+    "deletions": 20,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13147/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13147",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13189/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13189",
     "labels": [],
-    "merged": true,
-    "number": 13147,
+    "merged": false,
+    "number": 13189,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "fix guider ",
-    "updated_at": "2026-02-15T04:09:48Z"
+    "state": "open",
+    "title": "Fix for loading of Kohya's Musubi tuner's Flux.2 dev lora",
+    "updated_at": "2026-03-31T06:43:45Z"
   },
   {
     "additions": 2,
-    "author": "yiyixuxu",
-    "author_association": "MEMBER",
-    "body_excerpt": null,
+    "author": "christopher5106",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "Text encoder lora layers are dropped for some loras such as [this one](https://huggingface.co/scenario-labs/big-head-kontext-lora/blob/main/flux_kontext_lora.safetensors) A log message confirms it: `No LoRA keys associated to CLIPTextModel\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13146",
-    "created_at": "2026-02-14T19:17:32Z",
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13188",
+    "created_at": "2026-02-26T18:19:43Z",
     "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13146/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13146",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13188/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13188",
     "labels": [],
     "merged": true,
-    "number": 13146,
+    "number": 13188,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix MT5Tokenizer",
-    "updated_at": "2026-02-14T19:40:08Z"
+    "title": "Fixing Kohya loras loading: Flux.1-dev loras with TE (\"lora_te1_\" prefix)",
+    "updated_at": "2026-03-05T18:13:18Z"
   },
   {
-    "additions": 127,
-    "author": "yiyixuxu",
-    "author_association": "MEMBER",
-    "body_excerpt": "otherwise cannot load modular blocks like this with `ModularPipeline.from_pretrained()` directly https://huggingface.co/diffusers/flux2-modular",
-    "changed_files": 3,
+    "additions": 75,
+    "author": "Songrui625",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13145",
-    "created_at": "2026-02-14T03:03:31Z",
-    "deletions": 1,
+    "comments_count": 6,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13187",
+    "created_at": "2026-02-26T12:37:36Z",
+    "deletions": 3,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13145/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13145",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13187/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13187",
     "labels": [],
     "merged": true,
-    "number": 13145,
-    "review_comments_count": 0,
+    "number": 13187,
+    "review_comments_count": 2,
     "state": "closed",
-    "title": "[Modular] add different pipeine blocks to init",
-    "updated_at": "2026-02-14T04:36:51Z"
+    "title": "Fix LTX-2 image-to-video generation failure in two stages generation",
+    "updated_at": "2026-02-27T08:55:02Z"
   },
   {
-    "additions": 2,
-    "author": "asomoza",
+    "additions": 115,
+    "author": "DN6",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? LTX2 has the incorrect lora mixing which prevents using LTX2 lora with it. This PR fixes that. ## Who can review? @sayakpaul @dg845",
-    "changed_files": 1,
+    "body_excerpt": "# What does this PR do? Add `regisiter_for_auto_class` method to `ConfigMixin` so that we don't have to manually edit the `config.json` file to add the `auto_map` field that we need to load custom code models. <!-- Congratulations! You've\u2026",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13144",
-    "created_at": "2026-02-14T03:01:49Z",
-    "deletions": 2,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13186",
+    "created_at": "2026-02-26T11:59:09Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13144/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13144",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13186/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13186",
     "labels": [],
     "merged": true,
-    "number": 13144,
+    "number": 13186,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "[LTX2] Fix wrong lora mixin",
-    "updated_at": "2026-02-15T06:06:21Z"
+    "title": "[AutoModel] Allow registering `auto_map` to model config",
+    "updated_at": "2026-03-02T16:43:25Z"
   },
   {
-    "additions": 14,
-    "author": "DavidBert",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Fixes `PRXPipeline.from_pretrained()` failing with transformers 5.x due to `T5GemmaConfig` being refactored into a composite config. In transformers 5.x, `T5GemmaConfig` became a composite config with `encoder`/`dec\u2026",
-    "changed_files": 1,
+    "additions": 13,
+    "author": "jashshah999",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## What does this PR do? Fixes several latent bugs across the codebase: **Mutable default arguments** (shared state across calls): - `_replace_with_gguf_linear`: `modules_to_not_convert=[]` -> `None` with guard - `_dequantize_gguf_and_rest\u2026",
+    "changed_files": 5,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13143",
-    "created_at": "2026-02-13T19:22:52Z",
-    "deletions": 3,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13185",
+    "created_at": "2026-02-25T22:19:12Z",
+    "deletions": 8,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13143/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13143",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13185/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13185",
     "labels": [],
-    "merged": true,
-    "number": 13143,
+    "merged": false,
+    "number": 13185,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "Fix T5GemmaEncoder loading for transformers 5.x composite T5GemmaConfig",
-    "updated_at": "2026-02-23T10:15:46Z"
+    "state": "open",
+    "title": "Fix mutable default arguments, resource leak, and redundant assignment",
+    "updated_at": "2026-03-12T15:16:34Z"
   },
   {
     "additions": 6,
-    "author": "rustadadam",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Allows users to set the variance for the DDPMScheduler to be zero for deterministic generations. This helps in research for understanding the underlying model.",
+    "author": "dxqb",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "please see https://github.com/huggingface/diffusers/issues/12893#issuecomment-3961152727 this is a PR against release v.0.36.0, therefore currently not mergable. I can change that if and when this PR is agreed upon. - Pipelines and pipelin\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13141",
-    "created_at": "2026-02-13T18:20:08Z",
-    "deletions": 2,
+    "comments_count": 4,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13184",
+    "created_at": "2026-02-25T18:16:26Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13141/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13141",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13184/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13184",
     "labels": [],
     "merged": false,
-    "number": 13141,
+    "number": 13184,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "Allows DDPM Scheduler to have Zero Variance on Generation Runs",
-    "updated_at": "2026-02-13T18:22:18Z"
+    "state": "closed",
+    "title": "Z-Image: no attention mask if attention mask is a no-op",
+    "updated_at": "2026-03-24T18:01:33Z"
   },
   {
-    "additions": 18,
-    "author": "stevhliu",
-    "author_association": "MEMBER",
-    "body_excerpt": "adds a kernels section in the `Accelerate inference` docs with the results: - cross-linked to `Attention backends` docs which demonstrates support for loading attention kernels with `set_attention_backend` - defer to the blog post and pipe\u2026",
+    "additions": 1,
+    "author": "kirillsst",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "## Summary - Z-Image uses a different CFG formula: `pred = pos + guidance_scale * (pos - neg)`, where `guidance_scale = 0` corresponds to no guidance - The `do_classifier_free_guidance` property was checking `> 1` (standard diffusers conve\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13139",
-    "created_at": "2026-02-13T16:59:48Z",
-    "deletions": 0,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13183",
+    "created_at": "2026-02-24T12:11:43Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13139/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13139",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13183/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13183",
     "labels": [],
     "merged": true,
-    "number": 13139,
-    "review_comments_count": 4,
+    "number": 13183,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "[docs] kernels",
-    "updated_at": "2026-03-25T16:32:00Z"
+    "title": "Fix wrong `do_classifier_free_guidance` threshold in ZImagePipeline",
+    "updated_at": "2026-02-26T01:08:12Z"
   },
   {
-    "additions": 155,
-    "author": "aazeeem11",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "## Summary - Adds Diffusers Claude Skill + Claude Code plugin scaffold - Adds marketplace compatibility via marketplace.json - Tested locally: - plugin.json validates - marketplace add works - plugin installs and is enabled ## Reviewer Che\u2026",
+    "additions": 140,
+    "author": "sayakpaul",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? * Fixes support for ring attention for Flash and Flash 3 * Adds a test suite for CP backends x attention backends <img width=\"1024\" height=\"1024\" alt=\"image\" src=\"https://github.com/user-attachments/assets/2600632f-\u2026",
     "changed_files": 6,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13138",
-    "created_at": "2026-02-13T16:19:48Z",
-    "deletions": 0,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13182",
+    "created_at": "2026-02-24T11:39:24Z",
+    "deletions": 15,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13138/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13138",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/diffusers/pull/13182/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13182",
+    "labels": [
+      "models",
+      "tests",
+      "size/M"
+    ],
     "merged": false,
-    "number": 13138,
+    "number": 13182,
     "review_comments_count": 0,
     "state": "open",
-    "title": "Add Claude Diffusers skill and Claude Code plugin scaffold (#12935)",
-    "updated_at": "2026-02-13T20:37:29Z"
+    "title": "[attention backends] fix ring CP for flash and flash 3",
+    "updated_at": "2026-04-17T04:22:13Z"
   },
   {
-    "additions": 17,
-    "author": "jscaldwell55",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "## Summary Core issue stems from the fact that `ZImageControlNetModel.from_transformer()` shares module references with the source transformer but doesn't freeze them. This allows users to accidentally train/modify shared modules, corrupti\u2026",
+    "additions": 1,
+    "author": "sayakpaul",
+    "author_association": "MEMBER",
+    "body_excerpt": null,
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13136",
-    "created_at": "2026-02-13T11:56:21Z",
-    "deletions": 0,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13181",
+    "created_at": "2026-02-24T10:40:49Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13136/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13136",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13181/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13181",
     "labels": [],
-    "merged": false,
-    "number": 13136,
+    "merged": true,
+    "number": 13181,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "fix: freeze shared transformer modules in ZImageControlNet.from_transformer",
-    "updated_at": "2026-02-23T17:16:10Z"
+    "state": "closed",
+    "title": "[docs] Fix torchrun command argument order in docs",
+    "updated_at": "2026-02-24T16:31:40Z"
   },
   {
-    "additions": 2,
-    "author": "sayakpaul",
+    "additions": 7,
+    "author": "DN6",
     "author_association": "MEMBER",
-    "body_excerpt": null,
-    "changed_files": 1,
+    "body_excerpt": "# What does this PR do? Since Quanto is now in maintenance mode. I think we can deprecate it and remove by 1.0.0 <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in th\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13135",
-    "created_at": "2026-02-13T09:50:24Z",
-    "deletions": 2,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13180",
+    "created_at": "2026-02-24T09:14:48Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13135/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13135",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/diffusers/pull/13180/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13180",
+    "labels": [
+      "roadmap"
+    ],
     "merged": true,
-    "number": 13135,
+    "number": 13180,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "[docs] fix ltx2 i2v docstring.",
-    "updated_at": "2026-02-14T03:10:23Z"
+    "title": "[Quantization] Deprecate Quanto",
+    "updated_at": "2026-03-11T03:56:47Z"
   },
   {
     "additions": 1,
-    "author": "sayakpaul",
+    "author": "dg845",
     "author_association": "MEMBER",
-    "body_excerpt": "Fixes https://github.com/huggingface/diffusers/actions/runs/21974409178/job/63482724343?pr=12994#step:5:270",
+    "body_excerpt": "# What does this PR do? #13123 introduced an error where we use a Python `<3.9` style type hint with `Optional[...]` and `Union[...]` without importing them from `typing`. This PR fixes this by converting the type hint to a Python `>=3.9`\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13134",
-    "created_at": "2026-02-13T04:34:27Z",
-    "deletions": 1,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13178",
+    "created_at": "2026-02-24T00:04:08Z",
+    "deletions": 3,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13134/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13134",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13178/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13178",
     "labels": [],
     "merged": true,
-    "number": 13134,
+    "number": 13178,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix cosmos transformer typing.",
-    "updated_at": "2026-02-13T09:21:35Z"
+    "title": "Fix AutoModel `typing` Import Error",
+    "updated_at": "2026-02-24T02:47:41Z"
   },
   {
-    "additions": 1,
-    "author": "Minisal",
+    "additions": 41,
+    "author": "dhruvildarji",
     "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? This PR fixes a dimension mismatch in the attention mask handling for the HunyuanVideo1.5 VAE implementation in `autoencoder_kl_hunyuanvideo15.py`. #### Problem In the current implementation (line 148 of `autoencode\u2026",
+    "body_excerpt": "## Summary Closes #13104 Issue #13104 reports a `NameError: name 'logger' is not defined` that occurs at import time when `_update_torch_safe_globals()` in `torchao_quantizer.py` raises an `ImportError` (e.g. when `torchao.dtypes.uintx.uin\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13133",
-    "created_at": "2026-02-13T03:34:51Z",
-    "deletions": 1,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13175",
+    "created_at": "2026-02-23T10:03:12Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13133/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13133",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13175/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13175",
     "labels": [],
     "merged": false,
-    "number": 13133,
+    "number": 13175,
     "review_comments_count": 0,
     "state": "open",
-    "title": "Fix missing .unsqueeze(1) in attention mask for HunyuanVideo VAE",
-    "updated_at": "2026-02-13T04:37:41Z"
+    "title": "tests: add regression test for logger NameError in torchao_quantizer",
+    "updated_at": "2026-02-23T10:03:12Z"
   },
   {
-    "additions": 5,
-    "author": "dg845",
+    "additions": 304,
+    "author": "sayakpaul",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? This PR is a follow-up to #13129 which attempts to fix the same error for the `pr_tests_gpu.yml` workflow, which performs GPU tests for PRs. The fix is currently the same as in the previous PR, but I'm not sure how\u2026",
+    "body_excerpt": "# What does this PR do? Add attention backend tests.",
+    "changed_files": 5,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13174",
+    "created_at": "2026-02-23T06:01:10Z",
+    "deletions": 172,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/diffusers/pull/13174/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13174",
+    "labels": [
+      "tests",
+      "size/L"
+    ],
+    "merged": false,
+    "number": 13174,
+    "review_comments_count": 0,
+    "state": "open",
+    "title": "[tests] add attention backend tests.",
+    "updated_at": "2026-04-17T03:40:02Z"
+  },
+  {
+    "additions": 10,
+    "author": "veeceey",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "After [Dao-AILab/flash-attention@ed20940](https://github.com/Dao-AILab/flash-attention/commit/ed209409acedbb2379f870bbd03abce31a7a51b7), `flash_attn_3_func` no longer returns `(out, lse, ...)` by default -- it just returns `out`. This brea\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13132",
-    "created_at": "2026-02-13T02:24:35Z",
-    "deletions": 0,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13173",
+    "created_at": "2026-02-23T04:19:21Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13132/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13132",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13173/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13173",
     "labels": [],
     "merged": true,
-    "number": 13132,
-    "review_comments_count": 0,
+    "number": 13173,
+    "review_comments_count": 1,
     "state": "closed",
-    "title": "[CI] Fix `setuptools` `pkg_resources` Bug for PR GPU Tests",
-    "updated_at": "2026-02-13T04:41:21Z"
+    "title": "Fix Flash Attention 3 interface for new FA3 return format",
+    "updated_at": "2026-02-26T12:04:37Z"
   },
   {
-    "additions": 33,
-    "author": "Mr-Neutr0n",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "## What does this PR do? Fixes #12893 The attention masks in `ZImageTransformer2DModel` were not properly masking inner padding tokens. After `patchify_and_embed` pads each sequence to a multiple of `SEQ_MULTI_OF` (32), the sequence length\u2026",
-    "changed_files": 1,
+    "additions": 4,
+    "author": "veeceey",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "The dynamic CFG calculation in CogVideoX pipelines was using `t.item()` (the raw timestep value, which ranges from ~1 to ~999) instead of the loop index `i` (which ranges from 0 to `num_inference_steps - 1`). This made `((num_inference_ste\u2026",
+    "changed_files": 4,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13131",
-    "created_at": "2026-02-13T02:04:03Z",
-    "deletions": 9,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13172",
+    "created_at": "2026-02-23T04:17:38Z",
+    "deletions": 4,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13131/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13131",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13172/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13172",
     "labels": [],
     "merged": false,
-    "number": 13131,
+    "number": 13172,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "Fix broken attention masking for inner padding in ZImageTransformer2DModel",
-    "updated_at": "2026-02-13T08:16:07Z"
+    "state": "open",
+    "title": "Fix CogVideoX dynamic CFG to use loop index instead of timestep value",
+    "updated_at": "2026-02-23T04:17:50Z"
   },
   {
-    "additions": 48,
-    "author": "delmalih",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? This PR improves the docstrings and type hints in `src/diffusers/schedulers/scheduling_flow_match_heun_discrete.py` to follow the project conventions as described in #9567. ## Context This is part of a broader effor\u2026",
+    "additions": 34,
+    "author": "sayakpaul",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Now that the Hub-based attention backends (Flash, Flash 3, SAGE) have been fully feature-complete (`torch.compile`, CP, etc.), we can confidently remove the non-Hub variants. This has several advantages: * More Hub-\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/diffusers/pull/13130",
-    "created_at": "2026-02-12T20:25:56Z",
-    "deletions": 16,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/diffusers/pull/13171",
+    "created_at": "2026-02-22T06:58:31Z",
+    "deletions": 708,
     "draft": false,
-    "files_url": "https://github.com/huggingface/diffusers/pull/13130/files",
-    "html_url": "https://github.com/huggingface/diffusers/pull/13130",
+    "files_url": "https://github.com/huggingface/diffusers/pull/13171/files",
+    "html_url": "https://github.com/huggingface/diffusers/pull/13171",
     "labels": [],
-    "merged": true,
-    "number": 13130,
+    "merged": false,
+    "number": 13171,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "docs: improve docstring scheduling_flow_match_heun_discrete.py",
-    "updated_at": "2026-02-12T22:32:04Z"
+    "state": "open",
+    "title": "[attention backends] remove non-hub attention backends.",
+    "updated_at": "2026-02-23T23:08:31Z"
   }
 ]