gliner2-5-base-coreml / reports /classification-embedding-w8.json
alexwengg's picture
Add validated embedding W8 classification option
8f6aaf0 verified
Raw History Blame Contribute Delete
5.35 kB
{
"source_revision": "1a8bc24e00dc7300b9017c81d63e3dcdabb26596",
"shape": "L128/K8 classification",
"candidate_package": "gliner2_base_classification_embedding_w8_L128_K8.mlpackage",
"weight_sha256": "037f91cb176d31a38bc1795272d99c47531f3ac235b6f4533bab0df9ae4de327",
"compression": {
"source": "build/gliner2_base_classification_fp16_L128_K8.mlpackage",
"output": "build/gliner2_base_classification_embedding_w8_L128_K8.mlpackage",
"method": "embedding",
"granularity": "per_channel",
"selected_weights": [
"encoder_embeddings_word_embeddings_weight_to_fp16"
],
"source_bytes": 388981604,
"output_bytes": 291053994,
"compression_seconds": 5.819930917001329
},
"paired_first_manifest_all": {
"reference": "build/gliner2_base_classification_fp16_L128_K8.mlpackage",
"candidate": "build/gliner2_base_classification_embedding_w8_L128_K8.mlpackage",
"manifest": "../../gliner2-small/coreml/fixtures/optimization-selected20.jsonl",
"compute_units": "all",
"processor": "arm",
"machine": "arm64",
"macos": "27.0",
"power_audit_after_run": "Now drawing from 'AC Power'\n -InternalBattery-0 (id=35651683)\t50%; charging; 1:25 remaining present: true",
"reference_load_ms": 2044.181875011418,
"candidate_load_ms": 4582.357916980982,
"warmup_calls_per_model": 1,
"timed_order": "alternating reference-first/candidate-first by request",
"checked": 14,
"choice_agreement": 14,
"max_probability_error": 0.00992196798324585,
"probability_error_limit": 0.02,
"reference_p50_ms": 3.334374981932342,
"candidate_p50_ms": 3.261812496930361,
"reference_p95_ms": 9.934750036336482,
"candidate_p95_ms": 5.232541996520013,
"reference_total_p50_ms": 3.6611664691008627,
"candidate_total_p50_ms": 3.6319584760349244
},
"paired_first_manifest_cpu_ane": {
"reference": "build/gliner2_base_classification_fp16_L128_K8.mlpackage",
"candidate": "build/gliner2_base_classification_embedding_w8_L128_K8.mlpackage",
"manifest": "../../gliner2-small/coreml/fixtures/optimization-selected20.jsonl",
"compute_units": "cpu-ane",
"processor": "arm",
"machine": "arm64",
"macos": "27.0",
"power_audit_after_run": "Now drawing from 'AC Power'\n -InternalBattery-0 (id=35651683)\t51%; charging; 1:33 remaining present: true",
"reference_load_ms": 3068.8863749965094,
"candidate_load_ms": 5926.175208995119,
"warmup_calls_per_model": 1,
"timed_order": "alternating reference-first/candidate-first by request",
"checked": 14,
"choice_agreement": 14,
"max_probability_error": 0.0068359375,
"probability_error_limit": 0.02,
"reference_p50_ms": 7.578062970424071,
"candidate_p50_ms": 7.721708010649309,
"reference_p95_ms": 8.475541020743549,
"candidate_p95_ms": 9.883625025395304,
"reference_total_p50_ms": 9.168104501441121,
"candidate_total_p50_ms": 9.666354482760653
},
"native_selected100": {
"model": "fastino/gliner2.5-base-v1",
"revision": "1a8bc24e00dc7300b9017c81d63e3dcdabb26596",
"package": "build/gliner2_base_classification_embedding_w8_L128_K8.mlpackage",
"selected_manifest": "first eligible rows in source suite order; no gold labels used",
"counts": {
"checked": 100,
"too_many_options": 300,
"too_long": 7,
"duplicate_labels": 0,
"mismatches": 0
},
"maximum_confidence_error": 0.013702988624572754,
"mean_confidence_error": 0.001015886664390564,
"model_call_p50_ms": 10.512874985579401,
"model_call_p95_ms": 14.00475000264123,
"failures": []
},
"compute_plan_forced_cpu_ane": {
"total_ops": 516,
"ane_ops": 496,
"cpu_ops": 20,
"ane_percent": 96.1,
"reasons": [
{
"reason": "Unsupported tensor data type: int32",
"count": 17,
"estimated_cpu_runtime_ms": 2.8975,
"op_types": {
"ios17.expand_dims": 5,
"ios17.cast": 5,
"ios17.add": 2,
"select": 2,
"ios17.gather": 1,
"ios17.mul": 1,
"tile": 1
},
"ops": [
"add_0",
"select_0",
"add_0_1",
"select_0_1",
"inputs_embeds_cast_fp16",
"mask_3",
"mask_3_to_fp16",
"var_86",
"extended_attention_mask",
"extended_attention_mask_to_uint16",
"var_88_cast_int16_to_int32",
"var_89",
"attention_mask_1",
"var_272",
"var_2537",
"indices",
"indices_to_int16"
]
},
{
"reason": "Unable to resolve operation input \"y\".",
"count": 2,
"estimated_cpu_runtime_ms": 0.002,
"op_types": {
"ios17.greater_equal": 2
},
"ops": [
"greater_equal_0",
"greater_equal_0_1"
]
},
{
"reason": "ANE not available for this op",
"count": 1,
"estimated_cpu_runtime_ms": 0,
"op_types": {
"ios16.constexpr_affine_dequantize": 1
},
"ops": [
"encoder_embeddings_word_embeddings_weight_to_fp16_quantized"
]
}
]
},
"scope": "Selected real classification requests only; no Decision Index or 2048 score; extraction packages are separate."
}