diff --git a/analysis/current/analysis-report-hybrid.json b/analysis/current/analysis-report-hybrid.json index 5922bf552130b3bcaf1083f88e72ced121e7aebb..128cbd951af2cb041e4e5068b05dab7f764e8398 100644 --- a/analysis/current/analysis-report-hybrid.json +++ b/analysis/current/analysis-report-hybrid.json @@ -1,8 +1,8 @@ { "schema_version": "1.0", "repo": "huggingface/transformers", - "snapshot_id": "20260627T180026Z", - "generated_at": "2026-06-27T18:10:33Z", + "snapshot_id": "20260628T000017Z", + "generated_at": "2026-06-28T00:10:07Z", "evidence_quality": "full", "llm_enrichment": true, "meta_bugs": [ @@ -1050,26 +1050,122 @@ ] }, { - "cluster_id": "cluster-46894-7", - "summary": "Cluster of 3 issues and 4 PRs centered on issue #46897.", + "cluster_id": "cluster-41211-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #41211.", "status": "open", - "confidence": 0.8, + "confidence": 0.55, + "canonical_issue_number": 41211, + "canonical_pr_number": 44339, + "issue_numbers": [ + 41211 + ], + "pr_numbers": [ + 41356, + 44339 + ], + "evidence_types": [ + "closing_reference" + ], + "pr_comparisons": [ + { + "left_pr_number": 41356, + "right_pr_number": 44339, + "code_similarity": 0.155, + "size_similarity": 0.094, + "file_overlap": 0.273, + "area_overlap": 0.078, + "patch_similarity": 0.183, + "shared_filenames": [ + "docs/source/en/model_doc/deimv2.md", + "src/transformers/models/deimv2/__init__.py", + "src/transformers/models/deimv2/configuration_deimv2.py", + "src/transformers/models/deimv2/modeling_deimv2.py", + "tests/models/deimv2/__init__.py", + "tests/models/deimv2/test_modeling_deimv2.py" + ], + "shared_file_areas": [ + { + "filename": "docs/source/en/model_doc/deimv2.md", + "left_ranges": [ + [ + 1, + 132 + ] + ], + "right_ranges": [ + [ + 1, + 65 + ] + ] + }, + { + "filename": "src/transformers/models/deimv2/__init__.py", + "left_ranges": [ + [ + 1, + 15 + ] + ], + "right_ranges": [ + [ + 1, + 29 + ] + ] + }, + { + "filename": "src/transformers/models/deimv2/configuration_deimv2.py", + "left_ranges": [ + [ + 1, + 74 + ] + ], + "right_ranges": [ + [ + 1, + 266 + ] + ] + }, + { + "filename": "tests/models/deimv2/test_modeling_deimv2.py", + "left_ranges": [ + [ + 1, + 15 + ] + ], + "right_ranges": [ + [ + 1, + 1734 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-46894-5", + "summary": "Cluster of 2 issues and 3 PRs centered on issue #46897.", + "status": "open", + "confidence": 0.6, "canonical_issue_number": 46897, - "canonical_pr_number": 46903, + "canonical_pr_number": 46895, "issue_numbers": [ 46894, - 46897, - 46901 + 46897 ], "pr_numbers": [ 46895, 46898, - 46903, 46916 ], "evidence_types": [ "closing_reference", - "shared_issue_target", "soft_similarity" ], "pr_comparisons": [ @@ -1084,17 +1180,6 @@ "shared_filenames": [], "shared_file_areas": [] }, - { - "left_pr_number": 46895, - "right_pr_number": 46903, - "code_similarity": 0.24, - "size_similarity": 0.844, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.476, - "shared_filenames": [], - "shared_file_areas": [] - }, { "left_pr_number": 46895, "right_pr_number": 46916, @@ -1148,17 +1233,6 @@ } ] }, - { - "left_pr_number": 46898, - "right_pr_number": 46903, - "code_similarity": 0.125, - "size_similarity": 0.553, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.099, - "shared_filenames": [], - "shared_file_areas": [] - }, { "left_pr_number": 46898, "right_pr_number": 46916, @@ -1219,116 +1293,6 @@ ] } ] - }, - { - "left_pr_number": 46903, - "right_pr_number": 46916, - "code_similarity": 0.17, - "size_similarity": 0.737, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.154, - "shared_filenames": [], - "shared_file_areas": [] - } - ] - }, - { - "cluster_id": "cluster-41211-3", - "summary": "Cluster of 1 issues and 2 PRs centered on issue #41211.", - "status": "open", - "confidence": 0.55, - "canonical_issue_number": 41211, - "canonical_pr_number": 44339, - "issue_numbers": [ - 41211 - ], - "pr_numbers": [ - 41356, - 44339 - ], - "evidence_types": [ - "closing_reference" - ], - "pr_comparisons": [ - { - "left_pr_number": 41356, - "right_pr_number": 44339, - "code_similarity": 0.155, - "size_similarity": 0.094, - "file_overlap": 0.273, - "area_overlap": 0.078, - "patch_similarity": 0.183, - "shared_filenames": [ - "docs/source/en/model_doc/deimv2.md", - "src/transformers/models/deimv2/__init__.py", - "src/transformers/models/deimv2/configuration_deimv2.py", - "src/transformers/models/deimv2/modeling_deimv2.py", - "tests/models/deimv2/__init__.py", - "tests/models/deimv2/test_modeling_deimv2.py" - ], - "shared_file_areas": [ - { - "filename": "docs/source/en/model_doc/deimv2.md", - "left_ranges": [ - [ - 1, - 132 - ] - ], - "right_ranges": [ - [ - 1, - 65 - ] - ] - }, - { - "filename": "src/transformers/models/deimv2/__init__.py", - "left_ranges": [ - [ - 1, - 15 - ] - ], - "right_ranges": [ - [ - 1, - 29 - ] - ] - }, - { - "filename": "src/transformers/models/deimv2/configuration_deimv2.py", - "left_ranges": [ - [ - 1, - 74 - ] - ], - "right_ranges": [ - [ - 1, - 266 - ] - ] - }, - { - "filename": "tests/models/deimv2/test_modeling_deimv2.py", - "left_ranges": [ - [ - 1, - 15 - ] - ], - "right_ranges": [ - [ - 1, - 1734 - ] - ] - } - ] } ] }, @@ -1902,7 +1866,7 @@ "cluster_id": "cluster-46305-3", "summary": "Cluster of 1 issues and 2 PRs centered on issue #46305.", "status": "open", - "confidence": 0.55, + "confidence": 0.8, "canonical_issue_number": 46305, "canonical_pr_number": 46546, "issue_numbers": [ @@ -1913,7 +1877,9 @@ 46546 ], "evidence_types": [ - "closing_reference" + "closing_reference", + "shared_issue_target", + "soft_similarity" ], "pr_comparisons": [ { @@ -2080,6 +2046,229 @@ } ] }, + { + "cluster_id": "cluster-43612-2", + "summary": "Cluster of 2 related pull requests linked by soft_similarity.", + "status": "open", + "confidence": 0.5, + "canonical_issue_number": null, + "canonical_pr_number": 43612, + "issue_numbers": [], + "pr_numbers": [ + 43612, + 43613 + ], + "evidence_types": [ + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 43612, + "right_pr_number": 43613, + "code_similarity": 0.6, + "size_similarity": 0.822, + "file_overlap": 0.333, + "area_overlap": 0.698, + "patch_similarity": 0.608, + "shared_filenames": [ + "docs/source/en/_toctree.yml", + "docs/source/en/main_classes/pipelines.md", + "docs/source/en/model_doc/auto.md", + "src/transformers/__init__.py", + "src/transformers/models/auto/modeling_auto.py", + "src/transformers/pipelines/__init__.py", + "utils/check_docstrings.py", + "utils/update_metadata.py" + ], + "shared_file_areas": [ + { + "filename": "docs/source/en/_toctree.yml", + "left_ranges": [ + [ + 335, + 342 + ] + ], + "right_ranges": [ + [ + 309, + 316 + ] + ] + }, + { + "filename": "docs/source/en/main_classes/pipelines.md", + "left_ranges": [ + [ + 473, + 484 + ] + ], + "right_ranges": [ + [ + 473, + 484 + ] + ] + }, + { + "filename": "docs/source/en/model_doc/auto.md", + "left_ranges": [ + [ + 113, + 122 + ] + ], + "right_ranges": [ + [ + 113, + 122 + ] + ] + }, + { + "filename": "src/transformers/__init__.py", + "left_ranges": [ + [ + 155, + 161 + ], + [ + 664, + 670 + ] + ], + "right_ranges": [ + [ + 155, + 161 + ], + [ + 664, + 670 + ] + ] + }, + { + "filename": "src/transformers/models/auto/modeling_auto.py", + "left_ranges": [ + [ + 1629, + 1642 + ], + [ + 1807, + 1816 + ], + [ + 1840, + 1849 + ], + [ + 2186, + 2192 + ], + [ + 2233, + 2239 + ] + ], + "right_ranges": [ + [ + 1629, + 1646 + ], + [ + 1811, + 1820 + ], + [ + 1844, + 1853 + ], + [ + 2190, + 2196 + ], + [ + 2237, + 2243 + ] + ] + }, + { + "filename": "src/transformers/pipelines/__init__.py", + "left_ranges": [ + [ + 71, + 77 + ], + [ + 108, + 114 + ], + [ + 300, + 311 + ], + [ + 446, + 453 + ] + ], + "right_ranges": [ + [ + 71, + 77 + ], + [ + 108, + 114 + ], + [ + 300, + 311 + ], + [ + 446, + 453 + ] + ] + }, + { + "filename": "utils/check_docstrings.py", + "left_ranges": [ + [ + 109, + 115 + ] + ], + "right_ranges": [ + [ + 109, + 115 + ] + ] + }, + { + "filename": "utils/update_metadata.py", + "left_ranges": [ + [ + 74, + 84 + ] + ], + "right_ranges": [ + [ + 74, + 84 + ] + ] + } + ] + } + ] + }, { "cluster_id": "cluster-44018-2", "summary": "Cluster of 2 related pull requests linked by soft_similarity.", @@ -2358,13 +2547,12 @@ ], "duplicate_issues": [ { - "cluster_id": "cluster-46894-7", + "cluster_id": "cluster-46894-5", "canonical_issue_number": 46897, "duplicate_issue_numbers": [ - 46894, - 46901 + 46894 ], - "reason": "Issues in cluster-46894-7 are treated as duplicates because they share closing_reference, shared_issue_target, soft_similarity evidence." + "reason": "Issues in cluster-46894-5 are treated as duplicates because they share closing_reference, soft_similarity evidence." } ], "duplicate_prs": [ @@ -2386,6 +2574,15 @@ "target_issue_number": 43240, "reason": "PRs in cluster-43240-3 are treated as duplicates because they converge on issue #43240 with closing_reference, shared_issue_target evidence." }, + { + "cluster_id": "cluster-43612-2", + "canonical_pr_number": 43612, + "duplicate_pr_numbers": [ + 43613 + ], + "target_issue_number": null, + "reason": "PRs in cluster-43612-2 are treated as duplicates because they share soft_similarity evidence." + }, { "cluster_id": "cluster-43698-3", "canonical_pr_number": 43779, @@ -2464,7 +2661,7 @@ 46310 ], "target_issue_number": 46305, - "reason": "PRs in cluster-46305-3 are treated as duplicates because they converge on issue #46305 with closing_reference evidence." + "reason": "PRs in cluster-46305-3 are treated as duplicates because they converge on issue #46305 with closing_reference, shared_issue_target, soft_similarity evidence." }, { "cluster_id": "cluster-46869-3", @@ -2476,22 +2673,21 @@ "reason": "PRs in cluster-46869-3 are treated as duplicates because they converge on issue #46869 with closing_reference, shared_issue_target evidence." }, { - "cluster_id": "cluster-46894-7", - "canonical_pr_number": 46903, + "cluster_id": "cluster-46894-5", + "canonical_pr_number": 46895, "duplicate_pr_numbers": [ - 46895, 46898, 46916 ], "target_issue_number": 46897, - "reason": "PRs in cluster-46894-7 are treated as duplicates because they converge on issue #46897 with closing_reference, shared_issue_target, soft_similarity evidence." + "reason": "PRs in cluster-46894-5 are treated as duplicates because they converge on issue #46897 with closing_reference, soft_similarity evidence." } ], "best_issue": { - "cluster_id": "cluster-46894-7", + "cluster_id": "cluster-46894-5", "issue_number": 46897, - "reason": "Issue #46897 is the strongest global issue candidate because it is open, belongs to a cluster with 7 artifacts, and carries 0 discussion comments plus 9 inbound references.", - "score": 179.0 + "reason": "Issue #46897 is the strongest global issue candidate because it is open, belongs to a cluster with 5 artifacts, and carries 0 discussion comments plus 9 inbound references.", + "score": 159.0 }, "best_pr": { "cluster_id": "cluster-43979-11", diff --git a/analysis/current/analysis-report-hybrid.llm-reviews.json b/analysis/current/analysis-report-hybrid.llm-reviews.json index fc7802477634266b106a19c5e1877d0eba15fabe..08ddb89f2d0642bb166b42a6709f56bd7eaca65c 100644 --- a/analysis/current/analysis-report-hybrid.llm-reviews.json +++ b/analysis/current/analysis-report-hybrid.llm-reviews.json @@ -1,8 +1,8 @@ { "schema_version": "1.0", "repo": "huggingface/transformers", - "snapshot_id": "20260627T180026Z", - "generated_at": "2026-06-27T18:10:33Z", + "snapshot_id": "20260628T000017Z", + "generated_at": "2026-06-28T00:10:07Z", "model": "gpt-5.4-mini?service_tier=flex", "reviews": [ { @@ -43,7 +43,6 @@ "issue:44457", "issue:44509", "issue:44512", - "issue:44560", "issue:44805", "issue:44841", "issue:44918", @@ -89,7 +88,7 @@ "issue:44805|issue:46294", "issue:44805|issue:45161", "issue:46165|issue:46468", - "issue:44560|issue:46165", + "issue:43596|issue:46165", "issue:34634|issue:35545", "issue:30277|issue:43638", "issue:43596|issue:43638", @@ -114,8 +113,8 @@ "issue:43425|issue:44292", "issue:34928|issue:41720", "issue:43828|issue:45161", - "issue:43541|issue:43716", "issue:43828|issue:44292", + "issue:43541|issue:43716", "issue:43716|issue:46468", "issue:44292|issue:44918", "issue:45507|issue:45685", @@ -132,346 +131,412 @@ "issue:45237|issue:46015", "issue:43039|issue:44805" ], - "prepared_review_unit_hash": "8ebdbff0ab919b4f3c4340465a7be47be8f35f87016d502fdf116218b0323dd5", + "prepared_review_unit_hash": "517c12978990242cc8684f5400b8837991f5f08e1019f4aebf3d2c16da4cefe9", "component_budget": { "node_count": 502, "item_count": 502, - "soft_pair_count": 1842, - "serialized_chars": 956182, - "estimated_input_tokens": 239046, - "estimated_eval_tokens": 478348 + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 }, "budget": { - "node_count": 48, - "item_count": 48, + "node_count": 47, + "item_count": 47, "soft_pair_count": 63, - "serialized_chars": 46565, - "estimated_input_tokens": 11642, - "estimated_eval_tokens": 23540 + "serialized_chars": 46010, + "estimated_input_tokens": 11503, + "estimated_eval_tokens": 23262 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly heterogeneous. Only a couple of soft edges look like true duplicates: the DeepSpeed ZeRO-3/BERT index error pair and the torch 2.4.1 incompatibility pair. Most other candidates are related only at a broad subsystem level, not the same bug/change.", - "confidence": 0.69, - "canonical_issue_reason": "No single issue cleanly represents the whole cluster, but issue 43596 is the best canonical anchor because it is the broadest clear duplicate target in the strongest issue pair (43638 is its narrower BERT/ZeRO-3 variant).", + "summary": "This is a heterogeneous issue cluster, not one coherent duplicate set. Only a few soft edges look like true duplicates: the Bert/ZeRO-3 index error pair, the Qwen3.5 MoE rotary-embedding/offload pair, and the torch 2.4.1 compatibility pair.", + "confidence": 0.83, + "canonical_issue_reason": "No single canonical issue fits the whole cluster; 43596 is the best anchor because it is the broadest version of the clearest duplicate family (paired with 43638 on the same ZeRO-3/BERT IndexError).", "canonical_pr_reason": null, - "best_issue_reason": "Issue 43596 is the best representative issue to keep: it captures a concrete, duplicateable failure mode and is broader than its twin while remaining specific enough to triage.", + "best_issue_reason": "43596 is the strongest global representative among the duplicate-like items: it states the core ZeRO-3/BERT index-out-of-bounds bug in the most general way.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:44322", "right": "issue:45464", "accept": false, - "reason": "Different Qwen3.5 failures: missing config attribute vs streaming chat/completions error." + "reason": "Different failures: config attribute mismatch in Qwen3.5 vs streaming chat/completions API error." }, { "left": "issue:45237", "right": "issue:45685", "accept": false, - "reason": "Both involve model/runtime problems, but GPT-OSS AMD GPU failure and MPS histogram error are unrelated bugs." + "reason": "Different model/backend problems: AMD GPU failure vs an MPS histogram/int dtype issue." }, { "left": "issue:43716", "right": "issue:45237", "accept": false, - "reason": "Different model families and different failure modes: dtype mismatch in preprocessing vs AMD GPU runtime failure." + "reason": "Different code paths and symptoms: image preprocessor dtype mismatch vs GPT-OSS AMD GPU failure." }, { "left": "issue:43828", "right": "issue:45237", "accept": false, - "reason": "Autocast dtype mismatch in Phi-tiny-MoE is not the same bug as GPT-OSS failing on AMD GPUs." + "reason": "Autocast dtype mismatch in Phi-tiny-MoE is unrelated to GPT-OSS-on-AMD failure." }, { "left": "issue:45237", "right": "issue:46468", "accept": false, - "reason": "Different models and symptoms; no shared concrete code-path failure." + "reason": "Unrelated model/example failures; not the same underlying bug." }, { "left": "issue:45237", "right": "issue:45507", "accept": false, - "reason": "GPT-OSS AMD GPU failure and GraniteMoEHybrid invalid-method call are different underlying bugs." + "reason": "Different components: GPT-OSS GPU issue vs GraniteMoEHybrid invalid method call." }, { "left": "issue:30277", "right": "issue:35545", "accept": false, - "reason": "DeepSpeed collective mismatch and ONNX export failure are unrelated." + "reason": "DeepSpeed collective mismatch and ModernBERT ONNX export are unrelated." }, { "left": "issue:35545", "right": "issue:42915", "accept": false, - "reason": "Both are export/quantization-type issues at a high level, but they affect different models and code paths." + "reason": "ONNX export failure vs Qwen3Moe FP8 failure; different bug families." }, { "left": "issue:35545", "right": "issue:41720", "accept": false, - "reason": "ONNX export error and Qwen3 auto device map cudaErrorAssert are unrelated." + "reason": "Export-time error and Qwen3 auto device-map cuda assert are different problems." }, { "left": "issue:45901", "right": "issue:46015", "accept": false, - "reason": "Table QA crash and outdated RAG examples are separate issues; one is runtime, the other docs." + "reason": "Table-QA crash and outdated RAG docs are unrelated." }, { "left": "issue:44509", "right": "issue:44512", "accept": false, - "reason": "Both are v5 documentation cleanup items, but they reference different removed docs/features and are not the same change." + "reason": "Both are docs cleanups, but they refer to different removed docs features and are not the same concrete issue." }, { "left": "issue:35545", "right": "issue:36010", "accept": false, - "reason": "ModernBERT ONNX export and GenerationMixin import error are different problems." + "reason": "ONNX export and GenerationMixin import error are different code paths." }, { "left": "issue:44918", "right": "issue:46165", "accept": false, - "reason": "TRL SFT embedding unpacking and training-resume/trust-remote-code behavior are unrelated." + "reason": "TRL SFT embedding unpacking and remote-code resume training are unrelated." }, { "left": "issue:46165", "right": "issue:46869", "accept": false, - "reason": "Model resume/trust-remote-code issue and YOLO multi-GPU device-map issue are different bugs." + "reason": "Remote-code resume training and YOLO multi-GPU device-map failure are unrelated." }, { "left": "issue:44805", "right": "issue:45823", "accept": false, - "reason": "Mask shape mismatch and Gemma4 PLE device mismatch are different failure modes." + "reason": "Both mention mismatches, but the models and failure modes differ." }, { "left": "issue:43638", "right": "issue:44805", "accept": false, - "reason": "Both are IndexError-style reports, but the Bert zero3 loading issue and the mask/tensor mismatch are not the same bug." - }, - { - "left": "issue:43638", - "right": "issue:46165", - "accept": false, - "reason": "BERT ZeRO-3 loading error and training-resume/trust-remote-code issue are unrelated." + "reason": "ZeRO-3/BERT index error vs mask/tensor shape mismatch are different bugs." }, { "left": "issue:43596", "right": "issue:44805", "accept": false, - "reason": "Same broad IndexError class, but different model/setup and different code path." + "reason": "Same broad symptom class, but different model and error path." }, { "left": "issue:44805", "right": "issue:46294", "accept": false, - "reason": "Mask/tensor shape mismatch is unrelated to Qwen3.5 video token-feature mismatch." + "reason": "Mask shape mismatch and video token/template mismatch are unrelated." }, { "left": "issue:44805", "right": "issue:45161", "accept": false, - "reason": "Shape mismatch in masking is not the same as GPT-OSS MoE tensor-parallel failure." + "reason": "Different issues: tensor mask mismatch vs GPT-OSS tensor-parallel failure." }, { "left": "issue:46165", "right": "issue:46468", "accept": false, - "reason": "Resume/trust-remote-code and Nemotron example breakage are different issues." + "reason": "Unrelated training resume and Nemotron example failures." }, { - "left": "issue:44560", + "left": "issue:43596", "right": "issue:46165", "accept": false, - "reason": "Qwen3.5-VL video StopIteration and resume-training issue are unrelated." + "reason": "ZeRO-3/BERT init bug vs resume-training remote-code issue." }, { "left": "issue:34634", "right": "issue:35545", "accept": false, - "reason": "BarkProcessor voice preset and ModernBERT ONNX export are unrelated." + "reason": "BarkProcessor voice_preset and ModernBERT export are unrelated." }, { "left": "issue:30277", "right": "issue:43638", "accept": false, - "reason": "Both involve distributed loading/training, but the concrete failures and affected models differ." + "reason": "Both involve DeepSpeed, but one is a collective mismatch and the other is a ZeRO-3 empty-index error." }, { "left": "issue:43596", "right": "issue:43638", "accept": true, - "reason": "Same underlying DeepSpeed ZeRO-3 BERT loading bug; 43638 is the narrower non-pretrained BERT variant of the same out-of-bounds IndexError." + "reason": "Same underlying ZeRO-3/BERT IndexError family; 43638 is the narrower non-pretrained Bert variant of 43596." }, { "left": "issue:43039", "right": "issue:45161", "accept": false, - "reason": "Liger Kernel cross-entropy being called and GPT-OSS MoE tensor-parallel failure are unrelated." + "reason": "Liger Kernel cross-entropy and GPT-OSS tensor-parallel failure are unrelated." }, { "left": "issue:43975", "right": "issue:44457", "accept": false, - "reason": "Tokenizer detokenization bug and LoRA merge/save/load mismatch are different problems." + "reason": "Incorrect detokenization and LoRA save/reload mismatch are different bugs." }, { "left": "issue:44292", "right": "issue:44457", "accept": false, - "reason": "Qwen-3-8B NVFP4 runtime error and LoRA merge/save/load mismatch are unrelated." + "reason": "Qwen NVFP4 runtime error and LoRA save/reload mismatch are unrelated." }, { "left": "issue:44841", "right": "issue:46661", "accept": false, - "reason": "Processor failure for Voxtral and remote-code model_class config_class error are different issues." + "reason": "Processor failure for Voxtral and remote-code config_class import error are different problems." }, { "left": "issue:43716", "right": "issue:46869", "accept": false, - "reason": "Mistral dtype mismatch and YOLO multi-GPU device-map failure are unrelated." + "reason": "Different model families and failure modes." }, { "left": "issue:24643", "right": "issue:30277", "accept": false, - "reason": "DeepSpeed training errors, but different symptoms and models; not the same bug." + "reason": "Both are DeepSpeed-related, but the concrete errors and code paths differ." }, { "left": "issue:34928", "right": "issue:44805", "accept": false, - "reason": "Activation-checkpointing/FSDP recompute mismatch and mask shape mismatch are different failures." + "reason": "Activation-checkpointing/FSDP recompute mismatch is not the same as the mask-shape IndexError." }, { "left": "issue:45859", "right": "issue:45902", - "accept": false, - "reason": "Both mention Qwen3_5MoeTextRotaryEmbedding, but one is CPU-offload incompatibility and the other is uninitialized-memory after meta-to-empty materialization." + "accept": true, + "reason": "Same Qwen3.5 MoE rotary-embedding/materialization bug; CPU offload failure and uninitialized inv_freq point to the same root cause." }, { "left": "issue:45685", "right": "issue:46869", "accept": false, - "reason": "MPS histogram implementation error and YOLO multi-GPU device-map failure are unrelated." + "reason": "MPS histogram/int issue and YOLO multi-GPU device-map failure are unrelated." }, { "left": "issue:43638", "right": "issue:45161", "accept": false, - "reason": "BERT ZeRO-3 index error and GPT-OSS tensor-parallel failure are unrelated." + "reason": "ZeRO-3 Bert init IndexError and GPT-OSS tensor-parallel issue are unrelated." }, { "left": "issue:34928", "right": "issue:36331", "accept": false, - "reason": "FSDP activation-checkpointing issue and custom Trainer compute_loss signature breakage are different bugs." + "reason": "FSDP activation-checkpointing mismatch and CustomTrainer compute_loss signature change are unrelated." }, { "left": "issue:43596", "right": "issue:45161", "accept": false, - "reason": "BERT ZeRO-3 loading error and GPT-OSS MoE tensor-parallel failure are unrelated." + "reason": "ZeRO-3/BERT empty-index bug and GPT-OSS tensor-parallel failure are unrelated." }, { "left": "issue:43716", "right": "issue:45507", "accept": false, - "reason": "Different model families and different concrete failures." + "reason": "Mistral dtype mismatch and GraniteMoEHybrid invalid method call are different bugs." }, { "left": "issue:44292", "right": "issue:45237", "accept": false, - "reason": "Qwen NVFP4 runtime error and GPT-OSS AMD GPU issue are not the same bug." + "reason": "Qwen NVFP4 runtime error and GPT-OSS AMD GPU failure are unrelated." }, { "left": "issue:45874", "right": "issue:46294", "accept": false, - "reason": "Gemma4 reverse embedding lookup and Qwen3.5 video chat-template token mismatch are unrelated." + "reason": "Gemma4 reverse-embedding lookup and Qwen3.5 video token mismatch are unrelated." }, { "left": "issue:39290", "right": "issue:45081", "accept": false, - "reason": "Gemma3 config attribute error with vLLM and Mistral regex patch crash are different issues." + "reason": "Gemma3 sliding-window config error and Mistral regex patch crash are different issues." }, { "left": "issue:45800", "right": "issue:46515", "accept": true, - "reason": "Same torch 2.4.1 incompatibility report; 46515 is a later, shorter duplicate of the same version-mismatch problem." + "reason": "These are the same torch 2.4.1 incompatibility issue; 46515 is just the shorter duplicate report." }, { "left": "issue:45823", "right": "issue:45874", "accept": false, - "reason": "Device mismatch in Gemma4 forward and expensive reverse embedding lookup are different issues." + "reason": "Same model family, but device mismatch and expensive reverse embedding lookup are different bugs." }, { "left": "issue:43425", "right": "issue:44292", "accept": false, - "reason": "Generic torch incompatibility and Qwen NVFP4 runtime error are not the same bug." + "reason": "Torch-version incompatibility and Qwen NVFP4 runtime error are unrelated." }, { "left": "issue:34928", "right": "issue:41720", "accept": false, - "reason": "FSDP checkpointing mismatch and Qwen3 auto device-map cuda assert are unrelated." + "reason": "FSDP recompute mismatch and Qwen3 auto device-map cuda assert are different failures." + }, + { + "left": "issue:43828", + "right": "issue:45161", + "accept": false, + "reason": "Phi autocast dtype mismatch and GPT-OSS tensor-parallel failure are unrelated." + }, + { + "left": "issue:43828", + "right": "issue:44292", + "accept": false, + "reason": "Both are runtime errors, but the models and mechanisms differ." + }, + { + "left": "issue:43541", + "right": "issue:43716", + "accept": false, + "reason": "Mixtral grouped_mm tracing error and Mistral dtype mismatch are unrelated." + }, + { + "left": "issue:43716", + "right": "issue:46468", + "accept": false, + "reason": "Mistral dtype mismatch and Nemotron example failure are unrelated." + }, + { + "left": "issue:44292", + "right": "issue:44918", + "accept": false, + "reason": "Qwen NVFP4 runtime error and TRL SFT embedding unpacking are unrelated." + }, + { + "left": "issue:45507", + "right": "issue:45685", + "accept": false, + "reason": "GraniteMoEHybrid invalid method and MPS histogram/int error are unrelated." + }, + { + "left": "issue:43716", + "right": "issue:44292", + "accept": false, + "reason": "Different models and failure modes." + }, + { + "left": "issue:43716", + "right": "issue:45685", + "accept": false, + "reason": "Mistral dtype mismatch and MPS histogram/int issue are unrelated." + }, + { + "left": "issue:44918", + "right": "issue:45161", + "accept": false, + "reason": "TRL SFT embedding unpacking and GPT-OSS tensor-parallel failure are unrelated." + }, + { + "left": "issue:43039", + "right": "issue:44918", + "accept": false, + "reason": "Liger Kernel cross-entropy issue and Qwen3.5/TRl embedding unpacking are unrelated." + }, + { + "left": "issue:41762", + "right": "issue:43638", + "accept": false, + "reason": "Both are ZeRO-3 IndexErrors, but the models and exact code paths differ enough to avoid merging them." + }, + { + "left": "issue:43827", + "right": "issue:44512", + "accept": false, + "reason": "Both are docs regressions, but they concern different removed features and pages." }, { "left": "issue:43828", "right": "issue:45685", "accept": false, - "reason": "Phi autocast dtype mismatch and MPS histogram not implemented for Int are different bugs." + "reason": "Phi autocast dtype mismatch and MPS histogram/int error are unrelated." }, { "left": "issue:43039", "right": "issue:43828", "accept": false, - "reason": "Liger Kernel cross-entropy call and Phi autocast dtype mismatch are unrelated." + "reason": "Liger Kernel cross-entropy and Phi autocast dtype mismatch are unrelated." }, { "left": "issue:30277", "right": "issue:36331", "accept": false, - "reason": "DeepSpeed collective mismatch and custom trainer signature error are different classes of failure." + "reason": "DeepSpeed collective mismatch and compute_loss signature change are unrelated." }, { "left": "issue:34928", "right": "issue:36010", "accept": false, - "reason": "FSDP activation-checkpointing mismatch and GenerationMixin import error are unrelated." + "reason": "FSDP recompute mismatch and GenerationMixin import error are unrelated." }, { "left": "issue:45237", "right": "issue:46015", "accept": false, - "reason": "GPT-OSS AMD GPU failure and outdated RAG docs are unrelated." + "reason": "GPT-OSS AMD GPU failure and RAG docs cleanup are unrelated." }, { "left": "issue:43039", "right": "issue:44805", "accept": false, - "reason": "Liger Kernel cross-entropy call and tensor mask shape mismatch are not the same bug." + "reason": "Liger Kernel cross-entropy and mask-shape IndexError are unrelated." } ] }, "evaluator_result": { "accept": true, - "feedback": "The summary is well grounded in the packet, and the accepted soft edges are supported by strong title/body similarity. The rejection calls are mostly conservative and avoid overclaiming duplicate status for pairs that only share a broad subsystem or error class." + "feedback": "The summary is grounded in the packet and stays conservative about the soft edges. The three accepted duplicate-like pairs are well supported by the titles, and the rejected pairs are generally justified as different bug families or code paths." }, "evaluator_used": true, "retried": false, @@ -585,292 +650,292 @@ "issue:43572|issue:45070", "issue:45800|issue:45823" ], - "prepared_review_unit_hash": "d008c3d18cc418c28e8dea7b0b1fdb508958fc78f637cd44e715da0d2c4879d4", + "prepared_review_unit_hash": "9efb9a731d66e7097e6ff5b62f4e59e1a73167e88f95a8f1bdd5b3190de84c38", "component_budget": { "node_count": 502, "item_count": 502, - "soft_pair_count": 1842, - "serialized_chars": 956182, - "estimated_input_tokens": 239046, - "estimated_eval_tokens": 478348 + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 42, - "serialized_chars": 40045, - "estimated_input_tokens": 10012, - "estimated_eval_tokens": 20280 + "serialized_chars": 40044, + "estimated_input_tokens": 10011, + "estimated_eval_tokens": 20278 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly heterogeneous bug reports with no PRs. The only clear duplicate-looking pair is the v5 AutoTokenizer token-ID regression reports (Granite-specific vs multi-model), which appear to describe the same underlying tokenizer bug.", - "confidence": 0.87, - "canonical_issue_reason": "issue:45920 is the best canonical issue because it describes the broader v5 AutoTokenizer regression affecting multiple model families, making it more representative than the narrower Granite-only report in issue:45812.", + "summary": "This cluster is mostly heterogeneous: one real duplicate-like tokenizer regression pair stands out, and one ZeRO-3 loading pair is plausibly the same bug. The rest look like unrelated model-specific failures or docs issues.", + "confidence": 0.74, + "canonical_issue_reason": "Issue 45920 is the best canonical anchor: it describes the broad v5 AutoTokenizer token-ID regression across multiple models, while 45812 is the narrower Granite-specific report of the same underlying bug.", "canonical_pr_reason": null, - "best_issue_reason": "issue:45920 is the strongest issue to anchor the cluster: it names the regression clearly, spans multiple affected models, and is the most general statement of the underlying bug.", + "best_issue_reason": "45920 is the strongest representative because it captures the widest symptom set for the tokenizer regression and is the clearest issue to attach follow-up fixes to.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43866", "right": "issue:46402", "accept": false, - "reason": "Different models and failures: corrupted checkpoint vs missing text-encoder weight loading." + "reason": "Different failures: corrupted checkpoint vs missing loaded weights in a different model." }, { "left": "issue:45800", "right": "issue:46304", "accept": false, - "reason": "Both mention torch/transformers compatibility, but they are distinct version-specific incompatibilities with different failure modes." + "reason": "Both mention torch/transformers compatibility, but they describe different breakages and code paths." }, { "left": "issue:43296", "right": "issue:46304", "accept": false, - "reason": "PaddleOCR-VL loading in vLLM is unrelated to the distributed build/support regression in USE_DISTRIBUTED=0." + "reason": "Different runtime stacks and symptoms; no clear shared underlying bug." }, { "left": "issue:41762", "right": "issue:43596", - "accept": false, - "reason": "Both hit a ZeRO-3 IndexError, but the model/setup contexts differ enough that this looks like two separate reports of a generic symptom." + "accept": true, + "reason": "Both report the same ZeRO-3 loading path hitting an index-out-of-bounds size-0 error during model initialization." }, { "left": "issue:39290", "right": "issue:41720", "accept": false, - "reason": "Gemma3 config attribute error and Qwen3 auto device-map CUDA assert are unrelated failures." + "reason": "Unrelated failures: a Gemma3/vLLM config attribute error versus a Qwen3 device-mapping CUDA assert." }, { "left": "issue:44589", "right": "issue:46304", "accept": false, - "reason": "Float8 storage lookup failure is unrelated to the distributed build regression." + "reason": "Storage-backend/type lookup failure is unrelated to the distributed build regression in 46304." }, { "left": "issue:43901", "right": "issue:44512", "accept": false, - "reason": "Both are documentation issues, but they concern different docs and different stale content." + "reason": "Both are docs-related, but they target different documentation problems and should not be merged." }, { "left": "issue:45237", "right": "issue:45901", "accept": false, - "reason": "AMD GPU support for GPT-OSS-20B is unrelated to a table-question-answering crash." + "reason": "Completely different subsystems: AMD GPU support versus table-question-answering pipeline crash." }, { "left": "issue:43975", "right": "issue:46015", "accept": false, - "reason": "Incorrect detokenization and outdated RAG examples are unrelated issues." + "reason": "Tokenizer detokenization bug and outdated RAG docs are unrelated." }, { "left": "issue:43039", "right": "issue:44560", "accept": false, - "reason": "Liger Kernel cross-entropy routing and Qwen3-VL video StopIteration are unrelated code paths." + "reason": "Liger kernel cross-entropy routing is unrelated to the video StopIteration failure." }, { "left": "issue:45823", "right": "issue:46294", "accept": false, - "reason": "Device-mapping mismatch for Gemma4 PLE and video/chat-template token mismatch for Qwen3.5 are different bugs." + "reason": "Both are multimodal model issues, but one is device placement and the other is chat-template video token mismatch." }, { "left": "issue:44841", "right": "issue:45823", "accept": false, - "reason": "Processor failure for Voxtral and Gemma4 device mismatch are different multimodal issues." + "reason": "Different models and different failure modes; no shared code-path bug is evident." }, { "left": "issue:43866", "right": "issue:46772", "accept": false, - "reason": "Checkpoint corruption and DiffusionGemma ModelOpt compatibility are unrelated." + "reason": "Checkpoint corruption is not the same as NVFP4/ModelOpt compatibility." }, { "left": "issue:44863", "right": "issue:46015", "accept": false, - "reason": "NemotronH checkpoint loading and outdated RAG docs are unrelated." + "reason": "Model-loading failure and docs cleanup are unrelated." }, { "left": "issue:45020", "right": "issue:46304", "accept": false, - "reason": "Remote-code model breakage and USE_DISTRIBUTED=0 support regression are different broad compatibility problems." + "reason": "Remote-code breakage and USE_DISTRIBUTED=0 support regression are distinct issues." }, { "left": "issue:45812", "right": "issue:45920", "accept": true, - "reason": "Both report the same v5 AutoTokenizer regression causing wrong token IDs across models; 45920 is the broader instance of the same bug." + "reason": "Same underlying v5 AutoTokenizer regression: both report wrong token IDs across models, with 45920 being the broader report." }, { "left": "issue:43866", "right": "issue:44863", "accept": false, - "reason": "Corrupted Ovis checkpoint and NemotronH checkpoint-loading failure are not the same underlying issue." + "reason": "Corrupted checkpoint versus missing checkpoint-loading support; not the same bug." }, { "left": "issue:43872", "right": "issue:46304", "accept": false, - "reason": "bitsandbytes Int8Params incompatibility is unrelated to the distributed build regression." + "reason": "bitsandbytes Int8Params API incompatibility is unrelated to the PyTorch distributed regression." }, { "left": "issue:44292", "right": "issue:46015", "accept": false, - "reason": "Qwen-3-8B-NVFP4 runtime error and RAG docs drift are unrelated." + "reason": "NVFP4 runtime error and RAG docs are unrelated." }, { "left": "issue:39290", "right": "issue:43531", "accept": false, - "reason": "Both mention sliding_window, but they describe different models and different config/behavior failures." + "reason": "Both mention sliding-window behavior, but the affected models and concrete failures differ." }, { "left": "issue:41093", "right": "issue:43638", "accept": false, - "reason": "Both are ZeRO-3-related IndexErrors, but the reported symptoms and model contexts differ." + "reason": "Both are IndexError reports, but one is a mask/tensor shape mismatch and the other is a ZeRO-3 model-init failure." }, { "left": "issue:43792", "right": "issue:45901", "accept": false, - "reason": "Whisper runtime failure and table-question-answering crash are unrelated." + "reason": "Whisper load failure and table-question-answering crash are unrelated." }, { "left": "issue:43854", "right": "issue:43866", "accept": false, - "reason": "Model loading test failure for GLM-4.7-Flash is unrelated to an Ovis checkpoint corruption report." + "reason": "Different model families and different failure modes." }, { "left": "issue:43425", "right": "issue:45070", "accept": false, - "reason": "Torch version incompatibility and a pydantic PretrainedConfig regression are different compatibility issues." + "reason": "Torch version compatibility and pydantic PretrainedConfig regression are separate problems." }, { "left": "issue:43975", "right": "issue:45901", "accept": false, - "reason": "Incorrect detokenization and table-question-answering crash are unrelated." + "reason": "Tokenizer detokenization regression is unrelated to the table QA crash." }, { "left": "issue:43792", "right": "issue:45070", "accept": false, - "reason": "Whisper cannot run and pydantic config regression are unrelated." + "reason": "Whisper runtime failure does not match the pydantic model regression." }, { "left": "issue:43596", "right": "issue:45800", "accept": false, - "reason": "ZeRO-3 loading IndexError and torch/transformers version incompatibility are different failures." + "reason": "ZeRO-3 loading IndexError and torch/transformers incompatibility are not the same bug." }, { "left": "issue:45362", "right": "issue:45464", "accept": false, - "reason": "Qwen3.5 chat crash and streaming chat/completions API failure are different issues." + "reason": "Both involve Qwen3.5, but one is chat crashes and the other is streaming chat/completions API failure." }, { "left": "issue:43541", "right": "issue:46869", "accept": false, - "reason": "Grouped_mm tracing in Mixtral and YOLO multi-GPU device-map failure are unrelated." + "reason": "Grouped_mm tracing failure in Mixtral is unrelated to YOLO multi-GPU device-mapping." }, { "left": "issue:44292", "right": "issue:45901", "accept": false, - "reason": "Qwen NVFP4 runtime error and table QA crash are unrelated." + "reason": "NVFP4 model runtime issue and table QA crash are unrelated." }, { "left": "issue:39290", "right": "issue:45823", "accept": false, - "reason": "Gemma3 sliding_window config error and Gemma4 device mismatch are different bugs." + "reason": "Gemma3/vLLM config error and Gemma4 device-mismatch are different bugs." }, { "left": "issue:45464", "right": "issue:45685", "accept": false, - "reason": "Streaming inference API failure and MPS histogram integer support error are unrelated." + "reason": "Streaming chat API failure is unrelated to the MPS histogram Int error." }, { "left": "issue:36010", "right": "issue:39290", "accept": false, - "reason": "ImportError for GenerationMixin and Gemma3 config attribute error are unrelated." + "reason": "ImportError for GenerationMixin is unrelated to the Gemma3/vLLM config issue." }, { "left": "issue:43541", "right": "issue:45685", "accept": false, - "reason": "Mixtral dynamo tracing failure and MPS histogram integer limitation are unrelated." + "reason": "Different model families and different backend failures." }, { "left": "issue:43541", "right": "issue:43828", "accept": false, - "reason": "Grouped_mm tracing error and autocast dtype mismatch are different runtime problems." + "reason": "Torch dynamo grouped_mm tracing failure and autocast dtype mismatch are distinct issues." }, { "left": "issue:44841", "right": "issue:45084", "accept": false, - "reason": "Processor failure for Voxtral and non-template-node compilation error are unrelated." + "reason": "Processor failure and non-template-node compile error are unrelated." }, { "left": "issue:45507", "right": "issue:46869", "accept": false, - "reason": "GraniteMoEHybrid invalid method call and YOLO multi-GPU device-map failure are unrelated." + "reason": "Invalid method call in GraniteMoEHybrid is not the same as YOLO multi-GPU device-map failure." }, { "left": "issue:43404", "right": "issue:43866", "accept": false, - "reason": "Mistral3 weight-tying bug and corrupted Ovis checkpoint are not the same issue." + "reason": "Weight-tying bug in Mistral3 is unrelated to checkpoint corruption." }, { "left": "issue:41093", "right": "issue:41762", "accept": false, - "reason": "Different IndexError shapes and different model-loading contexts suggest separate bugs." + "reason": "Both are IndexErrors, but the observed failure surfaces and contexts do not match." }, { "left": "issue:45399", "right": "issue:46661", "accept": false, - "reason": "Flash-attn fallback gating and remote-code config_class failure are unrelated." + "reason": "Flash-attn fallback logic and remote-code model_class/config_class handling are unrelated." }, { "left": "issue:43572", "right": "issue:45070", "accept": false, - "reason": "Missing pad_token_idx and pydantic PretrainedConfig field breakage are different config regressions." + "reason": "Missing StableLmConfig field and pydantic PretrainedConfig regression are different configuration bugs." }, { "left": "issue:45800", "right": "issue:45823", "accept": false, - "reason": "Torch/transformers version incompatibility and Gemma4 device mismatch are unrelated." + "reason": "Torch version compatibility and Gemma4 device placement are unrelated." } ] }, "evaluator_result": { "accept": true, - "feedback": "The cluster summary is grounded in the packet, and the only accepted soft-edge pair (issue:45812 vs issue:45920) is well supported by the issue titles as the broader v5 AutoTokenizer regression. The rejected pairs are handled conservatively based on distinct titles/models/failure modes, with no PR-pair mergeability concerns." + "feedback": "The summary is grounded in the packet and stays conservative overall. The canonicalization around 45812/45920 is well supported by the issue titles, and the ZeRO-3 pair is a plausible duplicate-style match given the identical index-out-of-bounds/sized-0 failure pattern during model loading. No soft-edge verdicts appear overstated based on the available evidence." }, "evaluator_used": true, "retried": false, @@ -925,12 +990,12 @@ "issue:45183", "issue:45399", "issue:45464", + "issue:45507", "issue:45632", "issue:45698", "issue:45759", "issue:45800", "issue:45902", - "issue:46165", "issue:46294", "issue:46463", "issue:46468", @@ -956,7 +1021,6 @@ "issue:44479|issue:46294", "issue:46468|issue:46869", "issue:45759|issue:45902", - "issue:43596|issue:46165", "issue:46640|issue:46661", "issue:43716|issue:44918", "issue:39186|issue:44918", @@ -971,232 +1035,230 @@ "issue:44492|issue:44512", "issue:44863|issue:45005", "issue:45183|issue:46463", - "issue:46489|issue:46491" + "issue:46489|issue:46491", + "issue:44918|issue:45507" ], - "prepared_review_unit_hash": "a5524e2b73ec5f077dfb4e8776ab2faa837e7cbeb96ec61989e5e6c7209b0825", + "prepared_review_unit_hash": "311ee8df90ca91f2472bfe4d4dd60b368ea9f3628129752a7841e4f2b4bd6434", "component_budget": { "node_count": 502, "item_count": 502, - "soft_pair_count": 1842, - "serialized_chars": 956182, - "estimated_input_tokens": 239046, - "estimated_eval_tokens": 478348 + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 31, - "serialized_chars": 36884, - "estimated_input_tokens": 9221, - "estimated_eval_tokens": 18698 + "serialized_chars": 36846, + "estimated_input_tokens": 9212, + "estimated_eval_tokens": 18680 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster is mostly heterogeneous and does not form a single duplicate set. Most soft edges are only broad topical similarities, not the same concrete bug. The clearest true duplicate candidate is the shared input-validation/error-message regression between 45183 and 46463.", - "confidence": 0.84, - "canonical_issue_reason": "issue:45183 is the best canonical issue because it describes the broad shared validation/XOR error-message bug across multiple models, which is the most general root-cause framing in the set.", + "summary": "This cluster is heterogeneous: it spans unrelated model-loading, tokenizer, quantization, multimodal, and runtime errors. I do not see a single duplicate issue to collapse the set, and none of the soft pairs look like the same concrete bug or change.", + "confidence": 0.95, + "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "issue:45183 is the strongest representative of the cluster since it captures a reusable code-path bug rather than a single-model symptom.", + "best_issue_reason": "No issue is a strong canonical representative for the whole cluster because the topics diverge substantially. If forced, issue 45698 is one of the clearest self-contained bug reports, but it is not a true cluster anchor.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43296", "right": "issue:43366", "accept": false, - "reason": "Different models and failure modes: PaddleOCR-VL loading vs GGUF gpt-oss architecture support." + "reason": "Different failures: PaddleOCR-VL load error vs GGUF gpt-oss architecture support. Same broad area (model support), but not the same bug." }, { "left": "issue:44387", "right": "issue:45005", "accept": false, - "reason": "One is an int4/OOM regression, the other is tied-weights behavior on translation models; same broad v5 theme but not the same bug." + "reason": "Both mention v5 and weight/quantization-related regressions, but one is int4 CUDA reserved memory OOM and the other is tied weights on translation models. Not the same underlying defect." }, { "left": "issue:44805", "right": "issue:44918", "accept": false, - "reason": "Both are shape-related errors, but one is a mask/index mismatch and the other is Qwen3.5 embedding unpacking with TRL SFT; different code paths." + "reason": "Both involve shape/packing-style errors, but one is a mask-vs-tensor indexing mismatch and the other is Qwen3.5 input embedding unpacking with TRL SFT. Different code paths and symptoms." }, { "left": "issue:43541", "right": "issue:45161", "accept": false, - "reason": "Different MoE problems: torch dynamo grouped_mm tracing vs tensor-parallel support for GPT-OSS." + "reason": "Both are MoE/runtime issues, but one is grouped_mm tracing failure for Mixtral float32 and the other is TP not working for GPT-OSS MoE. Not the same bug." }, { "left": "issue:43828", "right": "issue:45464", "accept": false, - "reason": "Different models and APIs: autocast dtype mismatch in Phi-tiny-MoE vs streaming chat/completions failure in Qwen3.5-0.8B." + "reason": "Different modalities and failures: autocast dtype mismatch in Phi-tiny-MoE vs streaming chat/completions failure on Qwen3.5-0.8B." }, { "left": "issue:43366", "right": "issue:45464", "accept": false, - "reason": "Unrelated issues: GGUF architecture support versus Qwen3.5 streaming inference." + "reason": "GGUF gpt-oss architecture support vs Qwen3.5 streaming inference. Shared inference context only, not the same issue." }, { "left": "issue:44291", "right": "issue:45070", "accept": false, - "reason": "Both are compatibility regressions, but one is init_empty_weights handling and the other is a PretrainedConfig pydantic-field break." + "reason": "Both are v5 compatibility regressions, but one is init_empty_weights/_is_hf_initialized TypeError and the other is PretrainedConfig field breakage in pydantic models." }, { "left": "issue:43531", "right": "issue:45070", "accept": false, - "reason": "Qwen3-MoE sliding_window behavior is unrelated to the PretrainedConfig/pydantic issue." + "reason": "Sliding-window behavior in Qwen3-MoE is unrelated to the pydantic PretrainedConfig regression." }, { "left": "issue:41093", "right": "issue:43596", "accept": false, - "reason": "Both raise index errors, but the tensor-mask mismatch and the deepspeed zero3 BertModel out-of-bounds are distinct bugs." + "reason": "Both are IndexError reports, but the first is a mask-length mismatch and the second is zero-sized tensor indexing during deepspeed zero3 init. Different causes." }, { "left": "issue:45020", "right": "issue:45800", "accept": false, - "reason": "One is a remote_code breakage, the other is a torch 2.4.1 / transformers 5.8.0 incompatibility; not the same code defect." + "reason": "Both are broad compatibility complaints, but one is remote_code breakage and the other is torch 2.4.1 vs transformers 5.8.0 incompatibility. Too generic to merge." }, { "left": "issue:43716", "right": "issue:46294", "accept": false, - "reason": "Different multimodal paths: image-preprocessor dtype mismatch vs video feature/token mismatch in apply_chat_template." + "reason": "Both touch Qwen3.5/multimodal input handling, but one is image-preprocessor dtype mismatch and the other is video feature/token count mismatch." }, { "left": "issue:43054", "right": "issue:43994", "accept": false, - "reason": "Same model family, but the reports differ: worse text embeddings vs nonsensical results with AutoModel/pipeline." + "reason": "Both are SigLIP/SigLIP2 quality regressions, but one is text embedding quality and the other is nonsensical AutoModel/pipeline output. Related family, different bug." }, { "left": "issue:44479", "right": "issue:46294", "accept": false, - "reason": "Both touch video input, but one is a broad v5.3.0 regression and the other is a specific apply_chat_template token/feature mismatch; too coarse to merge." + "reason": "Both involve Qwen3.5 multimodal video, but one is a v5.3.0 video-input regression across several models and the other is a chat template feature/token mismatch. Not the same concrete defect." }, { "left": "issue:46468", "right": "issue:46869", "accept": false, - "reason": "Completely unrelated: Nemotron example breakage versus YOLO multi-GPU device-map auto failure." + "reason": "Different model stacks and failures: broken Nemotron example vs YOLO multi-GPU device_map auto." }, { "left": "issue:45759", "right": "issue:45902", "accept": false, - "reason": "Different Qwen-related bugs: config unwrapping in from_config versus uninitialized rotary embedding memory after meta->to_empty." - }, - { - "left": "issue:43596", - "right": "issue:46165", - "accept": false, - "reason": "Different root causes and workflows: deepspeed zero3/BertModel indexing versus trust_remote_code resume behavior." + "reason": "Both concern Qwen 3.5/3.6 internals, but one is from_config text_config unwrapping and the other is uninitialized rotary embedding memory after to_empty materialization." }, { "left": "issue:46640", "right": "issue:46661", "accept": false, - "reason": "Flash Attention fallback behavior and remote-code model_class/config_class handling are unrelated." + "reason": "Flash Attention 2 issue vs remote-code model_class.config_class AttributeError. No shared code-path." }, { "left": "issue:43716", "right": "issue:44918", "accept": false, - "reason": "Different failures: multimodal dtype mismatch versus TRL SFT unpacking of Qwen3.5 embeddings." + "reason": "Different multimodal/training failures: image preprocessor dtype mismatch vs TRL SFT unpacking of Qwen3.5 embeddings." }, { "left": "issue:39186", "right": "issue:44918", "accept": false, - "reason": "FSDP weight dimensionality error is unrelated to Qwen3.5 embedding unpacking." + "reason": "FSDP weight-dimension RuntimeError is unrelated to Qwen3.5 embedding unpacking." }, { "left": "issue:40990", "right": "issue:44960", "accept": false, - "reason": "Extremely high perplexity on gpt-oss-20b and a generic GLM5 issue do not describe the same bug." + "reason": "High perplexity on gpt-oss-20b and a generic GLM5 issue are unrelated and at very different abstraction levels." }, { "left": "issue:43645", "right": "issue:45399", "accept": false, - "reason": "Jupyter custom-model initialization breakage and flash-attn2 fallback gating are different issues." + "reason": "Jupyter custom-model initialization breakage is unrelated to flash-attn2 fallback gating." }, { "left": "issue:45632", "right": "issue:45698", "accept": false, - "reason": "Both involve remote-code/custom-module loading, but one is a cache-path collision for local models and the other is wrong custom-module loading after save_pretrained; related area, not clearly the same bug." + "reason": "Both involve remote-code/custom-module loading, but one is cache-path collision for local models sharing a leaf name and the other is loading the wrong custom module after save_pretrained. Similar area, not the same bug." }, { "left": "issue:43404", "right": "issue:45005", "accept": false, - "reason": "Both mention tied weights, but they concern different models and different regressions." + "reason": "Mistral3 tied lm_head weight bug vs translation-model tied weights in v5. Different model families and failure modes." }, { "left": "issue:42915", "right": "issue:45399", "accept": false, - "reason": "FP8 quantization support failure and flash-attn2 fallback checks are separate problems." + "reason": "Qwen3-MoE fine-grained FP8 config failure is unrelated to flash-attn2 fallback blocking." }, { "left": "issue:43873", "right": "issue:45005", "accept": false, - "reason": "Quantization offloading behavior is unrelated to translation-model tied weights." + "reason": "Quantization offloading behavior vs tied-weights regression. Both are v5-era infra issues, but not the same defect." }, { "left": "issue:44530", "right": "issue:45005", "accept": false, - "reason": "PagedAttentionCache linear_attention crash is a different subsystem and bug from tied weights." + "reason": "PagedAttentionCache linear_attention crash on Qwen3.5 is unrelated to tied weights on translation models." }, { "left": "issue:44360", "right": "issue:44512", "accept": false, - "reason": "A DSA indexer ReLU discussion is unrelated to a documentation typo." + "reason": "DSA indexer ReLU discussion is unrelated to docs mentioning a removed run command." }, { "left": "issue:44492", "right": "issue:44512", "accept": false, - "reason": "Cache-strategy typo and removed-run-command docs are different documentation issues." + "reason": "Cache-strategy typo and removed docs command are both documentation-ish, but not the same issue." }, { "left": "issue:44863", "right": "issue:45005", "accept": false, - "reason": "NemotronH checkpoint-loading failure is unrelated to tied-weight handling in translation models." + "reason": "NemotronH checkpoint-loading failure is unrelated to translation-model tied weights." }, { "left": "issue:45183", "right": "issue:46463", - "accept": true, - "reason": "These look like the same validation/error-message bug: a shared input-validation path producing wrong or misleading errors, with 46463 appearing to be the T5-specific manifestation of 45183." + "accept": false, + "reason": "Both are validation/error-message issues, but one is XOR validation for input_ids/inputs_embeds across models and the other is decoder input validation messaging for T5. Similar pattern, different target and likely different fix." }, { "left": "issue:46489", "right": "issue:46491", "accept": false, - "reason": "Both are tokenizer regressions, but one is DeepSeek-Coder v1 incorrect output and the other is a CodeLlama leading-space round-trip issue; not the same bug." + "reason": "Both are tokenizer regressions, but they affect different model families and different exact round-trip behaviors. Too broad to treat as one duplicate." + }, + { + "left": "issue:44918", + "right": "issue:45507", + "accept": false, + "reason": "Qwen3.5 embedding unpacking with TRL SFT is unrelated to GraniteMoEHybrid calling an invalid method." } ] }, - "evaluator_result": { - "accept": true, - "feedback": "The summary is well grounded in the packet and stays conservative on the soft edges. The only accepted pair, 45183/46463, is phrased cautiously and matches the shared validation/error-message theme in the titles. The rest of the cluster is summarized as heterogeneous without overclaiming duplicate status." - }, - "evaluator_used": true, + "evaluator_result": null, + "evaluator_used": false, "retried": false, - "accepted_nontrivial_soft_edge": true, + "accepted_nontrivial_soft_edge": false, "error_kind": null, "error_message": null }, @@ -1243,16 +1305,16 @@ "issue:44534", "issue:44589", "issue:44910", - "issue:44918", "issue:44928", "issue:44960", "issue:44991", "issue:45020", "issue:45399", - "issue:45507", "issue:45520", "issue:45663", + "issue:45685", "issue:45701", + "issue:45823", "issue:45859", "issue:45902", "issue:45920", @@ -1263,7 +1325,6 @@ "issue:46932" ], "soft_pairs": [ - "issue:44918|issue:45507", "issue:43975|issue:44292", "issue:43950|issue:44534", "issue:39692|issue:43054", @@ -1293,228 +1354,229 @@ "issue:44991|issue:45399", "issue:43824|issue:43872", "issue:43782|issue:44960", - "issue:43881|issue:43883" + "issue:43881|issue:43883", + "issue:45685|issue:45823" ], - "prepared_review_unit_hash": "adfd18fb982fa00d968c8cc5da62b2b00964a9eaf5088233e956ee62707fd53a", + "prepared_review_unit_hash": "7b75f5370cfc9d858b6aeeb348c9e27af4c72107f76aaf9e1736e51305a7ce83", "component_budget": { "node_count": 502, "item_count": 502, - "soft_pair_count": 1842, - "serialized_chars": 956182, - "estimated_input_tokens": 239046, - "estimated_eval_tokens": 478348 + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 31, - "serialized_chars": 37088, - "estimated_input_tokens": 9272, - "estimated_eval_tokens": 18800 + "serialized_chars": 37127, + "estimated_input_tokens": 9282, + "estimated_eval_tokens": 18820 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly heterogeneous Transformers bug reports, but there is a small duplicate core around the v5 non-persistent-buffer regression and a separate tokenizer-regression pair. Most other soft links are only broad subsystem similarity, not the same concrete bug.", - "confidence": 0.69, - "canonical_issue_reason": "#43950 is the clearest anchor for the non-persistent-buffer regression: it states the root corruption bug directly, and nearby issues like #43644, #44534, and #45902 read as manifestations of the same loading/materialization problem.", + "summary": "Mostly a heterogeneous batch with a few clear duplicate pairs: non-persistent buffer corruption in v5, SigLIP2 quantized embedder dtype casting, and cache length return-type regressions.", + "confidence": 0.78, + "canonical_issue_reason": "43950 is the best anchor for the non-persistent-buffer regression cluster: it states the concrete v5 corruption bug clearly, and the nearby reports 43644 and 44534 are the same issue phrased more tersely.", "canonical_pr_reason": null, - "best_issue_reason": "#43950 is the best representative issue overall for the accepted duplicate subset because it describes the core regression cleanly and is broad enough to subsume the closely related buffer-corruption reports; it does not represent the unrelated model-specific issues in the cluster.", + "best_issue_reason": "43950 is the most canonical issue overall because it is the most specific, technical description of a real regression and is supported by multiple near-identical reports.", "best_pr_reason": null, "soft_edge_verdicts": [ - { - "left": "issue:44918", - "right": "issue:45507", - "accept": false, - "reason": "Different failures: Qwen3.5 input embedding unpacking vs GraniteMoEHybrid calling an invalid method." - }, { "left": "issue:43975", "right": "issue:44292", "accept": false, - "reason": "Both are v5-era model issues, but one is a DeepSeek-Coder detokenization bug and the other is an NVFP4 loading failure; not the same bug." + "reason": "Different models and failure modes: tokenizer detokenization vs loading a Qwen NVFP4 model." }, { "left": "issue:43950", "right": "issue:44534", "accept": true, - "reason": "Same underlying regression: non-persistent buffers are being corrupted/filled with junk in v5." + "reason": "Same v5 regression corrupting non-persistent buffers; 44534 is just a shorter restatement." }, { "left": "issue:39692", "right": "issue:43054", "accept": false, - "reason": "Doc-example model/processor mismatch and quantization failure is unrelated to SigLIP2 embedding quality." + "reason": "Docs-example model/processor mismatch and quantization errors are not the same as a SigLIP2 text-embedding quality regression." }, { "left": "issue:43644", "right": "issue:43950", "accept": true, - "reason": "Same non-persistent-buffer v5 regression, phrased as junk-filled buffers vs silent corruption." + "reason": "Same non-persistent-buffer junk/corruption bug in Transformers 5.x." }, { "left": "issue:36331", "right": "issue:41762", "accept": false, - "reason": "Different trainer/API mismatch vs DeepSpeed ZeRO-3 loading crash." + "reason": "Unrelated training API signature issue vs DeepSpeed ZeRO-3 loading crash." }, { "left": "issue:43872", "right": "issue:44589", "accept": false, - "reason": "bitsandbytes parameter ctor incompatibility is unrelated to missing Float8 storage handling." + "reason": "Both are quantization/loading problems, but the error types and code paths are different." }, { "left": "issue:46899", "right": "issue:46932", "accept": true, - "reason": "Near-identical Gemma4 quantized vision/audio embedder dtype-casting bug; #46932 explicitly references the same fix family." + "reason": "Same Gemma4 quantized vision/audio embedder dtype-casting bug; 46932 explicitly calls it out as the same issue." }, { "left": "issue:43828", "right": "issue:44928", "accept": false, - "reason": "Different Qwen3.5 issues: autocast dtype mismatch vs RLHF gradient explosion from 3D position_ids/SDPA fallback." + "reason": "Different models and different root causes: autocast dtype mismatch vs 3D position_ids causing flash-attn/SDPA issues." }, { "left": "issue:45920", "right": "issue:46489", - "accept": true, - "reason": "Both are tokenizer-output regressions in v5; #46489 looks like a DeepSeek-Coder v1 edge case of the broader wrong-token-IDs bug." + "accept": false, + "reason": "Both concern tokenizer correctness, but 46489 is a model-specific follow-up and not clearly the same bug as the broader multi-model regression." }, { "left": "issue:45987", "right": "issue:46858", "accept": true, - "reason": "Same StaticCache.get_seq_length contract bug: returns a tensor instead of an int, causing downstream arithmetic failures." + "reason": "Same cache-seq-length contract bug: tensor return value instead of int, with 46858 being a backend-specific manifestation." }, { "left": "issue:43054", "right": "issue:45701", "accept": false, - "reason": "SigLIP2 text-embedding quality regression is not the same as a general tokenization-change report." + "reason": "SigLIP2 text-embedding quality issue is unrelated to a generic tokenizer-change report." }, { "left": "issue:38175", "right": "issue:43054", "accept": false, - "reason": "Unexpected zero probabilities and worse text embeddings are related to SigLIP2, but they describe different symptoms and likely different causes." + "reason": "Both are SigLIP2-related, but zero probabilities and degraded text embeddings are not clearly the same underlying bug." }, { "left": "issue:43950", "right": "issue:45902", - "accept": true, - "reason": "Both point to the same non-persistent-buffer/materialization problem; #45902 is a model-specific manifestation of uninitialized buffer data." + "accept": false, + "reason": "Shared theme of non-persistent buffers, but 45902 is a specific materialization/uninitialized-memory bug rather than the same from_pretrained corruption regression." }, { "left": "issue:41553", "right": "issue:43883", "accept": false, - "reason": "Bad AutoTokenizer error handling for Voxtral is unrelated to Molmo's missing all_tied_weights_keys attribute." + "reason": "Bad AutoTokenizer error message for Voxtral is unrelated to Molmo tied-weights loading failure." }, { "left": "issue:43054", "right": "issue:43493", - "accept": false, - "reason": "SigLIP2 embedding degradation vs HF/JAX implementation discrepancy are related only at a high level, not the same concrete bug." + "accept": true, + "reason": "Both point to the same SigLIP2 implementation discrepancy; the JAX-vs-HF mismatch plausibly explains the degraded text embeddings." }, { "left": "issue:44910", "right": "issue:45859", "accept": false, - "reason": "Both involve Qwen3.5, but one is a flash-attn packed-sequence misread and the other is CPU-offload incompatibility in rotary embedding." + "reason": "Different Qwen3.5 MoE issues: flash-attn packed-sequence misinterpretation vs CPU-offload incompatibility in rotary embedding." }, { "left": "issue:42947", "right": "issue:44387", "accept": false, - "reason": "Gradient checkpointing/LoRA behavior is not the same as int4 quantization causing higher reserved CUDA memory." + "reason": "Both are memory-related, but gradient checkpointing/LoRA inefficiency is not the same bug as int4 reserved-memory OOM." }, { "left": "issue:45663", "right": "issue:45902", "accept": false, - "reason": "Gemma4 FSDP2 shared_kv_states rebuilding is unrelated to the Qwen3_5Moe non-persistent-buffer memory read." + "reason": "Different Gemma4/FSDP and buffer-materialization problems with no clear shared code-path." }, { "left": "issue:41553", "right": "issue:43881", "accept": false, - "reason": "Different model-loading problems for Voxtral tokenizer error messaging vs glm-4v-9b load failure." + "reason": "Different models and failures: Voxtral tokenizer loading vs glm-4v-9b loading." }, { "left": "issue:44403", "right": "issue:44451", "accept": false, - "reason": "Generic loading noise vs a specific ScandiBERT loading failure are not the same bug." + "reason": "Generic loading noise is not the same underlying problem as a specific model load failure." }, { "left": "issue:43975", "right": "issue:45520", "accept": false, - "reason": "Tokenizer detokenization regression is unrelated to a flash_attn import error on Python 3.13." + "reason": "Tokenizer detokenization bug and flash-attn import KeyError are unrelated." }, { "left": "issue:44403", "right": "issue:44488", "accept": false, - "reason": "Same broad loading area, but different models and likely different failure modes." + "reason": "Again, generic loader noise vs a specific Sleng-BERT load failure." }, { "left": "issue:24643", "right": "issue:28530", "accept": false, - "reason": "DeepSpeed weight-dimension training error and early-stopping metric selection are unrelated trainer issues." + "reason": "Completely different training errors: DeepSpeed weight shape vs early-stopping metric selection." }, { "left": "issue:42947", "right": "issue:43856", "accept": false, - "reason": "LoRA gradient checkpointing behavior is unrelated to Qwen3 MoE memory usage." + "reason": "Memory usage symptoms are too broad; LoRA checkpointing inefficiency is not the same as Qwen3 MoE training memory behavior." }, { "left": "issue:41762", "right": "issue:42915", "accept": false, - "reason": "Gemma3 ZeRO-3 loading IndexError and Qwen3Moe FP8 failure are different model/backend bugs." + "reason": "Different model families and different failure mechanisms." }, { "left": "issue:43643", "right": "issue:45020", "accept": false, - "reason": "Missing fields from AutoConfig with trust_remote_code is a specific config-loading bug, not the broader remote_code breakage claim." + "reason": "43643 is a specific missing-fields bug in AutoConfig with trust_remote_code; 45020 is a broad remote_code regression report, not clearly the same underlying issue." }, { "left": "issue:44991", "right": "issue:45399", "accept": false, - "reason": "Tokenizer loading failure for one model is unrelated to flash-attn2 fallback gating." + "reason": "Tokenizer loading failure and flash-attn fallback logic are unrelated." }, { "left": "issue:43824", "right": "issue:43872", "accept": false, - "reason": "Missing Qwen2_5_VL import and bitsandbytes Int8Params ctor incompatibility are different subsystems." + "reason": "Importing a missing VL class is unrelated to bitsandbytes Int8Params constructor incompatibility." }, { "left": "issue:43782", "right": "issue:44960", "accept": false, - "reason": "Qwen3VL weight_only loading error and a GLM5 issue are different model-loading failures." + "reason": "Qwen3VL weight_only loading failure and a GLM5 issue are not the same bug." }, { "left": "issue:43881", "right": "issue:43883", "accept": false, - "reason": "glm-4v-9b load failure and Molmo missing all_tied_weights_keys are unrelated model-specific issues." + "reason": "Different model loading failures with different missing components." + }, + { + "left": "issue:45685", + "right": "issue:45823", + "accept": false, + "reason": "MPS histogram-int support error and Gemma4 device mismatch are distinct backend/model issues." } ] }, "evaluator_result": { "accept": true, - "feedback": "The summary is grounded in the packet and stays mostly conservative. The accepted duplicate cores are plausibly mergeable into single bugfix PRs, especially the non-persistent-buffer regression and the tokenizer-regression follow-up, and the rejected pairs are appropriately separated by concrete failure mode/model differences. Canonical issue choice #43950 is also well supported by the titles." + "feedback": "The summary is grounded in the packet and the accepted soft-edge calls are reasonably conservative. The main duplicate clusters called out are supported by the titles: non-persistent-buffer corruption (43950/43644/44534), SigLIP2-related discrepancy/casting issues (43054/43493 and 46899/46932), and cache-length return-type regressions (45987/46858). Rejections are also appropriately cautious." }, "evaluator_used": true, "retried": false, @@ -1572,12 +1634,12 @@ "issue:45399", "issue:45488", "issue:45588", - "issue:45685", "issue:45715", "issue:45759", "issue:45823", "issue:45923", "issue:46315", + "issue:46402", "issue:46489", "issue:46661", "issue:46869", @@ -1585,7 +1647,6 @@ "issue:46897" ], "soft_pairs": [ - "issue:45685|issue:45823", "issue:45081|issue:45399", "issue:45020|issue:45823", "issue:44661|issue:45020", @@ -1602,8 +1663,8 @@ "issue:43872|issue:44291", "issue:45020|issue:46661", "issue:43792|issue:44220", - "issue:41553|issue:44991", "issue:43257|issue:45305", + "issue:41553|issue:44991", "issue:36331|issue:41093", "issue:42947|issue:45923", "issue:43854|issue:44863", @@ -1613,198 +1674,223 @@ "issue:41093|issue:44805", "issue:44479|issue:44560", "issue:24643|issue:30064", - "issue:44530|issue:44863" + "issue:44530|issue:44863", + "issue:44863|issue:46402" ], - "prepared_review_unit_hash": "313a6359b301a90947df1085c43861b4f619a1c274a536320ba6d3bd0780b2d9", + "prepared_review_unit_hash": "e7d52ec639e9bf6032ed3a883fb19eef029eac328fc1110dcf078e9142c60bfa", "component_budget": { "node_count": 502, "item_count": 502, - "soft_pair_count": 1842, - "serialized_chars": 956182, - "estimated_input_tokens": 239046, - "estimated_eval_tokens": 478348 + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 29, - "serialized_chars": 37092, - "estimated_input_tokens": 9273, - "estimated_eval_tokens": 18802 + "serialized_chars": 37086, + "estimated_input_tokens": 9272, + "estimated_eval_tokens": 18800 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Mostly a heterogeneous set of Transformers regressions, but a few soft pairs are clear duplicates: docs cleanup for removed v5 pipelines, DeepSpeed ZeRO-3 weight loading on Qwen3-family models, mask-shape errors, and a couple of media/preprocessing regressions.", - "confidence": 0.71, - "canonical_issue_reason": "Issue 45020 is the broadest, most representative anchor in this set: it captures the general class of recent model-loading breakages around remote_code, which several other issues in the cluster resemble at a narrower level.", + "summary": "This cluster is mostly heterogeneous, with a few clear duplicate subgroups: docs for removed pipeline tasks, the `_is_hf_initialized` / `init_empty_weights` loading regression, a mask-shape IndexError, and a pair of nearly identical training-loss double-shift reports.", + "confidence": 0.84, + "canonical_issue_reason": "No single cluster-wide canonical issue fits the whole set; if choosing the clearest duplicate subgroup, issue 43827 is the best root-cause report for the v5 docs/pipeline-task removal regression.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 45020 is the best global issue candidate because it is the most umbrella-like regression in the cluster and is central to the broader model-loading/tokenizer breakage theme.", + "best_issue_reason": "Issue 43827 is the strongest issue candidate because it states the underlying docs regression precisely and matches a near-identical follow-up report (44509).", "best_pr_reason": null, "soft_edge_verdicts": [ - { - "left": "issue:45685", - "right": "issue:45823", - "accept": false, - "reason": "Different failures and code paths: MPS histogram/int handling vs device_map=\"auto\" device placement mismatch." - }, { "left": "issue:45081", "right": "issue:45399", "accept": false, - "reason": "Both touch attention/tokenizer-adjacent code, but one is a Mistral regex patch crash and the other is flash-attn2 fallback gating; not the same bug." + "reason": "Different failures: tokenizer regex crash vs flash-attn2 fallback gating; same broad area, not the same bug." }, { "left": "issue:45020", "right": "issue:45823", "accept": false, - "reason": "Generic remote_code model-loading breakage vs a specific Gemma4 PLE device mismatch during forward." + "reason": "`remote_code` compatibility regression is unrelated to a Gemma4 device-map/device placement mismatch." }, { "left": "issue:44661", "right": "issue:45020", "accept": false, - "reason": "Tokenizer mapping registration failure is a different concrete problem from the remote_code loading regressions in 45020." + "reason": "`TOKENIZER_MAPPING_NAMES` handling is a specific tokenizer registration bug, not the same as the broader remote-code regression." }, { "left": "issue:42491", "right": "issue:44368", "accept": false, - "reason": "Checkpoint compatibility across hf4/hf5 for qwen3_moe is not the same as the LoRA tie_word_embeddings warning during fine-tuning." + "reason": "Qwen3 MoE hf4\u2192hf5 LoRA incompatibility and a tie_word_embeddings warning are different issues." }, { "left": "issue:43828", "right": "issue:45823", "accept": false, - "reason": "Autocast dtype mismatch and device_map device mismatch are distinct runtime bugs." + "reason": "Autocast dtype mismatch on Phi-tiny-MoE is a different code-path from the Gemma4 device-map issue." }, { "left": "issue:45020", "right": "issue:45356", "accept": false, - "reason": "Different subsystems: remote_code loading regressions vs Kimi tokenizer codec handling / misleading warning." + "reason": "A broad remote-code regression is not the same as the Kimi-K2.5 tokenizer codec regression." }, { "left": "issue:45715", "right": "issue:46315", "accept": false, - "reason": "Both are tokenizer performance regressions, but convert_ids_to_tokens and convert_tokens_to_ids are separate code paths with different fixes." + "reason": "Both are tokenizer performance regressions, but they hit different methods and symptoms (`convert_ids_to_tokens` loop rebuild vs `convert_tokens_to_ids` slowdown)." }, { "left": "issue:45488", "right": "issue:46489", "accept": false, - "reason": "Both are tokenizer regressions, but one is a LlamaTokenizer pre-tokenizer override and the other is a DeepSeek-Coder output bug; not the same concrete defect." + "reason": "Both are DeepSeek tokenizer regressions, but the titles point to different affected families and one is framed as a gap in a prior fix; not clearly the same concrete bug." }, { "left": "issue:43827", "right": "issue:44509", "accept": true, - "reason": "Same underlying docs cleanup: both say the v5 docs still reference removed pipeline tasks after the pipeline removals." + "reason": "Both report the same v5 docs regression: obsolete pipeline task references after pipeline removals." }, { "left": "issue:43872", "right": "issue:46869", "accept": false, - "reason": "bitsandbytes constructor incompatibility is unrelated to YOLO failing with multi-GPU device_map auto." + "reason": "Bitsandbytes `_is_hf_initialized` incompatibility is unrelated to a YOLO multi-GPU `device_map='auto'` failure." }, { "left": "issue:46894", "right": "issue:46897", "accept": true, - "reason": "Same training-loss double-shift bug pattern affecting different model families; likely the same loss computation fix." + "reason": "Near-identical training-loss double-shift reports with the same failure pattern; only the example model differs." }, { "left": "issue:45759", "right": "issue:46661", "accept": false, - "reason": "Both concern AutoModel factories, but one is text_config unwrapping in from_config and the other is config_class on remote-code models in from_pretrained." + "reason": "`text_config` unwrapping for composite Qwen configs is not the same as the remote-code `config_class` AttributeError." }, { "left": "issue:43404", "right": "issue:43950", "accept": false, - "reason": "Tied lm_head weights in Mistral3 and silent corruption of non-persistent buffers are unrelated initialization/loading bugs." + "reason": "`lm_head` tying in Mistral3 and non-persistent buffer corruption are different model-loading bugs." }, { "left": "issue:43872", "right": "issue:44291", - "accept": false, - "reason": "Both involve unexpected constructor arguments, but they are different libraries and different entry points (_is_hf_initialized vs Int8Params)." + "accept": true, + "reason": "These describe the same `_is_hf_initialized` / `init_empty_weights` incompatibility; 43872 is the concrete bitsandbytes symptom and 44291 names the root cause." }, { - "left": "issue:43940", - "right": "issue:45313", - "accept": true, - "reason": "Both are DeepSpeed ZeRO-3 weight-loading failures in the Qwen3 family, pointing to the same missing-weights loading path." + "left": "issue:45020", + "right": "issue:46661", + "accept": false, + "reason": "The remote-code umbrella issue is broader; 46661 is one specific `config_class` failure mode, not enough to merge as a duplicate." }, { - "left": "issue:41553", - "right": "issue:44991", + "left": "issue:43792", + "right": "issue:44220", "accept": false, - "reason": "Voxtral AutoTokenizer error messaging and est-roberta tokenizer loading failure are different tokenizer issues." + "reason": "A model-level Whisper failure and a low-level fbank feature-extraction issue are related but not clearly the same bug." }, { "left": "issue:43257", "right": "issue:45305", "accept": false, - "reason": "One is Qwen3 MoE weights not converted under accelerate+deepspeed; the other is gradient averaging under GAS with DeepSpeed. Different bugs." + "reason": "DeepSpeed ZeRO-3 weight loading and DeepSpeed gradient averaging are different problems." + }, + { + "left": "issue:41553", + "right": "issue:44991", + "accept": false, + "reason": "Bad error messaging for Voxtral tokenizer loading and a tokenizer load failure for est-roberta are different regressions." }, { "left": "issue:36331", "right": "issue:41093", "accept": false, - "reason": "Unexpected compute_loss kwarg and mask/tensor length mismatch are different failure modes." + "reason": "CustomTrainer API breakage and a mask/tensor shape mismatch are unrelated." }, { "left": "issue:42947", "right": "issue:45923", "accept": false, - "reason": "PEFT LoRA gradient checkpointing being ineffective is not the same as a missing supports_gradient_checkpointing flag on a remote-code variant." + "reason": "Gradient checkpointing ineffective with PEFT LoRA is not the same as a missing `supports_gradient_checkpointing` flag on a trust_remote_code variant." }, { "left": "issue:43854", "right": "issue:44863", "accept": false, - "reason": "Different model-loading failures for different model families; no shared concrete code-path is evident." + "reason": "Different model-loading failures for different model families; no shared concrete bug is evident." + }, + { + "left": "issue:43940", + "right": "issue:45313", + "accept": false, + "reason": "Both mention ZeRO-3 weight loading, but one is Qwen3-Next all-params-missing and the other is Qwen3.5 language_model weights missing; too different to collapse." + }, + { + "left": "issue:41553", + "right": "issue:44488", + "accept": false, + "reason": "Voxtral tokenizer error messaging and sleng-bert load failure are unrelated." }, { "left": "issue:44928", "right": "issue:45588", "accept": false, - "reason": "RLHF NaN from 3D position_ids / SDPA fallback is unrelated to flash_attention sink-less handling of s_aux=None." + "reason": "RLHF gradient explosion from 3D position_ids / SDPA fallback is a different bug from the flash-attention `s_aux=None` crash." }, { "left": "issue:41093", "right": "issue:44805", "accept": true, - "reason": "Same shape-mismatch IndexError pattern; the titles differ only in the concrete sizes." + "reason": "Same mask-shape mismatch IndexError; one report gives concrete dimensions, the other is the same error class more generally." }, { "left": "issue:44479", "right": "issue:44560", - "accept": true, - "reason": "Both describe a v5.3 video-input regression in the Qwen VL family, likely the same preprocessing/video loader path." + "accept": false, + "reason": "Both are video-related Qwen regressions, but the reported failures differ (`video input` regression vs `StopIteration` in embedding)." }, { "left": "issue:24643", "right": "issue:30064", "accept": false, - "reason": "DeepSpeed training weight-shape error and void segmentation map processing are unrelated." + "reason": "DeepSpeed training `weight must be 2-D` and void segmentation-map processing are unrelated." }, { "left": "issue:44530", "right": "issue:44863", "accept": false, - "reason": "PagedAttentionCache linear_attention crash and NemotronH checkpoint loading are different defects." + "reason": "PagedAttentionCache `linear_attention` crash and NemotronH checkpoint loading are different failures." + }, + { + "left": "issue:43872", + "right": "issue:44291", + "accept": true, + "reason": "These describe the same `_is_hf_initialized` / `init_empty_weights` incompatibility; 43872 is the concrete bitsandbytes symptom and 44291 names the root cause." + }, + { + "left": "issue:44863", + "right": "issue:46402", + "accept": false, + "reason": "NemotronH checkpoint loading and Sam3Model text-encoder weights not loading are separate model-specific regressions." } ] }, "evaluator_result": { "accept": true, - "feedback": "The summary is grounded in the packet and stays mostly conservative. The accepted soft pairs are plausibly mergeable duplicates based on the issue titles and similarity signals, and the rejected pairs are not overstated. Minor wording like \"a couple of media/preprocessing regressions\" is a bit loose, but not enough to make the assessment ungrounded." + "feedback": "The summary is grounded in the packet and stays conservative: it treats the cluster as mostly heterogeneous and only marks the clearest duplicate subgroups as matches. The accepted soft pairs are well-supported by the titles (`43827`/`44509`, `43872`/`44291`, `41093`/`44805`, and the two double-shift reports). No obvious overstatement of evidence." }, "evaluator_used": true, "retried": false, @@ -1838,6 +1924,7 @@ "issue:43596", "issue:43632", "issue:43638", + "issue:43644", "issue:43645", "issue:43646", "issue:43716", @@ -1848,13 +1935,13 @@ "issue:43957", "issue:43994", "issue:44155", + "issue:44534", "issue:44560", "issue:44589", "issue:44661", "issue:44683", "issue:44756", "issue:44805", - "issue:44863", "issue:44877", "issue:44910", "issue:44918", @@ -1868,14 +1955,12 @@ "issue:45823", "issue:45834", "issue:45902", - "issue:46402", "issue:46489", "issue:46515", "issue:46858", "issue:46899" ], "soft_pairs": [ - "issue:44863|issue:46402", "issue:43366|issue:43828", "issue:43596|issue:46515", "issue:45325|issue:45902", @@ -1909,252 +1994,253 @@ "issue:42491|issue:43720", "issue:44661|issue:44805", "issue:43317|issue:44756", - "issue:38175|issue:39692" + "issue:38175|issue:39692", + "issue:43644|issue:44534" ], - "prepared_review_unit_hash": "11d80182ab4226b7fc16c12a1ca73fcd0f2b6270d6659ddbb2ebbf72a47626c0", + "prepared_review_unit_hash": "819b1752c1928ac84ddb0cd026f23f50f452b4c62f9712e02927cd3249d1f6a7", "component_budget": { "node_count": 502, "item_count": 502, - "soft_pair_count": 1842, - "serialized_chars": 956182, - "estimated_input_tokens": 239046, - "estimated_eval_tokens": 478348 + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 35, - "serialized_chars": 38361, - "estimated_input_tokens": 9591, - "estimated_eval_tokens": 19438 + "serialized_chars": 38345, + "estimated_input_tokens": 9587, + "estimated_eval_tokens": 19430 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a loose bundle of unrelated closed issues, many around Transformers v5 regressions and model-loading/runtime edge cases. Only one soft pair looks like a true duplicate: the custom model initialization regression in 43645/43646.", - "confidence": 0.74, - "canonical_issue_reason": "issue:43646 is the best canonical issue because it states the broader core regression ('Transformers 5.0.0 breaks custom model initialization'); issue:43645 reads like a narrower Jupyter-notebook-specific repro of the same problem.", + "summary": "This cluster is mostly heterogeneous: most pairs are related by broad subsystem or model family only, not the same bug. I only consider four pairs true duplicates: custom model initialization, SigLIP2 bad outputs, non-persistent buffer corruption, and Gemma4 4-bit quantization/dtype casting.", + "confidence": 0.88, + "canonical_issue_reason": "issue:43646 is the clearest general representative of a true duplicate subthread: it states the broad custom-model initialization regression without the notebook-specific framing of issue:43645.", "canonical_pr_reason": null, - "best_issue_reason": "issue:43646 is the strongest representative issue in the cluster: broad, concrete, and paired with a clear duplicate candidate.", + "best_issue_reason": "issue:43646 is the best issue to anchor the cluster because it is the most general and readable among the actual duplicate candidates; the rest of the cluster is too mixed to serve as one canonical bug.", "best_pr_reason": null, "soft_edge_verdicts": [ - { - "left": "issue:44863", - "right": "issue:46402", - "accept": false, - "reason": "Different models and failures: NemotronH checkpoint loading vs Sam3 text-encoder weights not loading." - }, { "left": "issue:43366", "right": "issue:43828", "accept": false, - "reason": "Both mention model/runtime problems, but one is GGUF support for gpt-oss and the other is a dtype mismatch in Phi-tiny-MoE." + "reason": "Different problems: GGUF gpt-oss support vs Phi autocast dtype mismatch." }, { "left": "issue:43596", "right": "issue:46515", "accept": false, - "reason": "Generic similarity around loading/runtime compatibility, but the concrete bugs are different (Bert zero3 index error vs torch 2.4.1 incompatibility)." + "reason": "Both are generic failure reports, but the concrete bugs differ: deepspeed zero3/Bert indexing vs torch 2.4.1 incompatibility." }, { "left": "issue:45325", "right": "issue:45902", "accept": false, - "reason": "Both involve Qwen rotary/position-related behavior, but one is a still-image RoPE scaling bug and the other is uninitialized memory after meta-to-empty materialization." + "reason": "Different models and code paths: Qwen2.5-VL rope scaling vs Qwen3_5 MoE rotary embedding materialization." }, { "left": "issue:43366", "right": "issue:43531", "accept": false, - "reason": "Unrelated problems: GGUF gpt-oss support vs Qwen3-MoE sliding-window behavior." + "reason": "Unrelated issues despite both mentioning Qwen3-MoE." }, { "left": "issue:43638", "right": "issue:46515", "accept": false, - "reason": "Both touch loading behavior, but one is a Bert zero3 training bug and the other is a broad torch-version incompatibility report." + "reason": "Separate bugs: zero3/non-pretrained Bert indexing vs a torch-version incompatibility report." }, { "left": "issue:45663", "right": "issue:45823", "accept": false, - "reason": "Different Gemma4 issues: FSDP2 shared_kv_states reconstruction vs device_map auto device mismatch." + "reason": "Both are Gemma4-related, but one is an FSDP2 KeyError in text attention and the other is a device_map auto placement mismatch." }, { "left": "issue:44155", "right": "issue:45584", "accept": false, - "reason": "AudioFlamingo3 batched inference token leakage is unrelated to Whisper empty-transcription generation failure." + "reason": "Different failure modes and models: AudioFlamingo3 batch leakage vs Whisper empty-transcription generation." }, { "left": "issue:43366", "right": "issue:44589", "accept": false, - "reason": "Different model families and failure modes: gpt-oss GGUF support vs missing Float8 storage type." + "reason": "GGUF gpt-oss support is unrelated to Float8 storage loading errors." }, { "left": "issue:43824", "right": "issue:43957", "accept": false, - "reason": "ImportError for a specific Qwen2.5-VL class is not the same as meta-device loading breakage across some models." + "reason": "Import error for Qwen2.5-VL is not the same as meta-device loading failures in other models." }, { "left": "issue:43950", "right": "issue:45834", "accept": false, - "reason": "Non-persistent buffer corruption during from_pretrained is a different class of bug from Kosmos2.5 long-OCR index errors." + "reason": "Non-persistent buffer corruption and Kosmos2.5 OCR index errors are unrelated bugs." }, { "left": "issue:42915", "right": "issue:43957", "accept": false, - "reason": "Qwen3Moe FP8 config failure is unrelated to meta-device loading regressions." + "reason": "Qwen3Moe FP8 config failure is distinct from meta-device loading regressions." }, { "left": "issue:43716", "right": "issue:44560", "accept": false, - "reason": "Mistral-3 image-preprocessor dtype mismatch and Qwen3-VL video StopIteration are different multimodal bugs." + "reason": "Mistral-3 image-preprocessor dtype mismatch is not the same as Qwen3-vl video StopIteration." }, { "left": "issue:44910", "right": "issue:46858", "accept": false, - "reason": "Both involve sequence/cache mechanics, but one is a FlashAttention position_ids bug and the other is static-cache slice arithmetic on non-CUDA backends." + "reason": "Both involve caching/position arithmetic, but one is a flash-attention illegal access on Qwen3.5 and the other is a static-cache torch.compile backend issue." }, { "left": "issue:43645", "right": "issue:43646", "accept": true, - "reason": "These describe the same custom-model initialization regression; 43645 is the notebook-specific repro, while 43646 is the broader issue." + "reason": "Same underlying custom-model initialization regression; 43645 is the notebook-specific manifestation of 43646." }, { "left": "issue:44560", "right": "issue:45685", "accept": false, - "reason": "Qwen3-VL video StopIteration and MPS histogram error in MoE are unrelated." + "reason": "Different subsystems and error types: Qwen3-vl video inference vs MPS histogram on int tensors." }, { "left": "issue:43550", "right": "issue:44910", "accept": false, - "reason": "Different code paths: torch.compile+SDPA in Bamba-9B-v2 vs Flash Attention / position_ids handling in Qwen3.5." + "reason": "Both mention attention/compile failures, but the concrete code paths differ: SDPA torch.compile on Bamba vs Qwen3.5 3D position_ids misread by flash attention." }, { "left": "issue:44683", "right": "issue:46515", "accept": false, - "reason": "Compiled flex_attention on torch>=2.9 is a separate problem from a generic torch 2.4.1 incompatibility report." + "reason": "torch>=2.9 flex_attention failure is unrelated to a torch 2.4.1 incompatibility report." }, { "left": "issue:42617", "right": "issue:43366", "accept": false, - "reason": "3d_parallel.py execution failure is unrelated to gpt-oss GGUF support." + "reason": "Generic runtime failure for 3d_parallel.py is unrelated to gpt-oss GGUF support." }, { "left": "issue:44560", "right": "issue:44918", "accept": false, - "reason": "Both are Qwen3.5/Qwen-related, but one is video StopIteration and the other is TRL SFT embedding unpacking." + "reason": "Both are Qwen3.x issues, but one is a video StopIteration bug and the other is SFT input-embedding unpacking." }, { "left": "issue:36010", "right": "issue:42915", "accept": false, - "reason": "ImportError for GenerationMixin and Qwen3Moe FP8 config are different regressions with no clear shared code path." + "reason": "ImportError for GenerationMixin is unrelated to Qwen3Moe FP8 failures." }, { "left": "issue:45812", "right": "issue:46489", "accept": false, - "reason": "Both are tokenizer regressions, but they affect different model families and different failure mechanisms." + "reason": "Different tokenizer regressions affecting different model families; not the same bug." }, { "left": "issue:45823", "right": "issue:46515", "accept": false, - "reason": "Gemma4 PLE device mismatch is a model-specific forward bug, not the same as a broad torch compatibility issue." + "reason": "Gemma4 device placement mismatch is separate from the torch 2.4.1 compatibility issue." }, { "left": "issue:44661", "right": "issue:45823", "accept": false, - "reason": "TOKENIZER_MAPPING_NAMES add-new-model-like failure is unrelated to Gemma4 device placement during forward." + "reason": "Tokenizer-mapping add-new-model bug is unrelated to Gemma4 PLE device mismatch." }, { "left": "issue:44560", "right": "issue:44805", "accept": false, - "reason": "Different symptoms and likely different paths: video StopIteration vs mask/tensor shape mismatch." + "reason": "Different symptoms and likely causes: video StopIteration vs mask/tensor shape mismatch." }, { "left": "issue:38175", "right": "issue:43994", - "accept": false, - "reason": "Both concern siglip2-base-patch16-224 output quality, but one is zero probabilities and the other is a broader docs/pipeline nonsensical-results report; not enough evidence of the same bug." + "accept": true, + "reason": "Same SigLIP2 bad-output regression; both describe nonsensical/zero-probability behavior for google/siglip2-base-patch16-224." }, { "left": "issue:45672", "right": "issue:46899", - "accept": false, - "reason": "Both are Gemma4 quantization-related, but one is uint8 torch.finfo() in audio modules and the other is unsafe dtype casting in multimodal embedders." + "accept": true, + "reason": "Same Gemma4 4-bit quantization/dtype-cast bug in multimodal embedders; the second issue is a broader restatement of the first." }, { "left": "issue:44877", "right": "issue:45030", "accept": false, - "reason": "Strict config validation for granite_speech and tiny-random glm4v config loading are distinct config-validation issues." + "reason": "Both mention config validation, but they concern different models and failure conditions." }, { "left": "issue:43632", "right": "issue:45834", "accept": false, - "reason": "The _is_hf_initialized flag regression and Kosmos2.5 long OCR index error are different bugs." + "reason": "_is_hf_initialized regression is unrelated to Kosmos2.5 long-OCR index errors." }, { "left": "issue:43638", "right": "issue:44661", "accept": false, - "reason": "Bert zero3 training index error and add-new-model-like/tokenizer mapping failure are unrelated." + "reason": "Zero3/Bert indexing failure is unrelated to tokenizer-mapping name handling." }, { "left": "issue:43596", - "right": "issue:44661", + "right": "issue:44805", "accept": false, - "reason": "Bert zero3 init index error is unrelated to the tokenizer-mapping add-new-model-like failure." + "reason": "Both are IndexError reports, but the shapes, APIs, and model contexts differ." }, { "left": "issue:42491", "right": "issue:43720", "accept": false, - "reason": "Qwen3 MoE LoRA hf4.x/hf5.x incompatibility and BitNet packed-weight loading are different model-loading regressions." + "reason": "LoRA/qwen3_moe hf4.x versus hf5.x incompatibility is a different problem from BitNet packed-weight loading." }, { "left": "issue:44661", "right": "issue:44805", "accept": false, - "reason": "Tokenizer-mapping bug and mask/tensor shape mismatch are different subsystems and failure modes." + "reason": "Tokenizer mapping bug and mask/tensor shape mismatch are unrelated." }, { "left": "issue:43317", "right": "issue:44756", "accept": false, - "reason": "device_map auto dequantized-model offload failure and Strix Halo mmap OOM are both memory/loading related but not the same bug." + "reason": "device_map offload failure and mmap OOM on Strix Halo are different memory/loading bugs." }, { "left": "issue:38175", "right": "issue:39692", "accept": false, - "reason": "Same model family, but one is incorrect outputs and the other is a documentation example with multiple setup errors (model/processor mismatch plus quantization failure)." + "reason": "Both involve SigLIP2, but one is a model output regression while the other is a documentation example with multiple issues." + }, + { + "left": "issue:43644", + "right": "issue:44534", + "accept": true, + "reason": "Same non-persistent-buffer corruption regression in Transformers 5.x; titles differ only in phrasing." } ] }, "evaluator_result": { "accept": true, - "feedback": "The assessment is grounded in the packet and stays conservative. The summary correctly reflects that most pairs are only loosely similar, and the sole accepted duplicate pair (43645/43646) is supported by the titles and cluster evidence. No soft-edge verdict appears overstated, and there are no PR-pair mergeability concerns here." + "feedback": "The cluster summary is grounded in the packet, and the soft-edge decisions stay conservative overall. The four accepted duplicate pairs are reasonably supported by the issue titles, while the rest are appropriately rejected as related but distinct." }, "evaluator_used": true, "retried": false, @@ -2198,7 +2284,6 @@ "issue:44402", "issue:44451", "issue:44488", - "issue:44534", "issue:44756", "issue:44863", "issue:44910", @@ -2225,7 +2310,6 @@ "issue:46901" ], "soft_pairs": [ - "issue:43644|issue:44534", "issue:42491|issue:45276", "issue:41720|issue:42915", "issue:42915|issue:44756", @@ -2241,8 +2325,8 @@ "issue:39692|issue:43994", "issue:43257|issue:43854", "issue:43856|issue:43873", - "issue:43856|issue:44387", "issue:43708|issue:44075", + "issue:43856|issue:44387", "issue:43854|issue:43931", "issue:42491|issue:45335", "issue:44960|issue:46050", @@ -2257,226 +2341,220 @@ "issue:39692|issue:44960", "issue:41720|issue:44155" ], - "prepared_review_unit_hash": "5ba1be4105f73010d1f34987d1aed343d1dee92da4b0a078a6cadf45495774fb", + "prepared_review_unit_hash": "4caa714ada8b33ec632f260e419d6d9e9af889c4ec71010529d5ff7604aade15", "component_budget": { "node_count": 502, "item_count": 502, - "soft_pair_count": 1842, - "serialized_chars": 956182, - "estimated_input_tokens": 239046, - "estimated_eval_tokens": 478348 + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 }, "budget": { - "node_count": 48, - "item_count": 48, - "soft_pair_count": 31, - "serialized_chars": 36952, - "estimated_input_tokens": 9238, - "estimated_eval_tokens": 18732 + "node_count": 47, + "item_count": 47, + "soft_pair_count": 30, + "serialized_chars": 36102, + "estimated_input_tokens": 9026, + "estimated_eval_tokens": 18308 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Most items are unrelated, but there are a few clear duplicate pockets: tokenizer/model-loading regressions for BERT-like models, the Transformers v5 non-persistent-buffer regression, and a training-loss double-shift pair.", - "confidence": 0.77, - "canonical_issue_reason": "issue:44488 is the most central issue in the tokenizer-loading regression pocket: it has related reports from multiple models and describes the broadest failure mode.", + "summary": "The cluster is mostly heterogeneous Transformers 5.x regressions spanning tokenizer loading, quantization/offloading, model-specific loading failures, and training bugs. Only the two BERT tokenizer-loading reports look like true duplicates; the rest are too model- or code-path-specific to merge.", + "confidence": 0.84, + "canonical_issue_reason": "No single issue cleanly represents the whole cluster; issue 43873 is the broadest still-open anchor, but it does not subsume the other reports.", "canonical_pr_reason": null, - "best_issue_reason": "issue:44488 is the best representative issue overall because it captures the broader loading regression and sits at the center of a small duplicate cluster.", + "best_issue_reason": "Issue 43873 is the best representative because it is open, broadly scoped, and has the most ongoing discussion/inbound references among the open items.", "best_pr_reason": null, "soft_edge_verdicts": [ - { - "left": "issue:43644", - "right": "issue:44534", - "accept": true, - "reason": "Same underlying Transformers v5 regression about non-persistent buffers being filled with junk; the titles differ only in wording." - }, { "left": "issue:42491", "right": "issue:45276", "accept": false, - "reason": "Different model families and different symptoms; one is a Qwen3 MoE LoRA compatibility issue, the other is a Gemma4 resize_token_embeddings bug." + "reason": "Different bugs: Qwen3 MoE LoRA incompatibility vs Gemma4 token-embedding resize behavior." }, { "left": "issue:41720", "right": "issue:42915", "accept": false, - "reason": "Both involve Qwen3, but the failures are different concrete bugs (device-mapping CUDA assert vs FineGrainedFP8Config load failure)." + "reason": "Both are Qwen3-related, but one is an A800 auto-device-map crash and the other is a FineGrainedFP8Config failure." }, { "left": "issue:42915", "right": "issue:44756", "accept": false, - "reason": "Unrelated problems: Qwen3Moe FP8 loading failure versus Strix Halo mmap/OOM behavior." + "reason": "Qwen3 FP8 config failure and Strix Halo mmap OOM are unrelated code paths." }, { "left": "issue:44451", "right": "issue:44488", "accept": true, - "reason": "Both report the same latest/current-version model-loading regression for BERT-style models; likely one underlying tokenizer/model loading bug." + "reason": "Both report the same generic model-loading failure for different BERT repos, with no sign of distinct root causes." }, { "left": "issue:44960", "right": "issue:44977", "accept": false, - "reason": "Different model families and different runtime failures; these do not look like the same bug." + "reason": "Different models and symptoms; too little evidence of a shared concrete bug." }, { "left": "issue:44863", "right": "issue:46846", "accept": false, - "reason": "Different code paths and symptoms: NemotronH checkpoint loading versus Qwen3.5 tensor-parallel planning issues." + "reason": "NemotronH checkpoint loading and Qwen3.5 TP-plan missing entries are separate issues." }, { "left": "issue:46459", "right": "issue:46512", "accept": false, - "reason": "Related dtype-handling theme, but different operations and failure modes: loading composite checkpoints versus from_config stale config dtype." + "reason": "Both involve dtype handling, but one is loading composite checkpoints and the other is from_config config-state drift." }, { "left": "issue:45313", "right": "issue:45357", "accept": false, - "reason": "One is a DeepSpeed ZeRO-3 load failure and the other is a save_pretrained key regression; not the same concrete bug." + "reason": "DeepSpeed ZeRO-3 weight loading and save_pretrained visual-key regression are different failures." }, { "left": "issue:44488", "right": "issue:44991", "accept": true, - "reason": "Same broad tokenizer/model-loading regression affecting different BERT models on Transformers >=5.0.0." + "reason": "Both are the same class of tokenizer-loading regression affecting specific BERT-like repos in transformers>=5.0." }, { "left": "issue:45507", "right": "issue:46468", "accept": false, - "reason": "Different models and different broken paths; one is an invalid method call in GraniteMoEHybrid, the other an official Nemotron example issue." + "reason": "GraniteMoEHybrid invalid method call and broken Nemotron example do not share the same concrete bug." }, { "left": "issue:43644", "right": "issue:43645", "accept": false, - "reason": "Both are Transformers v5 regressions, but the bugs are different: buffer initialization corruption versus Jupyter/custom-model initialization." + "reason": "Non-persistent buffer junk and Jupyter custom-model initialization are unrelated regressions." }, { "left": "issue:42491", "right": "issue:45292", "accept": false, - "reason": "Different bugs: Qwen3 MoE hf4.x/hf5.x incompatibility versus resize_token_embeddings not updating output embeddings." + "reason": "Unrelated: Qwen3 MoE LoRA reuse vs a generic resize_token_embeddings/output_embeddings bug." }, { "left": "issue:39692", "right": "issue:43994", "accept": false, - "reason": "Different SigLIP2 vs SigLIP model issues; one is a doc example mismatch/quantization failure, the other is incorrect outputs with AutoModel/pipeline." + "reason": "Different SigLIP2 example error vs SigLIP2 AutoModel/pipeline wrong results; not the same bug." }, { "left": "issue:43257", "right": "issue:43854", "accept": false, - "reason": "Both are loading problems, but they concern different models and different mechanisms (accelerate+deepspeed weight conversion vs unit-test loading)." + "reason": "Qwen3 MoE offloading/accelerate-deepspeed issue vs GLM-4.7-Flash unit-test load failure." }, { "left": "issue:43856", "right": "issue:43873", "accept": false, - "reason": "Both mention memory/quantization, but one is Qwen3 MoE training memory usage and the other is offloading behavior under quantization." + "reason": "Both concern Qwen3/quantization, but one is training memory usage and the other is offloading behavior." }, { - "left": "issue:43856", - "right": "issue:44387", + "left": "issue:43708", + "right": "issue:44075", "accept": false, - "reason": "Related to memory, but one is Qwen3 MoE training and the other is a Transformers 5.x int4 reserved-memory regression." + "reason": "Trainer checkpoint resume math and SGD argument handling are unrelated." }, { - "left": "issue:43708", - "right": "issue:44075", + "left": "issue:43856", + "right": "issue:44387", "accept": false, - "reason": "Different areas entirely: Trainer resume/max_steps calculation versus optimizer argument handling." + "reason": "Qwen3 MoE memory usage and int4 reserved-memory/OOM regression are different issues." }, { "left": "issue:43854", "right": "issue:43931", "accept": false, - "reason": "Different model-loading failures for different models; no clear shared underlying bug is evident." + "reason": "GLM-4.7-Flash test loading and Qwen3-VL weight-shape mismatch are separate model-loading problems." }, { "left": "issue:42491", "right": "issue:45335", "accept": false, - "reason": "Both involve model-specific embedding issues, but the affected architectures and failure modes are different." + "reason": "Different architectures and failures; not the same underlying bug." }, { "left": "issue:44960", "right": "issue:46050", "accept": false, - "reason": "Different problems: GLM5 model issue versus quantized model save/load breakage after sinq." + "reason": "GLM5 loading/inference and sinq save_pretrained/from_pretrained corruption are unrelated." }, { "left": "issue:46897", "right": "issue:46901", - "accept": true, - "reason": "Both describe the same double-shift training-loss bug: the model trains against labels shifted by one token instead of the intended labels." + "accept": false, + "reason": "Same symptom family, but these are separate model implementations and not clearly the same code-path bug." }, { "left": "issue:45488", "right": "issue:45812", "accept": false, - "reason": "Both are tokenizer regressions, but they affect different model families and different concrete tokenizer behaviors." + "reason": "Different tokenizer regressions affecting different model families; too broad to merge." }, { "left": "issue:45362", "right": "issue:45406", "accept": false, - "reason": "Different frontend/runtime failures: transformers chat crash versus serve crash from a missing processor attribute." + "reason": "Qwen3.5 chat crash and Gemma4Processor missing _tokenizer are separate issues." }, { "left": "issue:45276", "right": "issue:45335", "accept": false, - "reason": "Both are resize_token_embeddings bugs, but the affected model internals and symptoms differ; not clearly one underlying defect." + "reason": "Both involve resize_token_embeddings, but they are distinct model-specific embedding propagation bugs." }, { "left": "issue:44164", "right": "issue:44291", "accept": false, - "reason": "Different serialization/loading paths: extra_state handling versus init_empty_weights argument compatibility." + "reason": "save/from_pretrained extra_state handling and init_empty_weights argument mismatch are different loader bugs." }, { "left": "issue:44910", "right": "issue:45588", "accept": false, - "reason": "Both involve flash-attention crashes, but the causes are different (packed-sequence misinterpretation vs s_aux=None attribute error)." + "reason": "Different flash-attention failures with different triggers and models." }, { "left": "issue:44402", "right": "issue:44488", "accept": false, - "reason": "Both concern BERT-like models, but one is a tokenizer vocab-size mismatch and the other is a model-loading regression." + "reason": "Tokenizer vocab-size mismatch is a different bug from the BERT loading failure." }, { "left": "issue:43425", "right": "issue:43957", "accept": false, - "reason": "Different incompatibility surfaces: Torch 2.10 version support versus meta-device model loading breakage." + "reason": "Torch version incompatibility and meta-device loading failure are not the same defect." }, { "left": "issue:39692", "right": "issue:44960", "accept": false, - "reason": "Completely unrelated model families and failure modes." + "reason": "SigLIP2 documentation errors and GLM5 issue are unrelated." }, { "left": "issue:41720", "right": "issue:44155", "accept": false, - "reason": "Different Qwen3 device-mapping crash versus AudioFlamingo3 batched inference leakage." + "reason": "Qwen3 device-map assert and AudioFlamingo3 batched inference leak are unrelated." } ] }, "evaluator_result": { "accept": true, - "feedback": "The summary is conservative and mostly grounded in the packet. The accepted soft edges are well supported by the titles/body-level signals, especially the near-identical Transformers v5 buffer regression pair and the double-shift training-loss pair. The tokenizer-loading cluster around issue:44488 is plausible as a representative pocket. No pull-request mergeability concerns apply here." + "feedback": "The summary is grounded in the packet, and the pairwise judgments are generally conservative. The open-item anchor choice of issue 43873 is supported by its higher discussion/inbound-reference activity, and the accepted BERT-loading pair is plausibly the same regression class. No mergeability overreach is evident." }, "evaluator_used": true, "retried": false, @@ -2567,8 +2645,8 @@ "issue:43381|issue:44928", "issue:44164|issue:44756", "issue:45684|issue:45698", - "issue:42796|issue:46326", "issue:45588|issue:45663", + "issue:42796|issue:46326", "issue:43854|issue:45216", "issue:45292|issue:45335", "issue:46894|issue:46901", @@ -2579,226 +2657,226 @@ "issue:44589|issue:44756", "issue:43425|issue:43828" ], - "prepared_review_unit_hash": "2796840a475186e857f3d4fb685983a99af733f23b7f21d05d89bb2036cabd28", + "prepared_review_unit_hash": "c0dbae09dc1ca99f53e624f04dea6b358b5b70fad7ad89aeca6ab544714ff3b1", "component_budget": { "node_count": 502, "item_count": 502, - "soft_pair_count": 1842, - "serialized_chars": 956182, - "estimated_input_tokens": 239046, - "estimated_eval_tokens": 478348 + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 31, - "serialized_chars": 37131, - "estimated_input_tokens": 9283, - "estimated_eval_tokens": 18822 + "serialized_chars": 37128, + "estimated_input_tokens": 9282, + "estimated_eval_tokens": 18820 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The set is mostly heterogeneous bug reports, but there are two small duplicate-like mini-clusters: the `resize_token_embeddings` regressions and the training-loss double-shift reports. No pull requests are present.", - "confidence": 0.46, - "canonical_issue_reason": "Issue 45292 is the best representative of the resize_token_embeddings family because it states the core symptom in a generic way, while 45276 and 45335 look like model-specific manifestations of the same API-level embedding update problem.", + "summary": "This cluster is mostly a loose mix of unrelated Transformers issues. The only clear duplicate family is the resize_token_embeddings propagation bug across related Gemma/t5gemma reports; most other pairs share only a broad subsystem or symptom class and should stay separate.", + "confidence": 0.91, + "canonical_issue_reason": "Issue 45292 is the broadest, most neutral statement of the shared resize_token_embeddings/output_embeddings bug, while 45276 and 45335 read as model-specific variants of the same underlying defect.", "canonical_pr_reason": null, - "best_issue_reason": "45292 is the cleanest and most central issue in the cluster: it describes the shared resize_token_embeddings failure without extra model-specific noise, so it is the best anchor for the duplicate family.", + "best_issue_reason": "Issue 45292 is the best canonical issue because it describes the core bug without overfitting to one model family and cleanly subsumes the two related resize_token_embeddings reports.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:45663", "right": "issue:45759", "accept": false, - "reason": "Both are model/config bugs, but they affect different code paths (Gemma4 FSDP shared_kv_states vs Qwen multimodal config unwrapping) and do not look mergeable into one fix." + "reason": "Both involve model/config handling, but one is an FSDP2 KeyError in Gemma4 attention and the other is a Qwen multimodal config unwrap issue; different code paths and fixes." }, { "left": "issue:43404", "right": "issue:43454", "accept": false, - "reason": "Both mention untied lm_head weights, but they concern different architectures and likely different tying logic; too broad to treat as the same concrete bug." + "reason": "Both mention lm_head tying, but they affect different model families (Mistral3 vs AyaVision) and likely different tying logic." }, { "left": "issue:44079", "right": "issue:46829", "accept": false, - "reason": "These are different output-shape/data-structure bugs in unrelated components; the similarity is only at the very broad ModelOutput level." + "reason": "These are different bugs: a generic ModelOutput key assignment issue versus a VideoPrism classifier output field mix-up." }, { "left": "issue:44291", "right": "issue:44756", "accept": false, - "reason": "One is about init_empty_weights argument handling and the other about mmap/OOM on Strix Halo; they are not the same underlying problem." + "reason": "init_empty_weights/_is_hf_initialized and Strix Halo mmap/OOM are unrelated failure modes." }, { "left": "issue:42947", "right": "issue:45704", "accept": false, - "reason": "Gradient checkpointing with LoRA and an apex RMSNorm memory leak are unrelated fixes." + "reason": "Gradient checkpointing with PEFT LoRA and a T5 apex RMSNorm memory leak are separate bugs in different areas." }, { "left": "issue:43873", "right": "issue:44387", "accept": false, - "reason": "Both involve quantization/memory, but one is about offloading behavior and the other about int4 reserved-memory OOM; not the same code-path problem." + "reason": "Both concern quantization, but one is offloading behavior and the other is increased reserved memory/OOM under int4; not the same defect." }, { "left": "issue:44292", "right": "issue:44912", "accept": false, - "reason": "Both are quantization-loading regressions, but they target different quantization formats and model families, so they are not clearly one duplicate bug." + "reason": "Different quantization formats and symptoms: NVFP4 load error versus MXFP4 fallback to bf16." }, { "left": "issue:36010", "right": "issue:39404", "accept": false, - "reason": "Importing GenerationMixin and Whisper return_language pipeline behavior are unrelated issues." + "reason": "Importing GenerationMixin and Whisper return_language pipeline regression are unrelated issues." }, { "left": "issue:45663", "right": "issue:45859", "accept": false, - "reason": "Different model-specific runtime failures (Gemma4 FSDP vs Qwen3_5Moe CPU offload) with no shared concrete fix." + "reason": "Both are model-specific runtime problems, but one is Gemma4 FSDP2 state handling and the other is Qwen3.5-MoE CPU offload compatibility." }, { "left": "issue:44387", "right": "issue:44756", "accept": false, - "reason": "Int4 quantization OOM and disabling mmap on Strix Halo are distinct memory-related regressions, not the same bug." + "reason": "Int4 quantization OOM and disabling mmap on Strix Halo are different memory-path issues." }, { "left": "issue:42947", "right": "issue:43381", "accept": false, - "reason": "Both mention gradient checkpointing, but one is ineffective with PEFT LoRA while the other forbids eval-mode use; different behavior and likely different fixes." + "reason": "One is LoRA/gradient-checkpointing ineffectiveness; the other is eval-mode incompatibility. Similar theme, not same bug." }, { "left": "issue:44155", "right": "issue:45823", "accept": false, - "reason": "AudioFlamingo3 track leakage and Gemma4 PLE device mismatch are unrelated model-specific bugs." + "reason": "AudioFlamingo3 batched inference leakage and Gemma4 PLE device mismatch are unrelated model-specific bugs." }, { "left": "issue:43957", "right": "issue:44756", "accept": false, - "reason": "Model loading with torch.device('meta') and Strix Halo mmap/OOM are different failure modes." + "reason": "Meta-device loading failures and Strix Halo mmap OOM are distinct loading/runtime problems." }, { "left": "issue:43335", "right": "issue:45663", "accept": false, - "reason": "SwitchTransformers sparse-layer creation and Gemma4 FSDP attention state rebuilding do not share the same bug." + "reason": "SwitchTransformers sparse-layer construction bug and Gemma4 FSDP2 attention state bug are unrelated." }, { "left": "issue:45276", "right": "issue:45292", "accept": true, - "reason": "Both describe the same resize_token_embeddings regression affecting output embeddings in closely related model implementations; one fix could plausibly address both." + "reason": "Same underlying bug: resize_token_embeddings does not propagate updated embeddings correctly; 45276 is the Gemma4-specific manifestation and 45292 states the general failure." }, { "left": "issue:44863", "right": "issue:46468", "accept": false, - "reason": "Nemotron checkpoint loading and the official Nemotron example breaking are not clearly the same underlying defect." + "reason": "Both are Nemotron-related, but one is checkpoint loading and the other is a broken official example; insufficient evidence of the same code-path." }, { "left": "issue:44479", "right": "issue:44877", "accept": false, - "reason": "Both are loading/config regressions, but they affect different models and mechanisms." + "reason": "Video-input regression for Qwen VL variants and strict Granite Speech config loading are separate issues." }, { "left": "issue:43381", "right": "issue:44928", "accept": false, - "reason": "Gradient checkpointing and RLHF NaN explosions from SDPA/BF16 fallback are unrelated." + "reason": "Eval-mode gradient checkpointing and RLHF NaN explosion from 3D position_ids/SDPA fallback are different bugs." }, { "left": "issue:44164", "right": "issue:44756", "accept": false, - "reason": "Save/from_pretrained extra_state handling and mmap/OOM on Strix Halo are different subsystems." + "reason": "save/from_pretrained extra_state handling and Strix Halo mmap/OOM are unrelated." }, { "left": "issue:45684", "right": "issue:45698", "accept": false, - "reason": "Both touch custom-model save/load workflows, but one is about read-only permissions on copied source files and the other about loading the wrong custom module; not the same bug." + "reason": "Both involve custom-model save/load flows, but one is permissions propagation and the other is wrong module reloading; different defects." }, { - "left": "issue:42796", - "right": "issue:46326", + "left": "issue:45588", + "right": "issue:45663", "accept": false, - "reason": "Tokenizer BPE marker handling and continue_final_message behavior are distinct tokenizer bugs." + "reason": "flash_attention.py s_aux=None crash and Gemma4 FSDP2 shared_kv_states KeyError are unrelated." }, { - "left": "issue:45588", - "right": "issue:45663", + "left": "issue:42796", + "right": "issue:46326", "accept": false, - "reason": "A flash_attention AttributeError on sink-less models is unrelated to Gemma4 FSDP shared_kv_states rebuilding." + "reason": "Tokenizer BPE-marker handling and continue_final_message behavior are different tokenizer bugs." }, { "left": "issue:43854", "right": "issue:45216", "accept": false, - "reason": "Model loading failure in tests for GLM-4.7-Flash and Qwen3.5 save_pretrained regression are separate model-specific issues." + "reason": "GLM-4.7-Flash unit-test loading failure and Qwen3.5 save_pretrained regression are unrelated." }, { "left": "issue:45292", "right": "issue:45335", "accept": true, - "reason": "Both are the same resize_token_embeddings family: embeddings are not propagated correctly to model outputs/decoder embeddings after resizing." + "reason": "Same resize_token_embeddings propagation bug: both report that resized embeddings do not update the relevant output/decoder embedding weights." }, { "left": "issue:46894", "right": "issue:46901", - "accept": true, - "reason": "Both describe the same training-loss double-shift pattern, where labels are shifted incorrectly before loss computation; they appear mergeable into one fix." + "accept": false, + "reason": "Both describe a training-loss double-shift pattern, but they are on different models and lack evidence of a single shared code-path." }, { "left": "issue:43828", "right": "issue:43957", "accept": false, - "reason": "An autocast dtype mismatch in Phi-tiny-MoE and meta-device loading regressions are different problems." + "reason": "Autocast dtype mismatch and meta-device loading failures are distinct runtime issues." }, { "left": "issue:44402", "right": "issue:44451", "accept": false, - "reason": "Tokenizer vocab-size mismatch and failure to load a specific model are not the same underlying issue." + "reason": "Tokenizer vocab mismatch for specific tokens and a model load failure are not the same bug." }, { "left": "issue:43646", "right": "issue:43708", "accept": false, - "reason": "Custom model initialization in notebooks and resume_from_checkpoint max_steps miscalculation are unrelated." + "reason": "Custom model initialization breakage in Transformers 5.0 and Trainer resume/max_steps miscalculation are unrelated." }, { "left": "issue:43645", "right": "issue:44164", "accept": false, - "reason": "Notebook/custom model initialization breakage and extra_state save/load handling are distinct workflows." + "reason": "Notebook custom-model initialization and extra_state save/load handling are separate issues." }, { "left": "issue:44589", "right": "issue:44756", "accept": false, - "reason": "Float8 storage lookup failure and mmap/OOM on Strix Halo are different runtime issues." + "reason": "Float8 storage lookup failure and Strix Halo mmap/OOM are different problems." }, { "left": "issue:43425", "right": "issue:43828", "accept": false, - "reason": "Torch 2.10 incompatibility and an autocast dtype mismatch are different compatibility failures." + "reason": "Torch 2.10 compatibility and autocast dtype mismatch are not the same underlying bug." } ] }, "evaluator_result": { "accept": true, - "feedback": "The summary is grounded in the packet and stays conservative. The canonical issue choice is well-supported, the two accepted mini-clusters are the obvious near-duplicates in the set, and the rejected soft pairs are mostly justified by differing subsystems or fix paths. No PRs are present, as stated." + "feedback": "The summary is grounded in the packet and stays mostly conservative. The resize_token_embeddings family is the only clearly supported duplicate cluster, and the other pair rejections avoid overclaiming shared root causes. No obvious unsupported mergeability claims stand out." }, "evaluator_used": true, "retried": false, @@ -2869,8 +2947,8 @@ "issue:46899" ], "soft_pairs": [ - "issue:46315|issue:46491", "issue:45812|issue:46315", + "issue:46315|issue:46491", "issue:39404|issue:45701", "issue:44530|issue:45084", "issue:43611|issue:43645", @@ -2903,232 +2981,232 @@ "issue:44488|issue:45081", "issue:43611|issue:43644" ], - "prepared_review_unit_hash": "df643d9c537a7d4637102d31d1003f33049d0afe19b2cbccada71099410a93ae", + "prepared_review_unit_hash": "256aaa98dbec636a3c3becea3ad982a3520891ffab08f73535139c5b3869d20f", "component_budget": { "node_count": 502, "item_count": 502, - "soft_pair_count": 1842, - "serialized_chars": 956182, - "estimated_input_tokens": 239046, - "estimated_eval_tokens": 478348 + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 33, - "serialized_chars": 37737, - "estimated_input_tokens": 9435, - "estimated_eval_tokens": 19126 + "serialized_chars": 37736, + "estimated_input_tokens": 9434, + "estimated_eval_tokens": 19124 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a heterogeneous set of Transformers v5 regressions and model/tokenizer loading issues. The soft links are mostly superficial title similarity rather than the same underlying bug, so I would not merge any of the candidate pairs.", - "confidence": 0.82, - "canonical_issue_reason": "issue:43611 is the broadest representative of the cluster\u2019s dominant theme: a Transformers 5.0 model-loading regression. It is more general and central than the model-specific, docs-only, or unrelated runtime issues elsewhere in the cluster.", + "summary": "The items are mostly distinct Transformers v5 regressions across different model families and subsystems; none of the soft pairs look like true duplicates.", + "confidence": 0.93, + "canonical_issue_reason": "issue:45812 is the strongest representative anchor: it is a concrete, high-impact v4\u2192v5 tokenizer regression with multiple inbound references and a clear failure mode.", "canonical_pr_reason": null, - "best_issue_reason": "issue:43611 is the best standalone anchor because it describes a concrete, broadly applicable v5 loading break with a clear failure mode. It is still not a perfect umbrella for the whole set, but it is the least specific and most representative issue here.", + "best_issue_reason": "issue:45812 is the best standalone issue to surface because it is specific, well-scoped, and broadly relevant compared with the more specialized reports.", "best_pr_reason": null, "soft_edge_verdicts": [ { - "left": "issue:46315", - "right": "issue:46491", + "left": "issue:45812", + "right": "issue:46315", "accept": false, - "reason": "Different bugs: tokenizer conversion performance regression vs CodeLlama round-trip spacing behavior." + "reason": "Different bug class: wrong token IDs for Granite vs a tokenizer performance regression in convert_tokens_to_ids." }, { - "left": "issue:45812", - "right": "issue:46315", + "left": "issue:46315", + "right": "issue:46491", "accept": false, - "reason": "Different failure modes: Granite token IDs are wrong, while the other issue is about slow token conversion." + "reason": "Performance regression in added-token lookup vs a CodeLlama encode/decode spacing bug." }, { "left": "issue:39404", "right": "issue:45701", "accept": false, - "reason": "Whisper pipeline return_language regression is unrelated to a general tokenization-version change report." + "reason": "Whisper pipeline return_language failure is unrelated to a general tokenization regression report." }, { "left": "issue:44530", "right": "issue:45084", "accept": false, - "reason": "Different code paths and symptoms: Qwen3.5 linear_attention crash vs compile error on non-template nodes." + "reason": "Different code paths: PagedAttentionCache group-type crash vs template compilation error." }, { "left": "issue:43611", "right": "issue:43645", "accept": false, - "reason": "Both are v5 regressions, but one is model loading via base_model_prefix and the other is custom model initialization in notebooks." + "reason": "Base_model_prefix loading issue and Jupyter custom-model init issue are separate regressions." }, { "left": "issue:39404", "right": "issue:46665", "accept": false, - "reason": "Whisper pipeline behavior is unrelated to from_pretrained inside Python classes." + "reason": "Pipeline return_language breakage is not the same as from_pretrained behavior inside Python classes." }, { "left": "issue:44683", "right": "issue:45823", "accept": false, - "reason": "Torch-version flex_attention failure and Gemma4 device-map mismatch are distinct problems." + "reason": "Torch version incompatibility in flex_attention is unrelated to Gemma4 device-map forwarding." }, { "left": "issue:43425", "right": "issue:46869", "accept": false, - "reason": "General torch incompatibility is not the same bug as YOLO multi-GPU device-map failure." + "reason": "Torch 2.10 incompatibility is not the same failure as YOLO multi-GPU device-map handling." }, { "left": "issue:43425", "right": "issue:43716", "accept": false, - "reason": "Torch 2.10 incompatibility is unrelated to the Mistral image-preprocessor dtype mismatch." + "reason": "Generic torch incompatibility does not match the specific dtype mismatch in Mistral-3 preprocessing." }, { "left": "issue:43425", "right": "issue:45823", "accept": false, - "reason": "Different symptoms and root causes: torch compatibility vs Gemma4 device placement." + "reason": "Torch compatibility and Gemma4 device mismatch are different problems." }, { "left": "issue:44589", "right": "issue:45084", "accept": false, - "reason": "Float8 storage lookup failure is unrelated to the template-node compilation error." + "reason": "Float8 storage lookup failure is unrelated to non-template-node compilation." }, { "left": "issue:44479", "right": "issue:44805", "accept": false, - "reason": "Both are shape-related regressions, but they affect different models and different failure points." + "reason": "Qwen video-input regression and mask-shape IndexError are different model/input failures." }, { "left": "issue:43906", "right": "issue:44530", "accept": false, - "reason": "An isolated reproduction of another issue is not the same as the Qwen3.5 PagedAttentionCache crash." + "reason": "An isolated reproduction of another issue is not the same as the linear_attention cache crash." }, { "left": "issue:43901", "right": "issue:44509", "accept": false, - "reason": "Both are docs-related, but they cover different API/task deprecations and are not the same change." + "reason": "Both are docs-related, but they concern different pipeline/deprecation topics." }, { "left": "issue:44671", "right": "issue:44991", "accept": false, - "reason": "CamemBERT masked-LM wrong predictions and tokenizer load failure for est-roberta are unrelated." + "reason": "CamemBERT masked-LM predictions and est-roberta tokenizer loading are unrelated." }, { "left": "issue:43994", "right": "issue:44079", "accept": false, - "reason": "SigLIP2 inference quality regression and ModelOutput key assignment bug are different layers of the stack." + "reason": "Wrong SigLIP2 outputs are not the same as ModelOutput key assignment behavior." }, { "left": "issue:36032", "right": "issue:43452", "accept": false, - "reason": "Tokenizer special-token conflict and gguf_file loading break are different tokenizer/load-path bugs." + "reason": "T5 tokenizer method conflict and gguf_file loading breakage are different tokenizer/model-loading failures." }, { "left": "issue:44265", "right": "issue:44898", "accept": false, - "reason": "torch.export exportability with torch_compilable_check is unrelated to Perceiver interpolation failure." + "reason": "torch.export with torch_compilable_check is unrelated to Perceiver interpolation at non-default resolution." }, { "left": "issue:39404", "right": "issue:44912", "accept": false, - "reason": "Whisper pipeline regression does not match the MXFP4 quantization fallback issue." + "reason": "Whisper pipeline regression and MXFP4 quantization fallback are unrelated." }, { "left": "issue:46050", "right": "issue:46326", "accept": false, - "reason": "Quantized save/load break and mistral_common continue_final_message bug are separate features." + "reason": "Quantized model save/load failure is not the same as mistral_common continue_final_message behavior." }, { "left": "issue:44164", "right": "issue:45759", "accept": false, - "reason": "extra_state serialization and AutoModelForCausalLM.from_config text_config handling are distinct model-loading paths." + "reason": "extra_state save/load handling and composite config unwrapping are different APIs and bugs." }, { "left": "issue:43906", "right": "issue:46326", "accept": false, - "reason": "A reproduction issue is not the same bug as the mistral_common tokenizer behavior change." + "reason": "A reproduction issue and a tokenizer chat-template bug are not duplicates." }, { "left": "issue:44991", "right": "issue:45081", "accept": false, - "reason": "Different affected models and failure mechanisms; they only share that tokenizer loading is involved." + "reason": "Different tokenizer failures: est-roberta loading vs Mistral regex patching on a backend tokenizer object." }, { "left": "issue:43299", "right": "issue:43931", "accept": false, - "reason": "Both involve Qwen3-VL-family loading, but one is a v5 regression and the other is a weight-shape mismatch on a specific model." + "reason": "Qwen3VL model-loading breakage and Qwen3-VL weight-shape mismatch are related by family only, not the same concrete bug." }, { "left": "issue:45823", "right": "issue:46899", "accept": false, - "reason": "Gemma4 device_map mismatch and Gemma4 4-bit dtype-cast issues are separate multimodal bugs." + "reason": "Gemma4 device-map mismatch and Gemma4 4-bit multimodal dtype-cast issues are separate failure modes." }, { "left": "issue:45479", "right": "issue:46894", "accept": false, - "reason": "Sequence-classification zero-loss bug and ASR training double-shift bug are unrelated training objectives." + "reason": "Zero-loss classification bug is unrelated to the CohereASR training-loss double-shift issue." }, { "left": "issue:43856", "right": "issue:44756", "accept": false, - "reason": "Qwen3 MoE memory inefficiency and Strix Halo mmap OOM are not the same underlying issue." + "reason": "Qwen3 MoE training memory inefficiency and Strix Halo mmap OOM are different memory problems." }, { "left": "issue:43596", "right": "issue:44683", "accept": false, - "reason": "Deepspeed zero3/BertModel index error is unrelated to compiled flex_attention on newer torch." + "reason": "Zero3 init IndexError in BertModel is unrelated to compiled flex_attention on torch>=2.9." }, { "left": "issue:45479", "right": "issue:46897", "accept": false, - "reason": "Different model families and different loss-shift bugs." + "reason": "Sequence-classification loss bug does not match the Florence2 double-shift training-loss bug." }, { "left": "issue:44164", "right": "issue:46821", "accept": false, - "reason": "extra_state save/load failure is unrelated to the TimesFM window_size AttributeError." + "reason": "extra_state persistence is unrelated to the TimesFM window_size attribute error." }, { "left": "issue:44488", "right": "issue:44987", "accept": false, - "reason": "Both are loading failures, but they affect different models and likely different causes." + "reason": "Both are load failures, but they affect different models and likely different root causes." }, { "left": "issue:44488", "right": "issue:45081", "accept": false, - "reason": "Different tokenizer/model loading failures with no shared concrete code path." + "reason": "Generic model-loading failure is not the same as the Mistral regex patch crash." }, { "left": "issue:43611", "right": "issue:43644", "accept": false, - "reason": "Model loading with base_model_prefix and non-persistent buffer junk are separate v5 regressions." + "reason": "Loading models with base_model_prefix and non-persistent buffer junk are distinct regressions." } ] }, @@ -3216,15 +3294,15 @@ "issue:42915|issue:44530", "issue:43541|issue:45084", "issue:43644|issue:43646", + "issue:43582|issue:43632", "issue:42175|issue:46178", "issue:34567|issue:40444", - "issue:43582|issue:43632", "issue:43883|issue:44488", "issue:40444|issue:46326", "issue:44206|issue:44220", "issue:44811|issue:45584", - "issue:43452|issue:43482", "issue:36032|issue:43645", + "issue:43452|issue:43482", "issue:44987|issue:44991", "issue:43317|issue:43482", "issue:45084|issue:45507", @@ -3233,221 +3311,218 @@ "issue:43883|issue:44991", "issue:45084|issue:46869" ], - "prepared_review_unit_hash": "bd9d3fb6743b46823838098578fb394456c4de13dccd7c97974d80bb9d8f674f", + "prepared_review_unit_hash": "873c579097a1926b6d318b965c440e6f64dd1f3fd9364db68971111709f6b82a", "component_budget": { "node_count": 502, "item_count": 502, - "soft_pair_count": 1842, - "serialized_chars": 956182, - "estimated_input_tokens": 239046, - "estimated_eval_tokens": 478348 + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 30, - "serialized_chars": 36684, - "estimated_input_tokens": 9171, - "estimated_eval_tokens": 18598 + "serialized_chars": 36689, + "estimated_input_tokens": 9173, + "estimated_eval_tokens": 18602 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly heterogeneous; the soft pairs are mostly superficial similarity matches across different regressions. I did not find any strong duplicate-issue merges. The best umbrella representative is the broad GGUF loading regression.", + "summary": "Mostly unrelated issue set with a few likely duplicate pairs around v5 init regressions and audio feature extraction. I would keep this cluster centered on the generic Transformers v5 custom-model initialization regression.", "confidence": 0.61, - "canonical_issue_reason": "Issue 43452 is the broadest, most central bug report here: a generic `gguf_file`/`from_pretrained` loading regression that affects both tokenizer and CausalLM paths, making it the best umbrella representative among the loading-related issues.", + "canonical_issue_reason": "issue:43646 is the broadest, most representative regression here: it describes the general Transformers 5.0.0 custom-model initialization breakage that likely subsumes the more specific buffer-init symptom in issue:43644.", "canonical_pr_reason": null, - "best_issue_reason": "43452 is the most reusable cluster representative because it captures a generic loader-path failure rather than a model-specific symptom, so it is the best issue to anchor deduping nearby GGUF/loading reports.", + "best_issue_reason": "issue:43646 is the best global representative because it is the clearest high-level bug report in the set and maps to a concrete, broadly relevant v5 regression rather than a model-specific symptom.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:44462", "right": "issue:46489", "accept": false, - "reason": "Both are tokenizer-related, but one is AutoTokenizer ignoring `tokenizer.json` and the other is a DeepSeek-Coder output regression; not clearly the same bug." + "reason": "Both are tokenizer-loading regressions, but one is about ignoring tokenizer.json and the other is a DeepSeek-Coder-specific output mismatch; not the same concrete bug." }, { "left": "issue:42796", "right": "issue:43404", "accept": false, - "reason": "Tokenizer BPE-marker handling and Mistral3 `lm_head` tying are unrelated code paths." + "reason": "Different subsystems and failures: tokenizer BPE marker handling vs a tied lm_head weight issue in Mistral3." }, { "left": "issue:46285", "right": "issue:46402", "accept": false, - "reason": "Different regressions: CLIP text model structure vs Sam3 text encoder weight loading." + "reason": "Both mention model structure/loading, but CLIP structure inconsistency and Sam3 text-encoder weights missing are distinct regressions." }, { "left": "issue:36032", "right": "issue:42915", "accept": false, - "reason": "Tokenizer special-token method conflict and Qwen3Moe FP8 failure are unrelated." + "reason": "Unrelated: T5 tokenizer special-token API conflict vs Qwen3Moe FP8 config failure." }, { "left": "issue:43716", "right": "issue:45084", "accept": false, - "reason": "Image preprocessor dtype mismatch and template compilation failure are different subsystems." + "reason": "Different code paths: image preprocessor/model dtype mismatch vs a template compilation error." }, { "left": "issue:32090", "right": "issue:39404", "accept": false, - "reason": "Trainer GPU broadcast TypeError is unrelated to Whisper `return_language` behavior." + "reason": "Trainer GPU broadcast TypeError and Whisper return_language pipeline behavior are unrelated." }, { "left": "issue:44661", "right": "issue:44683", "accept": false, - "reason": "Tokenizer mapping/model-add helper failure and torch flex_attention compilation failure are unrelated." + "reason": "Different areas: tokenizer-mapping model addition vs compiled flex_attention on torch>=2.9." }, { "left": "issue:44737", "right": "issue:45859", "accept": false, - "reason": "Both involve device placement, but the XLNet CPU arange bug and Qwen offload rotary embedding bug are different code paths." + "reason": "Both involve device placement, but XLNet CPU arange and Qwen3_5Moe CPU offload are separate model-specific bugs." }, { "left": "issue:43643", "right": "issue:44530", "accept": false, - "reason": "AutoConfig trust_remote_code field loss and Qwen3.5 PagedAttentionCache crash are unrelated." + "reason": "Trust-remote-code config field loss is unrelated to Qwen3.5 PagedAttentionCache linear_attention crashes." }, { "left": "issue:44451", "right": "issue:44987", "accept": false, - "reason": "Both are model-loading failures, but they point to different models and likely different root causes." + "reason": "Both are model-loading complaints, but they target different models and likely different loader regressions." }, { "left": "issue:40990", "right": "issue:43421", "accept": false, - "reason": "Perplexity regression on gpt-oss-20b is unrelated to runtime post-processor updates in TokenizersBackend." + "reason": "Perplexity regression on GPT-OSS is unrelated to runtime post-processor updates in TokenizersBackend." }, { "left": "issue:42915", "right": "issue:44530", "accept": false, - "reason": "Both mention Qwen, but FP8 config failure and PagedAttentionCache linear_attention crash are distinct bugs." + "reason": "Both involve Qwen models, but FineGrainedFP8Config failure and linear_attention group-type crash are distinct problems." }, { "left": "issue:43541", "right": "issue:45084", "accept": false, - "reason": "Mixtral torch-dynamo grouped_mm tracing error is unrelated to template-node compilation." + "reason": "Mixtral torch-dynamo grouped_mm tracing error is unrelated to the non-template-node compilation error." }, { "left": "issue:43644", "right": "issue:43646", + "accept": true, + "reason": "Both describe the same v5 custom-model initialization regression; 43644 is a concrete symptom of the broader initialization breakage in 43646." + }, + { + "left": "issue:43582", + "right": "issue:43632", "accept": false, - "reason": "Both are Transformers 5 regressions, but one is about non-persistent buffers and the other about custom model initialization; too broad to treat as one bug." + "reason": "Apple Silicon allocator warmup TypeError and the _is_hf_initialized flag regression are unrelated." }, { "left": "issue:42175", "right": "issue:46178", "accept": false, - "reason": "Backend packaging omission and a missing `tensorflow_text` backend entry are related only at a high level." + "reason": "Both are backend/package-related, but missing tensorflow in torch extras and missing tensorflow_text in BACKENDS_MAPPING are different issues." }, { "left": "issue:34567", "right": "issue:40444", "accept": false, - "reason": "Trainer token accounting and Qwen2.5-VL iterable dataset failure are unrelated." - }, - { - "left": "issue:43582", - "right": "issue:43632", - "accept": false, - "reason": "AppleSilicon warmup TypeError and `_is_hf_initialized` flag regression are different failures." + "reason": "TrainerState token-count tracking and Qwen2.5-VL multi-image iterable dataset failures are unrelated." }, { "left": "issue:43883", "right": "issue:44488", "accept": false, - "reason": "Molmo tied-weights attribute error and a specific BERT loading failure are unrelated." + "reason": "Molmo tied-weights metadata and loading cjvt/sleng-bert are unrelated loading issues." }, { "left": "issue:40444", "right": "issue:46326", "accept": false, - "reason": "Multimodal iterable dataset failure and Whisper/Mistral-tokenizer chat-template behavior are different bugs." + "reason": "Different domains and failures: multimodal fine-tuning with multiple images vs mistral_common continue_final_message behavior." }, { "left": "issue:44206", "right": "issue:44220", - "accept": false, - "reason": "Both are audio-feature related, but the titles do not show the same concrete failure or shared root cause." + "accept": true, + "reason": "Both point to the same audio feature-extraction path; 44206 gives the concrete unsupported-center crash that fits the generic _torch_extract_fbank_features() report." }, { "left": "issue:44811", "right": "issue:45584", "accept": false, - "reason": "Whisper `batch_decode` skip_special_tokens handling and empty-transcription generation after align_special_tokens are distinct issues." - }, - { - "left": "issue:43452", - "right": "issue:43482", - "accept": false, - "reason": "Both are GGUF/load failures, but one is a broad `gguf_file` loader regression and the other is a model-specific transformers v5 loading failure; not enough evidence they are the same bug." + "reason": "Whisper batch_decode skip_special_tokens handling and empty-transcription generation after align_special_tokens are different bugs." }, { "left": "issue:36032", "right": "issue:43645", "accept": false, - "reason": "Tokenizer special-token naming conflict and notebook custom-model initialization regression are unrelated." + "reason": "Tokenizer special-token method conflict is unrelated to Jupyter custom-model initialization breakage." }, { - "left": "issue:44987", - "right": "issue:44991", + "left": "issue:43452", + "right": "issue:43482", "accept": false, - "reason": "Different model-loading breakages with different symptoms and likely different root causes." + "reason": "Both mention GGUF loading, but one is a generic gguf_file/from_pretrained breakage and the other is a specific Qwen2.5-GGUF load failure; not clearly the same concrete bug." }, { "left": "issue:43317", "right": "issue:43482", "accept": false, - "reason": "Dequantized offload loading and GGUF v5 loading failure are separate loader problems." + "reason": "Dequantized model offload failure and Qwen2.5-GGUF loading failure are different loader problems." }, { "left": "issue:45084", "right": "issue:45507", "accept": false, - "reason": "Template compilation error and GraniteMoEHybrid invalid-method call are not the same bug." + "reason": "Template compilation TypeError is unrelated to GraniteMoEHybrid calling an invalid method." }, { "left": "issue:43262", "right": "issue:43377", "accept": false, - "reason": "Sampling-rate defaulting in audio chat templates and MIMI batch-vs-single padding-mask mismatch are different bugs." + "reason": "Audio chat-template sampling-rate defaulting and MIMI batched-vs-single mismatch are distinct issues." }, { "left": "issue:43746", "right": "issue:46326", "accept": false, - "reason": "PEFT adapter local-checkpoint loading and Whisper continue_final_message behavior are unrelated." + "reason": "PEFT checkpoint loading for GraniteSpeech and mistral_common continue_final_message are unrelated." }, { "left": "issue:43883", "right": "issue:44991", "accept": false, - "reason": "Missing tied-weights metadata and tokenizer loading failure for a specific model are unrelated." + "reason": "Missing all_tied_weights_keys for Molmo and tokenizer loading failure for EMBEDDIA/est-roberta are unrelated." }, { "left": "issue:45084", "right": "issue:46869", "accept": false, - "reason": "Template-node compilation and multi-GPU device-map loading are different code paths." + "reason": "Non-template-node compilation failure and YOLO multi-GPU device-map loading failure are unrelated." } ] }, - "evaluator_result": null, - "evaluator_used": false, + "evaluator_result": { + "accept": true, + "feedback": "The summary is conservative and grounded in the packet. The choice of issue:43646 as the representative v5 custom-model initialization regression is reasonable, and the two accepted soft pairs (43644/43646 and 44206/44220) are plausibly same-bug clusters without overclaiming. The rejected pairs are mostly framed narrowly enough to avoid overstating duplicates." + }, + "evaluator_used": true, "retried": false, - "accepted_nontrivial_soft_edge": false, + "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null }, @@ -3520,8 +3595,8 @@ "issue:43645|issue:44534", "issue:44849|issue:45325", "issue:43377|issue:43819", - "issue:44464|issue:45987", "issue:43404|issue:44466", + "issue:44464|issue:45987", "issue:44987|issue:45081", "issue:43994|issue:44610", "issue:44451|issue:45081", @@ -3538,221 +3613,227 @@ "issue:43632|issue:46821", "issue:36032|issue:41720", "issue:45216|issue:45357", - "issue:44991|issue:46315", "issue:43531|issue:44877", + "issue:44991|issue:46315", "issue:43425|issue:44403", "issue:43296|issue:43531", "issue:44220|issue:44683", "issue:43611|issue:44534" ], - "prepared_review_unit_hash": "15d1b0883916baca2369dee1fcd584b4126cf4dc12b52cbf8bd64d8e616667c6", + "prepared_review_unit_hash": "2cdb6e76c9eaff568a31695c4add9b39fadce31028f7c36d1da54b9377875f3f", "component_budget": { "node_count": 502, "item_count": 502, - "soft_pair_count": 1842, - "serialized_chars": 956182, - "estimated_input_tokens": 239046, - "estimated_eval_tokens": 478348 + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 30, - "serialized_chars": 36752, - "estimated_input_tokens": 9188, - "estimated_eval_tokens": 18632 + "serialized_chars": 36746, + "estimated_input_tokens": 9187, + "estimated_eval_tokens": 18630 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Mostly unrelated issue cluster with one likely duplicate pair around Qwen3.5 `save_pretrained` checkpoint corruption. No PRs present.", - "confidence": 0.82, - "canonical_issue_reason": "issue:45216 is the broadest and earliest report in the only near-duplicate family, describing the Qwen3.5 `save_pretrained` checkpoint regression.", + "summary": "The set is mostly a heterogeneous mix of unrelated Transformers bugs. I only accept one soft-duplicate pair: the two Qwen3.5 save_pretrained regression reports. Everything else is too model-, path-, or symptom-specific to safely merge.", + "confidence": 0.71, + "canonical_issue_reason": "Issue 45216 is the clearest canonical issue because it states the broad Qwen3.5 save_pretrained regression, and 45357 reads like a narrower follow-up on the same serialization problem.", "canonical_pr_reason": null, - "best_issue_reason": "issue:45216 is the best representative issue for this cluster because it captures the clearest recurring bug theme and is more general than the follow-up report.", + "best_issue_reason": "45216 is the best global issue anchor here: it is the broadest actionable duplicate target among the only clearly overlapping pair.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43646", "right": "issue:44534", "accept": false, - "reason": "Both are Transformer v5 regressions, but one is custom model init and the other is non-persistent buffer serialization; different code paths and symptoms." + "reason": "Both are v5 regressions, but one is custom model initialization and the other is non-persistent buffer corruption; different bugs." }, { "left": "issue:43792", "right": "issue:46869", "accept": false, - "reason": "Different models and failures: Whisper load/run versus YOLO multi-GPU device-map crash." + "reason": "Whisper loading failure and YOLO multi-GPU device-map assertion are unrelated models and code paths." }, { "left": "issue:44403", "right": "issue:45081", "accept": false, - "reason": "Noise/logging on load is unrelated to a tokenizer regex crash in Mistral loading." + "reason": "Generic loading noise and a Mistral tokenizer regex patch crash are not the same underlying issue." }, { "left": "issue:43645", "right": "issue:44534", "accept": false, - "reason": "Notebook/custom-model init regression is not the same bug as junk-filled non-persistent buffers." + "reason": "Notebook custom-model initialization failure is distinct from junk-filled non-persistent buffers." }, { "left": "issue:44849", "right": "issue:45325", "accept": false, - "reason": "Different Qwen3.5 output_hidden_states bug versus Qwen2.5-VL rope/temporal position-id scaling bug." + "reason": "Qwen3.5 output_hidden_states bug and Qwen2.5-VL rope scaling bug are different model families and defects." }, { "left": "issue:43377", "right": "issue:43819", "accept": false, - "reason": "Both are audio-model issues, but one is batching/padding-mask mismatch and the other is DAC latent decoding/STE mismatch." + "reason": "MIMI batching/padding-mask mismatch and DAC from_latents/STE mismatch are unrelated audio-model bugs." }, { - "left": "issue:44464", - "right": "issue:45987", + "left": "issue:43404", + "right": "issue:44466", "accept": false, - "reason": "Chunked generation inconsistencies with compiled forward are not the same as a cache-length return-type contract bug." + "reason": "Both involve lm_head weight behavior, but one is a Mistral3 tying bug and the other is device-dependent serialization inconsistency." }, { - "left": "issue:43404", - "right": "issue:44466", + "left": "issue:44464", + "right": "issue:45987", "accept": false, - "reason": "Both mention tied `lm_head` weights, but one is model construction/tie failure and the other is save_pretrained serialization variance." + "reason": "Chunked generation inconsistency with compiled forward and StaticCache.get_seq_length returning a tensor are separate generation/caching issues." }, { "left": "issue:44987", "right": "issue:45081", "accept": false, - "reason": "Tokenizer/model load failure for one repo is not the same as the Mistral regex patch crash." + "reason": "A model-loading failure and a tokenizer regex patch crash do not look like the same bug." }, { "left": "issue:43994", "right": "issue:44610", "accept": false, - "reason": "Different vision bugs: nonsensical inference output versus incorrect processor image size." + "reason": "SigLIP2 nonsensical outputs and OmDet-Turbo processor size mismatch are different model-specific problems." }, { "left": "issue:44451", "right": "issue:45081", "accept": false, - "reason": "ScandiBERT load failure is unrelated to the Mistral tokenizer backend crash." + "reason": "Loading ScandiBERT and Mistral tokenizer regex patching are unrelated failures." }, { "left": "issue:42915", "right": "issue:44291", "accept": false, - "reason": "FineGrainedFP8Config failure and unexpected `_is_hf_initialized` argument are separate initialization/compatibility problems." + "reason": "Qwen3Moe FP8 config failure and init_empty_weights/_is_hf_initialized argument breakage are different paths." }, { "left": "issue:28530", "right": "issue:34567", "accept": false, - "reason": "Early stopping metric selection and token-count tracking are unrelated trainer issues." + "reason": "Early stopping metric lookup and num_input_tokens_seen not updating are unrelated trainer bugs." }, { "left": "issue:44008", "right": "issue:46899", "accept": false, - "reason": "Different Gemma multimodal regressions: variable-name collision in forward versus unsafe dtype casting in 4-bit embedders." + "reason": "Gemma3n variable collision and Gemma4 quantization dtype-cast issues are different multimodal bugs." }, { "left": "issue:36032", "right": "issue:43742", "accept": false, - "reason": "Tokenizer method-name conflict and MobileLLM key error are different loading bugs." + "reason": "T5 tokenizer special-token conflict and MobileLLM key error are not the same defect." }, { "left": "issue:44062", "right": "issue:44987", "accept": false, - "reason": "Duplicate `AddedToken(special=...)` kwargs is not the same as the physical-intelligence/fast tokenizer load failure." + "reason": "AddedToken keyword conflict and a model-loading failure for physical-intelligence/fast are unrelated." }, { "left": "issue:44265", "right": "issue:44610", "accept": false, - "reason": "Export failure with `torch_compilable_check` is unrelated to OmDet-Turbo processor image-size mismatch." + "reason": "torch.export with torch_compilable_check and OmDet-Turbo image-size mismatch are different issues." }, { "left": "issue:42915", "right": "issue:44589", "accept": false, - "reason": "Different failure modes: FP8 config handling versus missing Float8 storage type resolution." + "reason": "Qwen3Moe FP8 config failure and Float8 storage lookup failure are distinct runtime problems." + }, + { + "left": "issue:43089", + "right": "issue:45987", + "accept": false, + "reason": "Generation overhead/performance regression is not the same as a return-type contract bug in StaticCache." }, { "left": "issue:43322", "right": "issue:43540", "accept": false, - "reason": "Llava Next load segfault and Qwen3OmniMoe video-processing ValueError are distinct multimodal bugs." + "reason": "Llava Next loading segfault and Qwen3OmniMoe video-input ValueError are unrelated multimodal failures." }, { "left": "issue:44479", "right": "issue:45325", "accept": false, - "reason": "Both involve Qwen VL families, but one is broad video-input regression and the other is still-image temporal position scaling; not the same defect." + "reason": "A broad video-input regression across Qwen models is not the same as the still-image temporal position_ids bug." }, { "left": "issue:43632", "right": "issue:46821", "accept": false, - "reason": "`_is_hf_initialized` flag breakage and TimesFM `window_size` attribute error are unrelated transformer-v5 issues." + "reason": "_is_hf_initialized regression and TimesFM window_size AttributeError are unrelated." }, { "left": "issue:36032", "right": "issue:41720", "accept": false, - "reason": "Tokenizer load conflict and Qwen3 auto device-map CUDA assert do not share a code path." + "reason": "Tokenizer constructor conflict and Qwen3 auto device mapping CUDA assert are different bugs." }, { "left": "issue:45216", "right": "issue:45357", "accept": true, - "reason": "Both report the same Qwen3.5 `save_pretrained` checkpoint corruption regression; the later issue is a narrower follow-up on incorrect visual encoder keys." + "reason": "Both report the same Qwen3.5 save_pretrained serialization regression; 45357 is a narrower manifestation about incorrect visual encoder keys." }, { - "left": "issue:44991", - "right": "issue:46315", + "left": "issue:43531", + "right": "issue:44877", "accept": false, - "reason": "Tokenizer load failure and slow `convert_tokens_to_ids` are different tokenizer problems, one correctness and one performance." + "reason": "Qwen3-MoE sliding_window behavior and granite_speech config loading are unrelated config/model issues." }, { - "left": "issue:43531", - "right": "issue:44877", + "left": "issue:44991", + "right": "issue:46315", "accept": false, - "reason": "Sliding-window behavior in Qwen3-MoE is unrelated to strict config rejection for granite_speech." + "reason": "Tokenizer loading failure and convert_tokens_to_ids slowdown are different symptoms and fixes." }, { "left": "issue:43425", "right": "issue:44403", "accept": false, - "reason": "Torch version compatibility is not the same as loading-noise behavior." + "reason": "Torch 2.10 incompatibility and loading noise are unrelated concerns." }, { "left": "issue:43296", "right": "issue:43531", "accept": false, - "reason": "PaddleOCR-VL load failure in vLLM and Qwen3-MoE sliding-window behavior are unrelated." + "reason": "PaddleOCR-VL load failure and Qwen3-MoE sliding_window issue are distinct problems." }, { "left": "issue:44220", "right": "issue:44683", "accept": false, - "reason": "Audio feature extraction issue and compiled flex_attention torch-version failure are different subsystems." + "reason": "fbank feature extraction and compiled flex_attention failures are unrelated subsystems." }, { "left": "issue:43611", "right": "issue:44534", "accept": false, - "reason": "`base_model_prefix` loading breakage and non-persistent buffer corruption are separate v5 regressions." + "reason": "base_model_prefix loading breakage and non-persistent buffer junk are different v5 regressions." } ] }, "evaluator_result": { "accept": true, - "feedback": "The summary is grounded in the packet: it stays conservative, identifies the only clearly related issue pair (45216/45357), and does not overclaim any broader duplication. No PRs are present, and the soft-edge verdicts are appropriately cautious." + "feedback": "The summary is well grounded in the packet and stays conservative. The lone accepted pair, issue:45216 vs issue:45357, is a plausible same-bug family around Qwen3.5 `save_pretrained` serialization, and the remaining soft-edge rejections do not overstate duplication based on the titles/body summaries provided." }, "evaluator_used": true, "retried": false, @@ -3844,8 +3925,8 @@ "issue:43531|issue:44291", "issue:43632|issue:44534", "issue:42175|issue:45542", - "issue:44462|issue:45812", "issue:41720|issue:45081", + "issue:44462|issue:45812", "issue:43317|issue:43856", "issue:44079|issue:44534", "issue:44509|issue:46909", @@ -3853,215 +3934,218 @@ "issue:43377|issue:45491", "issue:43611|issue:43646" ], - "prepared_review_unit_hash": "3a4d2622e31476e4ca9e8c253d25895a5f7cfb0328b659e6c48e9321c8f4e549", + "prepared_review_unit_hash": "712dd4a58f61c34d18688e023090735e7db41f5a997507b490a21b3286f4a08b", "component_budget": { "node_count": 502, "item_count": 502, - "soft_pair_count": 1842, - "serialized_chars": 956182, - "estimated_input_tokens": 239046, - "estimated_eval_tokens": 478348 + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 29, - "serialized_chars": 36439, - "estimated_input_tokens": 9110, - "estimated_eval_tokens": 18476 + "serialized_chars": 36441, + "estimated_input_tokens": 9111, + "estimated_eval_tokens": 18478 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a grab-bag of unrelated Transformers/model-tokenizer regressions and docs issues. The soft pairs are only superficially similar; none look like the same underlying bug or change, and there are no PRs to merge here.", - "confidence": 0.87, - "canonical_issue_reason": "issue:43646 is the broadest v5 model-initialization regression in the set, so it is the least bad representative, though the cluster is not truly duplicate-heavy.", + "summary": "Most items are unrelated issue reports; the only plausible duplicate pair is the Transformers v5 model-initialization regression around `base_model_prefix` / custom model init. No PRs are present.", + "confidence": 0.67, + "canonical_issue_reason": "issue:43646 is the broadest and most representative report of the v5 model-initialization regression, and it plausibly subsumes the more specific loading failure in issue:43611.", "canonical_pr_reason": null, - "best_issue_reason": "issue:43646 is the most general and central of the v5 loading/init breakage reports, making it the best single issue to anchor the cluster.", + "best_issue_reason": "issue:43646 best serves as the cluster anchor because it describes the underlying API breakage at the right abstraction level and is likely to cover the closely related `base_model_prefix` loading symptom.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:40444", "right": "issue:45325", "accept": false, - "reason": "Both are Qwen2.5-VL-related, but one is multi-image IterableDataset finetuning and the other is still-image temporal position-id scaling; different bugs." + "reason": "Both are Qwen2.5-VL, but one is a dataloader/multimodal-training failure and the other is a RoPE position-id bug; different code paths." }, { "left": "issue:43716", "right": "issue:46899", "accept": false, - "reason": "Both involve multimodal dtype handling, but one is image preprocessor/model dtype mismatch and the other is unsafe 4-bit cast in embedders; not the same path." + "reason": "Both involve multimodal dtype issues, but they hit different models and different mechanisms (preprocessor/model dtype mismatch vs unsafe cast in 4-bit embedders)." }, { "left": "issue:45812", "right": "issue:46491", "accept": false, - "reason": "Both are tokenizer regressions, but Granite token-ID corruption and CodeLlama leading-space round-trip loss are unrelated behaviors." + "reason": "Both are tokenizer regressions, but Granite wrong IDs and CodeLlama leading-space round-trip are distinct bugs in different tokenizers." }, { "left": "issue:44403", "right": "issue:44991", "accept": false, - "reason": "Generic loading noise vs AutoTokenizer ignoring tokenizer.json for a specific model; different failure modes." + "reason": "Unnecessary loading noise and a tokenizer-load failure for a specific model are not the same bug." }, { "left": "issue:43381", "right": "issue:45923", "accept": false, - "reason": "Eval-mode gradient-checkpointing misuse vs missing supports_gradient_checkpointing on a trust_remote_code model; distinct issues." + "reason": "Gradient checkpointing eval-mode behavior and a missing capability flag in a trust_remote_code model are related but not the same defect." }, { "left": "issue:43632", "right": "issue:44756", "accept": false, - "reason": "_is_hf_initialized regression vs mmap OOM on Strix Halo; no shared code-path." + "reason": "A v5 initialization flag regression and an AppleSilicon mmap OOM workaround are unrelated." }, { "left": "issue:42796", "right": "issue:43746", "accept": false, - "reason": "Tokenizer BPE-marker handling and PEFT adapter checkpoint loading are unrelated problems." + "reason": "Tokenizer BPE marker handling and GraniteSpeech PEFT checkpoint loading are different issues." }, { "left": "issue:43792", "right": "issue:44589", "accept": false, - "reason": "Whisper runtime failure vs Float8 storage lookup TypeError; different bugs." + "reason": "Whisper runtime failure and missing Float8 storage registration are unrelated." }, { "left": "issue:33666", "right": "issue:40444", "accept": false, - "reason": "Same broad Qwen-VL family, but multi-GPU training and multi-image finetuning failure are separate issues." + "reason": "Both are Qwen VL training issues, but one is multi-GPU training and the other is multi-image iterable dataset failure; not the same bug." }, { "left": "issue:40990", "right": "issue:44811", "accept": false, - "reason": "Perplexity regression on gpt-oss-20b and Whisper batch_decode skip_special_tokens are unrelated." + "reason": "Perplexity on gpt-oss-20b and Whisper batch_decode skipping specials are unrelated." }, { "left": "issue:44291", "right": "issue:44589", "accept": false, - "reason": "Both are load/init failures, but unexpected _is_hf_initialized and missing Float8 storage are different concrete regressions." + "reason": "Both touch v5 loading/runtime regressions, but one is an init_empty_weights argument break and the other is a Float8 storage lookup failure." }, { "left": "issue:42796", "right": "issue:46710", "accept": false, - "reason": "Tokenizer BPE-marker handling and a DeepSeek accuracy regression after upgrade do not match." + "reason": "Tokenizer marker handling and DeepSeek accuracy regression are not the same underlying issue." }, { "left": "issue:43482", "right": "issue:43525", "accept": false, - "reason": "Qwen2.5-GGUF loading failure and Llama4Config missing pad_token_id are separate model-specific issues." + "reason": "Qwen2.5-GGUF loading failure and missing `pad_token_id` on Llama4Config are separate model-specific regressions." }, { "left": "issue:35707", "right": "issue:40444", "accept": false, - "reason": "Progressive generation with inputs_embeds/past_key_values is unrelated to Qwen2.5-VL multi-image dataset finetuning." + "reason": "Progressive generation with `inputs_embeds`/`past_key_values` is unrelated to Qwen2.5-VL multi-image finetuning." }, { "left": "issue:44079", "right": "issue:45663", "accept": false, - "reason": "ModelOutput key assignment and Gemma4 FSDP2 shared_kv_states rebuilding are unrelated internals." + "reason": "ModelOutput key assignment and Gemma4 FSDP2 shared_kv state rebuilding are unrelated." }, { "left": "issue:41553", "right": "issue:43582", "accept": false, - "reason": "A bad tokenizer error message and an Apple Silicon allocator TypeError are different bugs." + "reason": "A bad tokenizer-loading error message and an AppleSilicon allocator TypeError are unrelated." }, { "left": "issue:43482", "right": "issue:45538", "accept": false, - "reason": "Qwen2.5-GGUF loading and CLIPTokenizer model_max_length are not the same issue." + "reason": "GGUF loading failure and CLIPTokenizer max-length default are different tokenizer/model-loading issues." }, { "left": "issue:43421", "right": "issue:44568", "accept": false, - "reason": "Both touch special-token behavior, but runtime post-processor updates and missing BOS/EOS insertion are different code paths." + "reason": "Runtime special-token post-processing and mdeberta BOS/EOS insertion are distinct tokenizer behaviors." }, { "left": "issue:43531", "right": "issue:44291", "accept": false, - "reason": "Qwen3-MoE sliding_window behavior and init_empty_weights _is_hf_initialized TypeError are unrelated." + "reason": "Qwen3-MoE sliding window behavior and an init_empty_weights argument regression do not match." }, { "left": "issue:43632", "right": "issue:44534", "accept": false, - "reason": "Both are v5 regressions, but one is initialization-flag handling and the other is non-persistent buffer corruption." + "reason": "The `_is_hf_initialized` regression and non-persistent buffer corruption are different v5 bugs." }, { "left": "issue:42175", "right": "issue:45542", "accept": false, - "reason": "Both mention TensorFlow/backend packaging, but one is missing tf backend under .[torch] and the other is tensorboard-only causing undefined tf backend." + "reason": "Both involve TensorFlow/backend packaging, but they describe different install paths and symptoms; not clearly the same bug." }, { - "left": "issue:44462", - "right": "issue:45812", + "left": "issue:41720", + "right": "issue:45081", "accept": false, - "reason": "AutoTokenizer ignoring tokenizer.json and Granite token-ID regression are different tokenizer bugs." + "reason": "Qwen3 auto device-mapping asserts and a Mistral regex-patching AttributeError are unrelated." }, { - "left": "issue:41720", - "right": "issue:45081", + "left": "issue:44462", + "right": "issue:45812", "accept": false, - "reason": "Qwen3 device-map cudaErrorAssert and Mistral regex patch crash are unrelated." + "reason": "Ignoring repository tokenizer.json and wrong Granite token IDs are different AutoTokenizer regressions." }, { "left": "issue:43317", "right": "issue:43856", "accept": false, - "reason": "device_map=auto dequantized-model offload failure and Qwen3 MoE memory inefficiency are separate memory/device issues." + "reason": "GPU+CPU offload loading failure and Qwen3 MoE training memory inefficiency are not the same issue." }, { "left": "issue:44079", "right": "issue:44534", "accept": false, - "reason": "ModelOutput key assignment bug and junk-filled non-persistent buffers are unrelated." + "reason": "ModelOutput key handling and junk-filled non-persistent buffers are unrelated." }, { "left": "issue:44509", "right": "issue:46909", "accept": false, - "reason": "Both are docs problems, but removed pipeline-task docs and broken DIA example are not the same change." + "reason": "Removed v5 pipeline docs and broken DIA documentation examples are different documentation issues." }, { "left": "issue:43296", "right": "issue:46821", "accept": false, - "reason": "PaddleOCR-VL load failure in vLLM and TimesFM window_size AttributeError are different model-specific bugs." + "reason": "PaddleOCR-VL load failure and TimesFM window_size AttributeError are unrelated model-specific bugs." }, { "left": "issue:43377", "right": "issue:45491", "accept": false, - "reason": "Both involve batching/masking, but missing padding-mask support in MIMI and Gemma3 NaNs from sliding-window all-padding windows are different failures." + "reason": "MIMI batching/padding-mask mismatch and Gemma3 mixed-length NaN embeddings are both batching-related, but not the same underlying bug." }, { "left": "issue:43611", "right": "issue:43646", - "accept": false, - "reason": "Both are Transformer v5 loading/init regressions, but base_model_prefix loading breakage and custom model initialization failure are distinct concrete bugs." + "accept": true, + "reason": "Both report the same Transformers v5 model-loading/initialization regression; `base_model_prefix` failures are a concrete symptom of the broader custom-init breakage." } ] }, - "evaluator_result": null, - "evaluator_used": false, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet, and the soft-edge verdicts are generally conservative. The only accepted pair, issue:43611 with issue:43646, is a reasonable duplicate call at the v5 custom-init/model-loading level. No PRs are present, so there is no mergeability concern." + }, + "evaluator_used": true, "retried": false, - "accepted_nontrivial_soft_edge": false, + "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null }, @@ -4132,8 +4216,8 @@ "issue:42175|issue:43296", "issue:43644|issue:44079", "issue:43818|issue:45357", - "issue:43493|issue:43994", "issue:44265|issue:45072", + "issue:43493|issue:43994", "issue:40444|issue:45381", "issue:45325|issue:45381", "issue:45325|issue:45834", @@ -4158,20 +4242,20 @@ "issue:41720|issue:45823", "issue:43377|issue:44610" ], - "prepared_review_unit_hash": "20fc42297a818405df5ea42fe1fc8d3553da591823f11674eb1528bad9fe37b9", + "prepared_review_unit_hash": "e81492b087f7236954c284a39eaa5d15640127889642451f5b437c2f324e5173", "component_budget": { "node_count": 502, "item_count": 502, - "soft_pair_count": 1842, - "serialized_chars": 956182, - "estimated_input_tokens": 239046, - "estimated_eval_tokens": 478348 + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 30, - "serialized_chars": 36277, + "serialized_chars": 36280, "estimated_input_tokens": 9070, "estimated_eval_tokens": 18396 }, @@ -4180,199 +4264,202 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This set does not look like a duplicate cluster: the soft pairs span different models, subsystems, and failure modes, so every candidate edge should be rejected.", - "confidence": 0.96, - "canonical_issue_reason": "No good canonical issue: the items span unrelated bugs (tokenization, loading, vision, dtype, offloading, training regressions, etc.) rather than one underlying defect.", + "summary": "Most soft pairs are broad-similarity false positives across unrelated models/subsystems. The only clear duplicate-like pair is the SigLIP2 implementation mismatch vs. nonsensical AutoModel/pipeline outputs.", + "confidence": 0.62, + "canonical_issue_reason": "issue:43493 is the clearest, most specific statement of the underlying SigLIP2 bug and best matches the duplicate candidate that reports bad downstream outputs.", "canonical_pr_reason": null, - "best_issue_reason": "No single issue is a strong global representative for the same reason; the closest-looking pairs still describe different concrete code paths and symptoms.", + "best_issue_reason": "issue:43493 is the best representative issue in this set because it precisely names the root discrepancy and can subsume the more user-facing symptom report.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43818", "right": "issue:46402", "accept": false, - "reason": "Different bugs in different models: Video-LLaVA temporal-attention/weight-sharing issue vs Sam3Model text-encoder weight loading." + "reason": "Different models and failures: Video-LLaVA missing temporal attention/weight sharing vs Sam3 text encoder weights not loaded." }, { "left": "issue:44829", "right": "issue:45479", "accept": false, - "reason": "Both affect sequence classification, but one is flash_attention_3 training degeneration while the other is a num_labels=1 zero-loss issue." + "reason": "Both affect training, but the causes differ: flash_attention_3 behavior vs single-label classification with num_labels=1." }, { "left": "issue:42175", "right": "issue:43296", "accept": false, - "reason": "Unrelated failures: package/backend install metadata vs loading PaddleOCR-VL in vLLM." + "reason": "Unrelated areas: backend packaging for TensorFlow vs loading PaddleOCR-VL in vLLM." }, { "left": "issue:43644", "right": "issue:44079", "accept": false, - "reason": "Different code paths: non-persistent buffer initialization vs ModelOutput key assignment." + "reason": "One is a v5 buffer initialization issue; the other is ModelOutput key assignment behavior. Not the same bug." }, { "left": "issue:43818", "right": "issue:45357", "accept": false, - "reason": "Different models and defects: Video-LLaVA architecture problem vs Qwen3.5 save_pretrained key serialization." - }, - { - "left": "issue:43493", - "right": "issue:43994", - "accept": false, - "reason": "Same model family, but different symptoms and likely different causes: JAX parity discrepancy vs nonsensical AutoModel/pipeline outputs." + "reason": "Both mention visual models, but one is Video-LLaVA architecture problems and the other is Qwen3.5 save_pretrained key serialization." }, { "left": "issue:44265", "right": "issue:45072", "accept": false, - "reason": "Export-time torch_compilable_check failure is unrelated to bfloat16 dtype mismatches in SwitchTransformers/TimmWrapperModel." + "reason": "Different failure modes: torch.export with torch_compilable_check vs dtype mismatches in bfloat16 inference." + }, + { + "left": "issue:43493", + "right": "issue:43994", + "accept": true, + "reason": "Same underlying SigLIP2 implementation problem: HF/JAX discrepancy aligns with nonsensical outputs in AutoModel/pipeline." }, { "left": "issue:40444", "right": "issue:45381", "accept": false, - "reason": "Different multimodal bugs: multi-image IterableDataset finetuning failure vs video vision_position_ids regression." + "reason": "Different Qwen2.5-VL issues: finetuning with multi-image iterable datasets vs video vision_position_ids regression." }, { "left": "issue:45325", "right": "issue:45381", "accept": false, - "reason": "Both concern Qwen2.5-VL positions, but one is still-image temporal scaling and the other is video vision_position_ids; not the same defect." + "reason": "Both are position-id bugs in Qwen2.5-VL, but one is still-image temporal scaling and the other is video vision_position_ids; not clearly the same concrete bug." }, { "left": "issue:45325", "right": "issue:45834", "accept": false, - "reason": "Qwen2.5-VL rope/position handling is unrelated to Kosmos2.5 long OCR index errors." + "reason": "Different models and symptoms: Qwen2.5-VL rope indexing vs Kosmos2.5 OCR long-input index error." }, { "left": "issue:43454", "right": "issue:44466", "accept": false, - "reason": "Both mention lm_head weights, but one is missing weight tying in a config and the other is device-dependent serialization behavior." + "reason": "Related to tied lm_head weights, but one is a model config failing to tie weights and the other is inconsistent serialization behavior." }, { "left": "issue:42915", "right": "issue:43482", "accept": false, - "reason": "Different models and failures: Qwen3Moe FP8 config issue vs Qwen2.5-GGUF loading under transformers v5." + "reason": "Different models and problems: FineGrainedFP8Config failure vs GGUF loading on transformers v5." }, { "left": "issue:44291", "right": "issue:44877", "accept": false, - "reason": "Different loading problems: init_empty_weights argument handling vs strict config validation for granite_speech." + "reason": "Different loading failures: init_empty_weights argument mismatch vs strict config blocking granite_speech." }, { "left": "issue:43701", "right": "issue:45325", "accept": false, - "reason": "Checkpoint resume key mismatch is unrelated to Qwen2.5-VL rope index scaling." + "reason": "No meaningful overlap: checkpoint key mismatch vs Qwen2.5-VL rope-index scaling." }, { "left": "issue:43873", "right": "issue:45701", "accept": false, - "reason": "Quantized offloading behavior is unrelated to tokenizer-version-dependent tokenization changes." + "reason": "Quantization/offloading behavior is unrelated to tokenization changes." }, { "left": "issue:42796", "right": "issue:45381", "accept": false, - "reason": "Tokenizer BPE-marker handling is unrelated to Qwen2.5-VL video position-id bugs." + "reason": "Tokenizer BPE marker handling is unrelated to Qwen2.5-VL video position IDs." }, { "left": "issue:43606", "right": "issue:44265", "accept": false, - "reason": "CPU offload device mismatch and torch.export/compilable-check failures are different code paths." + "reason": "Bark CPU offload device mismatch and torch.export/torch_compilable_check are separate code paths." }, { "left": "issue:40990", "right": "issue:45584", "accept": false, - "reason": "Perplexity regression on GPT-OSS is unrelated to Whisper empty-transcription generation failure." + "reason": "Different tasks and bugs: perplexity on GPT-OSS vs Whisper empty-transcription failure." }, { "left": "issue:43262", "right": "issue:43746", "accept": false, - "reason": "Audio chat-template sample-rate defaulting is unrelated to GraniteSpeech PEFT checkpoint loading." + "reason": "Audio processor sampling-rate defaults and GraniteSpeech checkpoint loading are unrelated." }, { "left": "issue:44479", "right": "issue:45823", "accept": false, - "reason": "Different failures: Qwen video-input regression vs Gemma4 PLE device_map forward mismatch." + "reason": "Different regressions: Qwen video-input behavior vs Gemma4 PLE device mismatch during forward." }, { "left": "issue:43881", "right": "issue:44488", "accept": false, - "reason": "Different model-loading failures for unrelated models." + "reason": "Different model loading failures for unrelated models." }, { "left": "issue:43828", "right": "issue:46821", "accept": false, - "reason": "Autocast dtype mismatch in Phi-tiny-MoE is unrelated to TimesFM window_size AttributeError." + "reason": "Autocast dtype mismatch is unrelated to a TimesFM window_size AttributeError." }, { "left": "issue:35707", "right": "issue:41720", "accept": false, - "reason": "Progressive generation with inputs_embeds/past_key_values is unrelated to Qwen3 auto device-mapping CUDA asserts." + "reason": "Progressive generation with inputs_embeds/past_key_values is unrelated to Qwen3 auto device mapping cuda assert." }, { "left": "issue:34634", "right": "issue:35707", "accept": false, - "reason": "BarkProcessor voice_preset failure is unrelated to progressive generation with inputs_embeds and past_key_values." + "reason": "Bark voice_preset is unrelated to generation with inputs_embeds and past_key_values." }, { "left": "issue:39692", "right": "issue:40444", "accept": false, - "reason": "SigLIP2 doc-example problems are unrelated to Qwen2.5-VL multi-image finetuning failures." + "reason": "SigLIP2 doc-example errors are unrelated to Qwen2.5-VL multi-image finetuning failures." }, { "left": "issue:43454", "right": "issue:43550", "accept": false, - "reason": "AyaVision lm_head tying/serialization issue is unrelated to Bamba SDPA torch.compile failure." + "reason": "Tied-weight serialization/loading issue is unrelated to Bamba torch.compile SDPA failure." }, { "left": "issue:43606", "right": "issue:43819", "accept": false, - "reason": "CPU offload device mismatch is unrelated to DAC.from_latents vs forward-pass mismatch with missing STE." + "reason": "Device offload mismatch and DAC forward/from_latents STE mismatch are different bugs." }, { "left": "issue:44186", "right": "issue:44265", "accept": false, - "reason": "Tokenizer crashes on NER padding/truncation are unrelated to torch.export failures from torch_compilable_check." + "reason": "Tokenizer NER/padding crashes are unrelated to torch.export and torch_compilable_check." }, { "left": "issue:41720", "right": "issue:45823", "accept": false, - "reason": "Qwen3 CUDA assert on A800 is unrelated to Gemma4 PLE device mismatch during forward." + "reason": "Both are device-map related, but they affect different models and different errors." }, { "left": "issue:43377", "right": "issue:44610", "accept": false, - "reason": "MIMI padding-mask batch inconsistency is unrelated to OmDet-Turbo processor/model input-size mismatch." + "reason": "MIMI padding-mask batching bug and OmDet-Turbo processor/model size mismatch are unrelated." } ] }, - "evaluator_result": null, - "evaluator_used": false, + "evaluator_result": { + "accept": true, + "feedback": "The analysis is grounded in the packet and stays conservative on the soft-edge pairs. The SigLIP2 pair (43493/43994) is the only clearly duplicate-like match from the available titles, and the rest of the rejections are appropriately cautious rather than overstated. The representative/canonical choice is also well supported by the issue titles." + }, + "evaluator_used": true, "retried": false, - "accepted_nontrivial_soft_edge": false, + "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null }, @@ -4465,226 +4552,226 @@ "issue:44877|issue:44991", "issue:43065|issue:46402", "issue:40990|issue:45701", - "issue:26413|issue:28530", "issue:44186|issue:45072", "issue:41720|issue:42617", + "issue:26413|issue:28530", "issue:43550|issue:43720", "issue:43873|issue:44756" ], - "prepared_review_unit_hash": "7fe41f2e8fc5e2e3bf3f256a3ef198afdbc586ded43855601dc1431437184b88", + "prepared_review_unit_hash": "e28134e280f1359b6308cd258cfc54909d44a529eba2fd1943b33da8f73a24d7", "component_budget": { "node_count": 502, "item_count": 502, - "soft_pair_count": 1842, - "serialized_chars": 956182, - "estimated_input_tokens": 239046, - "estimated_eval_tokens": 478348 + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 31, - "serialized_chars": 36971, - "estimated_input_tokens": 9243, - "estimated_eval_tokens": 18742 + "serialized_chars": 36967, + "estimated_input_tokens": 9242, + "estimated_eval_tokens": 18740 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is heterogeneous: the soft pairs are mostly model-specific bugs, tokenizer regressions, or loading/offloading issues that only share broad subsystem terms. None look like true duplicates.", - "confidence": 0.94, - "canonical_issue_reason": null, + "summary": "This cluster is mostly a loose collection of Transformer v5 regressions across tokenization, loading/offloading, config handling, and multimodal model bugs. The soft edges look thematically similar in a few places, but they are not the same underlying defects, so I would not merge any of them as duplicates.", + "confidence": 0.79, + "canonical_issue_reason": "issue:45920 is the broadest, most representative issue here: it captures a v5 tokenizer regression affecting multiple model families, which is the closest shared theme among the tokenizer-related items in the cluster.", "canonical_pr_reason": null, - "best_issue_reason": "If a representative issue is needed, 43873 is the broadest user-facing regression here, but it still does not unify the rest of the cluster and should not be treated as a duplicate hub.", + "best_issue_reason": "issue:45920 is the best anchor for this cluster because it is concrete, well-scoped, and broad enough to represent the recurring tokenizer regression theme better than the more model-specific or narrow reports.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:46305", "right": "issue:46315", "accept": false, - "reason": "Different bugs: one is an off-by-one in sliding-window masking, the other is a tokenizer performance regression in convert_tokens_to_ids." + "reason": "Different subsystems: a streaming mask off-by-one bug vs a tokenizer performance regression." }, { "left": "issue:43257", "right": "issue:45923", "accept": false, - "reason": "Both mention Qwen3/Nemotron-style model plumbing, but the failures are different: MOE loading with accelerate+deepspeed vs a missing supports_gradient_checkpointing flag." + "reason": "Different failures: Qwen3 MOE weight conversion/loading vs a missing gradient-checkpointing flag." }, { "left": "issue:44568", "right": "issue:45920", "accept": false, - "reason": "Both are tokenizer-related v5 regressions, but they affect different tokenizers and different failure modes (special-token insertion vs wrong token IDs)." + "reason": "Both are tokenizer regressions, but they affect different model families and different concrete behaviors." }, { "left": "issue:42796", "right": "issue:46209", "accept": false, - "reason": "Unrelated: one is a tokenizer BPE marker handling bug, the other is FP8 inference failing on B200 hardware." + "reason": "Tokenizer BPE-marker handling is unrelated to FP8 inference failure on B200." }, { "left": "issue:43454", "right": "issue:45072", "accept": false, - "reason": "Different code paths: incorrect batched video processor shapes vs dtype mismatches in model inference." + "reason": "lm_head tying bug vs dtype mismatch in inference; not the same code-path." }, { "left": "issue:45823", "right": "issue:46821", "accept": false, - "reason": "Device-mapping/forward mismatch vs a TimesFM window_size AttributeError; no shared concrete bug." + "reason": "Device-map forward mismatch and a window_size AttributeError are unrelated." }, { "left": "issue:43377", "right": "issue:44898", "accept": false, - "reason": "Both are batch-vs-single inconsistencies, but they involve different models and root causes; not the same underlying bug." + "reason": "Both involve model input handling, but one is padding-mask batching in MIMI and the other is Perceiver positional interpolation." }, { "left": "issue:43454", "right": "issue:43720", "accept": false, - "reason": "Video processor shape bug and BitNet packed-weight loading bug are separate issues." + "reason": "Different defects: tied-weight serialization vs packed-weight unpacking during accelerate loading." }, { "left": "issue:40990", "right": "issue:42947", "accept": false, - "reason": "Perplexity/quality issue on one model versus gradient checkpointing ineffectiveness with PEFT LoRA; unrelated." + "reason": "Perplexity quality issue vs gradient checkpointing behavior; no shared bug." }, { "left": "issue:42617", "right": "issue:42915", "accept": false, - "reason": "A 3d_parallel.py runtime problem and a Qwen3MoE FineGrainedFP8Config failure are different code paths." + "reason": "A 3d_parallel.py runtime failure is unrelated to Qwen3Moe FineGrainedFP8 loading." }, { "left": "issue:43317", "right": "issue:43873", "accept": false, - "reason": "Both involve loading/offloading, but one is dequantized model loading with device_map=auto and the other is a broader quantization offload regression; not the same bug." + "reason": "Both mention offloading/quantization, but the concrete failures and code paths differ too much to treat as the same bug." }, { "left": "issue:43749", "right": "issue:44534", "accept": false, - "reason": "FSDP RAM-efficient loading breakage is distinct from non-persistent buffers being filled with junk." + "reason": "FSDP CPU RAM efficient loading vs non-persistent buffer initialization are different issues." }, { "left": "issue:43572", "right": "issue:45759", "accept": false, - "reason": "Missing pad_token_idx in a config is unrelated to AutoModelForCausalLM.from_config not unwrapping text_config." + "reason": "Missing config field vs incorrect unwrapping of text_config in composite multimodal configs." }, { "left": "issue:43526", "right": "issue:43883", "accept": false, - "reason": "BeitImageProcessorFast label reduction bug is unrelated to Molmo missing all_tied_weights_keys." + "reason": "Image processor label reduction bug vs missing all_tied_weights_keys attribute; unrelated." }, { "left": "issue:43322", "right": "issue:46050", "accept": false, - "reason": "Llava Next loading segfault and sinq save/load breakage are different failures." + "reason": "A Llava Next segmentation fault is not the same as a sinq save/load quantization regression." }, { "left": "issue:41720", "right": "issue:43322", "accept": false, - "reason": "Qwen3 auto-device-map cudaErrorAssert and Llava Next segmentation fault are different model-loading bugs." + "reason": "Different model families and different failure modes: cudaErrorAssert vs segmentation fault." }, { "left": "issue:43881", "right": "issue:44991", "accept": false, - "reason": "Both are tokenizer load problems, but they affect different models and have different symptoms/root causes." + "reason": "Model loading failure and tokenizer loading failure are unrelated." }, { "left": "issue:36331", "right": "issue:42617", "accept": false, - "reason": "Unexpected num_items_in_batch in CustomTrainer.compute_loss is unrelated to a 3d_parallel.py failure." + "reason": "Trainer API signature breakage is unrelated to the 3d_parallel.py execution failure." }, { "left": "issue:44387", "right": "issue:45704", "accept": false, - "reason": "Different memory issues: int4 quantization OOM in Transformers 5.x vs apex FusedRMSNorm memory leak in T5." + "reason": "Quantization memory regression vs apex RMSNorm memory leak are different problems." }, { "left": "issue:41720", "right": "issue:44977", "accept": false, - "reason": "Both involve Qwen3.5/Qwen3 and generation failures, but one is device-map CUDA assert and the other is flash-attention generation corruption; not clearly the same bug." - }, - { - "left": "issue:36331", - "right": "issue:42915", - "accept": false, - "reason": "Trainer API signature mismatch and Qwen3MoE FP8 loading failure are unrelated." + "reason": "Qwen3 auto device-map assertion and Qwen3.5 flash-attention generation issues are not the same concrete bug." }, { "left": "issue:44466", "right": "issue:45357", "accept": false, - "reason": "Both touch save_pretrained/serialization, but one is tied lm_head weight serialization and the other is incorrect visual encoder keys for Qwen3.5." + "reason": "Both concern serialization, but one is lm_head.weight and the other is visual encoder keys; different model parts and fixes." }, { "left": "issue:43450", "right": "issue:45701", "accept": false, - "reason": "Video processor batch-shape bug is not the same as a version-dependent tokenization change." + "reason": "Video batching shape bug vs tokenizer behavior change; unrelated." }, { "left": "issue:44877", "right": "issue:44991", "accept": false, - "reason": "Strict config loading for granite_speech and tokenizer loading failure for est-roberta are separate issues." + "reason": "Strict config loading vs tokenizer loading regression are different failure modes." }, { - "left": "issue:43065", - "right": "issue:46402", + "left": "issue:44186", + "right": "issue:45072", "accept": false, - "reason": "Dummy Conv2d in Sam3PixelDecoder and missing text encoder weight loading are unrelated." + "reason": "Tokenizer crashes with NER/padding are unrelated to dtype mismatches in bfloat16 inference." }, { - "left": "issue:40990", - "right": "issue:45701", + "left": "issue:26413", + "right": "issue:28530", "accept": false, - "reason": "A perplexity regression and a tokenizer-behavior change are too different to merge." + "reason": "Checkpoint resume iterator exhaustion and early-stopping metric lookup are different training-loop bugs." }, { - "left": "issue:26413", - "right": "issue:28530", + "left": "issue:43550", + "right": "issue:43720", "accept": false, - "reason": "resume_from_checkpoint epoch_iterator failure and early stopping metric lookup are different training-loop bugs." + "reason": "torch.compile/SDPA failure vs accelerate loading of packed weights are unrelated." }, { - "left": "issue:44186", - "right": "issue:45072", + "left": "issue:43873", + "right": "issue:44756", "accept": false, - "reason": "LayoutLMv2Tokenizer crashes with NER/batching and SwitchTransformers/TimmWrapperModel dtype mismatches are unrelated." + "reason": "Offloading/quantization problems are not the same as disabling mmap to avoid OOM on Strix Halo." }, { - "left": "issue:41720", - "right": "issue:42617", + "left": "issue:40990", + "right": "issue:45701", "accept": false, - "reason": "Different failures: Qwen3 auto device mapping vs inability to run 3d_parallel.py." + "reason": "Perplexity regression vs tokenization changes; no shared underlying defect." }, { - "left": "issue:43550", - "right": "issue:43720", + "left": "issue:43065", + "right": "issue:46402", "accept": false, - "reason": "Both are quantized-model related, but one is torch.compile/SDPA on Bamba and the other is packed-weight unpacking during accelerate loading." + "reason": "A dummy Conv2d in Sam3PixelDecoder is unrelated to missing text encoder weights in Sam3Model." }, { - "left": "issue:43873", - "right": "issue:44756", + "left": "issue:41720", + "right": "issue:42617", + "accept": false, + "reason": "Qwen3 auto device-map loading failure and 3d_parallel.py runtime failure are unrelated." + }, + { + "left": "issue:44898", + "right": "issue:45072", "accept": false, - "reason": "Offloading with quantization and disabling mmap on Strix Halo to avoid OOM are not the same concrete bug." + "reason": "Perceiver interpolation failure and dtype mismatch are different issues." } ] }, @@ -4767,8 +4854,8 @@ "issue:43646|issue:46665", "issue:43404|issue:45357", "issue:42915|issue:43317", - "issue:44734|issue:46858", "issue:44610|issue:44898", + "issue:44734|issue:46858", "issue:36010|issue:42617", "issue:44423|issue:45406", "issue:46821|issue:46869", @@ -4788,14 +4875,14 @@ "issue:42915|issue:43322", "issue:28530|issue:30064" ], - "prepared_review_unit_hash": "2a749b68ad0c90a7c7c2cc81744d4091b01096d48982baa9b4b16d94617e954c", + "prepared_review_unit_hash": "f46c4347ef9930f033409d77ebca8dd6769378c16dc849873dd88451b6796729", "component_budget": { "node_count": 502, "item_count": 502, - "soft_pair_count": 1842, - "serialized_chars": 956182, - "estimated_input_tokens": 239046, - "estimated_eval_tokens": 478348 + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 }, "budget": { "node_count": 48, @@ -4810,186 +4897,186 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a loose collection of transformers issues that only share broad vocabulary around generation, caching, loading, and device placement. The soft pairs are topical but not duplicate-level matches; none should be merged as the same bug.", - "confidence": 0.91, - "canonical_issue_reason": "issue:46858 is the closest representative because it is a concrete low-level generation/cache runtime bug, which is the nearest recurring theme across the soft matches.", + "summary": "All soft pairs are cross-subsystem or cross-model reports with different failure modes, so none look like the same underlying bug/change. I would reject every soft edge. The most suitable standalone issue anchor is the cache/torch.compile backend bug in 46858 because it is the most concrete and actionable report.", + "confidence": 0.97, + "canonical_issue_reason": "issue:46858 is the most concrete, narrowly scoped bug report: it names the exact failing method, the tensor/device mismatch, and the backend condition (static cache + torch.compile on non-CUDA).", "canonical_pr_reason": null, - "best_issue_reason": "46858 is the best single issue to represent the cluster\u2019s shared cache/generation regression theme, but it is still too specific to serve as a true umbrella duplicate.", + "best_issue_reason": "Issue 46858 is the strongest representative report overall because it has a precise failure signature and clear reproduction context, making it a good triage anchor.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:46612", "right": "issue:46858", "accept": false, - "reason": "Both involve generation/cache behavior, but one is beam-search cache reorder omission and the other is a device-tensor seq_length bug in prefill/static cache. Different code paths and failure modes." + "reason": "Both involve generation/cache logic, but one is cache reordering semantics across model families and the other is a device-tensor arithmetic bug in prefill; different code paths." }, { "left": "issue:43577", "right": "issue:43708", "accept": false, - "reason": "Blip2 dtype propagation during load is unrelated to Trainer resume/max_steps math after batch-size changes." + "reason": "One is a model dtype loading bug in Blip2; the other is Trainer checkpoint resumption max_steps math. Unrelated failure modes." }, { "left": "issue:43295", "right": "issue:45701", "accept": false, - "reason": "Processor/tokenizer image-handling regression is not the same bug as version-dependent tokenization changes; different symptoms and root cause." + "reason": "Different problems: processor/tokenizer regression for multimodal custom code vs a tokenization-version behavior change." }, { "left": "issue:45685", "right": "issue:46821", "accept": false, - "reason": "MPS histogram on Int in a MoE path and TimesFM window_size AttributeError are unrelated failures." + "reason": "MPS histogram implementation error for MoE internals vs a TimesFM window_size attribute error. No shared underlying bug." }, { "left": "issue:43550", "right": "issue:43606", "accept": false, - "reason": "Both are device/runtime issues, but one is torch.compile+SDPA on Bamba and the other is CPU offload mismatch on Bark; not the same bug." + "reason": "Both are runtime failures, but one is torch.compile/SDPA on Bamba and the other is CPU-offload device mismatch on bark-small." }, { "left": "issue:44186", "right": "issue:44610", "accept": false, - "reason": "LayoutLMv2 tokenizer crashes on NER/padding inputs, while OmDet-Turbo has an image-size mismatch between processor and model." + "reason": "Tokenizer crash on NER/batching is separate from an OmDet image-size mismatch in its processor." }, { "left": "issue:43646", "right": "issue:46665", "accept": false, - "reason": "Custom model initialization breakage and inconsistent from_pretrained behavior inside Python classes are different API problems." + "reason": "Transformers 5.0 custom model initialization breakage is not the same as from_pretrained behaving inconsistently inside Python classes." }, { "left": "issue:43404", "right": "issue:45357", "accept": false, - "reason": "Mistral3 lm_head tying and Qwen3.5 visual encoder save_pretrained keys are unrelated model-specific issues." + "reason": "Weight tying bug in Mistral3 is unrelated to Qwen3.5 save_pretrained writing incorrect visual encoder keys." }, { "left": "issue:42915", "right": "issue:43317", "accept": false, - "reason": "FineGrainedFP8Config failure for Qwen3Moe is not the same as device_map auto failing to load a dequantized model with cpu offload." + "reason": "Different loading/quantization failures on different models; one is FineGrainedFP8Config, the other is dequantized gpu+cpu offload." }, { - "left": "issue:44734", - "right": "issue:46858", + "left": "issue:44610", + "right": "issue:44898", "accept": false, - "reason": "Both touch cache/generation, but one is KV-cache continuation indexing in serve and the other is seq_length returning a device tensor for static cache/compile." + "reason": "Both touch image preprocessing, but one is a fixed wrong output resolution and the other is a non-default-resolution interpolation failure." }, { - "left": "issue:44610", - "right": "issue:44898", + "left": "issue:44734", + "right": "issue:46858", "accept": false, - "reason": "Processor output resolution mismatch and Perceiver non-default-resolution interpolation failure are similar only at a high level; different models and fixes." + "reason": "Both mention cache/indexing, but one crashes in serve KV-cache continuation and the other is a prefill slice arithmetic issue on non-CUDA backends." }, { "left": "issue:36010", "right": "issue:42617", "accept": false, - "reason": "GenerationMixin import error and 3d_parallel.py runtime failure are unrelated." + "reason": "ImportError for GenerationMixin is unrelated to failure running 3d_parallel.py." }, { "left": "issue:44423", "right": "issue:45406", "accept": false, - "reason": "Serve crashes in multimodal continuous batching due to a string/.to error versus Gemma4Processor missing _tokenizer; different objects and code paths." + "reason": "Both are serve-related multimodal crashes, but the concrete bugs differ: tensor conversion on one side and missing _tokenizer on the other." }, { "left": "issue:46821", "right": "issue:46869", "accept": false, - "reason": "TimesFM window_size AttributeError and YOLO multi-GPU device_map auto failure are unrelated." + "reason": "TimesFM window_size attribute error and YOLO multi-GPU device-map failure are different models and different bug classes." }, { "left": "issue:44008", "right": "issue:45823", "accept": false, - "reason": "Gemma3n variable collision causing audio_mel_mask errors is not the same as Gemma4 PLE device mismatch with device_map=auto." + "reason": "Variable-name collision causing audio_mel_mask AttributeError is unrelated to Gemma4 PLE device mismatch under auto device mapping." }, { "left": "issue:45632", "right": "issue:45684", "accept": false, - "reason": "trust_remote_code cache-path collision and save_pretrained permission propagation are different filesystem behaviors." + "reason": "Cache-path collision for trust_remote_code local models is unrelated to permission propagation in save_pretrained/register_for_auto_class." }, { "left": "issue:44464", "right": "issue:46858", "accept": false, - "reason": "Chunked generation inconsistency under compiled forward and seq_length device-tensor arithmetic are both compile-related, but not the same bug." + "reason": "Both involve compiled-generation/caching, but one is inconsistent chunked generation outputs while the other is a device tensor returned from cache length." }, { "left": "issue:41720", "right": "issue:45161", "accept": false, - "reason": "Qwen3 auto device-map cuda assert and GPT-OSS MoE tensor-parallel failure are distinct multi-GPU problems." + "reason": "Auto device mapping cudaErrorAssert on A800 is not the same as TP-only failure in GPT-OSS MoE." }, { "left": "issue:28530", "right": "issue:33666", "accept": false, - "reason": "Early stopping metric_for_best_model handling and Qwen2-VL multi-GPU training are unrelated." + "reason": "Early-stopping metric resolution and Qwen2-VL multi-GPU training are completely unrelated." }, { "left": "issue:44671", "right": "issue:44743", "accept": false, - "reason": "CamemBERT masked-LM prediction regression and recurrent-state reset with cache are different model behaviors." + "reason": "CamemBERT MLM prediction drift is a different bug from Qwen3.5 recurrent-state reset with cache." }, { "left": "issue:43531", "right": "issue:43572", "accept": false, - "reason": "Qwen3-MoE sliding_window issue and missing pad_token_idx in StableLmConfig are unrelated config/runtime bugs." + "reason": "Sliding-window behavior in Qwen3-MoE is unrelated to the missing pad_token_idx config regression in StableLm." }, { "left": "issue:42915", "right": "issue:46291", "accept": false, - "reason": "FineGrainedFP8Config failure and pipelines import requiring kernels version/revision are unrelated dependency/configuration issues." + "reason": "FineGrainedFP8Config model loading failure is unrelated to importing transformers.pipelines with kernels version/revision constraints." }, { "left": "issue:36331", "right": "issue:41720", "accept": false, - "reason": "Trainer compute_loss signature mismatch is not the same as Qwen3 auto device-map cuda asserts." + "reason": "Trainer compute_loss API mismatch is unrelated to Qwen3 auto device-map cudaErrorAssert." }, { "left": "issue:44186", "right": "issue:44361", "accept": false, - "reason": "Both are tokenizer crashes, but on different tokenizer classes and different inputs; not a shared underlying bug." + "reason": "Both are tokenizer-related AttributeErrors, but they affect different tokenizers, inputs, and failure points." }, { "left": "issue:43526", "right": "issue:43881", "accept": false, - "reason": "BeitImageProcessorFast reduce_labels bug and glm-4v-9b loading failure are unrelated." + "reason": "BeitImageProcessorFast label reduction bug does not match a glm-4v-9b loading failure." }, { "left": "issue:43316", "right": "issue:44913", "accept": false, - "reason": "Gemma3TextConfig API discrepancy and GPTNeoX rotary_pct reload regression are different configuration issues." + "reason": "Config API discrepancy for Gemma3TextConfig is a different issue from GPTNeoX rotary_pct not persisting on reload." }, { "left": "issue:43425", "right": "issue:46821", "accept": false, - "reason": "Torch 2.10 incompatibility and TimesFM window_size AttributeError do not describe the same problem." + "reason": "Torch 2.10 compatibility is a broad environment issue, not the same as a TimesFM window_size AttributeError." }, { "left": "issue:42915", "right": "issue:43322", "accept": false, - "reason": "Qwen3Moe FineGrainedFP8Config failure and Llava Next loading segfault are unrelated model-specific crashes." + "reason": "Qwen3Moe FineGrainedFP8Config loading failure is unrelated to Llava Next segmentation faults." }, { "left": "issue:28530", "right": "issue:30064", "accept": false, - "reason": "Early stopping metric selection and void segmentation-map processing are entirely different bugs." + "reason": "Trainer early-stopping metric handling and void segmentation-map processing are different subsystems and bugs." } ] }, @@ -5094,130 +5181,130 @@ "issue:44987|issue:45042", "issue:44625|issue:45479" ], - "prepared_review_unit_hash": "552619adf9f6a75425fba96d48cb9fd28fc88ef0543b6d02855c7cf97ffb3a3f", + "prepared_review_unit_hash": "3fb9efd8616dee1f442fe42f09f31252907ab136a4f7b059398de4fc8b5915c2", "component_budget": { "node_count": 502, "item_count": 502, - "soft_pair_count": 1842, - "serialized_chars": 956182, - "estimated_input_tokens": 239046, - "estimated_eval_tokens": 478348 + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 30, - "serialized_chars": 36833, - "estimated_input_tokens": 9209, - "estimated_eval_tokens": 18674 + "serialized_chars": 36842, + "estimated_input_tokens": 9211, + "estimated_eval_tokens": 18678 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly noisy and heterogeneous: tokenizer, multimodal, offload/quantization, docs, and trainer bugs are mixed together. None of the soft pairs look like true duplicates or the same concrete fix path.", - "confidence": 0.97, - "canonical_issue_reason": "issue:43873 is the best representative issue: it is a clearly scoped, high-impact regression with concrete reproduction and multiple inbound references around quantization/offloading.", + "summary": "This cluster is mostly a semantic bag of unrelated Transformers issues. The soft-similarity links are largely false positives across different models, code paths, and bug classes. No pull requests are present, and no soft edge should be merged.", + "confidence": 0.93, + "canonical_issue_reason": "Issue 43644 is the best canonical anchor because it is the clearest concrete regression in the set with downstream references, and it matches the only near-duplicate theme here (the later 44534 report about non-persistent buffers).", "canonical_pr_reason": null, - "best_issue_reason": "issue:43873 is the strongest standalone issue in the set because it describes a broad regression, has external references, and is more central than the other one-off bugs.", + "best_issue_reason": "43644 is the strongest representative issue: specific, reproducible, and the only clear duplicate-style theme in the cluster. Most other items are unrelated tokenization, loading, multimodal, or training bugs.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:33357", "right": "issue:35707", "accept": false, - "reason": "Different failures in unrelated areas: CLIP bus error on macOS vs generation with inputs_embeds/past_key_values." + "reason": "Both are runtime failures, but one is a CLIP/MacOS bus error and the other is a generation bug with inputs_embeds/past_key_values; different code paths and causes." }, { "left": "issue:29159", "right": "issue:36032", "accept": false, - "reason": "Both are tokenizer-related, but one is slow/fast behavior inconsistency and the other is a T5 tokenizer load-time AttributeError." + "reason": "Both involve tokenizers, but one is slow-vs-fast inconsistency and the other is a T5Tokenizer loading/name conflict; not the same underlying bug." }, { "left": "issue:43540", "right": "issue:43782", "accept": false, - "reason": "Both involve Qwen multimodal models, but one is video input processing and the other is from_pretrained weight loading." + "reason": "Different multimodal failures: video input processing in Qwen3OmniMoe vs weight_only loading in Qwen3VL; not mergeable as one fix." }, { "left": "issue:43550", "right": "issue:43653", "accept": false, - "reason": "Unrelated subsystems: Bamba torch.compile/SDPA failure vs BigBirdTokenizer special-token registration." + "reason": "Unrelated subsystems: torch.compile/SDPA on Bamba vs BigBirdTokenizer special-token registration." }, { "left": "issue:44387", "right": "issue:45701", "accept": false, - "reason": "Quantization memory regression and tokenization changes are different bugs with no shared code path." + "reason": "One is int4 quantization memory growth, the other is tokenization changes across versions; same project area but not the same bug." }, { "left": "issue:43873", "right": "issue:45704", "accept": false, - "reason": "Offloading/quantization regression vs T5 apex RMSNorm memory leak; same broad model-loading theme, but different concrete defects." + "reason": "Offloading/quantization behavior is unrelated to T5's apex.FusedRMSNorm memory leak." }, { "left": "issue:30064", "right": "issue:34634", "accept": false, - "reason": "Different processors and symptoms: segmentation-map handling vs Bark voice_preset behavior." + "reason": "Image segmentation-map processing vs BarkProcessor voice_preset handling are separate features with different failures." }, { "left": "issue:43606", "right": "issue:43720", "accept": false, - "reason": "Both touch loading/offload, but one is CPU offload device mismatch and the other is BitNet packed-weight unpacking during accelerate loading." + "reason": "Both touch loading/offloading, but Bark CPU offload device mismatch and BitNet packed-weight unpacking are distinct code paths." }, { "left": "issue:44492", "right": "issue:44509", "accept": false, - "reason": "A typo in cache-strategy docs is not the same as stale pipeline-task documentation." + "reason": "Both are docs/wording issues, but a cache-strategy typo is not the same as stale pipeline-task documentation." }, { "left": "issue:44534", "right": "issue:46665", "accept": false, - "reason": "Both mention loading behavior, but one is a non-persistent buffer regression and the other is inconsistent from_pretrained use inside Python classes." + "reason": "The non-persistent-buffer regression is unrelated to the from_pretrained behavior discrepancy inside Python classes." }, { "left": "issue:43531", "right": "issue:45759", "accept": false, - "reason": "Both are Qwen config/model issues, but sliding_window handling and text_config unwrapping are different code paths." + "reason": "Qwen3-MoE sliding_window behavior and Qwen 3.5/3.6 config unwrapping are different bugs." }, { "left": "issue:41720", "right": "issue:43746", "accept": false, - "reason": "Qwen3 auto device mapping CUDA assert is unrelated to GraniteSpeech PEFT local-checkpoint loading." + "reason": "CUDA assert under auto device mapping on Qwen3 is unrelated to GraniteSpeech PEFT checkpoint loading." }, { "left": "issue:42490", "right": "issue:43329", "accept": false, - "reason": "Path import warnings and multimodal token counting/video-branch undefined names are unrelated bugs." + "reason": "Local-path slash warning and undefined video-token helper names in multimodal token counting are unrelated." }, { "left": "issue:43388", "right": "issue:45059", "accept": false, - "reason": "Trainer metric gathering with tuple labels is unrelated to SAM3 text/bounding-box behavior." + "reason": "A metrics-gathering last-batch truncation bug is not the same as SAM3 text/bbox behavior." }, { "left": "issue:45406", "right": "issue:45464", "accept": false, - "reason": "Both affect inference/serve, but Gemma4Processor missing _tokenizer and Qwen3.5 streaming failures are different issues." + "reason": "Serve crashes due to missing _tokenizer and chat/completions streaming failures are different failure modes." }, { "left": "issue:43606", "right": "issue:43653", "accept": false, - "reason": "Device-mismatch offload bug vs tokenizer special-token registration bug; no shared fix path." + "reason": "Bark CPU offload mismatch and BigBirdTokenizer special-token registration share only a loose CI context." }, { "left": "issue:34567", @@ -5229,73 +5316,79 @@ "left": "issue:43531", "right": "issue:46821", "accept": false, - "reason": "Both mention window-like terms, but Qwen3-MoE sliding_window and TimesFM window_size are distinct model bugs." + "reason": "Qwen3-MoE sliding-window handling is unrelated to TimesFM window_size attribute access." }, { "left": "issue:34634", "right": "issue:41720", "accept": false, - "reason": "BarkProcessor voice preset and Qwen3 auto device mapping are unrelated." + "reason": "BarkProcessor voice_preset and Qwen3 device-mapping CUDA asserts are different components and symptoms." }, { "left": "issue:43540", "right": "issue:46858", "accept": false, - "reason": "Qwen3OmniMoe video processing and static-cache torch.compile arithmetic are different failures." + "reason": "Video-input ValueError in Qwen3OmniMoe is unrelated to static-cache slice arithmetic on non-CUDA backends." }, { "left": "issue:44451", "right": "issue:44991", "accept": false, - "reason": "Both are model-loading failures, but ScandiBERT and est-roberta fail for different tokenizer/model reasons." + "reason": "Both are model/tokenizer loading failures, but different models and likely different serialization/tokenizer issues." }, { "left": "issue:45759", "right": "issue:45859", "accept": false, - "reason": "Both are Qwen 3.5 issues, but one is config unwrapping and the other is CPU-offload compatibility in rotary embedding." + "reason": "Composite-config unwrapping and CPU-offload incompatibility in rotary embedding are separate Qwen 3.5/3.6 bugs." }, { "left": "issue:45357", "right": "issue:46402", "accept": false, - "reason": "Both concern multimodal weight handling, but save_pretrained visual keys and Sam3Model text encoder loading are different code paths." + "reason": "Incorrect visual-encoder save keys and missing text-encoder weight loading are different loading/saving problems." + }, + { + "left": "issue:43531", + "right": "issue:44589", + "accept": false, + "reason": "Sliding-window behavior in Qwen3-MoE does not match the Float8 storage lookup failure." }, { "left": "issue:44077", "right": "issue:44625", "accept": false, - "reason": "PatchTSMixer post_init policy and Qwen3.5 num_labels propagation are unrelated configuration issues." + "reason": "A patchtsmixer post_init policy issue is unrelated to Qwen3.5 num_labels propagation." }, { "left": "issue:43644", "right": "issue:46665", "accept": false, - "reason": "Non-persistent buffer junk during v5 loading is not the same as from_pretrained behavior inside Python classes." + "reason": "Non-persistent-buffer junk initialization is not the same as from_pretrained behavior inside Python classes." }, { "left": "issue:44062", "right": "issue:44451", "accept": false, - "reason": "AddedToken keyword collision and ScandiBERT loading failure are different tokenizer/model issues." + "reason": "AddedToken special-argument collision and ScandiBERT loading failure are separate tokenizer/load issues." }, { "left": "issue:44898", "right": "issue:45072", "accept": false, - "reason": "Perceiver image-resolution failure and dtype mismatch in SwitchTransformers/TimmWrapperModel are unrelated." + "reason": "Perceiver interpolation failure and bfloat16 dtype mismatch in SwitchTransformers/TimmWrapperModel are unrelated." }, { "left": "issue:44987", "right": "issue:45042", "accept": false, - "reason": "Loading a specific model and PIL backend incorrectly requiring torchvision are separate regressions." + "reason": "Physical-intelligence/fast loading failure and PIL backend torchvision requirement are different regressions." }, { "left": "issue:44625", "right": "issue:45479", "accept": false, - "reason": "num_labels propagation and zero-loss single-label classification are different classification bugs, not the same fix." + "reason": "Qwen3.5 num_labels propagation and single-label zero-loss classification are not the same bug." } ] }, @@ -5328,6 +5421,7 @@ "issue:43381", "issue:43421", "issue:43493", + "issue:43550", "issue:43653", "issue:43720", "issue:43746", @@ -5337,10 +5431,9 @@ "issue:43873", "issue:43927", "issue:43994", - "issue:44062", "issue:44186", + "issue:44361", "issue:44368", - "issue:44488", "issue:44514", "issue:44545", "issue:44610", @@ -5389,9 +5482,9 @@ "issue:46032|issue:46463", "issue:43421|issue:44186", "issue:43782|issue:45381", - "issue:45081|issue:45356", "issue:30064|issue:36010", "issue:44995|issue:46858", + "issue:45081|issue:45356", "issue:45356|issue:45907", "issue:43720|issue:43819", "issue:39692|issue:43873", @@ -5399,228 +5492,228 @@ "issue:45987|issue:46829", "issue:46829|issue:46832", "issue:43295|issue:45834", - "issue:44062|issue:44488" + "issue:43550|issue:44361" ], - "prepared_review_unit_hash": "1f2a396143800ac7a49f6b4a4ca33eca15f60a95d9bb09da4aa23b55ff96217d", + "prepared_review_unit_hash": "8ec04d4f33e9a087cc7c78188aeaac8051fa76bac97ea12684dac08a75645026", "component_budget": { "node_count": 502, "item_count": 502, - "soft_pair_count": 1842, - "serialized_chars": 956182, - "estimated_input_tokens": 239046, - "estimated_eval_tokens": 478348 + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 31, - "serialized_chars": 37489, - "estimated_input_tokens": 9373, - "estimated_eval_tokens": 19002 + "serialized_chars": 37468, + "estimated_input_tokens": 9367, + "estimated_eval_tokens": 18990 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Only one soft pair is a clear duplicate: the two Qwen2_5_VLProcessor.apply_chat_template crash reports with identical titles and the same padding=False batched-input failure. The other pairs are related by subsystem or theme, but not the same concrete bug/change.", - "confidence": 0.78, - "canonical_issue_reason": "Issue 44514 is the older report in the exact duplicate pair and states the crash most directly; 44545 repeats the same bug with the same title.", + "summary": "This cluster is mostly heterogeneous. I found two real duplicate subclusters: the Qwen2_5_VLProcessor batched `padding=False` crash, and an HF_HUB_OFFLINE tokenizer regression around `_patch_mistral_regex` / `model_info()`. The rest of the soft matches look like false positives across unrelated bugs.", + "confidence": 0.76, + "canonical_issue_reason": "issue:44545 is the best canonical issue here: it is an exact duplicate of issue:44514 with the same `Qwen2_5_VLProcessor.apply_chat_template` batched `padding=False` crash, and it has the stronger linkage signal.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 44514 is the best global issue anchor here because it is an exact duplicate match, has the cleanest repro description, and is the likely original report.", + "best_issue_reason": "issue:44545 is the clearest duplicate target in the set: precise title, same failure mode as its twin, and better reference signal than the alternative duplicates.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43381", "right": "issue:43844", "accept": false, - "reason": "Both mention training/gradients, but one is about gradient checkpointing in eval mode and the other is a ZeRO-3 gradient explosion regression; different code paths and symptoms." + "reason": "Different bugs: gradient checkpointing eval-mode misuse vs ZeRO-3 gradient growth during training." }, { "left": "issue:44368", "right": "issue:45357", "accept": false, - "reason": "One is a tied-word-embeddings warning during LoRA fine-tuning; the other is a save_pretrained regression saving incorrect visual encoder keys. Not the same bug." + "reason": "Unrelated Qwen warnings/regressions: tie_word_embeddings warning vs bad visual encoder keys on save." }, { "left": "issue:39692", "right": "issue:44610", "accept": false, - "reason": "Both involve vision models/processors, but one is a SigLIP2 docs example error and the other is an OmDet-Turbo image-size mismatch. Separate issues." + "reason": "Different model paths and failures: SigLIP2 doc example issues vs OmDet-Turbo image-size mismatch." }, { "left": "issue:46655", "right": "issue:46693", "accept": false, - "reason": "Video processor padding calculation versus FA2 long-context performance regression; unrelated failure modes." + "reason": "Different areas: SmolVLM video padding bug vs FA2 long-context performance regression." }, { "left": "issue:43421", "right": "issue:43927", "accept": false, - "reason": "Special-token post-processor update bug versus DiaConfig losing custom token IDs on save/load. Both touch tokens, but not the same underlying defect." + "reason": "Both involve tokenizer/config state, but the concrete bugs differ: runtime special-token updates vs save/load custom token IDs." }, { "left": "issue:41720", "right": "issue:45704", "accept": false, - "reason": "Qwen3 auto device mapping CUDA assert and T5 apex FusedRMSNorm memory leak are unrelated." + "reason": "Completely different problems: Qwen3 device-mapping CUDA assert vs T5 apex RMSNorm memory leak." }, { "left": "issue:43746", "right": "issue:46050", "accept": false, - "reason": "Loading PEFT adapters from local checkpoints versus quantized-sinq save/load breakage are different persistence paths and causes." + "reason": "Different load/save failures and different model families; not the same concrete code-path bug." }, { "left": "issue:43381", "right": "issue:45704", "accept": false, - "reason": "Gradient checkpointing/eval-mode behavior is unrelated to T5 RMSNorm memory leak." + "reason": "No overlap beyond being model/runtime issues; the failure modes are unrelated." }, { "left": "issue:42886", "right": "issue:44843", - "accept": false, - "reason": "Both mention offline mode, but 44843 is specifically an unconditional model_info() call in _patch_mistral_regex; 42886 is a broader tokenizer cache-load failure. Not the same concrete bug." + "accept": true, + "reason": "Same offline-tokenizer regression: 44843 identifies the concrete root cause (`model_info()` in `_patch_mistral_regex`) behind 42886's cache-load failure." }, { "left": "issue:45704", "right": "issue:45923", "accept": false, - "reason": "T5 RMSNorm memory leak versus missing supports_gradient_checkpointing flag on a trust_remote_code model; unrelated." + "reason": "Different issues in different models: T5 RMSNorm leak vs missing `supports_gradient_checkpointing` flag." }, { "left": "issue:44918", "right": "issue:44936", "accept": false, - "reason": "Unpacking Qwen3.5 input embeddings in TRL SFT trainer versus trainer.evaluate() failing after training are different training-state bugs." + "reason": "Different trainer problems: unpacking Qwen3.5 embeddings in TRL vs evaluate failing after train." }, { "left": "issue:44514", "right": "issue:44545", "accept": true, - "reason": "Exact same title and same batched-input padding=False crash in Qwen2_5_VLProcessor.apply_chat_template; clear duplicate." + "reason": "Exact duplicate: same `Qwen2_5_VLProcessor.apply_chat_template` batched `padding=False` crash." }, { "left": "issue:45759", "right": "issue:45923", "accept": false, - "reason": "AutoModelForCausalLM.from_config unwrap logic for composite Qwen multimodal configs is unrelated to a missing gradient-checkpointing flag." + "reason": "Unrelated multimodal config issue vs trust_remote_code capability flag issue." }, { "left": "issue:43421", "right": "issue:43653", "accept": false, - "reason": "Runtime post-processor refresh for special tokens versus BigBirdTokenizer mask token registration/empty decode are distinct tokenizer bugs." + "reason": "Both are tokenizer-related, but one is runtime post-processor updates and the other is BigBird mask-token registration; different bugs." }, { "left": "issue:43746", "right": "issue:43873", "accept": false, - "reason": "Local checkpoint loading for GraniteSpeech adapters versus offloading with quantization are different loading/offload paths." + "reason": "Different loading/offloading problems with different model scopes; not the same underlying bug." }, { "left": "issue:43994", "right": "issue:44671", "accept": false, - "reason": "Both report incorrect outputs, but on different models and different mechanisms; no shared code path is evident." + "reason": "Different models and symptoms: SigLIP2 nonsensical outputs vs CamemBERT masked-LM regression." }, { "left": "issue:39692", "right": "issue:43493", "accept": false, - "reason": "SigLIP2 docs example errors and SigLIP2 HF-vs-JAX discrepancy are related to the same model family but describe different bugs and fixes." + "reason": "Same broad SigLIP2 family, but different concrete problems: documentation/example failure vs HF/JAX implementation mismatch." }, { "left": "issue:46032", "right": "issue:46463", "accept": false, - "reason": "Mamba2Mixer cache correctness bug versus T5 decoder input validation error message; unrelated." + "reason": "Different code paths: Mamba2 cache misuse vs T5 decoder-input validation message wording." }, { "left": "issue:43421", "right": "issue:44186", "accept": false, - "reason": "Post-processor update on special-token changes versus LayoutLMv2Tokenizer crashes on NER/batched padding are different tokenizer failures." + "reason": "Both tokenizer issues, but one is post-processor updates and the other is LayoutLMv2 NER/padding crashes." }, { "left": "issue:43782", "right": "issue:45381", "accept": false, - "reason": "Qwen3VL weight_only loading error versus Qwen2.5-VL video vision_position_ids bug are separate multimodal loading/runtime issues." - }, - { - "left": "issue:45081", - "right": "issue:45356", - "accept": false, - "reason": "Both are about tokenizer regex fixes, but one is an AttributeError in _patch_mistral_regex and the other is codec-handling plus warning regression for Kimi-K2.5. Similar area, not the same bug." + "reason": "Different Qwen VL failures: `weight_only=True` load error vs wrong `vision_position_ids` on video input." }, { "left": "issue:30064", "right": "issue:36010", "accept": false, - "reason": "Void segmentation map processing and a GenerationMixin import error are unrelated." + "reason": "Unrelated issues: image processor segmentation-map handling vs GenerationMixin import error." }, { "left": "issue:44995", "right": "issue:46858", "accept": false, - "reason": "Both involve cache state, but one is a GlmMoeDsa second-forward stale indexer bug and the other is StaticCache.get_seq_length() returning a device tensor that breaks slice arithmetic." + "reason": "Both mention cache/state, but the concrete bugs differ: stale indexer cache on second forward vs static-cache seq-length return type in compile paths." + }, + { + "left": "issue:45081", + "right": "issue:45356", + "accept": false, + "reason": "Related tokenizer patch area, but different failures: AttributeError in `_patch_mistral_regex` vs codec-handling regression and warning text." }, { "left": "issue:45356", "right": "issue:45907", "accept": false, - "reason": "Tokenizer codec/fix_mistral_regex regression versus list_repo_files not forwarding revision; unrelated." + "reason": "Different problems: Kimi-K2.5 tokenizer regression vs `list_repo_files` not forwarding `revision`." }, { "left": "issue:43720", "right": "issue:43819", "accept": false, - "reason": "BitNet packed-weight loading bug versus DAC.from_latents/STE mismatch; different models and code paths." + "reason": "Different model families and bugs: packed-weight loading vs DAC latent/forward mismatch." }, { "left": "issue:39692", "right": "issue:43873", "accept": false, - "reason": "SigLIP2 doc-example/model mismatch is unrelated to quantization offloading behavior." + "reason": "No shared concrete failure mode; SigLIP2 doc/model mismatch is unrelated to quantization offloading behavior." }, { "left": "issue:43653", "right": "issue:43720", "accept": false, - "reason": "BigBird special-token registration issue versus BitNet packed-weight unpacking during accelerate loading; unrelated." + "reason": "Tokenizer special-token registration bug vs BitNet packed-weight loading bug; unrelated." }, { "left": "issue:45987", "right": "issue:46829", "accept": false, - "reason": "StaticCache sequence-length type contract issue versus VideoPrism hidden_states output mismatch; different components." + "reason": "Both are output-shape/interface issues, but one is cache seq-length type and the other is classifier-output hidden_states misassignment." }, { "left": "issue:46829", "right": "issue:46832", "accept": false, - "reason": "VideoPrism output field mismatch versus RT-DETR forward TypeError when feature levels exceed backbone outputs." + "reason": "Different model outputs and failure modes; no duplicate evidence." }, { "left": "issue:43295", "right": "issue:45834", "accept": false, - "reason": "Processor.tokenizer regression with custom model code versus Kosmos2.5 long-OCR index error; different processors and failure modes." + "reason": "Both are multimodal processor regressions, but the concrete bugs differ: processor.tokenizer/images regression vs Kosmos2.5 long-OCR index error." }, { - "left": "issue:44062", - "right": "issue:44488", + "left": "issue:43550", + "right": "issue:44361", "accept": false, - "reason": "AddedToken special-argument duplication versus failure to load a specific model; not the same underlying tokenizer bug." + "reason": "Completely unrelated: torch.compile+SDPA on Bamba vs MLukeTokenizer AttributeError." } ] }, "evaluator_result": { "accept": true, - "feedback": "The result is grounded in the packet and the soft-edge judgments are conservative. The 44514/44545 pair is clearly a duplicate from the titles and timestamps, and the remaining pairs are reasonably rejected as related-at-most rather than the same concrete bug." + "feedback": "The analysis is grounded in the packet and stays conservative. The two accepted duplicate pairs are well supported by the titles and linkage signals, and the rejected soft matches are generally justified by clearly different failure modes or model paths. No pull-request pairing issues apply here." }, "evaluator_used": true, "retried": false, @@ -5663,11 +5756,12 @@ "issue:43819", "issue:43927", "issue:44008", + "issue:44062", "issue:44186", "issue:44220", - "issue:44361", "issue:44442", "issue:44448", + "issue:44488", "issue:44661", "issue:44671", "issue:44821", @@ -5690,7 +5784,7 @@ "issue:46932" ], "soft_pairs": [ - "issue:43550|issue:44361", + "issue:44062|issue:44488", "issue:43550|issue:43927", "issue:43596|issue:44220", "issue:45381|issue:46693", @@ -5698,16 +5792,16 @@ "issue:43531|issue:43643", "issue:44008|issue:46858", "issue:43611|issue:46665", - "issue:44220|issue:44661", "issue:43653|issue:43819", + "issue:44220|issue:44661", "issue:45987|issue:46932", "issue:44186|issue:44898", "issue:41720|issue:44991", "issue:43618|issue:43697", "issue:30064|issue:36331", "issue:43643|issue:44877", - "issue:43335|issue:43756", "issue:43550|issue:43819", + "issue:43335|issue:43756", "issue:44843|issue:45081", "issue:44821|issue:45084", "issue:44186|issue:44442", @@ -5719,118 +5813,118 @@ "issue:45183|issue:45479", "issue:45584|issue:45834" ], - "prepared_review_unit_hash": "11658f275bef9695b1c0081590c406553e404ac8d503ed3108ba0d0ae1fc1780", + "prepared_review_unit_hash": "24909c2a8408db001127ec7614a3a5791741f9480c0202c6e4e5718a52433c01", "component_budget": { "node_count": 502, "item_count": 502, - "soft_pair_count": 1842, - "serialized_chars": 956182, - "estimated_input_tokens": 239046, - "estimated_eval_tokens": 478348 + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 }, "budget": { - "node_count": 47, - "item_count": 47, + "node_count": 48, + "item_count": 48, "soft_pair_count": 28, - "serialized_chars": 35700, - "estimated_input_tokens": 8925, - "estimated_eval_tokens": 18106 + "serialized_chars": 36286, + "estimated_input_tokens": 9072, + "estimated_eval_tokens": 18400 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a collection of unrelated Transformer bugs spanning tokenizers, multimodal inputs, caching, loading, and training. None of the soft pairs looks like the same underlying defect, so there is no true duplicate hub here.", - "confidence": 0.95, - "canonical_issue_reason": "No clear canonical duplicate: the cluster is heterogeneous and the soft links do not converge on one shared failure mode.", + "summary": "The cluster is largely heterogeneous: most pairs are only superficially similar (same model family, same subsystem, or same version-regression theme) but describe different code paths and fixes. No soft edge should be merged as a duplicate.", + "confidence": 0.97, + "canonical_issue_reason": "Issue 44843 is the clearest concrete regression in the set and the most actionable representative of the tokenizer-loading/_patch_mistral_regex family, with a direct failure mode and downstream references.", "canonical_pr_reason": null, - "best_issue_reason": "If one issue must represent the cluster, issue 43550 is the most connected node, but it is still just a standalone Bamba compile/SDPA bug rather than a duplicate anchor.", + "best_issue_reason": "44843 is the strongest standalone issue to anchor the cluster because it has a specific repro, a single code path, and the clearest connection to a nearby soft-candidate issue family.", "best_pr_reason": null, "soft_edge_verdicts": [ { - "left": "issue:43550", - "right": "issue:44361", + "left": "issue:44062", + "right": "issue:44488", "accept": false, - "reason": "Different models and failures: Bamba torch.compile/SDPA vs MLukeTokenizer AttributeError." + "reason": "Both involve tokenizers, but one is an AddedToken constructor error while the other is a model-loading failure for a specific checkpoint; different bugs and fixes." }, { "left": "issue:43550", "right": "issue:43927", "accept": false, - "reason": "Unrelated bugs: Bamba compile/SDPA failure vs DiaConfig save/load losing custom token IDs." + "reason": "Unrelated failures: torch.compile/SDPA on Bamba versus save/load of DiaConfig token IDs." }, { "left": "issue:43596", "right": "issue:44220", "accept": false, - "reason": "Both are IndexErrors in different areas, but the reported code paths are unrelated (Deepspeed ZeRO3/BertModel vs fbank extraction)." + "reason": "Different subsystems and symptoms: deepspeed zero3/BertModel initialization versus fbank feature extraction." }, { "left": "issue:45381", "right": "issue:46693", "accept": false, - "reason": "Different problem classes: wrong video vision_position_ids vs a performance regression in long-context generation." + "reason": "One is a qwen2.5-vl video-position-id bug; the other is a long-context FA2 performance regression. Same model area only superficially." }, { "left": "issue:43540", "right": "issue:46294", "accept": false, - "reason": "Both involve video inputs, but they are different Qwen variants and different failures (ValueError vs chat-template feature/token mismatch)." + "reason": "Both are video-related Qwen issues, but they fail in different stages with different root causes and are not the same bug." }, { "left": "issue:43531", "right": "issue:43643", "accept": false, - "reason": "Qwen3-MoE sliding_window behavior is unrelated to AutoConfig trust_remote_code field loss." + "reason": "Sliding-window behavior in Qwen3-MoE is unrelated to AutoConfig missing fields with trust_remote_code." }, { "left": "issue:44008", "right": "issue:46858", "accept": false, - "reason": "Gemma3n variable-name collision in forward is unrelated to StaticCache seq-length returning a device tensor." + "reason": "Gemma3n variable-name collision versus StaticCache sequence-length tensor behavior; different code paths." }, { "left": "issue:43611", "right": "issue:46665", "accept": false, - "reason": "Both concern loading, but one is base_model_prefix handling in Transformers 5.0 and the other is from_pretrained behavior inside Python classes." + "reason": "Base-model-prefix loading breakage is not the same as inconsistent from_pretrained behavior inside Python classes." }, { - "left": "issue:44220", - "right": "issue:44661", + "left": "issue:43653", + "right": "issue:43819", "accept": false, - "reason": "Audio feature extraction bug vs tokenizer-mapping/new-model scaffolding bug; no shared defect." + "reason": "Tokenizer special-token registration bug versus DAC.from_latents/STE mismatch; unrelated." }, { - "left": "issue:43653", - "right": "issue:43819", + "left": "issue:44220", + "right": "issue:44661", "accept": false, - "reason": "BigBirdTokenizer special-token registration issue is unrelated to DAC.from_latents missing STE behavior." + "reason": "Feature extraction failure versus tokenizer-mapping-name handling in add-new-model-like; not the same underlying bug." }, { "left": "issue:45987", "right": "issue:46932", "accept": false, - "reason": "StaticCache get_seq_length contract bug is unrelated to Gemma4 quantized vision/audio dtype casting." + "reason": "StaticCache.get_seq_length() returning a tensor is unrelated to Gemma4 quantized vision/audio embedder dtype casting." }, { "left": "issue:44186", "right": "issue:44898", "accept": false, - "reason": "Different model families and different failures: LayoutLMv2 tokenizer padding/NER crash vs Perceiver non-default resolution handling." + "reason": "LayoutLMv2Tokenizer padding/NER crash is a different bug than Perceiver image classification on non-default resolution." }, { "left": "issue:41720", "right": "issue:44991", "accept": false, - "reason": "Qwen3 auto device mapping CUDA assert is unrelated to tokenizer loading for EMBEDDIA/est-roberta." + "reason": "CUDA assert with auto device mapping is unrelated to tokenizer loading for est-roberta." }, { "left": "issue:43618", "right": "issue:43697", "accept": false, - "reason": "Both are v5 regression reports, but they affect different models and different outputs; not the same bug." + "reason": "CLIPOutput attentions assignment and RTDetrV2 output drift are different model regressions." }, { "left": "issue:30064", @@ -5842,61 +5936,61 @@ "left": "issue:43643", "right": "issue:44877", "accept": false, - "reason": "Both touch config loading, but one is missing fields from trust_remote_code and the other is strict config blocking granite_speech; different fixes." + "reason": "Both concern config loading, but one drops fields from trust_remote_code and the other rejects a strict config; different failure mechanisms." }, { - "left": "issue:43335", - "right": "issue:43756", + "left": "issue:43550", + "right": "issue:43819", "accept": false, - "reason": "SwitchTransformers sparse-layer creation and Smollm3 RoPE-layer count are unrelated model-configuration bugs." + "reason": "torch.compile/SDPA failure is unrelated to DAC.from_latents missing STE behavior." }, { - "left": "issue:43550", - "right": "issue:43819", + "left": "issue:43335", + "right": "issue:43756", "accept": false, - "reason": "Bamba torch.compile/SDPA failure is unrelated to DAC.from_latents vs forward mismatch." + "reason": "SwitchTransformers sparse-layer construction and Smollm3 RoPE-layer loss are unrelated model-architecture bugs." }, { "left": "issue:44843", "right": "issue:45081", "accept": false, - "reason": "Both hit _patch_mistral_regex, but one is an offline-mode network call and the other is an AttributeError on tokenizer structure; different concrete bugs." + "reason": "Same helper function, but different bugs: offline model_info() call versus missing backend_tokenizer attribute. Not the same underlying defect." }, { "left": "issue:44821", "right": "issue:45084", "accept": false, - "reason": "AutoImageProcessor URL loading and template-node compilation are unrelated." + "reason": "AutoImageProcessor URL loading and Jinja/non-template compilation are unrelated." }, { "left": "issue:44186", "right": "issue:44442", "accept": false, - "reason": "LayoutLMv2 tokenizer batching/NER crash is unrelated to AutoTokenizer failing on FastSpeech2ConformerTokenizer." + "reason": "Tokenizer crash on NER/padding is not the same as AutoTokenizer failing to load FastSpeech2ConformerTokenizer." }, { "left": "issue:39692", "right": "issue:45834", "accept": false, - "reason": "SigLIP2 docs example errors and Kosmos2.5 long OCR index error are distinct issues." + "reason": "Documentation example errors and long OCR index errors are different problem classes." }, { "left": "issue:44448", "right": "issue:44671", "accept": false, - "reason": "Different regression reports for different models; no shared root cause." + "reason": "Both are version-regression output changes, but for different models and different code paths." }, { "left": "issue:43540", "right": "issue:44008", "accept": false, - "reason": "Both are multimodal model bugs, but one is Qwen video-input handling and the other is Gemma3n audio mask variable collision." + "reason": "Qwen video-input ValueError and Gemma3n audio_mel_mask collision are unrelated." }, { "left": "issue:43644", "right": "issue:43749", "accept": false, - "reason": "Transformers 5.0 buffer initialization issue is unrelated to broken FSDP CPU RAM efficient loading." + "reason": "Non-persistent buffer junk initialization and FSDP_CPU_RAM_EFFICIENT_LOADING are separate loading/runtime issues." }, { "left": "issue:34634", @@ -5908,13 +6002,13 @@ "left": "issue:45183", "right": "issue:45479", "accept": false, - "reason": "Validation error-message logic for input_ids/inputs_embeds is unrelated to the num_labels=1 zero-loss classification bug." + "reason": "Input validation XOR messaging and single-label classification zero-loss are different bugs." }, { "left": "issue:45584", "right": "issue:45834", "accept": false, - "reason": "Whisper empty-transcription generation failure and Kosmos2.5 long-OCR index error are different code paths." + "reason": "Whisper empty transcription after align_special_tokens is unrelated to Kosmos2.5 long OCR input index errors." } ] }, @@ -5993,19 +6087,19 @@ "issue:43550|issue:45072", "issue:43388|issue:46032", "issue:43782|issue:44977", - "issue:43825|issue:43994", "issue:44479|issue:45834", + "issue:43825|issue:43994", "issue:44991|issue:46291", - "issue:43577|issue:43818", "issue:43582|issue:46512", + "issue:43577|issue:43818", "issue:43421|issue:44811", "issue:44561|issue:45020", "issue:43716|issue:44008", "issue:43931|issue:46846", "issue:46294|issue:46463", "issue:28218|issue:29159", - "issue:44805|issue:44936", "issue:44610|issue:45072", + "issue:44805|issue:44936", "issue:43540|issue:45381", "issue:43720|issue:45072", "issue:43825|issue:43827", @@ -6016,196 +6110,196 @@ "issue:42175|issue:45081", "issue:46175|issue:46932" ], - "prepared_review_unit_hash": "277b8c621e1eff00c5f993e88f5c234ea4d34f3f1b016ef3d23e5a42c9ecadc1", + "prepared_review_unit_hash": "8579c11b2a6a0d3f8196b2b6f061f4c600107e371bbe2ea0804966c24f2f32ce", "component_budget": { "node_count": 502, "item_count": 502, - "soft_pair_count": 1842, - "serialized_chars": 956182, - "estimated_input_tokens": 239046, - "estimated_eval_tokens": 478348 + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 27, - "serialized_chars": 36221, - "estimated_input_tokens": 9056, - "estimated_eval_tokens": 18368 + "serialized_chars": 36226, + "estimated_input_tokens": 9057, + "estimated_eval_tokens": 18370 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is highly heterogeneous, with many unrelated transformers regressions that only loosely share a v5/compatibility theme. I do not see any true duplicate PRs here; the best issue representative is a broad umbrella v5/remote_code regression report.", - "confidence": 0.46, - "canonical_issue_reason": "issue:45020 is the broadest and most central umbrella in the set: it captures the general class of recent-version/remote_code breakages that several other issues appear to fall under, even though it is not a perfect duplicate of all of them.", + "summary": "No soft-edge pair looks like the same concrete bug/change; most are only loosely related by model family or subsystem. The strongest standalone issue here is the FSDP CPU RAM-efficient loading regression, which is concrete and well-discussed.", + "confidence": 0.59, + "canonical_issue_reason": "issue:43749 is a concrete, user-facing regression with high discussion activity and inbound references, making it a strong duplicate anchor.", "canonical_pr_reason": null, - "best_issue_reason": "issue:45020 is the best global issue choice because it has the widest cross-item relevance and best represents the recurring v5 compatibility-regression pattern in this cluster.", + "best_issue_reason": "issue:43749 is the most representative standalone bug report: specific failure mode, broad impact, and enough engagement to serve as a stable triage target.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43606", "right": "issue:45072", "accept": false, - "reason": "Different failures: bark-small CPU offload device mismatch vs bfloat16 dtype mismatches in unrelated models." + "reason": "Different failures and code paths: CPU offload device mismatch vs bfloat16 dtype mismatch." }, { "left": "issue:43322", "right": "issue:45685", "accept": false, - "reason": "Segfault while loading Llava Next is unrelated to the MPS histogram Int error in MoE." + "reason": "Different subsystems and symptoms: Llava Next loading segfault vs an MPS histogram/int error." }, { "left": "issue:43550", "right": "issue:45072", "accept": false, - "reason": "Bamba torch.compile+SDPA regression and dtype-mismatch issues are different code paths and fixes." + "reason": "One is torch.compile + SDPA on Bamba-9B-v2; the other is a generic dtype mismatch regression." }, { "left": "issue:43388", "right": "issue:46032", "accept": false, - "reason": "One is a gather_for_metrics label-dropping bug; the other is a Mamba2 use_cache correctness bug." + "reason": "Different logic bugs: metric gathering truncation vs Mamba2 use_cache correctness." }, { "left": "issue:43782", "right": "issue:44977", "accept": false, - "reason": "Qwen3VL weight_only loading failure is not the same bug as Qwen3.5 flash-attention generation regression." + "reason": "Different models and bugs: Qwen3VL weight_only loading error vs Qwen3.5 flash-attention generation failure." }, { - "left": "issue:43825", - "right": "issue:43994", + "left": "issue:44479", + "right": "issue:45834", "accept": false, - "reason": "Translation-task error-message wording and SigLIP2 nonsensical outputs are unrelated issues." + "reason": "Both involve vision/video inputs, but the affected models and failure modes differ materially." }, { - "left": "issue:44479", - "right": "issue:45834", + "left": "issue:43825", + "right": "issue:43994", "accept": false, - "reason": "These are different video-input regressions in different model families with different symptoms." + "reason": "Error-message wording about pipeline support is unrelated to SigLIP2 producing bad outputs." }, { "left": "issue:44991", "right": "issue:46291", "accept": false, - "reason": "Tokenizer loading failure and kernels import/version-resolution failure do not share the same underlying bug." + "reason": "Tokenizer loading failure for a specific model vs an import-time kernels dependency error." }, { - "left": "issue:43577", - "right": "issue:43818", + "left": "issue:43582", + "right": "issue:46512", "accept": false, - "reason": "Blip2 dtype-staleness and Video-LLaVA temporal-attention/weight-sharing problems are unrelated." + "reason": "AppleSilicon allocator TypeError and stale config.dtype after from_config are different bugs." }, { - "left": "issue:43582", - "right": "issue:46512", + "left": "issue:43577", + "right": "issue:43818", "accept": false, - "reason": "Apple Silicon allocator TypeError is not the same as stale config.dtype after from_config(dtype=...)." + "reason": "Blip2 dtype propagation and Video-LLaVA architecture/state-shared issues are unrelated." }, { "left": "issue:43421", "right": "issue:44811", "accept": false, - "reason": "Runtime post-processor update bug and Whisper batch_decode skip_special_tokens behavior are different tokenizer issues." + "reason": "Runtime special-token post-processor updates are not the same bug as Whisper batch_decode ignoring skip_special_tokens." }, { "left": "issue:44561", "right": "issue:45020", "accept": false, - "reason": "Both involve v5 breakage, but one is a specific removed-helper regression while the other is a broad umbrella issue, not the same bug." + "reason": "Both mention remote_code/v5 breakage, but the concrete root causes are not shown to be the same." }, { "left": "issue:43716", "right": "issue:44008", "accept": false, - "reason": "Image preprocessor/model dtype mismatch and Gemma3n variable-name collision are separate failures." + "reason": "Different models and error mechanisms: dtype mismatch in Mistral-3 vs a Gemma3n variable-name collision." }, { "left": "issue:43931", "right": "issue:46846", "accept": false, - "reason": "Qwen3-VL shape mismatch at load time and missing TP-plan entries causing OOM/shape errors are different defects." + "reason": "Qwen3-VL weight shape mismatch is not the same as missing linear_attn tensor-parallel plan entries." }, { "left": "issue:46294", "right": "issue:46463", "accept": false, - "reason": "Video feature/token mismatch in Qwen3.5-9B is unrelated to T5 decoder-input validation error messaging." + "reason": "A video chat-template token/feature mismatch is unrelated to T5 decoder input validation messaging." }, { "left": "issue:28218", "right": "issue:29159", "accept": false, - "reason": "Added-token spacing and slow-vs-fast tokenizer inconsistency may be related in theme, but they are not clearly the same underlying bug." + "reason": "Both concern tokenizer behavior, but one is a specific added-token spacing bug and the other is a broad slow/fast inconsistency report." }, { - "left": "issue:44805", - "right": "issue:44936", + "left": "issue:44610", + "right": "issue:45072", "accept": false, - "reason": "Mask-shape IndexError and trainer.evaluate() failing after train are different training/runtime issues." + "reason": "OmDet-Turbo image-size mismatch is a different class of bug from bfloat16 dtype mismatches." }, { - "left": "issue:44610", - "right": "issue:45072", + "left": "issue:44805", + "right": "issue:44936", "accept": false, - "reason": "OmDet processor resolution mismatch and bfloat16 dtype mismatches are unrelated bugs." + "reason": "Mask/tensor shape mismatch is not the same underlying issue as trainer.evaluate() failing after train()." }, { "left": "issue:43540", "right": "issue:45381", "accept": false, - "reason": "Qwen3Omni video-input ValueError and Qwen2.5-VL vision_position_ids regression are distinct model-specific issues." + "reason": "Different Qwen video-input failures with different model families and error conditions." }, { "left": "issue:43720", "right": "issue:45072", "accept": false, - "reason": "BitNet packed-weight loading and dtype mismatches in inference are different failure modes." + "reason": "Packed-weights loading failure and dtype mismatch in inference are unrelated." }, { "left": "issue:43825", "right": "issue:43827", "accept": false, - "reason": "One is a runtime error-message issue; the other is stale documentation referencing removed pipeline APIs." + "reason": "One is an error message issue; the other is stale docs still referencing removed pipelines." }, { "left": "issue:43749", "right": "issue:44291", "accept": false, - "reason": "FSDP CPU RAM efficient loading and init_empty_weights unexpected-argument errors overlap on loading, but not the same code-path bug." + "reason": "Both are loading-related, but FSDP CPU RAM-efficient loading and init_empty_weights argument handling are distinct code paths." }, { "left": "issue:43645", "right": "issue:46665", "accept": false, - "reason": "Jupyter custom-model initialization breakage and from_pretrained-inside-classes inconsistency are related only at a high level." + "reason": "Jupyter custom-model initialization breakage is different from from_pretrained behavior inside Python classes." }, { "left": "issue:44671", "right": "issue:45701", "accept": false, - "reason": "CamemBERT masked-LM prediction regression and tokenizer-version behavior changes are not the same underlying defect." + "reason": "CamemBERT masked-LM regression and a generic tokenization-change report are not clearly the same defect." }, { "left": "issue:43577", "right": "issue:43582", "accept": false, - "reason": "Blip2 dtype handling and Apple Silicon allocator warmup TypeError are unrelated." + "reason": "A dtype propagation bug in BLIP2 is unrelated to an AppleSilicon allocator TypeError." }, { "left": "issue:42175", "right": "issue:45081", "accept": false, - "reason": "Missing TensorFlow in the torch extra and Mistral regex patch crashes are unrelated packaging/tokenizer issues." + "reason": "Packaging/backend dependency omission and a Mistral tokenizer regex crash are unrelated." }, { "left": "issue:46175", "right": "issue:46932", "accept": false, - "reason": "Gemma4 one-hot position embedding blowup and quantized embedder dtype casting are different Gemma4 bugs." + "reason": "Massive one-hot position embedding materialization is a different problem than quantized embedder dtype casting." } ] }, @@ -6284,15 +6378,15 @@ "issue:43653|issue:44912", "issue:46697|issue:46728", "issue:43782|issue:45325", - "issue:45081|issue:45406", "issue:43262|issue:43819", + "issue:45081|issue:45406", "issue:45372|issue:46326", "issue:43582|issue:44857", "issue:44589|issue:45042", "issue:43550|issue:44912", "issue:43742|issue:45020", - "issue:43329|issue:45600", "issue:45084|issue:45538", + "issue:43329|issue:45600", "issue:43452|issue:44991", "issue:44060|issue:44368", "issue:43526|issue:43697", @@ -6308,14 +6402,14 @@ "issue:42490|issue:43257", "issue:43720|issue:44610" ], - "prepared_review_unit_hash": "36c554ad211a69fb9867b4492586eb63112e7dbd6809b57f10a0789dab0d490a", + "prepared_review_unit_hash": "85db1b9ba79c3c01ef722254e1e44b04e06c5405499b5f8b026ab69f810531ca", "component_budget": { "node_count": 502, "item_count": 502, - "soft_pair_count": 1842, - "serialized_chars": 956182, - "estimated_input_tokens": 239046, - "estimated_eval_tokens": 478348 + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 }, "budget": { "node_count": 48, @@ -6330,180 +6424,180 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "These issues are largely heterogeneous and do not form a single duplicate set. The soft pairs mostly share only broad subsystem similarity (tokenizers, image processors, loading, training), but not the same concrete bug.", - "confidence": 0.94, - "canonical_issue_reason": null, + "summary": "The cluster is mostly a loose bundle of unrelated Transformers bug reports that only share broad subsystem or version-regression themes. I do not see any true duplicate pairs among the soft edges, so all proposed merges are rejected.", + "confidence": 0.84, + "canonical_issue_reason": "Issue 45020 is the broadest representative of the cluster\u2019s general theme of Transformers version regressions affecting loading/remote-code models, but it is not a true duplicate umbrella for the whole set.", "canonical_pr_reason": null, - "best_issue_reason": "issue:45020 is the broadest representative of a class of recent loading/regression reports, but it should be treated as an umbrella reference rather than a true duplicate canonical for this cluster.", + "best_issue_reason": "Issue 45020 is the most central and broadly scoped report here; the others are mostly model- or component-specific regressions that should stay separate.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:34634", "right": "issue:36331", "accept": false, - "reason": "Different components and failures: BarkProcessor preset handling vs trainer compute_loss signature mismatch." + "reason": "Different components and failure modes: BarkProcessor voice_preset vs CustomTrainer.compute_loss signature mismatch." }, { "left": "issue:45003", "right": "issue:46285", "accept": false, - "reason": "Unrelated bugs: sys.modules access in modeling_utils vs CLIP text model structure inconsistency." + "reason": "Both are model-internals regressions, but one is unsafe sys.modules access and the other is CLIP class structure inconsistency." }, { "left": "issue:43653", "right": "issue:44912", "accept": false, - "reason": "Tokenizer special-token registration vs MXFP4 quantization loading; no shared code-path bug." + "reason": "Tokenizer special-token decode bug vs MXFP4 quantization loading/fallback bug; unrelated issues." }, { "left": "issue:46697", "right": "issue:46728", "accept": false, - "reason": "Both are image-processing edge cases, but they affect different models and different logic (padding vs patch counting)." + "reason": "Both involve image-processing sizing logic, but they affect different models and different calculations." }, { "left": "issue:43782", "right": "issue:45325", "accept": false, - "reason": "Different Qwen VL bugs: weight_only loading error vs rope_index temporal scaling error." + "reason": "Both are Qwen VL-related, but one is a from_pretrained weight_only failure and the other is a rope-index temporal scaling bug." }, { - "left": "issue:45081", - "right": "issue:45406", + "left": "issue:43262", + "right": "issue:43819", "accept": false, - "reason": "Tokenizer regex patch crash and Gemma4 processor _tokenizer crash are separate loading failures." + "reason": "Audio processor sampling-rate default mismatch is unrelated to DAC latent/STE forward-pass mismatch." }, { - "left": "issue:43262", - "right": "issue:43819", + "left": "issue:45081", + "right": "issue:45406", "accept": false, - "reason": "Audio chat-template sampling-rate defaulting is unrelated to DAC latent reconstruction/STE behavior." + "reason": "Tokenizer regex patch crash and Gemma4Processor missing _tokenizer are distinct load-path failures." }, { "left": "issue:45372", "right": "issue:46326", "accept": false, - "reason": "Different mistral_common/Gemma import issue vs continue_final_message behavior bug." + "reason": "Both mention mistral_common, but one is an import error and the other is a continue_final_message behavior bug." }, { "left": "issue:43582", "right": "issue:44857", "accept": false, - "reason": "Apple Silicon allocator warmup TypeError is unrelated to CUDA AMP image loss crashing." + "reason": "Apple Silicon allocator TypeError and CUDA AMP image-loss crash are unrelated runtime/platform bugs." }, { "left": "issue:44589", "right": "issue:45042", "accept": false, - "reason": "Storage-object import failure vs PIL backend torchvision dependency regression; unrelated." + "reason": "Storage-object lookup failure is unrelated to PIL image processors incorrectly requiring torchvision." }, { "left": "issue:43550", "right": "issue:44912", "accept": false, - "reason": "torch.compile/SDPA model failure is unrelated to MXFP4 quantization fallback." + "reason": "torch.compile/SDPA failure for Bamba is unrelated to MXFP4 quantization loading behavior." }, { "left": "issue:43742", "right": "issue:45020", "accept": false, - "reason": "A specific MobileLLM loading key error is not the same bug as the broader remote_code regression." + "reason": "A model-specific key error when loading MobileLLM-125M is not the same as the broad remote_code regression." }, { - "left": "issue:43329", - "right": "issue:45600", + "left": "issue:45084", + "right": "issue:45538", "accept": false, - "reason": "Multimodal token counting bug and removed auto_mappings config references are different issues." + "reason": "Template compilation TypeError and CLIPTokenizer max_length bug are unrelated." }, { - "left": "issue:45084", - "right": "issue:45538", + "left": "issue:43329", + "right": "issue:45600", "accept": false, - "reason": "Template compilation TypeError and CLIPTokenizer max_length are unrelated." + "reason": "Multimodal token-count video-branch bug is unrelated to auto_mappings referencing removed configs." }, { "left": "issue:43452", "right": "issue:44991", "accept": false, - "reason": "gguf_file loading breakage and a tokenizer-loading regression are different concrete failures." + "reason": "Both are from_pretrained loading problems, but the affected artifacts and root causes differ." }, { "left": "issue:44060", "right": "issue:44368", "accept": false, - "reason": "Both mention tied embeddings, but one is an incorrect weight-tying warning bug and the other is a user-facing warning during LoRA fine-tuning." + "reason": "One is an incorrect tied-weights warning bug; the other is a user-facing config warning about tie_word_embeddings." }, { "left": "issue:43526", "right": "issue:43697", "accept": false, - "reason": "Different vision-model bugs: reduce_labels output and RTDetrV2 output divergence." + "reason": "BeitImageProcessorFast label reduction bug is unrelated to RTDetrV2 output divergence." }, { "left": "issue:43257", "right": "issue:43329", "accept": false, - "reason": "Qwen3 MoE accelerate/deepspeed loading and multimodal token counting are unrelated." + "reason": "Qwen3-MoE weight conversion/loading issue is unrelated to multimodal token counting in the video branch." }, { "left": "issue:43526", "right": "issue:44554", "accept": false, - "reason": "Beit label reduction bug and MPS attention correctness issue do not share the same code path." + "reason": "Different domains and code paths: image label reduction vs MPS attention correctness." }, { "left": "issue:43906", "right": "issue:46209", "accept": false, - "reason": "An isolated reproduction of an older issue is unrelated to FP8 inference on B200." + "reason": "An isolated reproduction of an older issue is unrelated to FP8 inference failure on B200." }, { "left": "issue:36010", "right": "issue:36331", "accept": false, - "reason": "Importing GenerationMixin and compute_loss keyword mismatch are separate regressions." + "reason": "ImportError for GenerationMixin and compute_loss signature mismatch are separate regressions." }, { "left": "issue:43819", "right": "issue:43927", "accept": false, - "reason": "DAC latent reconstruction mismatch and DiaConfig save/load token ID loss are unrelated." + "reason": "DAC latent consistency bug and DiaConfig save/load token-id loss are different audio-model issues." }, { "left": "issue:36010", "right": "issue:43645", "accept": false, - "reason": "ImportError for GenerationMixin and Jupyter custom-model initialization are different failures." + "reason": "Import path breakage and Jupyter custom-model initialization failure are not the same underlying bug." }, { "left": "issue:43531", "right": "issue:45042", "accept": false, - "reason": "Qwen3-MoE sliding_window behavior and PIL backend torchvision dependency are unrelated." + "reason": "Qwen3-MoE sliding_window and PIL/torchvision backend requirements are unrelated." }, { "left": "issue:42175", "right": "issue:46291", "accept": false, - "reason": "Backend extra installation issue and kernels import requirement change are unrelated." + "reason": "Packaging backend-selection issue and pipelines/kernels version requirement issue are unrelated." }, { "left": "issue:43638", "right": "issue:44936", "accept": false, - "reason": "Zero3/non-pretrained Bert index error is not the same as evaluate-after-train failure." + "reason": "Zero3/non-pretrained Bert index error and train-then-evaluate failure are separate trainer/deepspeed bugs." }, { "left": "issue:42490", "right": "issue:43257", "accept": false, - "reason": "Local-path import warning and Qwen3 MoE accelerate loading issue are unrelated." + "reason": "Trailing-slash local path warning is unrelated to Qwen3 MoE weight conversion/loading." }, { "left": "issue:43720", "right": "issue:44610", "accept": false, - "reason": "BitNet packed-weight loading and OmDet-Turbo image-size mismatch are different bugs." + "reason": "BitNet packed-weight loading and OmDet-Turbo processor input-size mismatch are unrelated." } ] }, @@ -6578,8 +6672,8 @@ ], "soft_pairs": [ "issue:39692|issue:44912", - "issue:43643|issue:45759", "issue:43854|issue:44977", + "issue:43643|issue:45759", "issue:43746|issue:44811", "issue:43257|issue:43696", "issue:38175|issue:44610", @@ -6587,10 +6681,10 @@ "issue:43262|issue:44811", "issue:43257|issue:43931", "issue:45459|issue:45803", - "issue:46421|issue:46424", "issue:43329|issue:43450", - "issue:43329|issue:44479", + "issue:46421|issue:46424", "issue:42915|issue:43742", + "issue:43329|issue:44479", "issue:43749|issue:44756", "issue:34567|issue:40990", "issue:43650|issue:43906", @@ -6605,203 +6699,206 @@ "issue:44568|issue:45356", "issue:44987|issue:45538" ], - "prepared_review_unit_hash": "7d23724f457734c2269b32102c75543fdf1c247e28d0d3ddbfb28d6fc4184e43", + "prepared_review_unit_hash": "ca64c1a518ef71a46b0a9b7d3bfc1d1f3574d2970a32b7a224bfc66184dbc4f9", "component_budget": { "node_count": 502, "item_count": 502, - "soft_pair_count": 1842, - "serialized_chars": 956182, - "estimated_input_tokens": 239046, - "estimated_eval_tokens": 478348 + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 27, - "serialized_chars": 35929, - "estimated_input_tokens": 8983, - "estimated_eval_tokens": 18222 + "serialized_chars": 35923, + "estimated_input_tokens": 8981, + "estimated_eval_tokens": 18218 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is heterogeneous: it mixes unrelated issues in loading, tokenizers, multimodal/video processing, quantization, cache behavior, trainer state, and memory/OOMs. None of the proposed soft edges looks like the same underlying bug or change, so all candidate merges should be rejected.", - "confidence": 0.95, - "canonical_issue_reason": "No clear canonical issue: the items span multiple unrelated subsystems and failure modes, so there is no single issue that cleanly represents the whole cluster.", + "summary": "Most candidate pairs are only loosely related by subsystem or symptom and do not look like the same underlying bug. The only strong duplicate candidate is the Qwen video regression pair, where the broader regression report is the better anchor.", + "confidence": 0.73, + "canonical_issue_reason": "Issue 44479 is the best canonical issue: it is the broader user-facing regression report across multiple Qwen VL variants, while issue 43329 is a narrower root-cause/debugging report in the same video-token path.", "canonical_pr_reason": null, - "best_issue_reason": "No global best issue for deduping this set; if forced, issue:45759 is the most precise self-contained report, but it is not a duplicate umbrella for the others.", + "best_issue_reason": "Issue 44479 is the most representative and actionable issue in the set because it describes the regression scope and affected models clearly.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:39692", "right": "issue:44912", "accept": false, - "reason": "Different problems in different models: SigLIP2 docs/example mistakes vs gpt-oss-20b MXFP4 quantization fallback." + "reason": "Both are model/report quality problems, but they concern different models and different failure modes (docs example errors vs MXFP4 load/quantization fallback)." }, { - "left": "issue:43643", - "right": "issue:45759", + "left": "issue:43854", + "right": "issue:44977", "accept": false, - "reason": "Both involve config loading, but the bugs are different: missing fields from trust_remote_code vs not unwrapping text_config for composite multimodal configs." + "reason": "Different models and different code paths: one is a load-shape mismatch, the other is a flash-attention generation issue." }, { - "left": "issue:43854", - "right": "issue:44977", + "left": "issue:43643", + "right": "issue:45759", "accept": false, - "reason": "Unrelated failures: GLM-4.7 loading/unit-test issue vs Qwen3.5 flash-attention generation regression." + "reason": "Both involve config/model loading, but one is missing fields from trust_remote_code AutoConfig, the other is AutoModelForCausalLM.from_config not unwrapping text_config for composite multimodal configs." }, { "left": "issue:43746", "right": "issue:44811", "accept": false, - "reason": "Different code paths and symptoms: GraniteSpeech PEFT checkpoint loading vs Whisper batch_decode skip_special_tokens handling." + "reason": "Unrelated bugs: PEFT adapter local checkpoint loading versus Whisper batch_decode skip_special_tokens handling." }, { "left": "issue:43257", "right": "issue:43696", "accept": false, - "reason": "One is a Qwen3 MoE loading/conversion bug; the other is a GPT-oss CUDA OOM. Same family of model size concerns, but not the same defect." + "reason": "One is Qwen3 MoE weight conversion with accelerate/deepspeed; the other is a GPT-oss CUDA OOM. Same broad area, different bugs." }, { "left": "issue:38175", "right": "issue:44610", "accept": false, - "reason": "Different models and root causes: unexpected zero probabilities in SigLIP2 vs OmDet-Turbo processor/model input-size mismatch." + "reason": "Different models and failures: zero probabilities in SigLIP2 versus a processor/model input-size mismatch in OmDet-Turbo." }, { "left": "issue:44387", "right": "issue:44617", "accept": false, - "reason": "Both mention OOM, but one is an int4 quantization memory regression and the other is a Sam3Video model-specific OOM." + "reason": "Both mention memory pressure, but they are different issues: int4 quantization reserved-memory regression versus Sam3Video OOM." }, { "left": "issue:43262", "right": "issue:44811", "accept": false, - "reason": "Both touch audio processing, but apply_chat_template sampling-rate defaults and Whisper batch_decode token skipping are unrelated bugs." + "reason": "Audio chat-template sampling-rate default versus Whisper decode token skipping are separate processor behaviors." }, { "left": "issue:43257", "right": "issue:43931", "accept": false, - "reason": "Both involve Qwen3-family loading, but one is MoE weight conversion with accelerate/deepspeed and the other is a weight-shape mismatch in Qwen3-VL." + "reason": "Both are Qwen-family loading problems, but the failures differ: MoE weight conversion under accelerate/deepspeed versus Qwen3-VL weight-shape mismatch." }, { "left": "issue:45459", "right": "issue:45803", "accept": false, - "reason": "Both concern exception handling, but they are different implementations and symptoms: hidden tokenizer errors vs bare except swallowing interrupts/errors in FuyuBatchFeature." - }, - { - "left": "issue:46421", - "right": "issue:46424", - "accept": false, - "reason": "Related to cache behavior, but one is spurious recompilation from lazy init and the other is max_cache_length only being honored for Quantized Cache." + "reason": "Same general anti-pattern (exception swallowing) but different components and bugs; not the same underlying issue." }, { "left": "issue:43329", "right": "issue:43450", "accept": false, - "reason": "Video-related, but not the same bug: undefined variables in multimodal token counting vs incorrect batched video processor output shape." + "reason": "Both involve video processors, but one is an undefined-variable bug in multimodal token counting and the other is a batched-shape bug in video processing." }, { - "left": "issue:43329", - "right": "issue:44479", + "left": "issue:46421", + "right": "issue:46424", "accept": false, - "reason": "Both mention video regressions, but one is an internal undefined-variable bug in token counting and the other is a broader generation regression across several models." + "reason": "Both are cache-related, but one is lazy-init recompilation with chunked prefill and the other is max_cache_length only being honored for Quantized Cache." }, { "left": "issue:42915", "right": "issue:43742", "accept": false, - "reason": "Different failures in different models: FineGrainedFP8Config breakage in Qwen3 MoE vs a key error when loading MobileLLM-125M." + "reason": "Different models and different failure classes: FineGrainedFP8Config failure versus a MobileLLM loading key error." + }, + { + "left": "issue:43329", + "right": "issue:44479", + "accept": true, + "reason": "These describe the same video-input regression in the Qwen VL family; 43329 names the broken helper/undefined variables in the video token path, and 44479 reports the user-facing regression across affected models." }, { "left": "issue:43749", "right": "issue:44756", "accept": false, - "reason": "Both are memory-related, but FSDP CPU RAM efficient loading and disabling mmap on Strix Halo are distinct platform/feature issues." + "reason": "Both can lead to memory issues, but one is FSDP CPU RAM efficient loading and the other is mmap on Strix Halo; different mechanisms and fixes." }, { "left": "issue:34567", "right": "issue:40990", "accept": false, - "reason": "Unrelated: TrainerState token accounting vs high perplexity on gpt-oss-20b." + "reason": "Trainer token accounting not updating is unrelated to GPT-oss perplexity on WikiText." }, { "left": "issue:43650", "right": "issue:43906", "accept": false, - "reason": "No relation: a placeholder-like issue title versus an isolated reproduction of a different upstream bug." + "reason": "The titles are not about the same bug; one is a placeholder/placeholder-like report, the other is an isolated reproduction of a separate upstream issue." }, { "left": "issue:44291", "right": "issue:45042", "accept": false, - "reason": "Different subsystems: init_empty_weights argument handling vs PIL image processors incorrectly requiring torchvision." + "reason": "Different regressions in different areas: init_empty_weights argument compatibility versus PIL backend requiring torchvision." }, { "left": "issue:42915", "right": "issue:44821", "accept": false, - "reason": "Different bugs: Qwen3 MoE FP8 loading failure vs AutoImageProcessor URL loading." + "reason": "Qwen3 MoE FP8 failure and AutoImageProcessor-from-URL loading failure are unrelated." }, { "left": "issue:29159", "right": "issue:34567", "accept": false, - "reason": "Tokenizer inconsistency and TrainerState token counting are unrelated." + "reason": "Tokenizer slow/fast inconsistency is unrelated to TrainerState num_input_tokens_seen not updating." }, { "left": "issue:43606", "right": "issue:45859", "accept": false, - "reason": "Both involve offload/device behavior, but bark-small CPU offload mismatch and Qwen3_5Moe rotary embedding CPU-offload incompatibility are different code paths." + "reason": "Both mention CPU offload, but they are different models and different code paths; not the same underlying bug." }, { "left": "issue:43927", "right": "issue:44912", "accept": false, - "reason": "Different root causes: DiaConfig losing custom token IDs on save/load vs gpt-oss-20b MXFP4 quantization falling back to bf16." + "reason": "DiaConfig save/load token-ID loss causing generation IndexError is unrelated to MXFP4 quantization fallback on gpt-oss-20b." }, { "left": "issue:28282", "right": "issue:29870", "accept": false, - "reason": "Completely different issues: missing PyTorch environment detection vs Bart generate UnboundLocalError on decoder_attention_mask." + "reason": "ImportError about missing PyTorch is unrelated to Bart generate() using decoder_attention_mask before assignment." }, { "left": "issue:43540", "right": "issue:46701", "accept": false, - "reason": "Both are multimodal pipeline failures, but they affect different models and different failure modes." + "reason": "Different multimodal pipelines and different failure modes: Qwen3OmniMoe video input ValueError versus DiffusionGemma image-text-to-text pipeline failure." }, { "left": "issue:43525", "right": "issue:46291", "accept": false, - "reason": "Unrelated: missing pad_token_id on Llama4Config vs import-time kernels dependency/version requirements." + "reason": "Missing pad_token_id on Llama4Config is unrelated to the transformers.pipelines import failure with kernels v0.15." }, { "left": "issue:44568", "right": "issue:45356", "accept": false, - "reason": "Both are tokenizer regressions, but they affect different tokenizers and different behaviors (special tokens vs codec/warning handling)." + "reason": "Both are tokenizer regressions, but they affect different models and different mechanisms (special-token insertion vs codec/warning behavior)." }, { "left": "issue:44987", "right": "issue:45538", "accept": false, - "reason": "Different issues: loading failure for physical-intelligence/fast in transformers>=5.1.0 vs CLIPTokenizer using an absurdly large model_max_length." + "reason": "Model-loading failure for physical-intelligence/fast is unrelated to CLIPTokenizer model_max_length behavior." } ] }, - "evaluator_result": null, - "evaluator_used": false, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet: the chosen canonical pair (43329, 44479) is supported by the shared Qwen video-token regression path, and the broader/narrower framing stays conservative. The other soft-edge verdicts mostly distinguish different models or failure modes without overclaiming duplication. No mergeability issues apply here since these are issue pairs, not PR pairs." + }, + "evaluator_used": true, "retried": false, - "accepted_nontrivial_soft_edge": false, + "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null }, @@ -6869,8 +6966,8 @@ ], "soft_pairs": [ "issue:43454|issue:43577", - "issue:28282|issue:34689", "issue:45538|issue:46869", + "issue:28282|issue:34689", "issue:44857|issue:46832", "issue:43697|issue:45357", "issue:43425|issue:46291", @@ -6887,8 +6984,8 @@ "issue:42915|issue:43749", "issue:43577|issue:46512", "issue:44849|issue:45701", - "issue:27957|issue:28282", "issue:43452|issue:43582", + "issue:27957|issue:28282", "issue:24643|issue:29159", "issue:44987|issue:45907", "issue:41720|issue:43540", @@ -6896,202 +6993,196 @@ "issue:44568|issue:45488", "issue:44792|issue:44964" ], - "prepared_review_unit_hash": "8605fbf1e8901614a435ac7bdaba4ce0de0b13f953c4a803593d5802d5edbcd6", + "prepared_review_unit_hash": "b8f8fb07eb6f8dd870a6ceb6a8e702e982d5db5ac1329601d3c2202b4f55dcc5", "component_budget": { "node_count": 502, "item_count": 502, - "soft_pair_count": 1842, - "serialized_chars": 956182, - "estimated_input_tokens": 239046, - "estimated_eval_tokens": 478348 + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 27, - "serialized_chars": 35726, - "estimated_input_tokens": 8932, - "estimated_eval_tokens": 18120 + "serialized_chars": 35732, + "estimated_input_tokens": 8933, + "estimated_eval_tokens": 18122 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is heterogeneous: it spans unrelated tokenizer regressions, multimodal/model-loading failures, training/runtime errors, and backend packaging issues. None of the soft-edge pairs look like the same concrete bug/change, so I rejected them all. If one issue must anchor the cluster, issue 45488 is the strongest high-signal regression report, but it is not a duplicate of the others.", - "confidence": 0.89, - "canonical_issue_reason": "Issue 45488 is the clearest, most actionable anchor in an otherwise mixed cluster: it has active discussion, inbound refs, and a concrete tokenizer-regression description.", + "summary": "This cluster is heterogeneous: the paired issues mostly concern different models, subsystems, and failure modes, so I do not see any confident duplicate matches. If a single representative is needed, issue 45488 is the strongest anchor because it is the most detailed and has inbound references.", + "confidence": 0.87, + "canonical_issue_reason": "No true canonical duplicate stands out; if forced to pick one representative, issue 45488 is the most substantial/open regression report with the best discussion signal.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 45488 is the best representative issue here because it describes a specific regression with broader downstream impact than the other more isolated reports.", + "best_issue_reason": "Issue 45488 is the most actionable and representative issue in this set: detailed repro, clear impact, and inbound references.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43454", "right": "issue:43577", "accept": false, - "reason": "Different bugs: Mistral3 lm_head tying vs Blip2 dtype propagation." + "reason": "Both are model-weight/dtype problems, but they affect different model families and different failure points; not the same bug." }, { - "left": "issue:28282", - "right": "issue:34689", + "left": "issue:45538", + "right": "issue:46869", "accept": false, - "reason": "Both involve loading, but one is a missing-PyTorch dependency error and the other is a model-specific loading regression." + "reason": "Unrelated failures: tokenizer max-length handling vs multi-GPU device-map crashes for YOLO." }, { - "left": "issue:45538", - "right": "issue:46869", + "left": "issue:28282", + "right": "issue:34689", "accept": false, - "reason": "Unrelated: CLIPTokenizer max_length bug vs YOLO multi-GPU device-map failure." + "reason": "Both are loading errors, but one is a missing PyTorch import path issue and the other is a model-specific loading regression." }, { "left": "issue:44857", "right": "issue:46832", "accept": false, - "reason": "Different detection-model crashes with different failure modes and code paths." + "reason": "Different detection-model crashes with different triggers and code paths; not a shared root cause." }, { "left": "issue:43697", "right": "issue:45357", "accept": false, - "reason": "RT-DETR output drift and Qwen3.5 save_pretrained key regression are distinct bugs." + "reason": "RT-DETR output mismatch and Qwen3.5 save_pretrained key regression are unrelated bugs." }, { "left": "issue:43425", "right": "issue:46291", "accept": false, - "reason": "Torch compatibility issue vs kernels revision/version requirement on import." + "reason": "Torch version incompatibility and kernels revision/version import failure are separate dependency issues." }, { "left": "issue:33357", "right": "issue:34567", "accept": false, - "reason": "Mac CLIP bus error is unrelated to TrainerState token counting." + "reason": "MacOS bus error in CLIP loading vs TrainerState token counter not updating are unrelated." }, { "left": "issue:42915", "right": "issue:45538", "accept": false, - "reason": "Qwen3Moe FP8 failure is unrelated to CLIPTokenizer max_length behavior." + "reason": "Fine-grained FP8 config failure and CLIP tokenizer max-length bug are different subsystems." }, { "left": "issue:43404", "right": "issue:45127", "accept": false, - "reason": "Both touch tied weights, but they are different model families and different failure mechanisms." + "reason": "Both involve tied weights, but they are different models and different scenarios; not the same concrete bug." }, { "left": "issue:45137", "right": "issue:45405", "accept": false, - "reason": "DeepSpeed ZeRO3 deque error is unrelated to the PEFT version pin regression." + "reason": "DeepSpeed ZeRO3 deque error and PEFT version pin issue are unrelated." }, { "left": "issue:42175", "right": "issue:43824", "accept": false, - "reason": "Packaging/backend omission and a specific model import error are not the same bug." + "reason": "Package/backend installation issue vs missing Qwen2.5-VL export are separate problems." }, { "left": "issue:28282", "right": "issue:29127", "accept": false, - "reason": "Generic missing-PyTorch import error vs layoutlmv3 box-validation messaging." + "reason": "Generic import error and LayoutLMv3 box-validation message clarity are not the same bug." }, { "left": "issue:44062", "right": "issue:44991", "accept": false, - "reason": "AddedToken keyword collision and est-roberta tokenizer loading failure are not clearly the same code-path bug." + "reason": "The tokenizer TypeError may be related in spirit, but the second issue is model-specific and the evidence is not enough to confirm the same root cause." }, { "left": "issue:44479", "right": "issue:46693", "accept": false, - "reason": "Video-input regression and FA2 long-context performance regression are different issues." + "reason": "Video-input regression and FA2 long-context performance regression are distinct failures." }, { "left": "issue:43742", "right": "issue:44291", "accept": false, - "reason": "Different loading failures: missing key vs unexpected _is_hf_initialized argument." + "reason": "KeyError loading a model and init_empty_weights argument mismatch are different loading paths." }, { "left": "issue:44936", "right": "issue:46821", "accept": false, - "reason": "Trainer lifecycle failure vs TimesFM window_size attribute error." + "reason": "Trainer lifecycle failure and TimesFM window_size AttributeError do not share a concrete code path." }, { "left": "issue:42915", "right": "issue:43749", "accept": false, - "reason": "Qwen3Moe FP8 config issue is unrelated to FSDP CPU RAM efficient loading." + "reason": "Qwen3Moe FP8 failure and FSDP CPU RAM efficient loading broken are separate distributed/memory-loading issues." }, { "left": "issue:43577", "right": "issue:46512", "accept": false, - "reason": "One is a model dtype-loading bug; the other is stale config dtype after from_config." + "reason": "Both mention dtype, but one is a component dtype bug in BLIP2 and the other is stale config dtype after from_config; not the same bug." }, { "left": "issue:44849", "right": "issue:45701", "accept": false, - "reason": "Qwen3.5 hidden-states bug is unrelated to tokenization-version changes." + "reason": "Qwen3.5 hidden-states bug and a broad tokenization regression are unrelated." }, { - "left": "issue:27957", - "right": "issue:28282", + "left": "issue:43452", + "right": "issue:43582", "accept": false, - "reason": "Flash Attention 2 behavior and missing-PyTorch import error are unrelated." + "reason": "GGUF loading breakage and Apple Silicon caching_allocator_warmup TypeError are unrelated." }, { - "left": "issue:43452", - "right": "issue:43582", + "left": "issue:27957", + "right": "issue:28282", "accept": false, - "reason": "gguf_file loading regression is unrelated to Apple Silicon caching_allocator_warmup TypeError." + "reason": "XLMRoberta/FlashAttention2 and missing-PyTorch import error are different issues." }, { "left": "issue:24643", "right": "issue:29159", "accept": false, - "reason": "DeepSpeed weight-shape runtime error is unrelated to tokenizer slow/fast inconsistency." + "reason": "DeepSpeed training weight-shape error and tokenizer slow/fast inconsistency are unrelated." }, { "left": "issue:44987", "right": "issue:45907", "accept": false, - "reason": "The titles suggest a possible shared revision-related area, but not the same concrete bug with enough evidence to merge." + "reason": "Model loading regression and list_repo_files revision forwarding bug are different code paths." }, { "left": "issue:41720", "right": "issue:43540", "accept": false, - "reason": "Auto device-map CUDA assert and Qwen3OmniMoe video-processing ValueError are different failures." + "reason": "Auto device-map CUDA assert and Qwen3OmniMoe video input ValueError are unrelated." }, { "left": "issue:42175", "right": "issue:44488", "accept": false, - "reason": "Backend packaging issue and sleng-bert loading failure do not match." + "reason": "Backend install dependency issue and model loading failure for a specific checkpoint are not the same bug." }, { "left": "issue:44568", "right": "issue:45488", "accept": false, - "reason": "Different tokenizer regressions with different mechanisms and model families." + "reason": "Different tokenizer regressions: one is missing BOS/EOS addition, the other overrides tokenizer.json pre-tokenizer behavior." }, { "left": "issue:44792", "right": "issue:44964", "accept": false, - "reason": "Janus generate-images test failure and Phi-4 multimodal loading failure are not the same code-path problem." - }, - { - "left": "issue:44936", - "right": "issue:46821", - "accept": false, - "reason": "Duplicate of the earlier trainer/timesfm mismatch: unrelated issues with no shared failure path." + "reason": "Failed image-generation test and Phi-4 multimodal loading regression are different model issues." } ] }, @@ -7166,11 +7257,11 @@ ], "soft_pairs": [ "issue:43540|issue:44560", - "issue:44479|issue:45381", "issue:44291|issue:44821", + "issue:44479|issue:45381", "issue:38175|issue:44898", - "issue:43257|issue:43756", "issue:34567|issue:36331", + "issue:43257|issue:43756", "issue:43540|issue:44479", "issue:44062|issue:45081", "issue:45042|issue:45081", @@ -7190,24 +7281,24 @@ "issue:44737|issue:46463", "issue:45823|issue:46463", "issue:43452|issue:44488", - "issue:43697|issue:46829", "issue:44625|issue:44877", + "issue:43697|issue:46829", "issue:45663|issue:46846" ], - "prepared_review_unit_hash": "d6d4366311af76387e8cf7c099f5eac38283d9583c8a1ecec180570c987f73e3", + "prepared_review_unit_hash": "22b256e4157719fa225ef6d047094f83815e1f6a6e8dcff7d9f7625690e17bf1", "component_budget": { "node_count": 502, "item_count": 502, - "soft_pair_count": 1842, - "serialized_chars": 956182, - "estimated_input_tokens": 239046, - "estimated_eval_tokens": 478348 + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 28, - "serialized_chars": 36480, + "serialized_chars": 36479, "estimated_input_tokens": 9120, "estimated_eval_tokens": 18496 }, @@ -7216,187 +7307,190 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a collection of unrelated Transformers bug reports across different models and subsystems. I did not find any pair that clearly looks like the same underlying defect, so all soft edges should stay split.", - "confidence": 0.79, - "canonical_issue_reason": "issue:44479 is the broadest, highest-signal report in the set: it describes a concrete v5.3.0 regression affecting multiple Qwen VL/MoE variants and is the best anchor for the small video-regression subset.", + "summary": "Most candidates are distinct one-off bugs in different model families or code paths. The only plausible duplicate edge is the Qwen2.5-VL 5.3.0 video-input position-id regression, where one issue is a broad umbrella report and the other is a specific symptom report.", + "confidence": 0.71, + "canonical_issue_reason": "Issue 44479 is the broadest and most representative report in the only strong duplicate family here: it captures the Transformers v5.3.0 Qwen video-input regression across multiple model variants, so it can serve as the umbrella canonical issue.", "canonical_pr_reason": null, - "best_issue_reason": "issue:44479 is the best standalone issue because it is well-scoped, version-specific, and covers the widest plausible duplicate space among the video-input reports.", + "best_issue_reason": "Issue 44479 is the best global issue candidate because it describes the widest affected surface and the clearest shared root cause among the near-duplicate Qwen video-input reports.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43540", "right": "issue:44560", "accept": false, - "reason": "Both are video-input bugs, but they hit different model families and different failure modes (Qwen3OmniMoe ValueError vs Qwen3-vl-embedding StopIteration)." - }, - { - "left": "issue:44479", - "right": "issue:45381", - "accept": false, - "reason": "These are the closest pair, but one is a broad multi-model regression report and the other is a narrower qwen2.5-vl position-id symptom; not clearly the same concrete bug from the titles alone." + "reason": "Different Qwen multimodal failure modes: one is Qwen3OmniMoe video-input processing, the other is a Qwen3-vl-embedding StopIteration during video handling." }, { "left": "issue:44291", "right": "issue:44821", "accept": false, - "reason": "Different subsystems: init_empty_weights / unexpected init arg versus AutoImageProcessor URL loading." + "reason": "Unrelated loader bugs: one is init_empty_weights / _is_hf_initialized, the other is AutoImageProcessor loading from URL." + }, + { + "left": "issue:44479", + "right": "issue:45381", + "accept": true, + "reason": "Both point to the same Transformers 5.3.0 Qwen2.5-VL video-input position-id regression; 44479 is the broader umbrella report and 45381 is a specific reproduction." }, { "left": "issue:38175", "right": "issue:44898", "accept": false, - "reason": "Both are vision-model issues, but the symptoms and model families differ completely (SigLIP2 zero probabilities vs Perceiver resize failure)." + "reason": "Different models and symptoms: SigLIP2 zero probabilities versus Perceiver non-default-resolution classification failure." }, { - "left": "issue:43257", - "right": "issue:43756", + "left": "issue:34567", + "right": "issue:36331", "accept": false, - "reason": "Different Qwen/SmolLM model-loading issues with unrelated mechanisms: MoE conversion under accelerate+deepspeed versus RoPE-layer dropping." + "reason": "These are separate Trainer API problems: token accounting not updating versus a custom compute_loss signature mismatch." }, { - "left": "issue:34567", - "right": "issue:36331", + "left": "issue:43257", + "right": "issue:43756", "accept": false, - "reason": "Both touch Trainer behavior, but one is token-count tracking and the other is a custom compute_loss signature break." + "reason": "Different model-specific bugs: Qwen3 MoE weight conversion with accelerate/deepspeed versus SmolLM3 RoPE-layer dropping." }, { "left": "issue:43540", "right": "issue:44479", "accept": false, - "reason": "Both mention video inputs, but they are different model lines and distinct error paths." + "reason": "Both are video-related, but they target different model families and distinct failures, so they are not the same underlying bug." }, { "left": "issue:44062", "right": "issue:45081", "accept": false, - "reason": "Tokenizer-related, but one is an AddedToken constructor conflict while the other is a Mistral regex patch crash." + "reason": "Tokenizer construction bugs in different places: AddedToken special-flag collision versus Mistral regex patching on a Tokenizer object." }, { "left": "issue:45042", "right": "issue:45081", "accept": false, - "reason": "Completely different areas: image processor backend dependency handling versus tokenizer regex patching." + "reason": "Image processor torchvision dependency regression is unrelated to the Mistral tokenizer backend_tokenizer AttributeError." }, { "left": "issue:43742", "right": "issue:44589", "accept": false, - "reason": "Different loading/runtime failures: MobileLLM key error versus missing Float8 storage type." + "reason": "Different loading errors: MobileLLM config/key failure versus a Float8 storage lookup error." }, { "left": "issue:43582", "right": "issue:46821", "accept": false, - "reason": "Different platforms and APIs: Apple Silicon allocator warmup TypeError versus TimesFM window_size attribute error." + "reason": "Apple Silicon allocator warmup TypeError and TimesFM window_size AttributeError are unrelated." }, { "left": "issue:43756", "right": "issue:46399", "accept": false, - "reason": "Both involve RoPE-ish model internals, but one is missing layers and the other is a kernel registration decorator crash." + "reason": "SmolLM3 RoPE-layer count bug and Qwen3.5 kernel registration crash are different code paths." }, { "left": "issue:43756", "right": "issue:45325", "accept": false, - "reason": "Both are Qwen-related, but one is a SmolLM3 RoPE-layer count bug and the other is a Qwen2.5-VL position-id scaling bug." + "reason": "These both touch Qwen-ish rotary logic, but one is SmolLM3 RoPE-layer dropping and the other is Qwen2.5-VL image temporal scaling; not the same bug." }, { "left": "issue:44568", "right": "issue:46491", "accept": false, - "reason": "Tokenizer behavior regressions, but one is missing BOS/EOS insertion and the other is a CodeLlama round-trip spacing issue." + "reason": "Different tokenizer regressions: missing BOS/EOS on add_special_tokens versus leading-space loss on CodeLlama round-trip." }, { "left": "issue:45405", "right": "issue:45600", "accept": false, - "reason": "Unrelated packaging/compatibility issues: PEFT version pin versus removed config references breaking CI." + "reason": "One is a version-pin problem for PEFT, the other is stale auto_mappings references causing CI breakage." }, { "left": "issue:45672", "right": "issue:45674", "accept": false, - "reason": "Both involve 4-bit inference, but the root causes differ: uint8 weight finfo error versus skip_modules clearing the default lm_head exclusion." + "reason": "Both involve quantized inference, but one is a Gemma4 uint8/torch.finfo audio-module failure and the other is a BitsAndBytesConfig skip-module exclusion bug." }, { "left": "issue:43531", "right": "issue:44821", "accept": false, - "reason": "Different domains entirely: Qwen3-MoE sliding_window bug versus image processor URL loading." + "reason": "Qwen3-MoE sliding_window behavior and AutoImageProcessor-from-URL loading are unrelated." }, { "left": "issue:43716", "right": "issue:46701", "accept": false, - "reason": "Both are multimodal, but they describe different model pipelines and different mismatch/dispatch failures." + "reason": "Different multimodal issues: Mistral-3 dtype mismatch in preprocessing versus DiffusionGemma image-text-to-text pipeline breakage." }, { "left": "issue:43577", "right": "issue:45072", "accept": false, - "reason": "Both are dtype-related, but the affected models and failure mechanisms are unrelated." + "reason": "Both mention dtype, but one is BLIP2 model/qformer dtype loading and the other is inference mismatches in SwitchTransformers/TimmWrapperModel." }, { "left": "issue:43906", "right": "issue:46468", "accept": false, - "reason": "A reproduction of an earlier issue is not enough here; the titles do not show the same concrete bug or code path." + "reason": "An isolated reproduction of another issue is not the same as a broken official Nemotron example." }, { "left": "issue:24643", "right": "issue:29870", "accept": false, - "reason": "Training DeepSpeed weight-shape error versus Bart generate() decoder_attention_mask UnboundLocalError; unrelated bugs." + "reason": "DeepSpeed training weight-shape RuntimeError and BART generate decoder_attention_mask unbound variable are separate failures." }, { "left": "issue:43906", "right": "issue:46693", "accept": false, - "reason": "Isolated reproduction versus FA2 long-context performance regression; not the same defect." + "reason": "Reproduction of an older issue is unrelated to the FA2 long-context performance regression." }, { "left": "issue:44737", "right": "issue:46463", "accept": false, - "reason": "XLNet CPU placement bug versus T5 error-message wording issue; different code paths and severity." + "reason": "XLNet CPU-device placement bug and T5 decoder-input validation error are unrelated." }, { "left": "issue:45823", "right": "issue:46463", "accept": false, - "reason": "Gemma4 device mismatch during forward is unrelated to a T5 decoder-input validation message bug." + "reason": "Gemma4 PLE device mismatch during forward is unrelated to the T5 decoder validation message issue." }, { "left": "issue:43452", "right": "issue:44488", "accept": false, - "reason": "Both are loading failures, but one is about gguf_file handling in from_pretrained while the other is a specific model that no longer loads." + "reason": "Tokenizer/model loading breakage around gguf_file is unrelated to the cjvt/sleng-bert load failure." }, { - "left": "issue:43697", - "right": "issue:46829", + "left": "issue:44625", + "right": "issue:44877", "accept": false, - "reason": "RTDetrV2 output mismatch and VideoPrism hidden_states reporting are distinct output-shape/semantic bugs." + "reason": "Qwen3.5 num_labels propagation and strict granite_speech config loading are different config-handling bugs." }, { - "left": "issue:44625", - "right": "issue:44877", + "left": "issue:43697", + "right": "issue:46829", "accept": false, - "reason": "Both are config-related, but one is num_labels propagation and the other is strict config loading for granite_speech." + "reason": "RTDetrV2 output changes and VideoPrism hidden_states packaging are separate model-output issues." }, { "left": "issue:45663", "right": "issue:46846", "accept": false, - "reason": "Different parallelism bugs: Gemma4 FSDP2 shared_kv_states KeyError versus Qwen3.5 TP plan omissions causing OOM/shape errors." + "reason": "Gemma4 FSDP2 shared_kv_states KeyError and Qwen3.5 tensor-parallel plan omissions are different training/distribution bugs." } ] }, - "evaluator_result": null, - "evaluator_used": false, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays appropriately conservative. The only accepted soft edge, issue:44479 with issue:45381, is well supported by the titles as the same Qwen2.5-VL 5.3.0 video-input regression. The remaining rejected pairs are phrased cautiously and do not overclaim duplicate status from the available evidence." + }, + "evaluator_used": true, "retried": false, - "accepted_nontrivial_soft_edge": false, + "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null }, @@ -7480,196 +7574,199 @@ "issue:43653|issue:44361", "issue:41762|issue:44936", "issue:43452|issue:44843", - "issue:43688|issue:43749", "issue:43819|issue:44811", "issue:38175|issue:43295", + "issue:43688|issue:43749", "issue:43994|issue:44898", - "issue:43525|issue:46701", "issue:46832|issue:46932", + "issue:43525|issue:46701", "issue:45372|issue:46170" ], - "prepared_review_unit_hash": "e6f8723f35f4436080fe905bcfd00ab826677edb96e3609424f47f50d6a3ad89", + "prepared_review_unit_hash": "0f9cae0ea6da7b42a487965945f9e877910785663b96b8af120231c3748ab1ad", "component_budget": { "node_count": 502, "item_count": 502, - "soft_pair_count": 1842, - "serialized_chars": 956182, - "estimated_input_tokens": 239046, - "estimated_eval_tokens": 478348 + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 25, - "serialized_chars": 35251, - "estimated_input_tokens": 8813, - "estimated_eval_tokens": 17882 + "serialized_chars": 35255, + "estimated_input_tokens": 8814, + "estimated_eval_tokens": 17884 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Most pairs are clearly unrelated despite superficial keyword overlap. The only plausible duplicate edge is the tied-weights serialization regression in v5: issue #44466 appears to be a concrete manifestation of the broader issue described in #45005.", - "confidence": 0.77, - "canonical_issue_reason": "#45005 is the broader, more central report: it describes tied-weight problems on translation models in v5, which appears to subsume the more specific device-dependent lm_head.weight serialization symptom in #44466.", + "summary": "This cluster is highly heterogeneous: the soft pairs mostly share broad wording but not the same concrete bug or code path. I did not find any true duplicate/mergeable pairs. I used issue #45920 as the representative issue because it is the broadest, most actionable regression report in the set.", + "confidence": 0.76, + "canonical_issue_reason": "#45920 is the best representative issue: it describes a concrete tokenizer regression with multiple affected model families and a clear failure mode.", "canonical_pr_reason": null, - "best_issue_reason": "#45005 is the best umbrella issue because it is phrased more generally and matches the shared tied-weights regression theme without being overly device-specific.", + "best_issue_reason": "#45920 is the strongest global issue candidate because it is broad, reproducible, and tied to a specific tokenizer path rather than a model-specific edge case.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:45357", "right": "issue:45920", "accept": false, - "reason": "Both concern model loading/saving behavior, but one is a Qwen3.5 visual-encoder key regression and the other is an AutoTokenizer token-id bug across many models; different code paths." + "reason": "Different bugs: visual-encoder key serialization in save_pretrained vs wrong token IDs from AutoTokenizer." }, { "left": "issue:43540", "right": "issue:46693", "accept": false, - "reason": "Qwen3OmniMoe video input validation and FA2 long-context performance are unrelated problems." + "reason": "Different subsystems and symptoms: video input processing error vs FA2 long-context performance regression." }, { "left": "issue:44913", "right": "issue:46121", "accept": false, - "reason": "Both touch rope/config handling, but one is a GPTNeoX config reload regression and the other is a TypeError in rope param conversion for JSON-loaded lists; different bugs." + "reason": "Both involve config/rope-related plumbing, but one is a reload default bug and the other is a TypeError on JSON-loaded lists." }, { "left": "issue:44737", "right": "issue:45588", "accept": false, - "reason": "XLNet CPU placement and flash_attention s_aux=None are separate code paths with different failures." + "reason": "Unrelated code paths: XLNet CPU device placement vs flash-attention crashing on s_aux=None." }, { "left": "issue:28282", "right": "issue:32090", "accept": false, - "reason": "PyTorch missing import error and Trainer GPU broadcast TypeError are unrelated." + "reason": "Missing PyTorch import vs Trainer GPU broadcast type error; no shared underlying bug." }, { "left": "issue:43697", "right": "issue:45216", "accept": false, - "reason": "RTDetrV2 output divergence and Qwen3.5 save_pretrained checkpoint corruption are different regressions." + "reason": "Different model families and failures: RT-DETR output mismatch vs Qwen3.5 save_pretrained checkpoint corruption." }, { "left": "issue:46463", "right": "issue:46899", "accept": false, - "reason": "Incorrect T5 validation message and Gemma4 4-bit dtype casting are unrelated." + "reason": "A T5 validation error message bug is unrelated to Gemma4 quantized multimodal dtype casting." }, { "left": "issue:43526", "right": "issue:45479", "accept": false, - "reason": "BEiT reduce_labels truncation and sequence-classification zero-loss behavior are distinct bugs." + "reason": "Different logic errors: BEiT label reduction bug vs sequence-classification zero-loss behavior across models." }, { "left": "issue:43720", "right": "issue:44898", "accept": false, - "reason": "BitNet packed-weight loading and Perceiver interpolation failure are different model-specific issues." + "reason": "BitNet packed-weight loading bug is not the same as Perceiver interpolation failure." }, { "left": "issue:43643", "right": "issue:45375", "accept": false, - "reason": "Both involve config fields being lost, but one is trust_remote_code AutoConfig loading and the other is a specific Qwen3_5MoeVisionConfig field dropped by @strict; not the same concrete bug." + "reason": "AutoConfig trust_remote_code missing-fields issue is distinct from Qwen vision config field dropping under @strict." }, { "left": "issue:45200", "right": "issue:46463", "accept": false, - "reason": "Gemma 4 token-type defaults and T5 decoder-input validation message are unrelated." + "reason": "Gemma4 token-type defaults and T5 decoder validation are different bugs." }, { "left": "issue:43761", "right": "issue:46829", "accept": false, - "reason": "Both mention hidden_states, but one returns None when requested and the other misplaces last_hidden_state into hidden_states; different output bugs." + "reason": "Both mention hidden_states, but the failures differ: missing hidden_states vs misassigned last_hidden_state in another model." }, { "left": "issue:44466", "right": "issue:45005", - "accept": true, - "reason": "Both report tied-weight serialization regressions in v5; #44466 looks like a specific device-dependent instance of the broader translation-model tied-weights issue in #45005." + "accept": false, + "reason": "Related theme of tied weights, but not enough evidence of the same concrete serialization bug or mergeable fix." }, { "left": "issue:44361", "right": "issue:44442", "accept": false, - "reason": "Different tokenizer classes and different failure modes." + "reason": "Different tokenizer classes and different failure modes; not the same underlying issue." + }, + { + "left": "issue:44493", + "right": "issue:46050", + "accept": false, + "reason": "Unexpected position-id keys and sinq quantized save/load breakage are unrelated." }, { "left": "issue:43653", "right": "issue:44361", "accept": false, - "reason": "BigBird special-token decode bug and MLukeTokenizer AttributeError are unrelated." + "reason": "BigBird special-token decode bug is unrelated to MLuke task AttributeError." }, { "left": "issue:41762", "right": "issue:44936", "accept": false, - "reason": "DeepSpeed ZeRO-3 loading and trainer.evaluate-after-train failure are unrelated." + "reason": "ZeRO-3 loading index error is not the same as trainer.evaluate() failing after training." }, { "left": "issue:43452", "right": "issue:44843", "accept": false, - "reason": "Both involve AutoTokenizer loading, but one is gguf_file support breaking from_pretrained and the other is an unconditional model_info() call breaking offline mode; distinct root causes." - }, - { - "left": "issue:43688", - "right": "issue:43749", - "accept": false, - "reason": "Aux-loss normalization and FSDP CPU RAM efficient loading are unrelated." + "reason": "Both touch AutoTokenizer/from_pretrained, but one is a gguf_file breakage and the other is an offline-mode regex patch bug." }, { "left": "issue:43819", "right": "issue:44811", "accept": false, - "reason": "DAC latent decoding mismatch and Whisper batch_decode skip_special_tokens handling are different issues." + "reason": "DAC latent reconstruction mismatch is unrelated to Whisper batch_decode skip_special_tokens handling." }, { "left": "issue:38175", "right": "issue:43295", "accept": false, - "reason": "Siglip2 zero probabilities and processor.tokenizer regression are not the same underlying bug." + "reason": "Zero-probability outputs for siglip2 are not the same as processor.tokenizer/image-input regression." }, { - "left": "issue:43994", - "right": "issue:44898", + "left": "issue:43688", + "right": "issue:43749", "accept": false, - "reason": "Both concern model behavior, but Siglip2 nonsense outputs and Perceiver non-default-resolution interpolation failure are unrelated." + "reason": "Auxiliary-loss normalization and FSDP CPU RAM efficient loading are different problems." }, { - "left": "issue:43525", - "right": "issue:46701", + "left": "issue:43994", + "right": "issue:44898", "accept": false, - "reason": "Llama4Config pad_token_id missing and DiffusionGemma image-text-to-text pipeline failure are unrelated." + "reason": "Different models and symptoms: nonsensical SigLIP2 outputs vs Perceiver interpolation failure." }, { "left": "issue:46832", "right": "issue:46932", "accept": false, - "reason": "RT-DETR feature-level mismatch and Gemma4 quantized embedder dtype casting are different problems." + "reason": "RT-DETR feature-level mismatch is unrelated to Gemma4 quantized embedder dtype casting." + }, + { + "left": "issue:43525", + "right": "issue:46701", + "accept": false, + "reason": "Missing pad_token_id in Llama4Config is a different bug from the DiffusionGemma pipeline failure." }, { "left": "issue:45372", "right": "issue:46170", "accept": false, - "reason": "Gemma4 processor import failure and Windows ROCm torch.distributed dependency issues are unrelated." + "reason": "Gemma4 processor import failure and Windows ROCm torch.distributed dependency issues are distinct." } ] }, - "evaluator_result": { - "accept": true, - "feedback": "The analysis is grounded in the packet and stays conservative. The only accepted soft edge, issue:44466 \u2194 issue:45005, is plausibly a specific tied-weights serialization regression under the broader umbrella issue, and the other pairs are rejected with appropriately cautious reasoning." - }, - "evaluator_used": true, + "evaluator_result": null, + "evaluator_used": false, "retried": false, - "accepted_nontrivial_soft_edge": true, + "accepted_nontrivial_soft_edge": false, "error_kind": null, "error_message": null }, @@ -7739,14 +7836,14 @@ "issue:45542|issue:46178", "issue:43295|issue:45356", "issue:43531|issue:43742", - "issue:42898|issue:46491", "issue:29159|issue:33357", + "issue:42898|issue:46491", "issue:43653|issue:44568", "issue:43646|issue:43749", "issue:43756|issue:46402", "issue:44843|issue:44991", - "issue:43688|issue:45704", "issue:43653|issue:43927", + "issue:43688|issue:45704", "issue:43452|issue:43577", "issue:44464|issue:44743", "issue:44484|issue:44704", @@ -7765,202 +7862,202 @@ "issue:44464|issue:44977", "issue:41720|issue:44464" ], - "prepared_review_unit_hash": "6dd6c2e4adab68b719cb22a3f96415ca736b06c6e30b50dc9eeea6a6ec5d5715", + "prepared_review_unit_hash": "3ecc479a82f19e883147fa8954d856c80483d265a13ce1645ee385beb948c008", "component_budget": { "node_count": 502, "item_count": 502, - "soft_pair_count": 1842, - "serialized_chars": 956182, - "estimated_input_tokens": 239046, - "estimated_eval_tokens": 478348 + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 28, - "serialized_chars": 36202, - "estimated_input_tokens": 9051, - "estimated_eval_tokens": 18358 + "serialized_chars": 36209, + "estimated_input_tokens": 9053, + "estimated_eval_tokens": 18362 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a false-positive similarity bucket: a few tokenizer-regression issues are loosely related, but most paired items are clearly different bugs across unrelated subsystems (generation, loading, training, model-specific regressions).", - "confidence": 0.78, - "canonical_issue_reason": "issue:45701 is the broadest tokenizer-regression report in the set and best fits the shared theme of version-dependent tokenization behavior, though it still does not truly unify the heterogeneous cluster.", + "summary": "This cluster is mostly a loose theme of tokenizer/model-loading regressions in Transformers, but the soft pairs are not the same underlying bug or change. None of the candidate pairs are strong enough to merge as duplicates.", + "confidence": 0.89, + "canonical_issue_reason": "issue:45701 is the best umbrella issue in this set: it is broad, version-regression oriented, and sits closest to the repeated tokenizer-behavior complaints without being overly model-specific.", "canonical_pr_reason": null, - "best_issue_reason": "issue:45701 is the most representative umbrella issue for the tokenizer subset because it is broad and regression-oriented, but the cluster as a whole is too diverse for a strong canonical issue.", + "best_issue_reason": "issue:45701 is the most suitable representative issue because it describes a general version-dependent tokenization regression and can serve as the broadest anchor for related tokenizer reports.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:45542", "right": "issue:46178", "accept": false, - "reason": "Both mention TensorFlow-related backend handling, but one is about tensorboard-only installs and the other is an undefined tensorflow_text backend mapping; different failure modes." + "reason": "Both are backend/configuration import errors, but they involve different missing-backend paths: TensorFlow/tensorboard vs tensorflow_text." }, { "left": "issue:43295", "right": "issue:45356", "accept": false, - "reason": "Different regressions: processor/tokenizer image-handling breakage vs Kimi-K2.5 codec handling and warning behavior." + "reason": "Different regressions: processor.tokenizer/image-handling breakage versus Kimi-K2.5 codec handling and mistral-regex warning behavior." }, { "left": "issue:43531", "right": "issue:43742", "accept": false, - "reason": "Qwen3-MoE sliding-window behavior and MobileLLM load key errors are unrelated code paths and symptoms." + "reason": "Unrelated failures in different models: Qwen3-MoE sliding-window behavior versus a KeyError loading MobileLLM-125M." }, { - "left": "issue:42898", - "right": "issue:46491", + "left": "issue:29159", + "right": "issue:33357", "accept": false, - "reason": "Both concern tokenization output, but one is v5 cleanup behavior and the other is a CodeLlama leading-space round-trip regression; not the same bug." + "reason": "Both mention tokenizers, but one is slow/fast inconsistency and the other is a MacOS bus error with a CLIP model." }, { - "left": "issue:29159", - "right": "issue:33357", + "left": "issue:42898", + "right": "issue:46491", "accept": false, - "reason": "Slow/fast tokenizer inconsistency is unrelated to a MacOS bus error loading a CLIP model." + "reason": "Both are tokenization regressions, but one is a general v5 behavior change and the other is a CodeLlama leading-space round-trip bug." }, { "left": "issue:43653", "right": "issue:44568", "accept": false, - "reason": "Both involve tokenizer special-token behavior, but they affect different models and different concrete failures (BigBird mask token registration vs mdeberta BOS/EOS insertion)." + "reason": "Special-token handling is the broad theme, but these are different tokenizers and different symptoms (mask decode output vs BOS/EOS insertion)." }, { "left": "issue:43646", "right": "issue:43749", "accept": false, - "reason": "Custom model initialization breakage and FSDP CPU RAM efficient loading are distinct issues in different loading flows." + "reason": "Both concern loading/runtime breakage, but one is custom model initialization in v5 and the other is FSDP CPU RAM-efficient loading." }, { "left": "issue:43756", "right": "issue:46402", "accept": false, - "reason": "Smollm3 RoPE-layer selection and Sam3 text-encoder weight loading are unrelated model-specific loading/config bugs." + "reason": "Different model-loading problems: Smollm3 RoPE layer loading versus Sam3 text-encoder weights not being loaded." }, { "left": "issue:44843", "right": "issue:44991", "accept": false, - "reason": "One is an offline-mode regex/model_info bug in AutoTokenizer, the other is a tokenizer load failure for EMBEDDIA/est-roberta; different root causes." + "reason": "Both are tokenizer loading issues, but one is offline-model_info patch behavior and the other is a specific tokenizer failure for EMBEDDIA/est-roberta." }, { - "left": "issue:43688", - "right": "issue:45704", + "left": "issue:43653", + "right": "issue:43927", "accept": false, - "reason": "Auxiliary-loss normalization and a T5 RMSNorm memory leak are unrelated despite both touching model internals." + "reason": "Different bugs: BigBird special-token registration versus DiaConfig losing custom token IDs after save/load." }, { - "left": "issue:43653", - "right": "issue:43927", + "left": "issue:43688", + "right": "issue:45704", "accept": false, - "reason": "Special-token registration in BigBird and loss of custom token IDs in DiaConfig are both token-related but not the same defect." + "reason": "These affect entirely different code paths: auxiliary-loss normalization versus a T5 RMSNorm memory leak." }, { "left": "issue:43452", "right": "issue:43577", "accept": false, - "reason": "GGUF load-path failures and Blip2 dtype propagation are unrelated loading problems." + "reason": "Loading gguf files and BLIP2 dtype propagation are unrelated issues." }, { "left": "issue:44464", "right": "issue:44743", "accept": false, - "reason": "Both involve generation/cache behavior, but one is compiled chunked generation inconsistency and the other is recurrent-state reset in Qwen3.5; different code paths." + "reason": "Both touch generation/cache, but one is chunked generation with compiled forward and the other is recurrent-state reset in Qwen3.5 cache handling." }, { "left": "issue:44484", "right": "issue:44704", "accept": false, - "reason": "A save_pretrained shard-size question is unrelated to missing kwargs being forwarded to cached_file." + "reason": "Unrelated APIs: save_pretrained shard sizing versus AutoProcessor kwargs not being passed to cached_file." }, { "left": "issue:43688", "right": "issue:45701", "accept": false, - "reason": "Auxiliary-loss normalization is unrelated to a broad tokenization-version regression." + "reason": "Aaux-loss normalization is unrelated to the broad tokenization regression described by issue:45701." }, { "left": "issue:45491", "right": "issue:46002", "accept": false, - "reason": "Gemma3 NaNs from sliding-window attention are not the same as NaNs in classification heads during checkout/init." + "reason": "Both mention NaNs, but they arise in different subsystems: Gemma3 sliding-window embeddings versus classification heads during checkout/init." }, { "left": "issue:44479", "right": "issue:44625", "accept": false, - "reason": "Video-input regression in Qwen VL models and num_labels propagation in Qwen3.5 are different bugs." + "reason": "One is a video-input regression across Qwen VL variants; the other is num_labels propagation in Qwen3.5 config." }, { "left": "issue:44821", "right": "issue:46821", "accept": false, - "reason": "AutoImageProcessor URL loading and TimesFM window_size AttributeError are unrelated." + "reason": "Different failures in different components: AutoImageProcessor URL loading versus TimesFM window_size AttributeError." }, { "left": "issue:43688", "right": "issue:45440", "accept": false, - "reason": "Auxiliary-loss normalization and DeepseekV3MoE divergence are different model-behavior issues." + "reason": "Unrelated model-specific issues: auxiliary-loss normalization versus DeepseekV3MoE divergence from remote implementation." }, { "left": "issue:33666", "right": "issue:36010", "accept": false, - "reason": "Qwen2-VL multi-GPU training and a GenerationMixin import error are unrelated." + "reason": "Multi-GPU training and a GenerationMixin import error are not the same bug." }, { "left": "issue:32090", "right": "issue:33666", "accept": false, - "reason": "Trainer GPU broadcast TypeError and Qwen2-VL multi-GPU training are distinct distributed-training problems." + "reason": "Both affect training, but the NoneType GPU broadcast error is distinct from Qwen2-VL multi-GPU training behavior." }, { "left": "issue:44466", "right": "issue:46050", "accept": false, - "reason": "Both concern save/load behavior, but one is device-dependent lm_head serialization and the other is quantized sinq reload failure; not the same bug." + "reason": "Different serialization/load issues: tied-weight serialization depending on device versus quantized sinq save/load breakage." }, { "left": "issue:44734", "right": "issue:46612", "accept": false, - "reason": "Serve KV-cache indexing and beam-search cache reorder omission are different generation-cache failures." + "reason": "Both are cache/generation related, but server-side KV continuation indexing is not the same as beam-search cache reorder being skipped." }, { "left": "issue:43504", "right": "issue:43720", "accept": false, - "reason": "Legacy-field model loading for Beit and packed-weight unpacking in BitNet accelerate loading are separate load-path bugs." + "reason": "Legacy-field loading for Beit is unrelated to BitNet packed-weight unpacking during accelerate loading." }, { "left": "issue:44977", "right": "issue:46693", "accept": false, - "reason": "Qwen3.5 flash-attention generation failure and FA2 long-context performance regression are related only at a very broad level; the concrete bugs differ." + "reason": "One is a correctness regression for Qwen3.5 flash-attention generation; the other is a performance regression with FA2 and long-context generation." }, { "left": "issue:43504", "right": "issue:43550", "accept": false, - "reason": "Beit loading failure and Bamba torch.compile SDPA failure are unrelated model-specific issues." + "reason": "Both are model-loading/runtime problems, but one is a pretrained preset legacy-field issue and the other is a torch.compile + SDPA failure." }, { "left": "issue:44464", "right": "issue:44977", "accept": false, - "reason": "Chunked generation inconsistency under compiled forward is not the same as Qwen3.5 flash-attention generation breakage." + "reason": "Chunked compiled-forward generation inconsistency and Qwen3.5 flash-attention generation failure are different code paths and symptoms." }, { "left": "issue:41720", "right": "issue:44464", "accept": false, - "reason": "cudaErrorAssert on auto device mapping and compiled-forward chunked generation inconsistency are different failure classes." + "reason": "Auto device mapping cudaErrorAssert on A800 is unrelated to compiled-forward chunked generation output inconsistency." } ] }, @@ -8039,9 +8136,9 @@ "issue:43611|issue:46846", "issue:43720|issue:43927", "issue:43504|issue:44610", + "issue:43606|issue:43927", "issue:33666|issue:36331", "issue:43278|issue:43688", - "issue:43606|issue:43927", "issue:43782|issue:46693", "issue:43262|issue:43295", "issue:43720|issue:44361", @@ -8049,8 +8146,8 @@ "issue:44964|issue:45405", "issue:43493|issue:43825", "issue:44554|issue:46365", - "issue:43525|issue:43572", "issue:44977|issue:46209", + "issue:43525|issue:43572", "issue:44625|issue:45584", "issue:43452|issue:44451", "issue:28218|issue:33357", @@ -8060,19 +8157,19 @@ "issue:43493|issue:43844", "issue:39692|issue:43295", "issue:43278|issue:43844", - "issue:46209|issue:46710", "issue:43646|issue:44466", + "issue:46209|issue:46710", "issue:44849|issue:46209", "issue:33357|issue:33666" ], - "prepared_review_unit_hash": "074287ec7111a4f889b241c2fc7a258d7d41f8bc967501c8d5993d032aa7014f", + "prepared_review_unit_hash": "e5fbcef063db724c09eec8711707e885c4de79388e1c950c18114b1e3cf79ab6", "component_budget": { "node_count": 502, "item_count": 502, - "soft_pair_count": 1842, - "serialized_chars": 956182, - "estimated_input_tokens": 239046, - "estimated_eval_tokens": 478348 + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 }, "budget": { "node_count": 48, @@ -8087,8 +8184,8 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster is heterogeneous: tokenizer quirks, model-loading regressions, training bugs, docs examples, and unrelated architecture-specific failures. The soft-similarity edges do not indicate true duplicates, so there is no single canonical issue or PR for this cluster.", - "confidence": 0.98, + "summary": "This cluster is a loose collection of unrelated Transformers bug reports, mostly across different models and regressions. The suggested soft edges share only broad subsystem similarity, not the same concrete bug, so none should be merged as duplicates.", + "confidence": 0.96, "canonical_issue_reason": null, "canonical_pr_reason": null, "best_issue_reason": null, @@ -8098,181 +8195,181 @@ "left": "issue:43643", "right": "issue:44561", "accept": false, - "reason": "Both mention trust_remote_code, but one is missing config fields from AutoConfig and the other is a v5 break from removing is_torch_fx_available; different failure modes and code paths." + "reason": "Both mention trust_remote_code, but one is missing config fields while the other is a v5 regression from removing is_torch_fx_available; different root causes." }, { "left": "issue:43577", "right": "issue:44554", "accept": false, - "reason": "Different problems: one is dtype handling for BLIP2 loading, the other is an MPS attention correctness issue when value/query head dims differ." + "reason": "BLIP2 dtype propagation and an MPS attention correctness issue are unrelated bugs." }, { "left": "issue:43611", "right": "issue:46846", "accept": false, - "reason": "Unrelated: base_model_prefix loading regression versus Qwen3.5 tensor-parallel plan missing linear_attn entries causing OOM/shape errors." + "reason": "One is a base_model_prefix loading regression; the other is a Qwen3.5 TP-plan OOM/shape issue." }, { "left": "issue:43720", "right": "issue:43927", "accept": false, - "reason": "Both are loading-related, but one is about unpacking packed BitNet weights during accelerate loading and the other is DiaConfig losing custom token IDs after save/load." + "reason": "Packed-weight loading failure and DiaConfig token-ID loss are different failure modes in different code paths." }, { "left": "issue:43504", "right": "issue:44610", "accept": false, - "reason": "Both involve model/processor mismatches, but they affect different models and different inputs (legacy field on BEiT vs processor image size mismatch on OmDet-Turbo)." + "reason": "Legacy preset loading and processor/model image-size mismatch are unrelated." + }, + { + "left": "issue:43606", + "right": "issue:43927", + "accept": false, + "reason": "CPU offload device mismatch is not the same bug as DiaConfig save/load token corruption." }, { "left": "issue:33666", "right": "issue:36331", "accept": false, - "reason": "Multigpu training for Qwen2-VL and a Trainer compute_loss signature break are unrelated issues." + "reason": "Multi-GPU training and a custom compute_loss signature break are unrelated." }, { "left": "issue:43278", "right": "issue:43688", "accept": false, - "reason": "Different bugs: embedding dtype drift in eval versus incorrect auxiliary-loss normalization in OLMoE/GPT OSS." - }, - { - "left": "issue:43606", - "right": "issue:43927", - "accept": false, - "reason": "Device-mismatch on CPU offload for bark-small is unrelated to DiaConfig token-ID persistence." + "reason": "Embedding dtype drift and auxiliary-loss normalization are different bugs." }, { "left": "issue:43782", "right": "issue:46693", "accept": false, - "reason": "Qwen3VL weight_only loading error and FA2 long-context performance regression are different code paths and symptoms." + "reason": "Qwen3VL weight_only loading and FA2 long-context performance regression are unrelated." }, { "left": "issue:43262", "right": "issue:43295", "accept": false, - "reason": "Both mention processor/chat-template behavior, but one is an audio sampling-rate default and the other is a regression around processor.tokenizer/images in custom model code." + "reason": "Audio apply_chat_template sampling-rate defaulting and processor.tokenizer/image handling regression are different issues." }, { "left": "issue:43720", "right": "issue:44361", "accept": false, - "reason": "BitNet packed-weight loading and MLukeTokenizer task AttributeError are unrelated." + "reason": "BitNet packed-weight loading and MLukeTokenizer attribute failure are unrelated." }, { "left": "issue:44514", "right": "issue:45290", "accept": false, - "reason": "Both are apply_chat_template crashes, but one is batched padding=False handling and the other is assistant tool-call messages with no content; not the same concrete bug." + "reason": "Both crash in apply_chat_template, but on different inputs and different edge cases." }, { "left": "issue:44964", "right": "issue:45405", "accept": false, - "reason": "Different issues: loading a specific multimodal model versus a MIN_PEFT_VERSION bump to an unreleased PyPI version." + "reason": "Phi-4 multimodal loading and an unreleased PEFT version bump are unrelated." }, { "left": "issue:43493", "right": "issue:43825", "accept": false, - "reason": "SigLIP2 implementation discrepancy is unrelated to a pipeline error message incorrectly advertising translation support in v5." + "reason": "A SigLIP2 implementation discrepancy is not the same as a pipeline error-message bug." }, { "left": "issue:44554", "right": "issue:46365", "accept": false, - "reason": "MPS attention correctness and importing Trainer failing until datasets is uninstalled are unrelated." + "reason": "An attention correctness issue and a Trainer import conflict are unrelated." }, { - "left": "issue:43525", - "right": "issue:43572", + "left": "issue:44977", + "right": "issue:46209", "accept": false, - "reason": "Both are missing token-id related config fields, but for different model families and different attributes (pad_token_id vs pad_token_idx)." + "reason": "Qwen3.5 flash-attention generation and FP8/B200 inference failures are different bugs." }, { - "left": "issue:44977", - "right": "issue:46209", + "left": "issue:43525", + "right": "issue:43572", "accept": false, - "reason": "Qwen3.5 flash-attention generation bug and FP8 inference failure on B200 are different hardware/kernel issues." + "reason": "Missing pad_token_id and missing pad_token_idx are similar in theme but on different configs and attributes." }, { "left": "issue:44625", "right": "issue:45584", "accept": false, - "reason": "Qwen3.5 num_labels propagation and Whisper empty-transcription failure after align_special_tokens are unrelated." + "reason": "Qwen3.5 num_labels propagation and Whisper empty-transcription failure are unrelated." }, { "left": "issue:43452", "right": "issue:44451", "accept": false, - "reason": "gguf_file loading breakage and inability to load vesteinn/ScandiBERT are different model-loading problems." + "reason": "gguf_file loading regression and a specific model load failure are not the same concrete bug." }, { "left": "issue:28218", "right": "issue:33357", "accept": false, - "reason": "Tokenizer extra-space behavior and a MacOS bus error on a CLIP model are not the same bug." + "reason": "Tokenizer spacing after added tokens is unrelated to the MacOS CLIP bus error." }, { "left": "issue:45823", "right": "issue:46050", "accept": false, - "reason": "Both involve model loading/saving, but one is a device mismatch with device_map='auto' and the other is quantized-model persistence after save/load." + "reason": "A device_map auto forward mismatch and a sinq save/load quantization break are different issues." }, { "left": "issue:42796", "right": "issue:44977", "accept": false, - "reason": "Both are tokenizer/model issues around Qwen/Ministral-family models, but one is BPE-marker handling and the other is flash-attention generation behavior." + "reason": "Ministral BPE marker handling and Qwen3.5 flash-attention generation are unrelated." }, { "left": "issue:43994", "right": "issue:46192", "accept": false, - "reason": "SigLIP2 nonsensical outputs and return_tensors being ignored when text_kwargs is passed are different API bugs." + "reason": "SigLIP2 model/pipeline correctness and return_tensors handling are different bugs." }, { "left": "issue:43493", "right": "issue:43844", "accept": false, - "reason": "Model implementation discrepancy versus a gradient explosion under ZeRO-3 are unrelated training/runtime problems." + "reason": "SigLIP2 implementation mismatch and a ZeRO-3 gradient regression are unrelated." }, { "left": "issue:39692", "right": "issue:43295", "accept": false, - "reason": "SigLIP2 docs example errors are documentation/example issues, while the other is a processor.tokenizer regression affecting custom model code." + "reason": "Doc-example mismatches/quantization errors are not the same as processor.tokenizer and image-tokenizer regressions." }, { "left": "issue:43278", "right": "issue:43844", "accept": false, - "reason": "Different failure modes: dtype mismatch in eval versus abnormal gradient growth during DeepSpeed training." + "reason": "Embedding dtype changes and abnormal gradient growth under ZeRO-3 are different problems." }, { - "left": "issue:46209", - "right": "issue:46710", + "left": "issue:43646", + "right": "issue:44466", "accept": false, - "reason": "FP8 inference on B200 and accuracy regression after upgrading transformers are separate regressions with different symptoms." + "reason": "Custom model initialization breakage and lm_head serialization inconsistency are not the same bug." }, { - "left": "issue:43646", - "right": "issue:44466", + "left": "issue:46209", + "right": "issue:46710", "accept": false, - "reason": "Custom model initialization breakage and inconsistent lm_head.weight serialization are distinct v5 regressions." + "reason": "FP8 inference on B200 and a DeepSeek accuracy regression after upgrade are unrelated." }, { "left": "issue:44849", "right": "issue:46209", "accept": false, - "reason": "Qwen3.5 output_hidden_states bug and FP8/B200 inference failure are unrelated." + "reason": "Qwen3.5 hidden-states bug and FP8 inference failure are different issues." }, { "left": "issue:33357", "right": "issue:33666", "accept": false, - "reason": "CLIP bus error on MacOS and Qwen2-VL multi-GPU training are unrelated." + "reason": "A MacOS bus error in CLIP loading is unrelated to Qwen2-VL multi-GPU training." } ] }, @@ -8349,24 +8446,24 @@ "issue:44568|issue:46489", "issue:43278|issue:43701", "issue:44488|issue:44843", - "issue:44568|issue:45812", - "issue:44792|issue:45405", "issue:43927|issue:44361", + "issue:44792|issue:45405", + "issue:44568|issue:45812", "issue:43824|issue:44589", "issue:43575|issue:45405", "issue:44849|issue:45357", "issue:43504|issue:43606", "issue:43257|issue:43299", "issue:44589|issue:45520", + "issue:43606|issue:44112", "issue:46399|issue:46846", "issue:43334|issue:43572", "issue:45245|issue:46710", "issue:44493|issue:46285", - "issue:43606|issue:44112", "issue:46821|issue:46832", "issue:43334|issue:45759", - "issue:43572|issue:44291", "issue:43504|issue:43653", + "issue:43572|issue:44291", "issue:45072|issue:46832", "issue:44464|issue:46424", "issue:43937|issue:46701", @@ -8375,20 +8472,20 @@ "issue:45200|issue:45229", "issue:29942|issue:30333" ], - "prepared_review_unit_hash": "77849d6ad3ffb50ac3c7f33970e1634851f46f4381bbed6914077e2f7255d74a", + "prepared_review_unit_hash": "f687c56af1acc1542a575618dc67b24542b573422b5d7e8568ca54559d898bab", "component_budget": { "node_count": 502, "item_count": 502, - "soft_pair_count": 1842, - "serialized_chars": 956182, - "estimated_input_tokens": 239046, - "estimated_eval_tokens": 478348 + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 28, - "serialized_chars": 36204, + "serialized_chars": 36202, "estimated_input_tokens": 9051, "estimated_eval_tokens": 18358 }, @@ -8397,180 +8494,180 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "These are mostly unrelated issue pairs that only share broad subsystem keywords (tokenizers, Qwen, offload, loading, CI). None look like the same underlying bug with a shared fix path, so I would reject all soft edges.", - "confidence": 0.84, - "canonical_issue_reason": "Issue 45812 is the strongest canonical issue candidate: it describes a broad, user-visible tokenizer regression across all Granite models, has multiple inbound references, and is phrased as a concrete, reproducible bug rather than a narrow model-specific symptom.", + "summary": "The cluster is highly heterogeneous: tokenizer regressions, model-loading failures, training/offload issues, generation bugs, and unrelated CI/test failures. No pair looks like the same concrete bug/change, so all soft edges should be rejected.", + "confidence": 0.97, + "canonical_issue_reason": "No single canonical issue fits this cluster because the items cover unrelated failures across many subsystems and models.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 45812 is the best standalone issue in this set because it is clearly scoped, high impact, and generalizes beyond one model instance; it is the most representative duplicate target among the candidates.", + "best_issue_reason": "No global best issue: none of the issues is representative of the others, and several are clearly unrelated.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:44568", "right": "issue:46489", "accept": false, - "reason": "Both concern tokenization regressions, but they affect different model families and failure modes: mdeberta special-token behavior vs DeepSeek-Coder v1 token IDs. Too broad to treat as the same bug." + "reason": "Both are tokenizer regressions, but they affect different models and symptoms; not the same underlying bug." }, { "left": "issue:43278", "right": "issue:43701", "accept": false, - "reason": "Embedding dtype drift during evaluate is unrelated to a resume_from_checkpoint key mismatch. Different code paths, symptoms, and fixes." + "reason": "One is a dtype drift in evaluate, the other a checkpoint key mismatch; unrelated failures." }, { "left": "issue:44488", "right": "issue:44843", "accept": false, - "reason": "One is a model-specific load failure for cjvt/sleng-bert, the other is an offline-mode bug caused by an unconditional hub call in tokenizer patching. Not the same issue." + "reason": "Different subsystems: model load failure for a specific tokenizer/model vs offline-mode hub access bug." }, { - "left": "issue:44568", - "right": "issue:45812", + "left": "issue:43927", + "right": "issue:44361", "accept": false, - "reason": "Both are tokenizer regressions, but one is specific to microsoft/mdeberta-v3-base add_special_tokens handling while the other affects Granite token IDs broadly. Different underlying defects." + "reason": "Both involve tokenizers/configs, but the affected models and failure modes are different." }, { "left": "issue:44792", "right": "issue:45405", "accept": false, - "reason": "Janus image generation test failure and an unreleased MIN_PEFT_VERSION bump are unrelated." + "reason": "Image generation test failure vs dependency version bump; no shared bug." }, { - "left": "issue:43927", - "right": "issue:44361", + "left": "issue:44568", + "right": "issue:45812", "accept": false, - "reason": "DiaConfig token ID persistence and MLukeTokenizer task AttributeError involve different config/tokenizer behavior and different failure surfaces." + "reason": "Both are tokenizer-related regressions, but they are distinct model-specific issues with different outputs." }, { "left": "issue:43824", "right": "issue:44589", "accept": false, - "reason": "A missing import for Qwen2.5-VL is unrelated to a Float8 storage lookup TypeError. No shared bug path." + "reason": "Completely unrelated: import error for a Qwen class vs float8 storage lookup failure." }, { "left": "issue:43575", "right": "issue:45405", "accept": false, - "reason": "Tensor-parallel OOM for Qwen2-57B and a dependency version bump are not the same issue." + "reason": "OOM during tensor-parallel model load is unrelated to the PEFT minimum-version bump." }, { "left": "issue:44849", "right": "issue:45357", "accept": false, - "reason": "Both involve Qwen3.5, but one is output_hidden_states behavior and the other is save_pretrained saving wrong visual encoder keys. Different bug classes." + "reason": "Different bugs in different Qwen models: hidden-states output handling vs save_pretrained key serialization." }, { "left": "issue:43504", "right": "issue:43606", "accept": false, - "reason": "Beit legacy-field loading and bark-small CPU offload device mismatch are unrelated." + "reason": "Legacy-field load bug for BEiT is unrelated to CPU offload device mismatch in bark-small." }, { "left": "issue:43257", "right": "issue:43299", "accept": false, - "reason": "Both mention Qwen3 MOE loading, but one is accelerate+deepspeed weight conversion and the other is a transformers dev regression loading Qwen3VL MoE models. Not enough to merge." + "reason": "Both mention Qwen MOE loading, but one is accelerate/deepspeed conversion and the other is a general transformers-dev load break; not clearly the same fix." }, { "left": "issue:44589", "right": "issue:45520", "accept": false, - "reason": "Float8 storage lookup and flash_attn import key errors are unrelated runtime/import issues." + "reason": "Distinct low-level import/storage problems with no shared code path." + }, + { + "left": "issue:43606", + "right": "issue:44112", + "accept": false, + "reason": "One is a real device mismatch bug, the other is a stale CI test; not the same underlying issue." }, { "left": "issue:46399", "right": "issue:46846", "accept": false, - "reason": "Both are Qwen3.5-related, but one is a missing decorator on apply_rotary_pos_emb registration and the other is an incomplete TP plan causing OOM/shape errors. Different concrete fixes." + "reason": "Both are Qwen3.5-related, but one is a kernel registration error and the other is a TP plan omission causing OOM/shape issues." }, { "left": "issue:43334", "right": "issue:43572", "accept": false, - "reason": "Both are missing config fields after v5 updates, but they are different models and different fields (pad_token_id vs pad_token_idx). Not the same bug." + "reason": "Both are post-v5 config-field problems, but they concern different models and different missing fields." }, { "left": "issue:45245", "right": "issue:46710", "accept": false, - "reason": "A category-count runtime limit is unrelated to an accuracy regression after upgrading Transformers." + "reason": "Category-count limit error is unrelated to an accuracy regression after upgrade." }, { "left": "issue:44493", "right": "issue:46285", "accept": false, - "reason": "Unexpected position-id keys across many models and CLIP text-model structure divergence are different regressions in different components." - }, - { - "left": "issue:43606", - "right": "issue:44112", - "accept": false, - "reason": "Both are CI/device-related, but one is a real CPU-offload mismatch in bark-small while the other is a stale GraniteSpeech test. Not the same underlying defect." + "reason": "Both are structure/keys regressions, but they affect different model families and different attributes." }, { "left": "issue:46821", "right": "issue:46832", "accept": false, - "reason": "TimesFM window_size attribute handling and RT-DETR feature-level mismatch are unrelated model bugs." + "reason": "Unrelated model errors: TimesFM window_size attribute issue vs RT-DETR forward TypeError." }, { "left": "issue:43334", "right": "issue:45759", "accept": false, - "reason": "Qwen3-VL pad_token_id missing and composite Qwen 3.5/3.6 text_config unwrapping are both multimodal config issues, but they are different failure modes and code paths." + "reason": "Both touch multimodal Qwen configs, but one is a missing pad_token_id and the other is text_config unwrapping; not the same bug." }, { - "left": "issue:43572", - "right": "issue:44291", + "left": "issue:43504", + "right": "issue:43653", "accept": false, - "reason": "Missing pad_token_idx in StableLmConfig is unrelated to init_empty_weights passing an unexpected _is_hf_initialized argument." + "reason": "Different components and symptoms: BEiT pretrained loading vs BigBirdTokenizer special-token registration." }, { - "left": "issue:43504", - "right": "issue:43653", + "left": "issue:43572", + "right": "issue:44291", "accept": false, - "reason": "Beit legacy-field loading and BigBirdTokenizer special-token registration are unrelated." + "reason": "Missing pad_token_idx and unexpected _is_hf_initialized arg are separate config/init compatibility issues." }, { "left": "issue:45072", "right": "issue:46832", "accept": false, - "reason": "Dtype mismatches in bfloat16 inference and RTDetr forward type errors do not share a code path or fix." + "reason": "Different inference bugs in different models; no shared code path." }, { "left": "issue:44464", "right": "issue:46424", "accept": false, - "reason": "Chunked generation inconsistencies with compiled forward and cache max length behavior are different generation/cache bugs." + "reason": "Both are generation/cache-related, but one is chunked generation with compiled forward and the other is cache_length handling for quantized cache only." }, { "left": "issue:43937", "right": "issue:46701", "accept": false, - "reason": "Invalid GenerationConfig for GLM-5 and a broken image-text-to-text pipeline for DiffusionGemma are unrelated." + "reason": "Different generation/pipeline failures with no evidence of a common cause." }, { "left": "issue:43844", "right": "issue:44964", "accept": false, - "reason": "ZeRO-3 gradient growth and Phi-4 multimodal loading failures are unrelated model/training issues." + "reason": "ZeRO-3 gradient behavior and multimodal model loading are unrelated." }, { "left": "issue:43873", "right": "issue:46050", "accept": false, - "reason": "Quantization offloading behavior and sinq save/load breakage both mention quantization, but they are distinct bugs with different lifecycle stages." + "reason": "Quantization/offloading themes overlap, but save/load breakage in sinq is not the same bug as offloading not working." }, { "left": "issue:45200", "right": "issue:45229", "accept": false, - "reason": "mm_token_type_ids defaults for text-only fine-tuning and a multi-GPU CUDA OOM are unrelated." + "reason": "Text-only fine-tuning defaults vs multi-GPU inference OOM are unrelated." }, { "left": "issue:29942", "right": "issue:30333", "accept": false, - "reason": "Flash Attention test failures and MLFlow reporting on job failure are unrelated CI issues." + "reason": "Flash Attention test failures and MLflow job-reporting issues are unrelated." } ] }, @@ -8644,8 +8741,8 @@ ], "soft_pairs": [ "issue:44869|issue:46752", - "issue:45229|issue:45823", "issue:43994|issue:44493", + "issue:45229|issue:45823", "issue:45588|issue:46050", "issue:42490|issue:43696", "issue:29159|issue:32090", @@ -8654,8 +8751,8 @@ "issue:28218|issue:34567", "issue:43065|issue:46369", "issue:44554|issue:45446", - "issue:44423|issue:44734", "issue:15354|issue:38175", + "issue:44423|issue:44734", "issue:43493|issue:43701", "issue:41720|issue:46710", "issue:45406|issue:46121", @@ -8671,196 +8768,196 @@ "issue:43295|issue:44610", "issue:45823|issue:46192" ], - "prepared_review_unit_hash": "39ccf932931c61c3b5890c114a7103b7952daab8a666e56859ba8a3f111b7f0e", + "prepared_review_unit_hash": "7d300313cced9717d3ef40dfbce810226e0ca378179ab48aa0bb545f01be7985", "component_budget": { "node_count": 502, "item_count": 502, - "soft_pair_count": 1842, - "serialized_chars": 956182, - "estimated_input_tokens": 239046, - "estimated_eval_tokens": 478348 + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 27, - "serialized_chars": 35264, - "estimated_input_tokens": 8816, - "estimated_eval_tokens": 17888 + "serialized_chars": 35267, + "estimated_input_tokens": 8817, + "estimated_eval_tokens": 17890 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This set is mostly a heterogeneous mix of unrelated transformer bugs. The closest recurring theme is processor/tokenizer API regressions, but none of the soft pairs look like the same underlying fix.", - "confidence": 0.41, - "canonical_issue_reason": "issue:43295 is the broadest representative of the cluster\u2019s processor/tokenizer regression theme, and it is the most user-facing API break among the loosely related candidates.", + "summary": "This cluster is heterogeneous: the soft-linked issues share broad library areas but not the same concrete bug, model, or failing code path. I would not merge any of the suggested pairs as duplicates.", + "confidence": 0.94, + "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "issue:43295 is the strongest standalone issue here: clear regression, concrete failure mode, and the best fit for the small subset of processor-related reports. The rest are mostly distinct bugs.", + "best_issue_reason": "Issue 43295 is the strongest standalone anchor: it describes a clear, user-facing regression with a concrete breakage pattern and broad impact on custom code using processor/tokenizer access.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:44869", "right": "issue:46752", "accept": false, - "reason": "Different code paths and symptoms: Whisper timestamp decoding crash vs empty-list chat template index error." + "reason": "Different failure modes in different subsystems: Whisper word-timestamp decoding vs chat template handling on empty input." }, { - "left": "issue:45229", - "right": "issue:45823", + "left": "issue:43994", + "right": "issue:44493", "accept": false, - "reason": "Same model family, but OOM during multi-GPU inference is not the same bug as a device-mismatch forward failure." + "reason": "One is a SigLIP2 inference/output quality problem; the other is a position-id/key regression affecting many models. No shared code-path bug." }, { - "left": "issue:43994", - "right": "issue:44493", + "left": "issue:45229", + "right": "issue:45823", "accept": false, - "reason": "SigLIP2 output quality regression and position-id unexpected-key reports are unrelated failures." + "reason": "Both are Gemma4-related but the bugs differ: multi-GPU CUDA OOM versus a device mismatch during forward with auto device mapping." }, { "left": "issue:45588", "right": "issue:46050", "accept": false, - "reason": "Flash-attention sink-less model crash is a separate issue from sinq quantization save/load breakage." + "reason": "Unrelated: flash-attention sinkless-model attribute error versus quantized save/load corruption after sinq. Different code paths and symptoms." }, { "left": "issue:42490", "right": "issue:43696", "accept": false, - "reason": "Import warning on trailing-slash local paths is unrelated to GPT-oss CUDA OOM." + "reason": "Path parsing/import warning is unrelated to a CUDA out-of-memory report for GPT-oss-20b." }, { "left": "issue:29159", "right": "issue:32090", "accept": false, - "reason": "Tokenizer slow/fast inconsistency does not match a Trainer broadcast NoneType error." + "reason": "Tokenizer slow/fast inconsistency is unrelated to a Trainer distributed broadcast TypeError." }, { "left": "issue:44977", "right": "issue:46701", "accept": false, - "reason": "Qwen3.5 flash-attention generation bug and DiffusionGemma image-text-to-text pipeline failure are different problems." + "reason": "Different models and bugs: flash-attention generation issue versus an image-text-to-text pipeline failure for DiffusionGemmaGenerationOutput." }, { "left": "issue:45084", "right": "issue:45520", "accept": false, - "reason": "Compilation error and Python 3.13 flash_attn import KeyError are separate import/runtime issues." + "reason": "A template-node compilation TypeError is not the same as the Python 3.13 flash_attn import KeyError." }, { "left": "issue:28218", "right": "issue:34567", "accept": false, - "reason": "Extra tokenizer spacing after added tokens is unrelated to TrainerState token-count tracking." + "reason": "Tokenizer spacing behavior and TrainerState token-count tracking are unrelated." }, { "left": "issue:43065", "right": "issue:46369", "accept": false, - "reason": "Dummy Conv2d architecture detail and SAM3 memory overflow are not the same underlying bug." + "reason": "Both mention SAM3, but one is a dummy Conv2d implementation detail and the other is a memory overflow in sam3-litetext. Not the same bug." }, { "left": "issue:44554", "right": "issue:45446", "accept": false, - "reason": "Attention correctness on MPS and a PyTorch version-check import bug in flex_attention are unrelated." + "reason": "MPS attention correctness with mismatched head dims is unrelated to a flex_attention version-check import bug." }, { - "left": "issue:44423", - "right": "issue:44734", + "left": "issue:15354", + "right": "issue:38175", "accept": false, - "reason": "Continuous-batching multimodal crash and KV-cache tensor-indexing crash are different serve path failures." + "reason": "TorchScript GeneratorExp export failure is unrelated to SigLIP2 zero-probability inference behavior." }, { - "left": "issue:15354", - "right": "issue:38175", + "left": "issue:44423", + "right": "issue:44734", "accept": false, - "reason": "TorchScript GeneratorExp export limitation is unrelated to SigLIP2 zero-probability outputs." + "reason": "Continuous-batching multimodal crash from string tensor handling is different from KV-cache continuation tensor indexing." }, { "left": "issue:43493", "right": "issue:43701", "accept": false, - "reason": "SigLIP2 HF/JAX discrepancy and resume_from_checkpoint key mismatch do not share the same root cause." + "reason": "SigLIP2 implementation discrepancy is unrelated to resume_from_checkpoint key mismatch." }, { "left": "issue:41720", "right": "issue:46710", "accept": false, - "reason": "Auto device-map CUDA assert and an accuracy regression after version upgrade are distinct issues." + "reason": "CUDA assert on auto device mapping for Qwen3 is not the same as an accuracy regression after a Transformers upgrade." }, { "left": "issue:45406", "right": "issue:46121", "accept": false, - "reason": "Gemma4Processor missing _tokenizer in serve is unrelated to rope-parameter conversion TypeError." + "reason": "Gemma4Processor _tokenizer attribute crash is unrelated to rope-parameter list type handling." }, { "left": "issue:44977", "right": "issue:46710", "accept": false, - "reason": "Qwen3.5 flash-attention generation failure and DeepSeek accuracy regression are not the same bug." + "reason": "Flash-attention generation failure for Qwen3.5 is a different issue from a model accuracy regression after upgrade." }, { "left": "issue:45072", "right": "issue:45198", "accept": false, - "reason": "bfloat16 dtype mismatch bugs in different models/code paths are not the same concrete defect." + "reason": "dtype mismatch in CI inference and Wav2Vec2 save_pretrained/tokenization failure are unrelated." }, { "left": "issue:43295", "right": "issue:43450", "accept": false, - "reason": "Processor.tokenizer API regression and batched video shape handling are different processor bugs." + "reason": "Both involve processors, but one is a regression in processor.tokenizer access and image passing; the other is batched video shape handling. Different bugs." }, { "left": "issue:37428", "right": "issue:43957", "accept": false, - "reason": "Flash-attention import symbol failure and meta-device model-loading breakage are unrelated loader issues." + "reason": "Flash-attention utility import failure and meta-device model loading breakage are separate loading/runtime issues." }, { "left": "issue:45588", "right": "issue:45759", "accept": false, - "reason": "Sink-less flash-attention crash and composite-config text_config unwrapping are different multimodal/model-init problems." + "reason": "AttributeError on s_aux=None in flash_attention integration is unrelated to AutoModelForCausalLM.from_config not unwrapping text_config." }, { "left": "issue:43540", "right": "issue:45200", "accept": false, - "reason": "Video-input processing error in Qwen3OmniMoe and Gemma4 text-only mm_token_type_ids handling are unrelated." + "reason": "Video input validation in Qwen3OmniMoe is unrelated to Gemma4 mm_token_type_ids defaults for text-only fine-tuning." }, { "left": "issue:43295", "right": "issue:43873", "accept": false, - "reason": "Processor API regression and quantization offloading behavior are different layers of the stack." + "reason": "Processor/tokenizer regression and offloading with quantization are different failure modes and code paths." }, { "left": "issue:44186", "right": "issue:46192", "accept": false, - "reason": "Tokenizer crash on NER/batched truncation and silently ignored return_tensors are related to tokenization, but not the same underlying bug." + "reason": "LayoutLMv2Tokenizer NER/padding crash is unrelated to return_tensors being ignored when text_kwargs is passed." }, { "left": "issue:43572", "right": "issue:46821", "accept": false, - "reason": "StableLmConfig missing pad_token_idx and TimesFM window_size AttributeError are different config-access failures." + "reason": "Missing StableLmConfig pad_token_idx after v5 update is unrelated to TimesFM window_size AttributeError." }, { "left": "issue:43295", "right": "issue:44610", "accept": false, - "reason": "Processor.tokenizer regression and OmDet-Turbo processor output-size mismatch are distinct processor issues." + "reason": "Both are processor-related, but one is a tokenizer-access regression and the other is an OmDet-Turbo image size mismatch. Not the same underlying bug." }, { "left": "issue:45823", "right": "issue:46192", "accept": false, - "reason": "Gemma4 PLE device mismatch and return_tensors being ignored are unrelated API/forward-path bugs." + "reason": "Gemma4 device mismatch with auto device mapping is unrelated to return_tensors being ignored in multimodal kwargs handling." } ] }, @@ -8942,8 +9039,8 @@ "issue:43572|issue:44987", "issue:44554|issue:45356", "issue:37428|issue:43824", - "issue:43232|issue:44849", "issue:42490|issue:43493", + "issue:43232|issue:44849", "issue:43295|issue:43493", "issue:44466|issue:45127", "issue:39692|issue:46192", @@ -8961,20 +9058,20 @@ "issue:43262|issue:43450", "issue:44442|issue:45198" ], - "prepared_review_unit_hash": "6d69853f8140bb11f3cc01f0a1cc85ee96fa115220f09c77749a85d25b70a0af", + "prepared_review_unit_hash": "b695e4e7af99b7139fb30aad8b05060c40f5a6edcff52fa5dd78a998a0a45fa4", "component_budget": { "node_count": 502, "item_count": 502, - "soft_pair_count": 1842, - "serialized_chars": 956182, - "estimated_input_tokens": 239046, - "estimated_eval_tokens": 478348 + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 26, - "serialized_chars": 35882, + "serialized_chars": 35883, "estimated_input_tokens": 8971, "estimated_eval_tokens": 18198 }, @@ -8983,162 +9080,168 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is not a true duplicate set: it mixes unrelated issues across tokenization, generation, multimodal processors, import/export failures, config regressions, and training/runtime bugs. All soft pairs should be rejected.", - "confidence": 0.91, - "canonical_issue_reason": "No canonical issue: the candidates are too heterogeneous to represent one underlying bug or change.", - "canonical_pr_reason": "No PRs are present in this cluster.", - "best_issue_reason": "No single issue is a clear representative; the items do not form one cohesive duplicate family.", - "best_pr_reason": "No PRs are present.", + "summary": "Most pairs are only superficially related by subsystem or keyword and do not look like the same underlying bug. I would keep issue 43950 as the strongest representative issue: it describes a clear, core regression with a direct reproduction path and broad impact.", + "confidence": 0.71, + "canonical_issue_reason": "Issue 43950 is the most concrete, regression-framed core bug in the set, with a clear failure mode and strong user impact.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 43950 is the best overall issue to keep as the representative triage target because it is specific, actionable, and clearly describes a real regression.", + "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:29870", "right": "issue:29942", "accept": false, - "reason": "Different problems: decoder_attention_mask generation error vs Flash Attention 2 CI test failures." + "reason": "Different failures: generation-time decoder_attention_mask bug vs Flash Attention 2 test failures. Same broad area, not the same bug." }, { "left": "issue:43550", "right": "issue:43844", "accept": false, - "reason": "Unrelated code paths: torch.compile/SDPA failure vs ZeRO-3 gradient growth during training." + "reason": "One is a torch.compile/SDPA model failure, the other is a DeepSpeed ZeRO-3 gradient anomaly. Distinct code paths and symptoms." }, { "left": "issue:45520", "right": "issue:46178", "accept": false, - "reason": "Different import/backend issues: flash_attn on Python 3.13 vs tensorflow_text backend mapping." + "reason": "Import error for flash_attn on Python 3.13 is unrelated to a backend mapping ValueError for tensorflow_text." }, { "left": "issue:45200", "right": "issue:45823", "accept": false, - "reason": "Gemma4 input-default bug vs a forward-pass device mismatch; same model family but not the same bug." + "reason": "Gemma 4 token type ID defaulting bug and PLE device mismatch are different issues in different parts of the stack." }, { "left": "issue:43450", "right": "issue:43575", "accept": false, - "reason": "Video processor batch-shape bug vs Qwen2-57B tensor-parallel OOM are unrelated." + "reason": "Video processor batch-shape bug vs tensor-parallel OOM while loading Qwen2-57B. Not the same underlying problem." }, { "left": "issue:43572", "right": "issue:44987", "accept": false, - "reason": "Different models and failure modes: missing pad_token_idx vs a separate load regression in transformers>=5.1.0." + "reason": "Both are load-time regressions, but the titles point to different models and different root causes." }, { "left": "issue:44554", "right": "issue:45356", "accept": false, - "reason": "Attention correctness on MPS is unrelated to Kimi tokenizer codec handling/regression." + "reason": "MPS attention correctness issue and Kimi tokenizer regression are unrelated." }, { "left": "issue:37428", "right": "issue:43824", "accept": false, - "reason": "Both are import errors, but they concern different missing symbols/classes and different modules." + "reason": "Separate import errors for different symbols/models; no shared code-path bug is evident." }, { - "left": "issue:43232", - "right": "issue:44849", + "left": "issue:42490", + "right": "issue:43493", "accept": false, - "reason": "Generation kwargs handling after sync_gpus vs a Qwen3.5 output_hidden_states bug are separate issues." + "reason": "Path-with-slash import warning is unrelated to SigLIP2 implementation discrepancy." }, { - "left": "issue:42490", - "right": "issue:43493", + "left": "issue:43232", + "right": "issue:44849", "accept": false, - "reason": "Local-path trailing-slash warning is unrelated to SigLIP2 implementation drift from JAX." + "reason": "Generation kwargs sync issue and Qwen3.5 hidden-states bug are distinct generation behaviors." }, { "left": "issue:43295", "right": "issue:43493", "accept": false, - "reason": "Processor.tokenizer regression vs SigLIP2 model/processor mismatch are different code paths." + "reason": "Processor/tokenizer regression is not the same as SigLIP2 JAX-vs-HF discrepancy." }, { "left": "issue:44466", "right": "issue:45127", "accept": false, - "reason": "Both touch tied weights, but one is device-dependent serialization and the other is LoRA merge collapse with extended vocab; not the same bug." + "reason": "Both involve weights and vocabulary/tie behavior, but one is serialization inconsistency and the other is LoRA merge collapse. Different bugs." }, { "left": "issue:39692", "right": "issue:46192", "accept": false, - "reason": "Documentation/example and quantization issues are unrelated to text_kwargs overriding return_tensors." + "reason": "Documentation example errors and return_tensors being ignored are unrelated." }, { "left": "issue:44521", "right": "issue:45290", "accept": false, - "reason": "Both involve apply_chat_template, but one is assistant_masks for multimodal inputs and the other is a crash on tool-call messages with no content." + "reason": "Both touch apply_chat_template, but one is incorrect assistant masks for multimodal inputs and the other is a crash on tool-call messages without content." }, { "left": "issue:43653", "right": "issue:44112", "accept": false, - "reason": "BigBird special-token registration bug vs a GraniteSpeech CI device override test failure are unrelated." + "reason": "Tokenizer special-token registration bug vs CI failure in a stale device override test. Not a duplicate." }, { "left": "issue:45381", "right": "issue:46209", "accept": false, - "reason": "Qwen2.5-VL video position ids bug vs FP8 inference failure on B200 are different problems." + "reason": "Wrong vision_position_ids for qwen2.5-vl video input is unrelated to FP8 inference failure on B200." }, { "left": "issue:28218", "right": "issue:35141", "accept": false, - "reason": "Tokenizer spacing after added tokens is unrelated to embedding reinitialization during post_init." + "reason": "Tokenizer spacing after added tokens and embedding reinitialization after resize are separate behaviors." }, { "left": "issue:43278", "right": "issue:43856", "accept": false, - "reason": "BF16-to-FP32 eval dtype drift vs MoE training memory inefficiency are not the same issue." + "reason": "Dtype drift during evaluate and inefficient Qwen3 MoE training memory use are different problems." }, { "left": "issue:44737", "right": "issue:46894", "accept": false, - "reason": "CPU torch.arange in XLNet vs CohereASR double-shift training loss bug are unrelated." + "reason": "CPU-device placement bug in XLNet positional encoding is unrelated to a training-loss double-shift bug." }, { "left": "issue:42915", "right": "issue:43824", "accept": false, - "reason": "FP8 config failure in Qwen3Moe is unrelated to an import error for a Qwen2.5-VL class." + "reason": "FineGrainedFP8Config failure in Qwen3MoE and a Qwen2_5_VL import error do not share the same root cause." }, { "left": "issue:43950", "right": "issue:46285", "accept": false, - "reason": "Non-persistent buffer corruption on from_pretrained is unrelated to CLIPTextModel structural inconsistency." + "reason": "Non-persistent buffer corruption on from_pretrained is unrelated to CLIPTextModel structure inconsistency." }, { "left": "issue:43824", "right": "issue:43975", "accept": false, - "reason": "Importing a missing model class and incorrect detokenization are different failure modes." + "reason": "Importing a missing Qwen2_5_VL class and incorrect detokenization in DeepSeek-Coder are unrelated." }, { "left": "issue:43452", "right": "issue:46512", "accept": false, - "reason": "gguf_file loading breakage is unrelated to from_config(dtype=...) leaving config.dtype stale." + "reason": "Tokenizer/model loading failure with gguf_file is not the same as stale config dtype after from_config(dtype=...)." }, { "left": "issue:43334", "right": "issue:43525", "accept": false, - "reason": "Both mention pad_token_id, but they affect different config classes and model families with no shared concrete code path." + "reason": "Both mention missing pad_token_id, but they are model-specific load failures for different configs; not enough evidence of one underlying bug." }, { "left": "issue:43262", "right": "issue:43450", "accept": false, - "reason": "Audio chat-template sampling-rate defaulting and batched video shape errors are unrelated processor bugs." + "reason": "Audio chat-template sample-rate defaulting and video batch-shape handling are different processor bugs." + }, + { + "left": "issue:44442", + "right": "issue:45198", + "accept": false, + "reason": "FastSpeech2ConformerTokenizer loading and Wav2Vec2 save/tokenization failure are different tokenizer/model issues." } ] }, @@ -9226,8 +9329,8 @@ "issue:43388|issue:43526", "issue:44857|issue:45593", "issue:43819|issue:45083", - "issue:46424|issue:46858", "issue:44610|issue:44857", + "issue:46424|issue:46858", "issue:45759|issue:46285", "issue:45520|issue:46170", "issue:43606|issue:45083", @@ -9242,208 +9345,208 @@ "issue:43611|issue:43931", "issue:44448|issue:44743" ], - "prepared_review_unit_hash": "a2c8e9635137fd1f0ec680e719042af93794278cb395b49fdc2153d6f5f2942a", + "prepared_review_unit_hash": "e961155c6b2d15856b65cfd48bda0e33471af7ec9298881509a26f8ef9cd7b76", "component_budget": { "node_count": 502, "item_count": 502, - "soft_pair_count": 1842, - "serialized_chars": 956182, - "estimated_input_tokens": 239046, - "estimated_eval_tokens": 478348 + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 29, - "serialized_chars": 36675, - "estimated_input_tokens": 9169, - "estimated_eval_tokens": 18594 + "serialized_chars": 36683, + "estimated_input_tokens": 9171, + "estimated_eval_tokens": 18598 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "No soft pair looks like the same underlying bug/change. The cluster is mostly a loose topical grouping with many unrelated model, cache, import, CI, and docs issues. There are no PRs in the set.", - "confidence": 0.94, - "canonical_issue_reason": "Issue #45200 is the closest thing to a hub because it has the most soft links and is the most central Gemma4-related item, but the group is not a true duplicate cluster.", + "summary": "This cluster is mostly a grab-bag of unrelated issues, with only a small Gemma4 quantization-related subgroup. None of the soft pairs look like the same underlying bug/change, so I reject all soft edges.", + "confidence": 0.29, + "canonical_issue_reason": "issue:46899 is the clearest, most concrete bug report in the only closely related subgroup (Gemma4 multimodal quantization). It describes an actionable failing code path more directly than the neighboring Gemma4 items.", "canonical_pr_reason": null, - "best_issue_reason": "#45200 is the best representative only in a weak sense: it is concrete, recent, and connected to multiple Gemma4 multimodal-related items. It is still not actually duplicate-equivalent to the others.", + "best_issue_reason": "issue:46899 is the best representative issue here because it is specific, reproducible, and tightly scoped; the other items are either different bugs or broader/less directly actionable.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:45200", "right": "issue:46932", "accept": false, - "reason": "Same product area (Gemma4) but different bugs: default mm_token_type_ids vs unsafe dtype casting in quantized embedders." + "reason": "Different bugs: mm_token_type_ids defaults for text-only fine-tuning vs quantized vision/audio embedder dtype casting." }, { "left": "issue:44792", "right": "issue:45563", "accept": false, - "reason": "Both are generate-related failures, but one is a Janus image test and the other is a stale warning about num_return_sequences; not the same bug." + "reason": "Different failures: janus image generation test vs stale num_return_sequences warning in paged generate()." }, { "left": "issue:45200", "right": "issue:46899", "accept": false, - "reason": "Both mention Gemma4 multimodal embedders, but one is about mm_token_type_ids defaults and the other about dtype casts under 4-bit quantization." + "reason": "Different bugs in the same model family, but one is about token type ids and the other about 4-bit multimodal embedder casting." }, { "left": "issue:43819", "right": "issue:44112", "accept": false, - "reason": "Different models and failure modes: DAC from_latents mismatch vs a stale GraniteSpeech CI device-override test." + "reason": "Different code paths: DAC latent forward mismatch vs GraniteSpeech stale device-override CI test." }, { "left": "issue:44898", "right": "issue:45412", "accept": false, - "reason": "Perceiver image resolution failure and RT-DETR memory leak are unrelated bugs." + "reason": "Different subsystems: Perceiver resolution handling vs RT-DETR memory release/GC." }, { "left": "issue:43526", "right": "issue:46192", "accept": false, - "reason": "Both concern preprocessing behavior, but one is reduce_labels corruption and the other is return_tensors being ignored when text_kwargs is passed." + "reason": "Different bugs: BeitImageProcessorFast label reduction vs return_tensors being ignored with explicit text_kwargs." }, { "left": "issue:45200", "right": "issue:45325", "accept": false, - "reason": "Both are multimodal-model issues, but one is Gemma4 token-type defaults and the other is Qwen2.5-VL temporal position scaling." + "reason": "Different multimodal bugs: Gemma4 token-type defaults vs Qwen2.5-VL rope index scaling." }, { "left": "issue:43827", "right": "issue:43901", "accept": false, - "reason": "Both are documentation-related, but they concern different docs and different deprecations/behavior mismatches." + "reason": "Both are docs issues, but they concern different pipeline behaviors and are not the same change." }, { "left": "issue:41720", "right": "issue:46209", "accept": false, - "reason": "Different hardware/precision crashes: A800 auto device mapping assert vs FP8 inference on B200." + "reason": "Different hardware/model failures: Qwen3 auto device mapping cudaErrorAssert vs FP8 inference failure on B200." }, { "left": "issue:43335", "right": "issue:43550", "accept": false, - "reason": "Different concrete problems in different models: SwitchTransformers config creates an unwanted sparse layer, while Bamba-9B-v2 fails under torch.compile with SDPA." + "reason": "Different problems in different models: SwitchTransformers sparse layer construction vs Bamba torch.compile SDPA failure." }, { "left": "issue:44561", "right": "issue:46234", "accept": false, - "reason": "Both are import-utils/API surface issues, but they are about different missing symbols and different breakages." + "reason": "Different missing-helper/import problems; same utility module does not make them the same bug." }, { "left": "issue:43388", "right": "issue:43526", "accept": false, - "reason": "Both touch label handling, but one drops label elements in gather_for_metrics and the other reduces labels incorrectly in BeitImageProcessorFast." + "reason": "Different label-handling bugs in different components and failure modes." }, { "left": "issue:44857", "right": "issue:45593", "accept": false, - "reason": "Different training-side bugs in different detection models: float16 AMP crash vs auxiliary losses missing when denoising is off." + "reason": "Different object-detection issues: AMP/CUDA crash in LwDetrImageLoss vs missing auxiliary losses in D-FINE." }, { "left": "issue:43819", "right": "issue:45083", "accept": false, - "reason": "Completely different code paths: DAC latent conversion vs qwen3_omni_moe feature-length helper behavior." + "reason": "Different models and failure modes: DAC latent mismatch vs qwen3_omni_moe length helper behavior." }, { - "left": "issue:46424", - "right": "issue:46858", + "left": "issue:44610", + "right": "issue:44857", "accept": false, - "reason": "Both involve cache internals, but one is a max_cache_length config limitation and the other is a static-cache seq_length tensor bug under torch.compile." + "reason": "Different bugs: OmDet-Turbo preprocessing size mismatch vs LwDetrImageLoss AMP crash." }, { - "left": "issue:44610", - "right": "issue:44857", + "left": "issue:46424", + "right": "issue:46858", "accept": false, - "reason": "One is an input-size mismatch in OmDet-Turbo processing; the other is a mixed-precision loss crash in LwDETR." + "reason": "Both touch cache internals, but one is max_cache_length semantics and the other is a device-tensor seq_length arithmetic bug." }, { "left": "issue:45759", "right": "issue:46285", "accept": false, - "reason": "Different model-structure regressions: unwrapping text_config in composite Qwen multimodal configs vs CLIP text-model structure inconsistency." + "reason": "Different model-loading regressions: composite Qwen config unwrapping vs CLIPTextModel structure change." }, { "left": "issue:45520", "right": "issue:46170", "accept": false, - "reason": "Different import/runtime dependency failures on different platforms: flash_attn lookup on Python 3.13 vs torch.distributed dependency on Windows ROCm." + "reason": "Different import/platform issues: flash_attn lookup on Python 3.13 vs Windows/ROCm torch.distributed dependency problem." }, { "left": "issue:43606", "right": "issue:45083", "accept": false, - "reason": "CPU-offload device mismatch in bark-small and qwen3_omni_moe output-length helper behavior are unrelated." + "reason": "Different bugs: CPU offload device mismatch in Bark vs qwen3_omni_moe feature-length helper behavior." }, { "left": "issue:46512", "right": "issue:46821", "accept": false, - "reason": "One is dtype state mismatch after from_config, the other is a TimesFM window_size AttributeError." + "reason": "Different issues: stale config dtype after from_config vs TimesFM window_size AttributeError." }, { "left": "issue:44206", "right": "issue:45584", "accept": false, - "reason": "Different feature regressions: unsupported center arg in LasrFeatureExtractor vs empty transcription failure after align_special_tokens." + "reason": "Different regressions in unrelated components: feature extractor center arg vs Whisper empty transcription handling." }, { "left": "issue:38175", "right": "issue:45412", "accept": false, - "reason": "Unexpected zero probabilities in SigLIP2 and RT-DETR memory not released are unrelated." + "reason": "Different issues: unexpected zero probabilities in SigLIP2 vs RT-DETR memory not released." }, { "left": "issue:43335", "right": "issue:43653", "accept": false, - "reason": "Different areas entirely: SwitchTransformers config bug vs BigBirdTokenizer special-token registration issue." + "reason": "Different bugs: SwitchTransformers sparse-layer creation vs BigBirdTokenizer special-token registration." }, { "left": "issue:37428", "right": "issue:43425", "accept": false, - "reason": "Both are compatibility/import issues, but one is a missing internal flash-attn helper and the other is a broader Torch 2.10 incompatibility claim." + "reason": "One is a missing flash-attention helper import; the other is broad Torch 2.10 incompatibility." }, { "left": "issue:43550", "right": "issue:44112", "accept": false, - "reason": "A model compile/SDPA failure and a stale CI device-override test are not the same bug." + "reason": "Different code paths and symptoms: Bamba torch.compile/SDPA failure vs GraniteSpeech CI stale device override test." }, { "left": "issue:29942", "right": "issue:32090", "accept": false, - "reason": "Flash Attention 2 test failures and a trainer gpu_broadcast NoneType error are different failures with different code paths." + "reason": "Different failures: Flash Attention 2 test breakage vs Trainer broadcast TypeError on NoneType." }, { "left": "issue:28282", "right": "issue:33357", "accept": false, - "reason": "PyTorch-not-found ImportError and a MacOS bus error with a community CLIP model are unrelated." + "reason": "Different environments and bugs: AutoModel/PyTorch missing import vs MacOS bus error on community CLIP model." }, { "left": "issue:43611", "right": "issue:43931", "accept": false, - "reason": "Both are model-loading problems, but one is a v5 base_model_prefix regression and the other is a Qwen3-VL weight-shape mismatch." + "reason": "Different model-loading regressions: base_model_prefix breakage vs Qwen3-VL weight-shape mismatch." }, { "left": "issue:44448", "right": "issue:44743", "accept": false, - "reason": "Pegasus v4/v5 output differences and recurrent cache state reset in modular_qwen3_5 are different bugs." + "reason": "Different bugs: output drift across v4/v5 for Pegasus vs recurrent state reset in qwen3_5 cache handling." } ] }, @@ -9520,241 +9623,235 @@ "issue:43335|issue:43504", "issue:44855|issue:45083", "issue:43278|issue:43825", - "issue:43335|issue:45083", "issue:29127|issue:32090", "issue:43720|issue:44112", + "issue:43335|issue:45083", "issue:37428|issue:44351", - "issue:43335|issue:44112", "issue:43335|issue:45593", + "issue:43335|issue:44112", "issue:44514|issue:44521", "issue:45290|issue:46326", - "issue:45127|issue:45245", "issue:34689|issue:35141", "issue:45588|issue:45859", "issue:42490|issue:43854", "issue:46431|issue:46566", "issue:45310|issue:45478", + "issue:45127|issue:45245", "issue:44448|issue:44779", "issue:43819|issue:44493", - "issue:26413|issue:31501", "issue:39692|issue:43526", "issue:43606|issue:44857", + "issue:26413|issue:31501", "issue:28218|issue:32090", "issue:45459|issue:46234", - "issue:43701|issue:45083", "issue:44617|issue:46697", + "issue:43701|issue:45083", "issue:39692|issue:44493", "issue:44206|issue:45823", "issue:32090|issue:33357" ], - "prepared_review_unit_hash": "f776511cea14317d01e0c525b936e86ed30d8858dcd2239507df59d7fdd16dcb", + "prepared_review_unit_hash": "b97a8e6b3173156e5cbcaa14633dc3c99a624a1a8ea6b33cb8b37efd9ed47f42", "component_budget": { "node_count": 502, "item_count": 502, - "soft_pair_count": 1842, - "serialized_chars": 956182, - "estimated_input_tokens": 239046, - "estimated_eval_tokens": 478348 + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 29, - "serialized_chars": 36597, - "estimated_input_tokens": 9150, - "estimated_eval_tokens": 18556 + "serialized_chars": 36595, + "estimated_input_tokens": 9149, + "estimated_eval_tokens": 18554 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Most items are unrelated issues that only share broad Transformer subsystems or similar wording. The only clear duplicate is the Qwen3.5 MoE `from_pretrained` regression, where two issues describe the same failure and should be clustered together.", - "confidence": 0.89, - "canonical_issue_reason": "Issue 45310 is the clearest canonical: it is the same Qwen3.5 MoE `from_pretrained` failure as 45478, has the fuller report, and is the better tracker for the duplicate pair.", + "summary": "This cluster is mostly heterogeneous and does not look like one duplicate family. The only clear duplicate pair is the Qwen3.5 MoE from_pretrained error reports; most other soft edges are just broad subsystem overlap or shared regression themes, not the same concrete bug.", + "confidence": 0.93, + "canonical_issue_reason": "issue:45310 is the best canonical anchor because it is the earlier, clearer report of the exact same Qwen3.5 MoE from_pretrained failure; issue:45478 repeats the same title and appears to be a later duplicate.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 45310 is the best representative of the cluster because it captures the only strong duplicate pair and the concrete regression title most precisely.", + "best_issue_reason": "issue:45310 is the strongest issue to keep as the representative bug report among the items because it has the clearest duplicate counterpart and a precise, concrete failure description.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43335", "right": "issue:43504", "accept": false, - "reason": "Different models and failures: sparse-layer config creation vs BEiT legacy-field loading." + "reason": "Different models and failures: SwitchTransformers config bug versus BEiT legacy-field load failure." }, { "left": "issue:44855", "right": "issue:45083", "accept": false, - "reason": "Different bugs; one is a Python 3.13 import parsing error, the other is a feature-extraction helper behavior issue." + "reason": "Unrelated bugs: a Python import/IndentationError versus a qwen3_omni_moe helper-function behavior issue." }, { "left": "issue:43278", "right": "issue:43825", "accept": false, - "reason": "Same library area only; dtype drift in eval is unrelated to a pipeline error-message regression." - }, - { - "left": "issue:43335", - "right": "issue:45083", - "accept": false, - "reason": "Unrelated model/config bug vs helper-function behavior change." + "reason": "Different code paths and symptoms: dtype drift in evaluate versus an incorrect pipeline error message." }, { "left": "issue:29127", "right": "issue:32090", "accept": false, - "reason": "Different code paths: LayoutLMv3 error clarity vs Trainer GPU broadcast NoneType failure." + "reason": "One is a LayoutLMv3 error-message clarity issue; the other is a Trainer/NCCL broadcast type error." }, { "left": "issue:43720", "right": "issue:44112", "accept": false, - "reason": "Both are CI/device-related, but one is packed-weight loading and the other is a stale test issue." + "reason": "Different underlying problems: packed weights not unpacked during Accelerate loading versus a stale CI test in GraniteSpeech." + }, + { + "left": "issue:43335", + "right": "issue:45083", + "accept": false, + "reason": "No shared underlying bug; they concern different architectures and different failure modes." }, { "left": "issue:37428", "right": "issue:44351", "accept": false, - "reason": "Both are import errors, but they involve different missing symbols and different modules." + "reason": "Both are import failures, but for different symbols and different modules, so they are not the same bug." }, { "left": "issue:43335", - "right": "issue:44112", + "right": "issue:45593", "accept": false, - "reason": "No shared underlying bug; configuration generation and CI test flakiness are unrelated." + "reason": "Different model/config issues: sparse-layer config creation versus auxiliary-loss behavior when denoising is off." }, { "left": "issue:43335", - "right": "issue:45593", + "right": "issue:44112", "accept": false, - "reason": "Different behavior in different model families; not the same concrete bug." + "reason": "No evidence they share the same code path; one is config construction, the other is a CI override regression." }, { "left": "issue:44514", "right": "issue:44521", "accept": false, - "reason": "Both involve `apply_chat_template`, but one crashes on batched padding=False and the other returns wrong assistant masks." + "reason": "Both involve Qwen2.5 VL chat templating, but one is a padding-related crash and the other is incorrect assistant masks; not the same concrete bug." }, { "left": "issue:45290", "right": "issue:46326", "accept": false, - "reason": "Both are chat-template related, but they affect different tokenizer logic and failure modes." - }, - { - "left": "issue:45127", - "right": "issue:45245", - "accept": false, - "reason": "LoRA/vocab tie-weight collapse is unrelated to the category-count runtime limit." + "reason": "Both are chat-template related, but they target different tokenizer features and different failure modes." }, { "left": "issue:34689", "right": "issue:35141", "accept": false, - "reason": "Different loading/initialization regressions for different models." + "reason": "Different problems: model loading regression for Llama 3.2 Vision versus embedding reinitialization after resizing." }, { "left": "issue:45588", "right": "issue:45859", "accept": false, - "reason": "Both touch MoE/flash/offload areas, but one is an `s_aux=None` crash and the other is CPU-offload incompatibility in rotary embedding." + "reason": "Different subsystems and failures: flash_attention integration attribute error versus CPU-offload incompatibility in a rotary embedding implementation." }, { "left": "issue:42490", "right": "issue:43854", "accept": false, - "reason": "Local-path import warning is not the same as a specific model-loading failure in tests." + "reason": "Both involve loading behavior, but the local-path import warning and the GLM-4.7-Flash unit-test load failure are not the same bug." }, { "left": "issue:46431", "right": "issue:46566", "accept": false, - "reason": "Processor chat-template absence and meta-device loading are unrelated bugs." + "reason": "Chat-template absence and meta-device loading are unrelated issues." }, { "left": "issue:45310", "right": "issue:45478", "accept": true, - "reason": "Same Qwen3.5 MoE `from_pretrained` error on `transformers>=5.4.0`; the titles and described failure match closely." + "reason": "Exact same Qwen3.5 MoE from_pretrained error; the titles match and the later issue appears to be a duplicate report." + }, + { + "left": "issue:45127", + "right": "issue:45245", + "accept": false, + "reason": "Different failures: LoRA merge collapse with tied embeddings versus a category-count runtime limit." }, { "left": "issue:44448", "right": "issue:44779", "accept": false, - "reason": "Both are v5 regressions, but one is Pegasus output drift and the other is a Deepseek tokenizer regression." + "reason": "Both are v5 regressions, but one is Pegasus output drift and the other is Deepseek tokenizer correctness; not the same bug." }, { "left": "issue:43819", "right": "issue:44493", "accept": false, - "reason": "DAC latent mismatch and unexpected position-id keys are different bugs." - }, - { - "left": "issue:26413", - "right": "issue:31501", - "accept": false, - "reason": "Training/checkpoint resume failure is not the same as FineWeb startup issues." + "reason": "Different code paths: DAC from_latents mismatch versus unexpected position-id keys across models." }, { "left": "issue:39692", "right": "issue:43526", "accept": false, - "reason": "Documentation example problems vs an image-processor label reduction bug." + "reason": "Documentation example errors are not the same issue as BeitImageProcessorFast reduce_labels returning the wrong shape/result." }, { "left": "issue:43606", "right": "issue:44857", "accept": false, - "reason": "Different runtime failures: CPU offload device mismatch vs AMP/CUDA loss crash." + "reason": "Different runtime failures in different models/components: CPU-offload device mismatch versus AMP/CUDA loss crash." + }, + { + "left": "issue:26413", + "right": "issue:31501", + "accept": false, + "reason": "Both mention training startup/resume, but one is a checkpoint-resume iterator failure and the other is a FineWeb training start problem; too vague to merge." }, { "left": "issue:28218", "right": "issue:32090", "accept": false, - "reason": "Tokenizer spacing bug is unrelated to Trainer broadcast failures." + "reason": "Tokenizer spacing after added tokens is unrelated to a Trainer GPU broadcast NoneType error." }, { "left": "issue:45459", "right": "issue:46234", "accept": false, - "reason": "Tokenizer exception masking and a missing utility import are different issues." - }, - { - "left": "issue:43701", - "right": "issue:45083", - "accept": false, - "reason": "Checkpoint-key mismatch is unrelated to the qwen3_omni_moe helper behavior." + "reason": "Tokenizer error masking and a missing import_utils helper are unrelated issues." }, { "left": "issue:44617", "right": "issue:46697", "accept": false, - "reason": "CUDA OOM in Sam3Video is not the same bug as Swin2SR over-padding." + "reason": "CUDA OOM in Sam3Video and over-padding in Swin2SR image processing are unrelated." }, { - "left": "issue:39692", - "right": "issue:44493", + "left": "issue:43701", + "right": "issue:45083", "accept": false, - "reason": "SigLIP2 docs errors and unexpected position-id keys are unrelated." + "reason": "Resume-from-checkpoint key mismatch and qwen3_omni_moe feature-length behavior are unrelated." }, { "left": "issue:44206", "right": "issue:45823", "accept": false, - "reason": "Unsupported center arg in LASR feature extraction is unrelated to Gemma4 PLE device mismatch." + "reason": "Different failures in different models: unsupported feature-extractor arg versus device-map forward mismatch." }, { "left": "issue:32090", "right": "issue:33357", "accept": false, - "reason": "Trainer `_gpu_broadcast_one` NoneType error is not the same as the MacOS CLIP bus error." + "reason": "A NoneType GPU broadcast error and a MacOS bus error in CLIP loading are not the same underlying bug." } ] }, "evaluator_result": { "accept": true, - "feedback": "The summary is well grounded in the packet and stays conservative on the soft-edge pairs. The only clearly supported duplicate is issue:45310 vs issue:45478, which have identical titles and matching described failure. The canonical/best-issue choice is reasonable, and the other pair rejections do not overstate similarity." + "feedback": "The summary is well grounded in the packet. The representative/canonical choice for issue:45310 over issue:45478 is supported by identical titles and earlier creation time, and the cluster summary stays conservative by treating the rest as non-duplicates unless they clearly share the same concrete bug. No overstatement stands out." }, "evaluator_used": true, "retried": false, @@ -9832,16 +9929,16 @@ "issue:29942|issue:35141", "issue:38175|issue:44493", "issue:29870|issue:32090", - "issue:43278|issue:43329", "issue:44448|issue:44977", + "issue:43278|issue:43329", "issue:45920|issue:46192", "issue:45290|issue:45381", "issue:10000|issue:15354", "issue:44977|issue:45229", - "issue:43278|issue:43381", - "issue:43927|issue:45698", "issue:29870|issue:33290", + "issue:43278|issue:43381", "issue:44855|issue:46399", + "issue:43927|issue:45698", "issue:45290|issue:46655", "issue:40444|issue:43526", "issue:37428|issue:42175", @@ -9849,220 +9946,220 @@ "issue:32090|issue:33290", "issue:46421|issue:46858", "issue:44561|issue:44779", - "issue:43873|issue:45229", "issue:44448|issue:44466", - "issue:43976|issue:45440", + "issue:43873|issue:45229", "issue:25251|issue:32090", + "issue:43976|issue:45440", "issue:42490|issue:43873" ], - "prepared_review_unit_hash": "0740997dd6e43fd44d27d635de449b9ac0e091f846ea8fd1f0b7e9b0aca49fd9", + "prepared_review_unit_hash": "dd52108c39aeaa880081bdf8638eaa2926a922b44630661e2fb5d03d277037f2", "component_budget": { "node_count": 502, "item_count": 502, - "soft_pair_count": 1842, - "serialized_chars": 956182, - "estimated_input_tokens": 239046, - "estimated_eval_tokens": 478348 + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 30, - "serialized_chars": 36357, - "estimated_input_tokens": 9090, - "estimated_eval_tokens": 18436 + "serialized_chars": 36350, + "estimated_input_tokens": 9088, + "estimated_eval_tokens": 18432 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster is highly heterogeneous: the items span unrelated bugs across tokenization, generation, flash-attention, serialization, training, and model-loading. I don\u2019t see any soft pair that is the same underlying issue or change, so all soft edges should be rejected.", - "confidence": 0.93, - "canonical_issue_reason": "issue:43873 is the best anchor in this mixed cluster because it is open, has the highest discussion/inbound-reference activity, and is phrased as a broadly scoped reproducible bug report.", + "summary": "The soft pairs are mostly false positives from broad semantic similarity: they mention shared themes like loading, generation, Flash Attention, dtype, or compile issues, but they point to different models, different code paths, or different failure modes. I would not merge any of these pairs as duplicates.", + "confidence": 0.95, + "canonical_issue_reason": "No clear canonical issue: this set is heterogeneous and the apparent matches are not the same underlying bug.", "canonical_pr_reason": null, - "best_issue_reason": "issue:43873 has the strongest signal for triage value among the issues: most activity, cross-references, and a clearly stated user-facing regression. It is the most suitable representative issue despite the cluster being non-duplicate overall.", + "best_issue_reason": "No single issue is a safe representative for the cluster; the items span unrelated regressions and model-specific failures.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:42490", "right": "issue:43575", "accept": false, - "reason": "Different problems: local-path slash warning vs tensor-parallel OOM when loading a large Qwen model." + "reason": "Both involve loading, but one is a local path trailing-slash warning and the other is a tensor-parallel OOM on a specific Qwen model; different failures and fixes." }, { "left": "issue:44360", "right": "issue:44561", "accept": false, - "reason": "Unrelated subsystems: DSA indexer ReLU discussion vs v5 trust_remote_code breakage from removing is_torch_fx_available." + "reason": "DSA indexer/ReLU discussion is unrelated to removal of `is_torch_fx_available` and `trust_remote_code` breakage." }, { "left": "issue:44857", "right": "issue:45072", "accept": false, - "reason": "Both mention dtype/AMP, but one is a CUDA float16 loss crash and the other is a bfloat16 inference mismatch in different models/code paths." + "reason": "Both mention dtype/AMP-style failures, but one is a loss crash on CUDA fp16 and the other is a bfloat16 inference mismatch in different models; not the same bug." }, { "left": "issue:43550", "right": "issue:45083", "accept": false, - "reason": "Different model families and failures: torch.compile with SDPA on Bamba vs helper length calculation bug in qwen3_omni_moe." + "reason": "`torch.compile`/SDPA on Bamba-9B-v2 is a model/runtime bug; `_get_feat_extract_output_lengths` in qwen3_omni_moe is a separate helper-function behavior issue." }, { "left": "issue:27957", "right": "issue:34689", "accept": false, - "reason": "Both involve models failing, but one is Flash Attention 2 support and the other is a model-loading regression for Llama 3.2 Vision." + "reason": "Both concern model loading/attention support, but they are different models with different compatibility problems." }, { "left": "issue:29942", "right": "issue:35141", "accept": false, - "reason": "Flash Attention 2 test failures are unrelated to embedding resizing reinitializing lm_head when weights are untied." + "reason": "Flash Attention test failures are unrelated to token-embedding resizing and `post_init` reinitialization." }, { "left": "issue:38175", "right": "issue:44493", "accept": false, - "reason": "Different symptoms and likely causes: zero probabilities in SigLIP2 vs unexpected position-id keys in many models." + "reason": "Unexpected zero probabilities in SigLIP2 is a model-specific output issue; unexpected position-id keys is a different serialization/config problem." }, { "left": "issue:29870", "right": "issue:32090", "accept": false, - "reason": "Bart generate() UnboundLocalError and Trainer _gpu_broadcast_one NoneType error are separate execution paths." + "reason": "BART `generate()` decoder mask bug and Trainer `_gpu_broadcast_one` NoneType error are separate runtime paths." }, { - "left": "issue:43278", - "right": "issue:43329", + "left": "issue:44448", + "right": "issue:44977", "accept": false, - "reason": "Embedding dtype drift in eval is unrelated to undefined variables in the video branch of multimodal token counting." + "reason": "Pegasus v4/v5 output divergence and Qwen3.5 flash-attention generation failure are different model regressions." }, { - "left": "issue:44448", - "right": "issue:44977", + "left": "issue:43278", + "right": "issue:43329", "accept": false, - "reason": "Different regressions: Pegasus output changes across v4/v5 vs Qwen3.5 flash-attention generation failure." + "reason": "Embedding dtype drift in evaluate is unrelated to multimodal token counting bugs in the video branch." }, { "left": "issue:45920", "right": "issue:46192", "accept": false, - "reason": "Tokenizer ID regression and return_tensors being ignored are distinct preprocessing bugs." + "reason": "AutoTokenizer wrong token IDs is a tokenizer regression; `return_tensors` being ignored with explicit `text_kwargs` is an input-processing API bug." }, { "left": "issue:45290", "right": "issue:45381", "accept": false, - "reason": "Chat-template crash with tool-call assistant messages is unrelated to Qwen2.5-VL video vision_position_ids." + "reason": "Chat template crashes on tool-call assistant messages; the other is incorrect video `vision_position_ids` in Qwen2.5-VL." }, { "left": "issue:10000", "right": "issue:15354", "accept": false, - "reason": "German DistilBertModel issue and GeneratorExp TorchScript export failure are completely different model/runtime problems." + "reason": "German DistilBertModel and a GeneratorExp/torch.jit.script export error are unrelated model/export failures." }, { "left": "issue:44977", "right": "issue:45229", "accept": false, - "reason": "Flash-attention generation bug and multi-GPU CUDA OOM are not the same code-path problem." + "reason": "Flash-attention generation issue for Qwen3.5 is not the same as multi-GPU CUDA OOM for Gemma4 31B." }, { - "left": "issue:43278", - "right": "issue:43381", + "left": "issue:29870", + "right": "issue:33290", "accept": false, - "reason": "Eval-time embedding dtype mismatch is unrelated to gradient checkpointing being disallowed in eval mode." + "reason": "Decoder attention mask generation bug is unrelated to Adafactor/deepspeed OOM." }, { - "left": "issue:43927", - "right": "issue:45698", + "left": "issue:43278", + "right": "issue:43381", "accept": false, - "reason": "Both involve save/load behavior, but one loses custom token IDs in config while the other resolves the wrong custom module after save_pretrained." + "reason": "BF16/FP32 dtype drift in evaluate is unrelated to gradient checkpointing being unusable in eval mode." }, { - "left": "issue:29870", - "right": "issue:33290", + "left": "issue:44855", + "right": "issue:46399", "accept": false, - "reason": "A generate-time decoder_attention_mask bug is unrelated to Deepspeed Adafactor OOM." + "reason": "Python 3.13 `IndentationError`/`torch.jit.script` parsing issue is unrelated to Qwen3.5 rotary-pos-emb registration failure." }, { - "left": "issue:44855", - "right": "issue:46399", + "left": "issue:43927", + "right": "issue:45698", "accept": false, - "reason": "Both mention TorchScript/decorator issues, but they are different failures in different models and with different root causes." + "reason": "Custom token IDs lost after save/load causing generation IndexError is a different persistence bug from loading the wrong custom module after `save_pretrained`." }, { "left": "issue:45290", "right": "issue:46655", "accept": false, - "reason": "Tool-call chat template crash and SmolVLM video padding bug are separate multimodal preprocessing issues." + "reason": "Chat-template tool-call crash and SmolVLM video padding miscalculation affect different subsystems." }, { "left": "issue:40444", "right": "issue:43526", "accept": false, - "reason": "Iterabledataset multi-image finetuning failure is unrelated to BeitImageProcessorFast reduce_labels truncating labels." + "reason": "IterableDataset multi-image finetuning failure and BeitImageProcessorFast label reduction bug are unrelated multimodal preprocessing issues." }, { "left": "issue:37428", "right": "issue:42175", "accept": false, - "reason": "ImportError for a flash-attention utility and missing tensorflow in a torch install extra are unrelated packaging/runtime issues." + "reason": "Flash-support import regression and missing TensorFlow in a torch-only install are different packaging/import problems." }, { "left": "issue:44857", "right": "issue:45588", "accept": false, - "reason": "Both touch flash-attention-adjacent code, but one is an AMP loss crash and the other is an AttributeError on s_aux=None in sink-less models." + "reason": "These are both crashes, but on different components: LwDetrImageLoss AMP/CUDA vs flash_attention integration with `s_aux=None`." }, { "left": "issue:32090", "right": "issue:33290", "accept": false, - "reason": "Trainer broadcast NoneType and Adafactor/Deepspeed OOM are distinct training failures." + "reason": "Trainer broadcast NoneType error and deepspeed Adafactor OOM are distinct training/runtime issues." }, { "left": "issue:46421", "right": "issue:46858", "accept": false, - "reason": "Cache lazy initialization causing recompilation is different from seq_length returning a device tensor and breaking static-cache arithmetic." + "reason": "Lazy cache initialization causing recompilation and `_prefill` seq-length device-tensor arithmetic are different cache/compile bugs." }, { "left": "issue:44561", "right": "issue:44779", "accept": false, - "reason": "Removal of is_torch_fx_available breaks trust_remote_code; DeepSeek tokenizer regression is a separate tokenizer bug." - }, - { - "left": "issue:43873", - "right": "issue:45229", - "accept": false, - "reason": "Quantization offloading malfunction and Gemma4 multi-GPU OOM are different memory-management problems." + "reason": "Removal of `is_torch_fx_available` breaks trust-remote-code models, while DeepSeek tokenizer regression is a separate tokenizer behavior change." }, { "left": "issue:44448", "right": "issue:44466", "accept": false, - "reason": "Output divergence for Pegasus across versions is unrelated to inconsistent lm_head.weight serialization tied to device placement." + "reason": "Model output divergence across versions and inconsistent `lm_head.weight` serialization are both v5 regressions, but they concern different mechanisms and are not the same bug." }, { - "left": "issue:43976", - "right": "issue:45440", + "left": "issue:43873", + "right": "issue:45229", "accept": false, - "reason": "Python-version compatibility issue for Transformers 5.1.0 is unrelated to DeepSeekV3MoE diverging from remote implementation." + "reason": "Quantization offloading behavior and Gemma4 multi-GPU OOM are unrelated memory-management issues." }, { "left": "issue:25251", "right": "issue:32090", "accept": false, - "reason": "Pipeline top_k nesting change has nothing to do with the Trainer _gpu_broadcast_one NoneType error." + "reason": "Pipeline `top_k` nesting behavior and Trainer broadcast NoneType error are unrelated API/runtime bugs." + }, + { + "left": "issue:43976", + "right": "issue:45440", + "accept": false, + "reason": "Python version compatibility for Transformers 5.1.0 is unrelated to native DeepseekV3MoE diverging from the remote implementation." }, { "left": "issue:42490", "right": "issue:43873", "accept": false, - "reason": "Local-path import warning with trailing slash and quantization offloading failure are not the same bug." + "reason": "Local-path import warning and quantization offloading behavior are different loading/offload concerns with no shared code-path evidence." } ] }, @@ -10141,8 +10238,8 @@ "issue:43381|issue:43701", "issue:41720|issue:45229", "issue:45588|issue:45823", - "issue:29127|issue:34567", "issue:33357|issue:39692", + "issue:29127|issue:34567", "issue:44977|issue:45563", "issue:45198|issue:45803", "issue:44361|issue:45198", @@ -10163,188 +10260,188 @@ "issue:44265|issue:45198", "issue:29127|issue:33357", "issue:43065|issue:43994", - "issue:45245|issue:45563", "issue:43065|issue:45412", + "issue:45245|issue:45563", "issue:43575|issue:43701" ], - "prepared_review_unit_hash": "5bf2f1423f9932f6a34371ce92a65320cb72d48c95fa7aaafa240dba94038176", + "prepared_review_unit_hash": "8a94729b7886b94a8f6ff418a569b9ef4bd76d3a418c310ab3e1f52f6a9222cb", "component_budget": { "node_count": 502, "item_count": 502, - "soft_pair_count": 1842, - "serialized_chars": 956182, - "estimated_input_tokens": 239046, - "estimated_eval_tokens": 478348 + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 31, - "serialized_chars": 36570, - "estimated_input_tokens": 9143, - "estimated_eval_tokens": 18542 + "serialized_chars": 36562, + "estimated_input_tokens": 9141, + "estimated_eval_tokens": 18538 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster is very heterogeneous: the soft pairs mostly share only a broad subsystem theme (tokenization, generation, multimodal, or memory) but not the same concrete bug. I reject all proposed duplicates and see no PRs in the cluster. The most representative issue is the Wav2Vec2 save/tokenization failure, but even that does not truly unify the set.", - "confidence": 0.96, - "canonical_issue_reason": "issue:45198 is the most representative concrete bug report in the set and appears in multiple soft-similarity candidates, but the overall cluster is too mixed for a strong duplicate canonical.", + "summary": "This cluster is mostly a grab-bag of unrelated bug reports that only share broad wording around model loading, generation, tokenizers, or attention. None of the suggested soft links look like true duplicates; they differ in model family, failing code path, or symptom.", + "confidence": 0.68, + "canonical_issue_reason": "issue:46693 is the most broadly scoped and currently active report in the set, with the clearest regression framing and enough detail to serve as an anchor issue, though the cluster itself is heterogeneous.", "canonical_pr_reason": null, - "best_issue_reason": "issue:45198 is the strongest single issue candidate by specificity and recurrence in the soft graph; it is a real runtime failure with clear repro details, even though it is not a true umbrella duplicate.", + "best_issue_reason": "issue:46693 is the strongest standalone issue here: it is specific enough to reproduce, broad enough to represent a real regression class, and has the most obvious tracking value among the listed issues.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:44568", "right": "issue:45491", "accept": false, - "reason": "Different failures: missing BOS/EOS insertion in one tokenizer vs NaN embeddings from sliding-window attention in Gemma3." + "reason": "Different failures: BOS/EOS token insertion for mdeberta-v3 vs Gemma3 NaN embeddings with mixed-length GPU batching." }, { "left": "issue:43278", "right": "issue:43575", "accept": false, - "reason": "Unrelated bugs: BF16->FP32 dtype change during eval vs tensor-parallel loading OOM for a large Qwen2 model." + "reason": "Unrelated bugs: BF16->FP32 dtype drift in eval vs tensor-parallel loading OOM for Qwen2-57B." }, { "left": "issue:44871", "right": "issue:44977", "accept": false, - "reason": "Different concrete problems: eos_token_id config mismatch vs Qwen3.5 flash-attention generation failure." + "reason": "Gemma eos_token_id config mismatch is not the same as Qwen3.5 flash-attention generation failure." }, { "left": "issue:43381", "right": "issue:43701", "accept": false, - "reason": "Eval-mode gradient checkpointing restriction and resume_from_checkpoint key mismatch are separate training-state issues." + "reason": "Gradient checkpointing in eval mode and resume_from_checkpoint key mismatch are different code paths and symptoms." }, { "left": "issue:41720", "right": "issue:45229", "accept": false, - "reason": "Both are GPU memory/device-map related, but one is a cudaErrorAssert on Qwen3 and the other is multi-GPU OOM on Gemma4." + "reason": "Both involve large-model inference, but one is a cudaErrorAssert on Qwen3 auto device mapping and the other is a CUDA OOM on Gemma4 multi-GPU." }, { "left": "issue:45588", "right": "issue:45823", "accept": false, - "reason": "Flash-attention integration crash on sink-less models is not the same as Gemma4 PLE device mismatch during forward." + "reason": "Flash-attention crash on s_aux=None is distinct from Gemma4 PLE device mismatch during forward." }, { - "left": "issue:29127", - "right": "issue:34567", + "left": "issue:33357", + "right": "issue:39692", "accept": false, - "reason": "LayoutLMv3 error-message clarity and TrainerState token-count tracking are unrelated." + "reason": "MacOS bus error for a community CLIP model is not the same as SigLIP2 documentation/example errors." }, { - "left": "issue:33357", - "right": "issue:39692", + "left": "issue:29127", + "right": "issue:34567", "accept": false, - "reason": "Both involve CLIP/SigLIP-ish vision models, but one is a MacOS bus error and the other is a broken documentation example." + "reason": "LayoutLMv3 error-message clarity and TrainerState num_input_tokens_seen not updating are unrelated issues." }, { "left": "issue:44977", "right": "issue:45563", "accept": false, - "reason": "Flash-attention generation regression and a stale warning about num_return_sequences are not the same bug." + "reason": "Qwen3.5 flash-attention generation breakage is unrelated to a stale warning in paged generate()." }, { "left": "issue:45198", "right": "issue:45803", "accept": false, - "reason": "Both mention tokenization/errors, but one is a Wav2Vec2 save/tokenization failure and the other is a bare-except swallowing errors in Fuyu batch conversion." + "reason": "Wav2Vec2 save/tokenization failures and a bare-except swallowing errors in FuyuBatchFeature are different bugs." }, { "left": "issue:44361", "right": "issue:45198", "accept": false, - "reason": "MLukeTokenizer AttributeError on tasks is a different tokenizer failure from Wav2Vec2 save_pretrained/tokenization failure." + "reason": "MLukeTokenizer AttributeError is unrelated to Wav2Vec2 save_pretrained/tokenization failure." }, { "left": "issue:44186", "right": "issue:45198", "accept": false, - "reason": "LayoutLMv2 NER/padding crash and Wav2Vec2 save/tokenization fail are distinct tokenizer bugs." + "reason": "LayoutLMv2Tokenizer NER/padding crash does not match the Wav2Vec2 save/tokenization bug." }, { "left": "issue:43653", "right": "issue:44743", "accept": false, - "reason": "BigBirdTokenizer empty decode output and Qwen3.5 recurrent-state reset are unrelated issues." + "reason": "BigBirdTokenizer special-token registration and Qwen3.5 recurrent-state reset are separate components with different root causes." }, { "left": "issue:43611", "right": "issue:46121", "accept": false, - "reason": "base_model_prefix loading regression and rope-validation TypeError are different model/config handling bugs." + "reason": "base_model_prefix loading regression and rope-parameter JSON list type error are different failures." }, { "left": "issue:29870", "right": "issue:34689", "accept": false, - "reason": "BART generate decoder_attention_mask initialization bug is unrelated to Llama 3.2 Vision loading regression." + "reason": "Bart generate decoder_attention_mask bug and Llama 3.2 Vision loading regression are unrelated." }, { "left": "issue:45372", "right": "issue:45588", "accept": false, - "reason": "Gemma4 processor import failure and flash-attention s_aux=None crash are separate problems." + "reason": "Both touch Gemma4, but one is an import error for ReasoningEffort during processor loading and the other is a flash_attention.py AttributeError on s_aux=None." }, { "left": "issue:42915", "right": "issue:42994", "accept": false, - "reason": "Both are quantization-adjacent, but one is FineGrainedFP8Config failure on Qwen3MoE and the other is quantized model saving failure." + "reason": "Qwen3Moe FineGrainedFP8Config failure and quantized model saving failure are different bugs." }, { "left": "issue:43756", "right": "issue:45440", "accept": false, - "reason": "Smollm3 RoPE-layer discrepancy and DeepseekV3MoE divergence from remote implementation are different model parity bugs." + "reason": "Smollm3 RoPE-layer mismatch and DeepseekV3MoE divergence from remote implementation are not the same issue." }, { "left": "issue:43653", "right": "issue:44442", "accept": false, - "reason": "BigBirdTokenizer special-token decode bug and AutoTokenizer load failure for FastSpeech2ConformerTokenizer are unrelated." + "reason": "BigBirdTokenizer special-token bug and AutoTokenizer loading FastSpeech2ConformerTokenizer are unrelated." }, { "left": "issue:29942", "right": "issue:30064", "accept": false, - "reason": "Failing Flash Attention 2 tests and void segmentation-map processing are unrelated." + "reason": "Flash Attention 2 test failures and void segmentation-map processing are different subsystems and symptoms." }, { "left": "issue:44206", "right": "issue:44479", "accept": false, - "reason": "Unsupported center arg in LasrFeatureExtractor is not the same as the video-input regression across Qwen VL variants." + "reason": "Unsupported center arg crash in LasrFeatureExtractor is not the same as the Qwen video-input regression." }, { "left": "issue:45372", "right": "issue:45406", "accept": false, - "reason": "Both touch Gemma4 processor loading, but one is an import dependency issue and the other is a missing _tokenizer attribute in serve." + "reason": "Both are Gemma4-related, but the first is a processor import/loading problem while the second is a missing _tokenizer attribute in transformers serve." }, { "left": "issue:43412", "right": "issue:45588", "accept": false, - "reason": "Executorch export guard/erfinv support and flash-attention sink-less crash are separate code paths." + "reason": "Gemma3n export/training-guard and erfinv support issues are unrelated to the flash_attention s_aux=None crash." }, { "left": "issue:45459", "right": "issue:45712", "accept": false, - "reason": "Protobuf exception masking tokenizer errors is unrelated to leftover dummy classes leaking into dir(transformers)." + "reason": "Tokenizer exception masking when protobuf is missing and leftover dummy classes in dummy_pt_objects are different maintenance bugs." }, { "left": "issue:45491", "right": "issue:46693", "accept": false, - "reason": "NaN embeddings in mixed-length Gemma3 batching is a correctness bug, while 5.2.0 FA2 long-context regression is a performance issue." + "reason": "These are both FlashAttention/Gemma-adjacent, but one is a NaN-embedding correctness bug and the other is a long-context performance regression; not the same underlying defect." }, { "left": "issue:44265", @@ -10356,31 +10453,31 @@ "left": "issue:29127", "right": "issue:33357", "accept": false, - "reason": "LayoutLMv3 error messaging and MacOS CLIP bus error are unrelated." + "reason": "LayoutLMv3 error-message clarity and CLIP bus error on MacOS are unrelated." }, { "left": "issue:43065", "right": "issue:43994", "accept": false, - "reason": "Dummy Conv2d in Sam3PixelDecoder and SigLIP2 nonsensical outputs are different bugs in different models." + "reason": "Dummy Conv2d in Sam3PixelDecoder and SigLIP2 nonsensical AutoModel/pipeline outputs are different model issues." }, { - "left": "issue:45245", - "right": "issue:45563", + "left": "issue:43065", + "right": "issue:45412", "accept": false, - "reason": "Category-cardinality RuntimeError and a stale warning in paged generate() are unrelated." + "reason": "Sam3PixelDecoder dummy-layer cleanup is unrelated to RT-DETR memory not being released." }, { - "left": "issue:43065", - "right": "issue:45412", + "left": "issue:45245", + "right": "issue:45563", "accept": false, - "reason": "Dummy Sam3PixelDecoder class issue and RT-DETR memory leak are unrelated." + "reason": "Category-count runtime error and paged-generate stale warning are different bug classes." }, { "left": "issue:43575", "right": "issue:43701", "accept": false, - "reason": "Tensor-parallel OOM when loading Qwen2-57B and resume_from_checkpoint key mismatch are not the same underlying bug." + "reason": "Tensor-parallel OOM when loading Qwen2-57B and resume_from_checkpoint key mismatch are unrelated." } ] }, @@ -10466,15 +10563,15 @@ "issue:44464|issue:45584", "issue:42915|issue:43931", "issue:43526|issue:45834", - "issue:43526|issue:46693", "issue:43824|issue:44279", + "issue:43526|issue:46693", "issue:27957|issue:29127", "issue:27957|issue:33290", "issue:45250|issue:46640", "issue:43994|issue:46369", "issue:33290|issue:34689", - "issue:43526|issue:44186", "issue:33290|issue:35141", + "issue:43526|issue:44186", "issue:43334|issue:43531", "issue:46121|issue:46399", "issue:45588|issue:45600", @@ -10483,208 +10580,208 @@ "issue:43646|issue:44077", "issue:43761|issue:45859" ], - "prepared_review_unit_hash": "41ac83fe73f9195ed947df375bf7452832147ee04e67db2ef9bcbfdaaf1ecd8e", + "prepared_review_unit_hash": "9f918b078dd5f69f359e788795648790464e4debd3f2477ee1d8d0abd4ea2cd9", "component_budget": { "node_count": 502, "item_count": 502, - "soft_pair_count": 1842, - "serialized_chars": 956182, - "estimated_input_tokens": 239046, - "estimated_eval_tokens": 478348 + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 29, - "serialized_chars": 35674, - "estimated_input_tokens": 8919, - "estimated_eval_tokens": 18094 + "serialized_chars": 35677, + "estimated_input_tokens": 8920, + "estimated_eval_tokens": 18096 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is heterogeneous: most items are distinct Transformers bugs across model loading, tokenizer behavior, quantization, and attention/performance paths. I did not find any soft-pair duplicates that look like the same concrete bug/change.", - "confidence": 0.26, - "canonical_issue_reason": "issue:43873 is the broadest, most active report in the set (open, many discussion entries and inbound references) and is a plausible anchor for a quantization/offloading-related bug cluster.", + "summary": "The set is a grab bag of unrelated Transformers issues with only superficial overlap in terms like Flash Attention, OOM, or loading errors, so I rejected all soft pairs as duplicates. If one issue must serve as the representative entry, 43873 is the strongest standalone report, but it is not a duplicate of the others.", + "confidence": 0.94, + "canonical_issue_reason": "Issue 43873 is the best standalone representative here: it is detailed, actively discussed, and has inbound references, but the overall set is not a true duplicate cluster.", "canonical_pr_reason": null, - "best_issue_reason": "issue:43873 has the strongest triage signal: active discussion, inbound references, and a general failure mode that could subsume multiple related runtime-memory complaints better than the narrower model-specific reports.", + "best_issue_reason": "Issue 43873 has the strongest combination of specificity, discussion activity, and external references, making it the most usable issue entry in this mixed set.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43066", "right": "issue:44462", "accept": false, - "reason": "Different problems: tokenizer decoder-type mismatch in v5 versus AutoTokenizer ignoring tokenizer.json from a repo." + "reason": "Different tokenizer bugs: v5 decoder-type mismatch vs AutoTokenizer ignoring repo tokenizer.json." }, { "left": "issue:45440", "right": "issue:45593", "accept": false, - "reason": "Unrelated code paths and symptoms: DeepSeekV3 MoE divergence versus D-FINE losing auxiliary losses when denoising is off." + "reason": "Different model-specific failures: DeepSeekV3 implementation divergence vs D-FINE auxiliary-loss behavior." }, { "left": "issue:45372", "right": "issue:45459", "accept": false, - "reason": "Both mention import-related failures, but they target different modules and root causes: mistral_common import regression versus protobuf masking tokenizer errors." + "reason": "Import/dependency error vs tokenizer error masking; same area, but not the same underlying bug." }, { "left": "issue:42490", "right": "issue:43381", "accept": false, - "reason": "No shared bug: local-path trailing-slash warning versus gradient checkpointing being disallowed in eval mode." + "reason": "Local-path import warning and eval-mode gradient checkpointing are unrelated." }, { "left": "issue:29127", "right": "issue:35141", "accept": false, - "reason": "LayoutLMv3 box validation messaging and token-embedding reinitialization are separate bugs in different subsystems." + "reason": "LayoutLMv3 box-validation messaging and embedding reinitialization are different code paths." }, { "left": "issue:43421", "right": "issue:45491", "accept": false, - "reason": "Tokenizer post-processor runtime updates and Gemma3 NaN embeddings are distinct issues with different failure modes." + "reason": "Tokenizer post-processor sync bug and Gemma3 NaNs are unrelated." }, { "left": "issue:43994", "right": "issue:45412", "accept": false, - "reason": "Different models and different failure classes: nonsensical SigLIP2 outputs versus RT-DETR memory not being released." + "reason": "Different symptoms and models: wrong outputs vs memory not being released." }, { "left": "issue:45412", "right": "issue:46369", "accept": false, - "reason": "Both are memory-related, but one is a deletion/GC leak in RT-DETR while the other is a runtime overflow in sam3-litetext." + "reason": "Both are memory-related, but they affect different models and appear to be different bugs." }, { "left": "issue:43575", "right": "issue:44387", "accept": false, - "reason": "Both are OOM reports, but one is tensor-parallel load OOM for a specific model and the other is a broader int4 quantization reserved-memory regression." + "reason": "Both mention OOM, but one is tensor-parallel loading and the other is int4 quantization memory growth." }, { "left": "issue:44442", "right": "issue:45072", "accept": false, - "reason": "AutoTokenizer loading failure and dtype mismatches in inference are unrelated bugs." + "reason": "Tokenizer loading failure and bf16 dtype mismatch are not the same issue." }, { "left": "issue:44464", "right": "issue:45584", "accept": false, - "reason": "Chunked generation with compiled forward and Whisper empty-transcription handling are different generation-path bugs." + "reason": "Compiled-forward chunked generation inconsistency is unrelated to Whisper empty-transcription handling." }, { "left": "issue:42915", "right": "issue:43931", "accept": false, - "reason": "Qwen3Moe FineGrainedFP8Config failure and Qwen3-VL weight-shape mismatch are separate model/config issues." + "reason": "FineGrainedFP8Config failure and Qwen3-VL weight-shape mismatch are distinct bugs." }, { "left": "issue:43526", "right": "issue:45834", "accept": false, - "reason": "BeitImageProcessorFast label reduction bug and Kosmos2.5 OCR index error do not share the same code path." + "reason": "BEiT label reduction and Kosmos2.5 OCR index errors are different processor bugs." }, { - "left": "issue:43526", - "right": "issue:46693", + "left": "issue:43824", + "right": "issue:44279", "accept": false, - "reason": "Label reduction in image preprocessing is unrelated to FA2 long-context performance regression." + "reason": "Specific missing-class ImportError and a generic dependency issue are too different to merge." }, { - "left": "issue:43824", - "right": "issue:44279", + "left": "issue:43526", + "right": "issue:46693", "accept": false, - "reason": "ImportError for a specific Qwen2.5-VL class is not the same as a generic dependency issue report." + "reason": "Image-processor label handling and FA2 long-context regression are unrelated." }, { "left": "issue:27957", "right": "issue:29127", "accept": false, - "reason": "Flash Attention 2 support for XLM-Roberta and LayoutLMv3 error-message clarity are unrelated." + "reason": "Flash Attention 2 support and LayoutLMv3 box-error clarity do not share the same underlying bug." }, { "left": "issue:27957", "right": "issue:33290", "accept": false, - "reason": "FA2 model support and Adafactor+DeepSpeed OOM are different bugs." + "reason": "FA2 support issue vs DeepSpeed+AdaFactor OOM are unrelated." }, { "left": "issue:45250", "right": "issue:46640", "accept": false, - "reason": "Both are broadly about Flash Attention 2, but the titles are too generic to show the same concrete bug/change." + "reason": "Both are generic Flash Attention 2 titles, but there is no concrete evidence they describe the same bug." }, { "left": "issue:43994", "right": "issue:46369", "accept": false, - "reason": "Different model families and different symptoms; no evidence of a shared underlying defect." + "reason": "Different model families and failure modes: nonsensical outputs vs memory overflow." }, { "left": "issue:33290", "right": "issue:34689", "accept": false, - "reason": "Adafactor OOM in DeepSpeed and Llama 3.2 Vision loading regression are unrelated." + "reason": "Optimizer OOM and model-loading regression are unrelated." }, { - "left": "issue:43526", - "right": "issue:44186", + "left": "issue:33290", + "right": "issue:35141", "accept": false, - "reason": "BEiT image label reduction and LayoutLMv2 tokenizer crashes are different tokenizer/processor issues." + "reason": "Optimizer memory blowup and embedding reinitialization are different problems." }, { - "left": "issue:33290", - "right": "issue:35141", + "left": "issue:43526", + "right": "issue:44186", "accept": false, - "reason": "OOM with Adafactor/DeepSpeed does not match embedding reinitialization during resize." + "reason": "BEiT fast processor label reduction is unrelated to LayoutLMv2 tokenizer padding/NER crashes." }, { "left": "issue:43334", "right": "issue:43531", "accept": false, - "reason": "Qwen3-VL missing pad_token_id and Qwen3-MoE sliding-window behavior are distinct bugs." + "reason": "Missing pad_token_id in Qwen3-VL and Qwen3-MoE sliding-window issues are different failures." }, { "left": "issue:46121", "right": "issue:46399", "accept": false, - "reason": "Rope-parameter validation TypeError and missing decorator for apply_rotary_pos_emb registration are separate rope-related failures." + "reason": "Rope-validation TypeError and missing-decorator registration error are unrelated." }, { "left": "issue:45588", "right": "issue:45600", "accept": false, - "reason": "Flash-attention sink-less crash and auto_mappings stale config references are unrelated." + "reason": "Flash-attention s_aux crash and Sam3LiteText auto-mapping cleanup are not the same bug." }, { "left": "issue:43381", "right": "issue:43873", "accept": false, - "reason": "Gradient checkpointing in eval mode and quantization offloading failures are separate issues." + "reason": "Eval-mode gradient checkpointing and quantization/offloading behavior are different issues." }, { "left": "issue:44514", "right": "issue:45381", "accept": false, - "reason": "Batched chat-template padding crash and incorrect vision_position_ids are different Qwen2.5-VL problems." + "reason": "Both are Qwen2.5-VL bugs, but batched chat-template padding and video vision_position_ids are distinct code paths." }, { "left": "issue:43646", "right": "issue:44077", "accept": false, - "reason": "Custom model initialization breakage in Transformers 5.0 and patchtsmixer post_init policy are not the same bug." + "reason": "Custom model initialization breakage and patchtsmixer post_init constraints are not the same concrete defect." }, { "left": "issue:43761", "right": "issue:45859", "accept": false, - "reason": "CLIP hidden_states regression and Qwen3.5 MoE CPU-offload incompatibility are unrelated." + "reason": "CLIPVisionModel hidden_states regression and Qwen3.5 MoE CPU-offload incompatibility are unrelated." } ] }, @@ -10780,26 +10877,26 @@ "issue:44190|issue:44496", "issue:46829|issue:46932", "issue:44987|issue:46178", - "issue:44442|issue:45803", "issue:43761|issue:44188", + "issue:44442|issue:45803", "issue:43761|issue:44079", "issue:43931|issue:45216", "issue:43232|issue:43388" ], - "prepared_review_unit_hash": "6a3960c9f05b7478572753e57f7bdda3f8ad0e05a59dce1125ce8a6816499a0e", + "prepared_review_unit_hash": "28fda421d669eb707715e7ea3eb7689eff203bb3765afe84264bb31e23c02b3f", "component_budget": { "node_count": 502, "item_count": 502, - "soft_pair_count": 1842, - "serialized_chars": 956182, - "estimated_input_tokens": 239046, - "estimated_eval_tokens": 478348 + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 27, - "serialized_chars": 36182, + "serialized_chars": 36183, "estimated_input_tokens": 9046, "estimated_eval_tokens": 18348 }, @@ -10808,174 +10905,168 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster is made up of mostly unrelated issues that only share broad surface-level themes (generation, loading, configs, vision/audio). None of the soft pairs look like the same underlying bug or change, so all should remain separate.", - "confidence": 0.97, - "canonical_issue_reason": "Issue #43873 is the most broadly scoped and actively discussed item in the set, with substantial conversation and references, so it is the best standalone representative if one issue must be chosen.", + "summary": "This set is mostly a collection of unrelated bugs across different models, generation paths, loading code, and docs/backend issues. None of the soft pairs looks like the same underlying defect, so I would not collapse any of them as duplicates.", + "confidence": 0.98, + "canonical_issue_reason": "Issue 43873 is the most active open bug in the set and the best available representative, even though the cluster itself does not form a true duplicate group.", "canonical_pr_reason": null, - "best_issue_reason": "Issue #43873 is detailed, high-activity, and still open, making it the strongest single issue to keep as the representative canonical item among this mixed set.", + "best_issue_reason": "43873 has the strongest ongoing signal (open, high discussion, inbound references) and is the most suitable issue to keep as the placeholder canonical record.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:44977", "right": "issue:45245", "accept": false, - "reason": "Different failures and code paths: Qwen3.5 flash-attention generation vs a category-count runtime error." + "reason": "Flash-attention generation failure and a save_pretrained checkpoint regression are different bugs with different code paths." }, { "left": "issue:44743", "right": "issue:44811", "accept": false, - "reason": "Both mention sequence behavior, but one is Qwen recurrent state reset and the other is Whisper decoding skip_special_tokens handling." + "reason": "Recurrent-state reset with cache is unrelated to Whisper batch_decode skipping special tokens." }, { "left": "issue:44442", "right": "issue:44898", "accept": false, - "reason": "Tokenizer loading failure and Perceiver image-classification interpolation are unrelated bugs." + "reason": "Tokenizer loading failure and Perceiver image interpolation failure affect different subsystems." }, { "left": "issue:43262", "right": "issue:44206", "accept": false, - "reason": "Audio chat-template sample-rate defaults and LaserFeatureExtractor argument handling are different components and defects." + "reason": "Audio chat-template sampling-rate defaults and a feature extractor argument crash are separate issues." }, { "left": "issue:42490", "right": "issue:44743", "accept": false, - "reason": "Local-path import warning is unrelated to Qwen recurrent-state reset during cached generation." + "reason": "Local-path trailing-slash import warning is not the same as the cache/recurrent-state bug." }, { "left": "issue:45003", "right": "issue:45440", "accept": false, - "reason": "Unsafe sys.modules access is a different issue from DeepseekV3MoE implementation divergence." + "reason": "Unsafe sys.modules access and DeepSeekV3MoE divergence are different problems in different areas." }, { "left": "issue:43720", "right": "issue:44442", "accept": false, - "reason": "Packed-weight loading under accelerate and AutoTokenizer loading failure are separate loading problems." + "reason": "BitNet packed-weight loading and FastSpeech tokenizer loading failure do not share the same code path." }, { "left": "issue:42175", "right": "issue:46234", "accept": false, - "reason": "Missing TensorFlow backend in a pip extra install and a missing import_utils availability helper are not the same bug." + "reason": "Missing tensorflow backend in extras and a missing import_utils availability helper are distinct packaging/API issues." }, { "left": "issue:27957", "right": "issue:29870", "accept": false, - "reason": "Both involve model generation paths, but they concern different models and distinct errors." + "reason": "XLMRoberta Flash Attention 2 support and BART generate decoder_attention_mask assignment are unrelated." }, { "left": "issue:43316", "right": "issue:43525", "accept": false, - "reason": "Both are config-field issues, but Gemma3TextConfig API differences and Llama4 pad_token_id absence are separate model-specific bugs." + "reason": "Gemma3TextConfig API mismatch and Llama4Config missing pad_token_id are different config-level bugs." }, { "left": "issue:43493", "right": "issue:43827", "accept": false, - "reason": "SigLIP2 implementation mismatch and pipeline() documentation fallout are unrelated." + "reason": "SigLIP2 implementation discrepancy and a docs reference to pipeline() are not the same change." }, { "left": "issue:44062", "right": "issue:45907", "accept": false, - "reason": "AddedToken keyword duplication and list_repo_files revision forwarding are different API bugs." + "reason": "AddedToken special-argument duplication and list_repo_files revision forwarding are unrelated API bugs." }, { "left": "issue:44077", "right": "issue:44479", "accept": false, - "reason": "PatchTSMixer post_init policy and Qwen video-input regression affect different models and code paths." + "reason": "PatchTSMixer post_init validation and Qwen video-input regression are different model-specific problems." }, { "left": "issue:45325", "right": "issue:46697", "accept": false, - "reason": "Both are preprocessing/positioning issues, but Qwen2.5-VL rope indexing and Swin2SR padding are not the same defect." + "reason": "Qwen2.5-VL rope-index scaling and Swin2SR over-padding are separate preprocessing/positioning bugs." }, { "left": "issue:43873", "right": "issue:44871", "accept": false, - "reason": "Quantized offloading behavior and Gemma-3 eos_token_id config inconsistency are unrelated." + "reason": "Quantized offloading failure and Gemma-3 eos_token_id inconsistency are unrelated." }, { "left": "issue:44855", "right": "issue:45137", "accept": false, - "reason": "Python 3.13 import parsing failure and DeepSpeed ZeRO3 deque exhaustion are unrelated runtime errors." + "reason": "Python 3.13 import parsing error and DeepSpeed ZeRO3 deque underflow are different failure modes." }, { "left": "issue:43643", "right": "issue:44913", "accept": false, - "reason": "Missing fields after trust_remote_code loading and GPTNeoX rotary_pct reset on reload are different serialization/config bugs." + "reason": "trust_remote_code field loss and GPTNeoX rotary_pct reload fallback are different serialization/config bugs." }, { "left": "issue:43749", "right": "issue:43931", "accept": false, - "reason": "FSDP CPU RAM efficient loading and Qwen3-VL weight-shape mismatch are separate model loading failures." - }, - { - "left": "issue:43316", - "right": "issue:45759", - "accept": false, - "reason": "Composite-config API discrepancy and AutoModelForCausalLM text_config unwrapping are related only loosely; they are different bugs in different model families." + "reason": "FSDP CPU RAM efficient loading breakage and Qwen3-VL weight-shape mismatch are not the same issue." }, { "left": "issue:44190", "right": "issue:44496", "accept": false, - "reason": "A local dataset loading script failure is unrelated to an unrecognized model_type config error." + "reason": "Local dataset loading in a training script and an unrecognized model config error are unrelated." }, { "left": "issue:46829", "right": "issue:46932", "accept": false, - "reason": "VideoPrism hidden_states output wiring and Gemma4 quantized embedders dtype casting are unrelated model issues." + "reason": "VideoPrism hidden_states output bug and Gemma4 quantized embedder dtype casting are different model paths." }, { "left": "issue:44987", "right": "issue:46178", "accept": false, - "reason": "Loading a specific model under transformers>=5.1.0 and backend mapping for tensorflow_text are different problems." + "reason": "Model loading failure for a specific repo and tensorflow_text backend mapping are unrelated problems." }, { - "left": "issue:44442", - "right": "issue:45803", + "left": "issue:43761", + "right": "issue:44188", "accept": false, - "reason": "Tokenizer loading and FuyuBatchFeature exception handling are different code paths." + "reason": "CLIPVisionModel hidden_states regression and torch.compile attention-kernel divergence are different bugs." }, { - "left": "issue:43761", - "right": "issue:44188", + "left": "issue:44442", + "right": "issue:45803", "accept": false, - "reason": "CLIPVision hidden_states regression and attention-kernel divergence under torch.compile are different bugs." + "reason": "FastSpeech2 tokenizer loading and Fuyu batch feature exception handling are unrelated." }, { "left": "issue:43761", "right": "issue:44079", "accept": false, - "reason": "The hidden_states regression is not the same as ModelOutput key assignment when a key was previously None." + "reason": "CLIP hidden_states regression and ModelOutput None-key assignment are not the same underlying defect." }, { "left": "issue:43931", "right": "issue:45216", "accept": false, - "reason": "Model weight-shape mismatch on load and incorrect save_pretrained checkpoint contents are distinct issues." + "reason": "Qwen3-VL loading shape mismatch and Qwen3.5 save_pretrained regression are separate issues at different stages." }, { "left": "issue:43232", "right": "issue:43388", "accept": false, - "reason": "Generation kwargs updates after sync_gpus and metric gathering dropping tuple labels are unrelated utilities." + "reason": "Generation kwarg handling after sync_gpus and gather_for_metrics label trimming are unrelated code paths." } ] }, @@ -11078,202 +11169,202 @@ "issue:43650|issue:46202", "issue:44610|issue:46369" ], - "prepared_review_unit_hash": "4d88066bb4ef943c7573b7e4d32296b8828a4498d3dd37c316ea0b0470975c20", + "prepared_review_unit_hash": "f52e09d5fb9164115067c3e57d1b5af581d35b16ea7dd73b56a446bfff737208", "component_budget": { "node_count": 502, "item_count": 502, - "soft_pair_count": 1842, - "serialized_chars": 956182, - "estimated_input_tokens": 239046, - "estimated_eval_tokens": 478348 + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 28, - "serialized_chars": 35738, - "estimated_input_tokens": 8935, - "estimated_eval_tokens": 18126 + "serialized_chars": 35731, + "estimated_input_tokens": 8933, + "estimated_eval_tokens": 18122 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster is mostly a set of unrelated issues that only share broad Transformers themes (loading, caching, models, training). I found no soft pair that looks like the same underlying bug or change, so all candidate edges are rejected.", - "confidence": 0.94, - "canonical_issue_reason": "issue:45698 is the strongest representative: it is specific, reproducible, still open, and concerns a core save_pretrained/from_pretrained custom-module loading failure with broad user impact.", + "summary": "This cluster is mostly a grab bag of unrelated Transformers issues; the soft-similarity links are generally too weak to be true duplicates. I\u2019d treat issue 45698 as the best hub candidate only because it is the broadest active custom-module loading regression, not because it matches the rest closely.", + "confidence": 0.61, + "canonical_issue_reason": "Issue 45698 is the broadest, most active open report here (custom module loading after save_pretrained/from_pretrained) and is the closest thing to a hub for nearby reports.", "canonical_pr_reason": null, - "best_issue_reason": "issue:45698 has the clearest high-signal bug report and the broadest practical relevance among the issues in this set.", + "best_issue_reason": "Issue 45698 is the strongest triage anchor: actionable, recently active, and centered on a core loading path that could plausibly gather related custom-model reports.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43701", "right": "issue:46506", "accept": false, - "reason": "Different failures: checkpoint key mismatch during resume vs CPU/CUDA output divergence in a model. No shared code-path bug." + "reason": "Checkpoint-key mismatch in training resume is unrelated to PPDocLayoutV3 CPU/CUDA output divergence." }, { "left": "issue:42907", "right": "issue:45698", "accept": false, - "reason": "One is about saving dequantized models; the other is custom-module loading after save_pretrained. Similar serialization theme, but not the same defect." + "reason": "Both involve save/load, but one is dequantized-model saving and the other is custom-module resolution after reload." }, { "left": "issue:42617", "right": "issue:43575", "accept": false, - "reason": "Different problems: script execution failure vs OOM during tensor-parallel loading. Not mergeable as one fix." + "reason": "Different failures: running 3d_parallel.py vs Qwen2 tensor-parallel OOM." }, { "left": "issue:27957", "right": "issue:32090", "accept": false, - "reason": "FlashAttention support for XLM-Roberta is unrelated to Trainer distributed broadcast failing on NoneType." + "reason": "Flash Attention support issue for XLMRoberta is unrelated to the Trainer NoneType broadcast error." }, { "left": "issue:43575", "right": "issue:43742", "accept": false, - "reason": "OOM on tp load and key error loading a different model are distinct load-time issues." + "reason": "OOM during tensor-parallel load is not the same bug as a KeyError while loading MobileLLM-125M." }, { "left": "issue:43550", "right": "issue:43701", "accept": false, - "reason": "torch.compile/SDPA model bug vs checkpoint key mismatch on resume; unrelated code paths." + "reason": "torch.compile/SDPA failure and resume-from-checkpoint key mismatch affect different code paths." }, { "left": "issue:43606", "right": "issue:43701", "accept": false, - "reason": "CPU offload device mismatch is a different runtime issue from resume checkpoint key handling." + "reason": "CPU offload device mismatch is unrelated to checkpoint key-name mismatch." }, { "left": "issue:25251", "right": "issue:28282", "accept": false, - "reason": "Pipeline output shape regression and missing PyTorch import are unrelated." + "reason": "Pipeline top_k output shape regression is unrelated to missing PyTorch import in AutoModel." }, { "left": "issue:46234", "right": "issue:46291", "accept": false, - "reason": "Missing import_utils helper vs pipelines import failure from kernels dependency constraints; different fixes." + "reason": "Missing helper symbol in import_utils and kernels dependency-version validation are different import-time problems." }, { "left": "issue:43295", "right": "issue:43825", "accept": false, - "reason": "Processor/tokenizer regression is unrelated to pipeline translation error messaging." + "reason": "Processor/tokenizer regression and a pipeline error-message/support regression are distinct v5 issues." }, { "left": "issue:16998", "right": "issue:22355", "accept": false, - "reason": "Model max length behavior for DeBERTa-V3 has nothing to do with missing transformers.onnx import." + "reason": "Model_max_length behavior for DeBERTa-V3 has nothing to do with a missing transformers.onnx module." }, { "left": "issue:43295", "right": "issue:44871", "accept": false, - "reason": "Custom processor/tokenizer regression and Gemma eos_token_id mismatch are separate configuration bugs." + "reason": "Both touch config regression, but one is processor.tokenizer access and the other is eos_token_id mismatch in Gemma." }, { "left": "issue:44464", "right": "issue:46421", "accept": false, - "reason": "Both mention chunked generation/compile/cache, but one is incorrect outputs and the other is spurious recompilation. Not the same bug." + "reason": "Chunked generation inconsistency with compiled forward is not the same as lazy cache initialization causing recompilation." }, { "left": "issue:44871", "right": "issue:45584", "accept": false, - "reason": "Gemma eos_token_id config mismatch vs Whisper empty-transcription failure after token alignment are different model-specific issues." + "reason": "Gemma eos_token_id mismatch and Whisper empty-transcription behavior are different model-specific token handling bugs." }, { "left": "issue:43316", "right": "issue:43716", "accept": false, - "reason": "API discrepancy in Gemma3TextConfig is unrelated to a dtype mismatch between image preprocessor and model." + "reason": "Gemma3TextConfig API discrepancy and Mistral-3 dtype mismatch are unrelated model-specific issues." }, { "left": "issue:42175", "right": "issue:43316", "accept": false, - "reason": "Packaging/backend dependency issue is not the same as a Gemma config API mismatch." + "reason": "Packaging/backend install behavior is unrelated to a Gemma3TextConfig API discrepancy." }, { "left": "issue:44077", "right": "issue:44534", "accept": false, - "reason": "Optional post_init validation and junk-filled non-persistent buffers are different v5 regression classes." + "reason": "Optional post_init deprecation and non-persistent buffer junk values are different initialization problems." }, { "left": "issue:44871", "right": "issue:46050", "accept": false, - "reason": "Eos token config inconsistency is unrelated to quantized save/load breakage." + "reason": "Token ID config mismatch and quantized save/load breakage are separate serialization issues." }, { "left": "issue:43646", "right": "issue:44913", "accept": false, - "reason": "Custom model initialization breakage and rotary_pct reload persistence are different config/init bugs." + "reason": "Custom model initialization breakage and GPTNeoX rotary_pct reload loss are not the same underlying bug." }, { "left": "issue:45803", "right": "issue:46365", "accept": false, - "reason": "A swallowed exception in preprocessing is unrelated to Trainer import failure caused by datasets." + "reason": "Bare except swallowing interrupts in tensor conversion is unrelated to importing Trainer being blocked by datasets." }, { "left": "issue:44162", "right": "issue:44792", "accept": false, - "reason": "ESM2 breakage and a Janus image-generation test failure are separate model issues." + "reason": "ESM2 regression and janus generate-images test failure are different model-specific failures." }, { "left": "issue:43452", "right": "issue:44496", "accept": false, - "reason": "gguf_file loading regression and unrecognized model_type config failure are unrelated from_pretrained paths." + "reason": "gguf_file argument handling and missing model_type config validation are both load-path issues but not the same bug." }, { "left": "issue:44242", "right": "issue:46894", "accept": false, - "reason": "Both are training-loss bugs, but they affect different models and likely different loss plumbing; too broad to merge." + "reason": "These are both loss-related, but they concern different models and different loss-calculation mistakes, so they should not be merged." }, { "left": "issue:43404", "right": "issue:43441", "accept": false, - "reason": "Weight tying bug in Mistral3ForConditionalGeneration is distinct from FlashAttention failure in Ministral-3." + "reason": "Untied lm_head weights and FlashAttention failures are distinct Mistral-family issues with different code paths." }, { "left": "issue:45632", "right": "issue:46424", "accept": false, - "reason": "Cache path collision for trust_remote_code is unrelated to max_cache_length being honored only for Quantized Cache." + "reason": "Cache-path collision for local trust_remote_code models is unrelated to max_cache_length semantics." }, { "left": "issue:45803", "right": "issue:46170", "accept": false, - "reason": "Bare except swallowing errors is a different bug from Windows ROCm import/dependency failure." + "reason": "A bare except in Fuyu batch conversion is unrelated to Windows ROCm/torch.distributed import errors." }, { "left": "issue:43650", "right": "issue:46202", "accept": false, - "reason": "Placeholder issue title vs Gemma pipeline roundtrip failure are clearly unrelated." + "reason": "ADD THE DATA is non-technical noise and not a duplicate of the Gemma pipeline roundtrip bug." }, { "left": "issue:44610", "right": "issue:46369", "accept": false, - "reason": "Processor size mismatch for OmDet-Turbo and memory overflow in sam3-litetext are different model/runtime problems." + "reason": "OmDet-Turbo input-size mismatch and sam3-litetext memory overflow are different model/runtime issues." } ] }, @@ -11350,11 +11441,11 @@ "issue:43761|issue:46932", "issue:45542|issue:46234", "issue:43504|issue:45698", - "issue:43352|issue:46640", "issue:43540|issue:44514", "issue:43572|issue:46178", - "issue:44792|issue:45600", + "issue:43352|issue:46640", "issue:43065|issue:44617", + "issue:44792|issue:45600", "issue:44792|issue:44977", "issue:43388|issue:43688", "issue:45127|issue:45200", @@ -11362,195 +11453,195 @@ "issue:41950|issue:43475", "issue:44190|issue:44964", "issue:43525|issue:43531", - "issue:44534|issue:44913", "issue:35141|issue:42490", + "issue:44534|issue:44913", "issue:44734|issue:45406", "issue:46082|issue:46697", "issue:45632|issue:46617", "issue:45072|issue:46932", - "issue:45245|issue:46858", "issue:44855|issue:45588", + "issue:45245|issue:46858", "issue:44869|issue:46832", "issue:43827|issue:43994" ], - "prepared_review_unit_hash": "038178e7760fb46dbabe375d6d02ea069857a2c0435862c33f5791da34670e19", + "prepared_review_unit_hash": "6560d6a7aa3be34975d9d2f38ec76cc6df6e1b4134dc5f3af2a0fc14048e47bf", "component_budget": { "node_count": 502, "item_count": 502, - "soft_pair_count": 1842, - "serialized_chars": 956182, - "estimated_input_tokens": 239046, - "estimated_eval_tokens": 478348 + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 25, - "serialized_chars": 35405, - "estimated_input_tokens": 8852, - "estimated_eval_tokens": 17960 + "serialized_chars": 35402, + "estimated_input_tokens": 8851, + "estimated_eval_tokens": 17958 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "No convincing duplicate cluster here: the items span unrelated bugs, regressions, docs, and config issues. The soft matches are mostly broad topical similarity (e.g. cache/loading, flash-attention, pipelines) without the same concrete failure mode.", - "confidence": 0.9, - "canonical_issue_reason": "No single canonical issue stands out because the set does not contain a true duplicate cluster; choosing one would be arbitrary.", + "summary": "This cluster is a loose collection of unrelated Transformers issues rather than a duplicate set. None of the soft pairs look like the same underlying bug/change, so all soft edges should be rejected. Issue 45698 is the best standalone representative because it has the strongest discussion signal and a broadly impactful loading bug.", + "confidence": 0.92, + "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "issue:45698 is the strongest standalone representative because it has the clearest user-facing loading/caching impact and the most discussion, but it is still not a duplicate of the others.", + "best_issue_reason": "issue:45698 is the strongest standalone issue in the set: it has active discussion, inbound references, and a clear user-facing regression in `from_pretrained`/`save_pretrained` custom module loading.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43761", "right": "issue:46932", "accept": false, - "reason": "Different regressions: missing hidden_states in CLIP vision forward vs dtype casting for Gemma4 quantized embedders." + "reason": "Different regressions: CLIPVision hidden states vs Gemma4 quantized embedder dtype casting." }, { "left": "issue:45542", "right": "issue:46234", "accept": false, - "reason": "Unrelated backend/import issues; one is TensorFlow/tensorboard fallback, the other is a missing import_utils symbol." + "reason": "Unrelated backend/import problems in different modules." }, { "left": "issue:43504", "right": "issue:45698", "accept": false, - "reason": "Different failure modes: legacy field while loading BEiT preset vs wrong custom module after save/load." - }, - { - "left": "issue:43352", - "right": "issue:46640", - "accept": false, - "reason": "Both mention Flash Attention 2, but one is explicit model unsupported behavior and the other is too vague to be the same bug." + "reason": "Different loading failures: legacy preset field vs wrong custom module after save/load." }, { "left": "issue:43540", "right": "issue:44514", "accept": false, - "reason": "Different processors and code paths: video input handling in Qwen3OmniMoe vs batched chat template padding in Qwen2_5_VLProcessor." + "reason": "Both are multimodal processor bugs, but they affect different models and code paths." }, { "left": "issue:43572", "right": "issue:46178", "accept": false, - "reason": "No shared code path; one is a config field regression, the other is an import backend mapping error." + "reason": "Different concerns: config field migration vs backend mapping validation." }, { - "left": "issue:44792", - "right": "issue:45600", + "left": "issue:43352", + "right": "issue:46640", "accept": false, - "reason": "Separate failures: Janus image generation test vs stale auto_mappings references to removed configs." + "reason": "Too broad; one is Nemotron-specific FlashAttention support, the other is a generic flash-attention issue." }, { "left": "issue:43065", "right": "issue:44617", "accept": false, - "reason": "Different SAM3 problems: a dummy Conv2d placeholder vs CUDA OOM in video inference." + "reason": "Different SAM3 bugs: a dummy layer vs CUDA out-of-memory." + }, + { + "left": "issue:44792", + "right": "issue:45600", + "accept": false, + "reason": "Different subsystems: janus test failure vs auto_mappings config reference breakage." }, { "left": "issue:44792", "right": "issue:44977", "accept": false, - "reason": "Not the same bug; Janus image test failure is unrelated to Qwen3.5 flash-attention generation issues." + "reason": "Different model failures and symptoms; only share that they involve generation." }, { "left": "issue:43388", "right": "issue:43688", "accept": false, - "reason": "Both involve training/eval metrics or losses, but they are distinct bugs in different subsystems and not the same code-path problem." + "reason": "Unrelated training/evaluation bugs with no shared code path." }, { "left": "issue:45127", "right": "issue:45200", "accept": false, - "reason": "Different model behaviors: LoRA/vocabulary collapse with tied embeddings vs missing default mm_token_type_ids." + "reason": "Different model behaviors: LoRA/extended vocab collapse vs missing mm_token_type_ids default." }, { "left": "issue:43673", "right": "issue:44743", "accept": false, - "reason": "Both concern cache/state handling, but one is a missing GenerationMixin cache in v5 chunked_prefill and the other is recurrent state reset in Qwen modular code." + "reason": "Both mention cache, but they are different bugs in different model families and logic." }, { "left": "issue:41950", "right": "issue:43475", "accept": false, - "reason": "One is pipeline processor selection, the other is a missing attribute on a SAM3 vision encoder output." + "reason": "Pipeline processor lookup bug is unrelated to Sam3 vision output attributes." }, { "left": "issue:44190", "right": "issue:44964", "accept": false, - "reason": "Different loading failures in different scripts/models; no shared underlying bug is evident." + "reason": "Different loading paths and failure modes; no evidence of the same bug." }, { "left": "issue:43525", "right": "issue:43531", "accept": false, - "reason": "Both are config-related, but one is a missing pad_token_id attribute and the other is a sliding_window issue in Qwen3-MoE." + "reason": "Different config issues: missing pad_token_id vs sliding_window handling." }, { - "left": "issue:44534", - "right": "issue:44913", + "left": "issue:35141", + "right": "issue:42490", "accept": false, - "reason": "Unrelated v5/config regressions: non-persistent buffers junk vs GPTNeoX rotary_pct reload defaulting." + "reason": "Embedding resize reinit bug is unrelated to trailing-slash local model warning/import behavior." }, { - "left": "issue:35141", - "right": "issue:42490", + "left": "issue:44534", + "right": "issue:44913", "accept": false, - "reason": "Different mechanics entirely: embedding resize reinitialization vs import warning for slash-terminated local paths." + "reason": "Different persistence/regression issues: non-persistent buffers vs config reload defaults." }, { "left": "issue:44734", "right": "issue:45406", "accept": false, - "reason": "Distinct transformers serve crashes: KV-cache continuation indexing vs Gemma4Processor missing _tokenizer." + "reason": "Serve KV-cache indexing crash and Gemma4Processor missing attribute are different defects." }, { "left": "issue:46082", "right": "issue:46697", "accept": false, - "reason": "Different domains and failures: LlamaConfig head_dim validation vs Swin2SR image over-padding." + "reason": "Different validation domains: Llama attention config vs Swin2SR image padding." }, { "left": "issue:45632", "right": "issue:46617", "accept": false, - "reason": "Both touch trust_remote_code/cache loading, but one is a leaf-directory cache collision and the other is a FileNotFoundError; not enough to call them the same bug." + "reason": "Both involve trust_remote_code caching, but one is a leaf-dir collision and the other is a missing cached file; not the same bug." }, { "left": "issue:45072", "right": "issue:46932", "accept": false, - "reason": "Both mention dtype issues, but they affect different models and contexts, so they are not mergeable as one fix." + "reason": "Both are dtype-related, but they affect different models and inference paths." }, { - "left": "issue:45245", - "right": "issue:46858", + "left": "issue:44855", + "right": "issue:45588", "accept": false, - "reason": "Different arithmetic/runtime bugs: category count limit vs device tensor returned from cache length breaking static-cache slicing." + "reason": "Python parsing/import error vs flash-attention None handling are unrelated." }, { - "left": "issue:44855", - "right": "issue:45588", + "left": "issue:45245", + "right": "issue:46858", "accept": false, - "reason": "One is a Python 3.13 import parsing issue, the other is a flash_attention.py AttributeError on sink-less models." + "reason": "No shared underlying problem: category-count limit vs cache sequence-length arithmetic." }, { "left": "issue:44869", "right": "issue:46832", "accept": false, - "reason": "Different inference paths and errors: Whisper word timestamp decode crash vs RT-DETR feature-level mismatch TypeError." + "reason": "Different crashes in different models and code paths." }, { "left": "issue:43827", "right": "issue:43994", "accept": false, - "reason": "Shared word 'pipeline' is not enough; one is docs referencing removed APIs, the other is a model producing bad outputs." + "reason": "Docs referring to removed pipeline APIs is unrelated to SigLIP2 inference quality." } ] }, @@ -11628,14 +11719,14 @@ "issue:44242|issue:46897", "issue:29870|issue:35141", "issue:44617|issue:45584", - "issue:44933|issue:45362", "issue:43644|issue:44913", + "issue:44933|issue:45362", "issue:25251|issue:27957", "issue:43976|issue:44855", "issue:45563|issue:45584", "issue:43550|issue:44190", - "issue:45381|issue:46369", "issue:44077|issue:44877", + "issue:45381|issue:46369", "issue:41720|issue:45245", "issue:44479|issue:44734", "issue:43335|issue:44190", @@ -11648,23 +11739,23 @@ "issue:44779|issue:46489", "issue:44617|issue:46369", "issue:43295|issue:45276", - "issue:45290|issue:46752", - "issue:44462|issue:45907" + "issue:44462|issue:45907", + "issue:45290|issue:46752" ], - "prepared_review_unit_hash": "7373411fd0ab1e327f6af3aa425dbe4b288b7f0ad2d43e35d36b1597e55d6c07", + "prepared_review_unit_hash": "5159dc4fca32783f2f86fcd75d1e5c1445f98cd63c208c109abab273809f609a", "component_budget": { "node_count": 502, "item_count": 502, - "soft_pair_count": 1842, - "serialized_chars": 956182, - "estimated_input_tokens": 239046, - "estimated_eval_tokens": 478348 + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 27, - "serialized_chars": 35239, + "serialized_chars": 35238, "estimated_input_tokens": 8810, "estimated_eval_tokens": 17876 }, @@ -11673,184 +11764,181 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a set of weak semantic matches across unrelated bugs. Only the DeepSeek tokenizer pair looks like a true duplicate; the rest differ by model, code path, or failure mode.", - "confidence": 0.9, - "canonical_issue_reason": "No strong single canonical issue for the whole cluster: most links are false positives from broad similarity, with only one clearly duplicate-like tokenizer regression pair.", + "summary": "These items are not duplicates of one another; they span many unrelated Transformers regressions and model-specific bugs. I did not find any soft-edge pair that looks like the same underlying issue.", + "confidence": 0.91, + "canonical_issue_reason": "issue:43644 is the broadest and most central bug report in the set, with inbound references and a clear, high-impact regression description, making it the best anchor if one issue must represent this cluster.", "canonical_pr_reason": null, - "best_issue_reason": "issue:44779 is the best representative because it describes a concrete, reproducible regression and is the only item with a close follow-on duplicate candidate.", + "best_issue_reason": "issue:43644 is the strongest standalone issue: it describes a concrete framework-level regression, is broadly applicable, and has evidence of being referenced by others.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43673", "right": "issue:46421", "accept": false, - "reason": "Both mention chunked prefill/cache, but one is a missing cache bug and the other is lazy cache init causing recompilation; related area, not the same defect." + "reason": "Both mention cache/lazy initialization and chunked prefill, but they describe different failure modes: missing GenerationMixin cache in v5.0.0 vs spurious recompilation. Not the same bug." }, { "left": "issue:43931", "right": "issue:44060", "accept": false, - "reason": "Different models and different failures: a Qwen3-VL load mismatch versus an incorrect tied-weights warning in Qwen3-Next." + "reason": "These are unrelated model-specific bugs: Qwen3-VL weight shape mismatch vs Qwen3-Next tied-weights warning. Different code paths and symptoms." }, { "left": "issue:44242", "right": "issue:46897", "accept": false, - "reason": "Both concern training loss behavior, but the router-logits loss omission and the Florence2 double-shift bug are distinct code paths." + "reason": "Both involve training loss behavior, but one is about MoE load-balancing loss gating and the other is about a Florence2 training-loss double-shift bug. Not the same underlying defect." }, { "left": "issue:29870", "right": "issue:35141", "accept": false, - "reason": "Bart generate() attention-mask bug and token-embedding resize reinitialization are unrelated issues." + "reason": "One is a BART generate() decoder_attention_mask assignment error; the other is token embedding reinitialization after resize when tie_word_embedding is False. Different mechanisms." }, { "left": "issue:44617", "right": "issue:45584", "accept": false, - "reason": "CUDA OOM in Sam3Video and empty-transcription failure in Whisper are completely different problems." + "reason": "CUDA OOM in Sam3Video and Whisper empty-transcription failure are completely different problems with different models and failure modes." }, { - "left": "issue:44933", - "right": "issue:45362", + "left": "issue:43644", + "right": "issue:44913", "accept": false, - "reason": "Nonexistent image_utils import and Qwen3.5 chat crash are unrelated." + "reason": "Both are reload/init-state related, but 43644 is about junk-filled non-persistent buffers while 44913 is a GPTNeoXConfig field not persisting on reload. Different bug classes." }, { - "left": "issue:43644", - "right": "issue:44913", + "left": "issue:44933", + "right": "issue:45362", "accept": false, - "reason": "Generic v5 buffer initialization junk and GPTNeoX rotary_pct reload regression do not share the same bug." + "reason": "Both concern failures around chat/image-related APIs, but one is a missing import from image_utils and the other is Qwen3.5 chat crashing. Not the same issue." }, { "left": "issue:25251", "right": "issue:27957", "accept": false, - "reason": "Pipeline top_k nesting change and XLMRoberta Flash Attention 2 are unrelated." + "reason": "Pipeline top_k nesting bug and XLMRoberta Flash Attention 2 support are unrelated." }, { "left": "issue:43976", "right": "issue:44855", "accept": false, - "reason": "Python-version compatibility failure and a Python 3.13 IndentationError in DebertaV2Model are different defects." + "reason": "These are both Python-version-related issues, but one is a package support/version mismatch and the other is a TorchScript parsing IndentationError. Different root causes." }, { "left": "issue:45563", "right": "issue:45584", "accept": false, - "reason": "A stale generate warning and a Whisper empty-transcription crash are not the same issue." + "reason": "Stale num_return_sequences warning in paged generate() is unrelated to Whisper failing on empty transcription after token alignment." }, { "left": "issue:43550", "right": "issue:44190", "accept": false, - "reason": "torch.compile+SDPA failure in Bamba is unrelated to local dataset loading in the image classification script." + "reason": "torch.compile + SDPA failure in Bamba-9B-v2 and local dataset loading failure in an example script are unrelated." }, { - "left": "issue:45381", - "right": "issue:46369", + "left": "issue:44077", + "right": "issue:44877", "accept": false, - "reason": "Qwen2.5-VL vision-position-id regression and Sam3-LiteText memory overflow are different models and failure modes." + "reason": "Optional post_init in patchtsmixer vs strict config blocking granite_speech loading are different configuration issues." }, { - "left": "issue:44077", - "right": "issue:44877", + "left": "issue:45381", + "right": "issue:46369", "accept": false, - "reason": "One is about optional post_init handling, the other about strict config loading; similar theme, not same bug." + "reason": "Both involve video/image models, but one is a Qwen2.5-VL vision_position_ids regression and the other is Sam3-LiteText memory overflow. Not the same bug." }, { "left": "issue:41720", "right": "issue:45245", "accept": false, - "reason": "Auto device mapping CUDA assert and category-count overflow are unrelated runtime errors." + "reason": "Qwen3 auto device mapping cudaErrorAssert and a categories-count limit runtime error are unrelated." }, { "left": "issue:44479", "right": "issue:44734", "accept": false, - "reason": "Video-input regression in Qwen VL models and KV-cache continuation crash in serve are different code paths." + "reason": "Video input regression for Qwen VL models and serve KV-cache continuation tensor indexing crash are different code paths." }, { "left": "issue:43335", "right": "issue:44190", "accept": false, - "reason": "Sparse-layer config bug and local dataset loading bug are unrelated." + "reason": "A SwitchTransformers sparse-layer config bug and a dataset-loading example bug are unrelated." }, { "left": "issue:43065", "right": "issue:44610", "accept": false, - "reason": "Both involve model/processor mismatch, but for different models and different dimensional assumptions." + "reason": "Sam3PixelDecoder dummy Conv2d and OmDet-Turbo processor/model input-size mismatch are different model implementation issues." }, { "left": "issue:44464", "right": "issue:45563", "accept": false, - "reason": "Compiled-forward chunked generation inconsistency and a stale warning about num_return_sequences are not the same defect." + "reason": "Chunked generation inconsistency with compiled forward and a stale paged-generate warning are not the same defect." }, { "left": "issue:39692", "right": "issue:44190", "accept": false, - "reason": "SigLIP2 doc example errors and local dataset loading failure are unrelated." + "reason": "SigLIP2 doc example errors and local dataset loading in a training script are unrelated documentation/example failures." }, { "left": "issue:45127", "right": "issue:45357", "accept": false, - "reason": "LoRA merge with extended vocab collapse and wrong visual encoder keys on save_pretrained are different bugs." + "reason": "Both involve embedding/weight handling after LoRA or saving, but one is model collapse after merging LoRA and the other is incorrect visual encoder keys on save. Different bugs." }, { "left": "issue:45381", "right": "issue:46655", "accept": false, - "reason": "Both are video-processor issues, but the incorrect vision_position_ids and padding-calculation bugs are distinct." + "reason": "Both are video-processing related, but one is a Qwen VL vision_position_ids regression and the other is SmolVLM padding calculation. Different root causes." }, { "left": "issue:44112", "right": "issue:45600", "accept": false, - "reason": "Both surfaced in CI, but one is a stale device override test and the other is an auto_mappings reference to removed configs." + "reason": "CI stale device override test and auto_mappings references to removed configs are separate CI breakages." }, { "left": "issue:44779", "right": "issue:46489", - "accept": true, - "reason": "Same DeepSeek tokenizer regression in v5+: 46489 reads as the unresolved gap in the fix for the incorrect tokenizer output described in 44779." + "accept": false, + "reason": "Deepseek tokenizer regression in v5 and DeepSeek-Coder v1 tokenizer gap in PR #44801 are related only by tokenizer area, not the same bug." }, { "left": "issue:44617", "right": "issue:46369", "accept": false, - "reason": "Whisper generation OOM and Sam3-LiteText memory overflow are different models and different resource failures." + "reason": "CUDA OOM in Sam3Video and memory overflow in sam3-litetext are both memory issues but for different models and causes." }, { "left": "issue:43295", "right": "issue:45276", "accept": false, - "reason": "Processor/tokenizer regression and gemma4 resize_token_embeddings behavior are not the same issue." + "reason": "Processor/tokenizer API regression in v4.57.5 and gemma4 resize_token_embeddings not updating output embeddings are different issues." }, { - "left": "issue:45290", - "right": "issue:46752", + "left": "issue:44462", + "right": "issue:45907", "accept": false, - "reason": "Both hit apply_chat_template edge cases, but tool-call assistant messages and empty lists are different triggers." + "reason": "Tokenizer.json being ignored by AutoTokenizer and list_repo_files missing revision forwarding are unrelated repository/file-loading bugs." }, { - "left": "issue:44462", - "right": "issue:45907", + "left": "issue:45290", + "right": "issue:46752", "accept": false, - "reason": "Tokenizer.json selection and list_repo_files revision forwarding are separate repository/file-resolution bugs." + "reason": "Both are chat-template crashes on edge cases, but one is assistant messages with tool calls and no content, the other is an empty list IndexError. Different inputs and failure paths." } ] }, - "evaluator_result": { - "accept": true, - "feedback": "The summary is grounded in the packet and stays conservative. The only clearly duplicate-like pair is the DeepSeek tokenizer regression (44779/46489); the rest of the soft-edge rejections are reasonable given the differing models, code paths, or failure modes. No mergeability issues arise because there are no PR pairs here." - }, - "evaluator_used": true, + "evaluator_result": null, + "evaluator_used": false, "retried": false, - "accepted_nontrivial_soft_edge": true, + "accepted_nontrivial_soft_edge": false, "error_kind": null, "error_message": null }, @@ -11925,8 +12013,8 @@ "issue:44521|issue:45381", "issue:43673|issue:46858", "issue:43854|issue:44964", - "issue:45632|issue:46234", "issue:44246|issue:45003", + "issue:45632|issue:46234", "issue:44964|issue:45600", "issue:43643|issue:44496", "issue:44112|issue:44964", @@ -11944,196 +12032,196 @@ "issue:43650|issue:44220", "issue:43673|issue:46424" ], - "prepared_review_unit_hash": "7831d982c7ef1b734e6e4c108acacb7362e8ced418416bf9f0fa912707e35e4f", + "prepared_review_unit_hash": "7ef8e6db4aafa145a106bfdfbc205fe90adcfff933bd828869db81bdca066233", "component_budget": { "node_count": 502, "item_count": 502, - "soft_pair_count": 1842, - "serialized_chars": 956182, - "estimated_input_tokens": 239046, - "estimated_eval_tokens": 478348 + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 27, - "serialized_chars": 35417, - "estimated_input_tokens": 8855, - "estimated_eval_tokens": 17966 + "serialized_chars": 35413, + "estimated_input_tokens": 8854, + "estimated_eval_tokens": 17964 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "These are mostly false-positive similarity matches: they share broad themes like import/load failures, multimodal processing, or cache handling, but not the same concrete bug or fix. I would not collapse any of the proposed soft pairs.", + "summary": "The cluster is highly heterogeneous: it groups many unrelated single-issue reports across pipelines, model loading, generation/cache, multimodal processing, packaging, and notebook/docs. I would not merge any of the soft pairs; they are only superficially similar. If one issue must anchor the cluster, issue 44521 is the strongest representative because it has the most ongoing discussion and inbound references, but even that is only a loose anchor.", "confidence": 0.93, - "canonical_issue_reason": null, + "canonical_issue_reason": "issue:44521 is the most active, well-scoped, and externally referenced report in the set, so it is the best available anchor even though the cluster is not a true duplicate set.", "canonical_pr_reason": null, - "best_issue_reason": null, + "best_issue_reason": "issue:44521 has the clearest actionable bug description and the strongest signal (discussion activity and inbound references), making it the best representative issue in this heterogeneous cluster.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43526", "right": "issue:46697", "accept": false, - "reason": "Both touch image processors, but one is about label reduction and the other about padding logic; different code paths and symptoms." + "reason": "Both are image-processor bugs, but one is about reduce_labels collapsing labels and the other is about over-padding divisible images; different code paths and fixes." }, { "left": "issue:45276", "right": "issue:45584", "accept": false, - "reason": "Unrelated bugs: token embedding resize/tied weights vs Whisper empty-transcription generation after token alignment." + "reason": "Embedding resize/tie-word-embedding behavior is unrelated to Whisper empty-transcription generation after align_special_tokens." }, { "left": "issue:44938", "right": "issue:45542", "accept": false, - "reason": "Both are environment/import complaints, but one is a Python 3.14 load failure and the other is a missing TensorFlow backend dependency." + "reason": "Python 3.14 import loading failure is a different problem from the missing TensorFlow backend in a torch-only install." }, { "left": "issue:44737", "right": "issue:45803", "accept": false, - "reason": "XLNet device placement bug and Fuyu exception handling bug are unrelated." + "reason": "XLNet CPU tensor creation in forward pass and Fuyu's bare except swallowing errors are unrelated bugs." }, { "left": "issue:43540", "right": "issue:44734", "accept": false, - "reason": "Both involve multimodal/generation failures, but one is video input processing in Qwen3OmniMoe and the other is KV-cache continuation indexing in transformers serve." + "reason": "Qwen3OmniMoe video-input validation and transformers serve KV-cache continuation indexing are different failure modes in different paths." }, { "left": "issue:25251", "right": "issue:29870", "accept": false, - "reason": "Pipeline output nesting and BART decoder_attention_mask initialization are unrelated bugs." + "reason": "A pipeline returning a nested list and BART's decoder_attention_mask unbound error are not the same bug." }, { "left": "issue:44521", "right": "issue:45381", "accept": false, - "reason": "Both are multimodal issues, but one is assistant mask generation and the other is Qwen2.5-VL vision position ids; not the same defect." + "reason": "Both are multimodal, but assistant_masks all-zero and wrong vision_position_ids are separate bugs with different internals and fixes." }, { "left": "issue:43673", "right": "issue:46858", "accept": false, - "reason": "Both mention cache/prefill, but one reports missing cache in chunked_prefill while the other is a static-cache tensor typing bug on non-CUDA backends." + "reason": "Both involve cache handling, but one is a missing cache in chunked_prefill and the other is a device-tensor seq_length arithmetic issue for static cache; not mergeable as one fix." }, { "left": "issue:43854", "right": "issue:44964", "accept": false, - "reason": "Both are model loading failures, but for different models and likely different incompatibilities." + "reason": "These are load failures for different models with different symptoms; they do not look like the same underlying bug." }, { - "left": "issue:45632", - "right": "issue:46234", + "left": "issue:44246", + "right": "issue:45003", "accept": false, - "reason": "A trust_remote_code cache-path collision is unrelated to a missing import_utils availability helper." + "reason": "Slow import and unsafe sys.modules access are distinct import-time concerns with different fixes." }, { - "left": "issue:44246", - "right": "issue:45003", + "left": "issue:45632", + "right": "issue:46234", "accept": false, - "reason": "Slow imports sometimes and unsafe sys.modules access are not the same bug, even if both relate to import behavior." + "reason": "A trust_remote_code cache-path collision is unrelated to a missing import_utils availability helper." }, { "left": "issue:44964", "right": "issue:45600", "accept": false, - "reason": "Different failures: model loading compatibility versus stale auto_mappings entries breaking CI." + "reason": "Model loading failure and stale auto_mappings references are different problems." }, { "left": "issue:43643", "right": "issue:44496", "accept": false, - "reason": "Both are config-loading issues, but one loses fields under trust_remote_code and the other lacks model_type for an unrecognized model." + "reason": "Missing fields when trust_remote_code=True and unrecognized model_type in config.json are both config-loading issues, but they are not the same concrete bug." }, { "left": "issue:44112", "right": "issue:44964", "accept": false, - "reason": "A stale CI test failure is unrelated to a model-loading incompatibility." + "reason": "A stale CI test in GraniteSpeech is unrelated to Phi-4 multimodal model loading." }, { "left": "issue:27957", "right": "issue:35141", "accept": false, - "reason": "Flash Attention support and token-embedding resize reinitialization are unrelated." + "reason": "Flash Attention 2 support for XLMRoberta and output embedding reinitialization after resize_token_embeddings are unrelated." }, { "left": "issue:42175", "right": "issue:44279", "accept": false, - "reason": "One is a packaging extra omission; the other is a generic dependency problem report, not the same dependency bug." + "reason": "The first is specifically about TensorFlow not being included with pip install '.[torch]'; the second is a generic dependency complaint." }, { "left": "issue:46082", "right": "issue:46121", "accept": false, - "reason": "Both are validation/config issues, but one is Llama head_dim rejection and the other is rope param parsing for JSON-loaded lists." + "reason": "LlamaConfig head_dim validation and convert_rope_params_to_dict type handling are different configuration-validation bugs." }, { "left": "issue:36246", "right": "issue:41628", "accept": false, - "reason": "Both are import errors for processor classes, but they reference different missing symbols and likely different export paths." + "reason": "Both are import errors, but they involve different symbols and modules; not the same missing-export bug." }, { "left": "issue:42796", "right": "issue:44521", "accept": false, - "reason": "Tokenizer BPE marker handling and multimodal chat-template assistant masks are different bugs." + "reason": "Tokenizer BPE-marker handling and multimodal assistant_masks are separate bugs in different subsystems." }, { "left": "issue:36246", "right": "issue:38617", "accept": false, - "reason": "Different missing imports from different modules; no shared underlying change." + "reason": "Different missing imports from different modules; no clear shared root cause." }, { "left": "issue:43404", "right": "issue:45216", "accept": false, - "reason": "Model-specific weight-tying regression and save_pretrained checkpoint regression are not the same defect." + "reason": "Mistral3 weight-tying and a Qwen3.5 save_pretrained regression are different model-specific issues." }, { "left": "issue:32090", "right": "issue:35141", "accept": false, - "reason": "Trainer gpu_broadcast NoneType error is unrelated to token embedding resizing/post_init reinitialization." + "reason": "Trainer broadcast of NoneType and embedding reinitialization after resize are unrelated." }, { "left": "issue:44295", "right": "issue:46506", "accept": false, - "reason": "A buffer/position_ids error and CPU-vs-CUDA output divergence are different issues." + "reason": "Registering position_ids as a buffer causing read errors is not the same as CPU/CUDA output divergence." }, { "left": "issue:41950", "right": "issue:44016", "accept": false, - "reason": "Video-classification processor lookup and a notebook syntax error are unrelated." + "reason": "A video-classification pipeline looking for image processors and a notebook syntax error are unrelated." }, { "left": "issue:43531", "right": "issue:45600", "accept": false, - "reason": "Qwen3-MoE sliding_window behavior and auto_mappings CI breakage do not share the same code-path bug." + "reason": "Qwen3-MoE sliding_window behavior and stale auto_mappings config references are distinct bugs." }, { "left": "issue:43650", "right": "issue:44220", "accept": false, - "reason": "Placeholder issue text is not a duplicate of the fbank feature extraction bug." + "reason": "A placeholder issue title and a _torch_extract_fbank_features bug are not the same underlying report." }, { "left": "issue:43673", "right": "issue:46424", "accept": false, - "reason": "Both concern cache settings, but one is a missing cache in chunked_prefill while the other is max_cache_length handling for quantized cache only." + "reason": "Both mention cache config, but one is a chunked_prefill cache regression and the other is about max_cache_length only applying to Quantized Cache; too different to merge." } ] }, @@ -12216,42 +12304,42 @@ "issue:44683|issue:45446", "issue:43976|issue:44279", "issue:43526|issue:43901", - "issue:43479|issue:45375", "issue:43381|issue:45593", + "issue:43479|issue:45375", "issue:44496|issue:45042", "issue:43901|issue:44496", "issue:43673|issue:44336", "issue:44792|issue:46424", "issue:44016|issue:44062", - "issue:43824|issue:44933", "issue:44496|issue:45759", + "issue:43824|issue:44933", "issue:43441|issue:45588", "issue:46123|issue:46369", "issue:44327|issue:46752", "issue:44521|issue:46693", "issue:44279|issue:46515", - "issue:43299|issue:45216", "issue:45335|issue:45584", + "issue:43299|issue:45216", "issue:44279|issue:45405", "issue:46082|issue:46399", "issue:43295|issue:44779", "issue:44075|issue:44368", "issue:43976|issue:45405" ], - "prepared_review_unit_hash": "12e1f07aaa316409185deade4f240513ad2e33935b9ebed26ec8d3853fb5c31d", + "prepared_review_unit_hash": "815a07a11faa3bfa7a280c3519f577b7660bd236d349d5634a27fc1ecc5e3253", "component_budget": { "node_count": 502, "item_count": 502, - "soft_pair_count": 1842, - "serialized_chars": 956182, - "estimated_input_tokens": 239046, - "estimated_eval_tokens": 478348 + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 30, - "serialized_chars": 36684, + "serialized_chars": 36682, "estimated_input_tokens": 9171, "estimated_eval_tokens": 18598 }, @@ -12260,186 +12348,192 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is semantically loose: the soft pairs mostly share broad vocabulary (model/config/cache/flash-attention/import/error) but describe different bugs, model families, or even unrelated docs/notebook issues. I would not collapse any of these pairs as duplicates.", - "confidence": 0.98, - "canonical_issue_reason": "No single canonical issue stands out because the items span unrelated problems across loading, generation, configs, pipelines, docs, and performance. The safest triage is to keep them separate.", + "summary": "This cluster is heterogeneous: the issues span unrelated bugs across models, configs, tokenizers, pipelines, and environment compatibility. None of the soft-similarity edges look like true duplicates, so no PR candidate applies and only one issue can serve as a loose anchor.", + "confidence": 0.94, + "canonical_issue_reason": "Issue 44521 is the strongest anchor by discussion/inbound refs and is a concrete user-facing multimodal regression; however, the surrounding items are not true duplicates of it.", "canonical_pr_reason": null, - "best_issue_reason": "No global best issue for deduplication: none of the reports is sufficiently central to represent the others, and the titles indicate distinct code paths.", + "best_issue_reason": "44521 is the best single issue to represent the cluster because it has the clearest active impact signal, but it should not be treated as subsuming the rest of the set.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43976", "right": "issue:44987", "accept": false, - "reason": "Python version compatibility is unrelated to loading a specific model package." + "reason": "Both are version/loading failures, but one is Python compatibility and the other is a specific model load failure; different bugs." }, { "left": "issue:43582", "right": "issue:46932", "accept": false, - "reason": "Apple Silicon allocator TypeError and Gemma4 quantized-embedder dtype casting are different bugs." + "reason": "Apple Silicon TypeError in warmup vs Gemma4 quantized vision/audio dtype casting; unrelated code paths." }, { "left": "issue:44016", "right": "issue:44220", "accept": false, - "reason": "Notebook syntax error and fbank feature extraction failure are unrelated." + "reason": "Notebook syntax error and audio feature-extraction helper issue are distinct." }, { "left": "issue:43295", "right": "issue:45335", "accept": false, - "reason": "Processor/tokenizer image-handling regression is not the same as resize_token_embeddings not updating decoder embeddings." + "reason": "Processor/tokenizer image-handling regression is unrelated to t5gemma embedding resize behavior." }, { "left": "issue:45083", "right": "issue:46506", "accept": false, - "reason": "Qwen3 Omni helper length behavior and PPDocLayoutV3 CPU/CUDA divergence are different code paths." + "reason": "Different bugs: qwen3_omni_moe feature-length helper vs PPDocLayoutV3 CPU/CUDA result mismatch." }, { "left": "issue:42907", "right": "issue:43441", "accept": false, - "reason": "Saving dequantized models and FlashAttention failure in Ministral-3 are distinct issues." + "reason": "Same model family, but save_pretrained dequantization failure is not the FlashAttention runtime failure described here." }, { "left": "issue:44683", "right": "issue:45446", "accept": false, - "reason": "Both touch flex_attention, but one is a compile/runtime failure on torch>=2.9 and the other is a version-check import bug." + "reason": "Both involve flex_attention, but one is compiled-kernel failure on torch>=2.9 and the other is a version-check/import guard bug." }, { "left": "issue:43976", "right": "issue:44279", "accept": false, - "reason": "Python support mismatch is unrelated to a generic dependency issue." + "reason": "General dependency/version complaints, but they refer to different failures and causes." }, { "left": "issue:43526", "right": "issue:43901", "accept": false, - "reason": "BeitImageProcessorFast label reduction and pipeline docs wording are different topics." + "reason": "BeitImageProcessorFast label reduction bug vs TextClassificationPipeline docs mismatch; unrelated." }, { - "left": "issue:43479", - "right": "issue:45375", + "left": "issue:43381", + "right": "issue:45593", "accept": false, - "reason": "Both are config-related, but they affect different model families and different missing/default fields." + "reason": "Gradient checkpointing eval-mode restriction is unrelated to D-FINE auxiliary-loss behavior." }, { - "left": "issue:43381", - "right": "issue:45593", + "left": "issue:43479", + "right": "issue:45375", "accept": false, - "reason": "Gradient checkpointing in eval mode and D-FINE auxiliary-loss behavior are unrelated." + "reason": "Both are config issues, but different models and different missing/default fields." }, { "left": "issue:44496", "right": "issue:45042", "accept": false, - "reason": "Unrecognized model config and unwanted torchvision requirement are separate loading problems." + "reason": "Unrecognized model/config import error vs torchvision requirement in PIL image processors; different problems." }, { "left": "issue:43901", "right": "issue:44496", "accept": false, - "reason": "Docs mention a deprecated parameter, while the other issue is a model-loading config error." + "reason": "Docs mismatch in a pipeline vs model/config recognition failure; no shared bug." }, { "left": "issue:43673", "right": "issue:44336", "accept": false, - "reason": "Generation cache handling and ANSI escape output are unrelated subsystems." + "reason": "Generation cache issue and ANSI logging bug are unrelated." }, { "left": "issue:44792", "right": "issue:46424", "accept": false, - "reason": "Janus image-generation test failure and cache length semantics are not the same bug." + "reason": "Failed janus image-generation test vs cache max length semantics; different code paths." }, { "left": "issue:44016", "right": "issue:44062", "accept": false, - "reason": "Notebook syntax error and AddedToken keyword collision are unrelated." + "reason": "Notebook syntax error is unrelated to AddedToken keyword collision." }, { - "left": "issue:43824", - "right": "issue:44933", + "left": "issue:44496", + "right": "issue:45759", "accept": false, - "reason": "Both are import errors, but they point to different missing symbols in different modules." + "reason": "Unrecognized model error vs AutoModelForCausalLM text_config unwrap bug; distinct loading failures." }, { - "left": "issue:44496", - "right": "issue:45759", + "left": "issue:43824", + "right": "issue:44933", "accept": false, - "reason": "Unrecognized model type and missing text_config unwrapping are different model-construction failures." + "reason": "Both are import errors, but for different symbols/modules and different root causes." }, { "left": "issue:43441", "right": "issue:45588", "accept": false, - "reason": "Both mention flash attention, but they target different failure conditions and code paths." + "reason": "Both touch flash attention, but one is model-specific failure and the other is a sink-less s_aux=None crash; not the same bug." }, { "left": "issue:46123", "right": "issue:46369", "accept": false, - "reason": "Partial-batch flag handling and model memory overflow are unrelated." + "reason": "Partial-batch result dropping vs memory overflow; unrelated issues." }, { "left": "issue:44327", "right": "issue:46752", "accept": false, - "reason": "QA span decoding top-k bounds and empty chat-template input are different crashes." + "reason": "QA span decoding top-k bounds error and empty apply_chat_template IndexError are different failures." }, { "left": "issue:44521", "right": "issue:46693", "accept": false, - "reason": "Assistant-mask computation and long-context FA2 performance regression are not the same issue." + "reason": "Assistant-mask generation bug is unrelated to FA2 long-context performance regression." }, { "left": "issue:44279", "right": "issue:46515", "accept": false, - "reason": "A dependency problem and a torch-version incompatibility are too broad and different to merge." + "reason": "Generic dependency issue vs torch 2.4.1 incompatibility; different causes." + }, + { + "left": "issue:45335", + "right": "issue:45584", + "accept": false, + "reason": "Embedding resize bug in t5gemma is unrelated to Whisper empty transcription after token alignment." }, { "left": "issue:43299", "right": "issue:45216", "accept": false, - "reason": "Qwen3VL loading regression and Qwen3.5 save_pretrained checkpoint corruption are different lifecycle bugs." + "reason": "Qwen3VL Moe load failure and Qwen3.5 save_pretrained regression are related only by family, not the same concrete bug." }, { - "left": "issue:45335", - "right": "issue:45584", + "left": "issue:44279", + "right": "issue:45405", "accept": false, - "reason": "Embedding resizing and Whisper empty-transcription generation failure are unrelated." + "reason": "Unrelated dependency problems: package dependency issue vs unreleased MIN_PEFT_VERSION bump." }, { "left": "issue:46082", "right": "issue:46399", "accept": false, - "reason": "Llama head_dim validation and Qwen3.5 kernel-registration failure are separate config/runtime problems." + "reason": "LlamaConfig head_dim validation and Qwen3.5 kernel registration crash are different issues." }, { "left": "issue:43295", "right": "issue:44779", "accept": false, - "reason": "Custom processor/tokenizer image input regression is not the same as Deepseek tokenizer output corruption." + "reason": "Custom processor.tokenizer/image handling regression is not the same as Deepseek tokenizer incorrect outputs." }, { "left": "issue:44075", "right": "issue:44368", "accept": false, - "reason": "SGD argument handling and a Qwen3.5 warning about tied embeddings are unrelated." + "reason": "SGD argument handling bug and Qwen3.5 warning about tie_word_embeddings are unrelated." }, { "left": "issue:43976", "right": "issue:45405", "accept": false, - "reason": "Python support is unrelated to the PEFT minimum-version bump." + "reason": "Python-version support issue is unrelated to the PEFT minimum-version bump." } ] }, @@ -12516,8 +12610,8 @@ "issue:39692|issue:43901", "issue:43577|issue:45216", "issue:44351|issue:45278", - "issue:46082|issue:46512", "issue:42796|issue:43650", + "issue:46082|issue:46512", "issue:44779|issue:45812", "issue:41950|issue:45020", "issue:43299|issue:45357", @@ -12540,20 +12634,20 @@ "issue:44261|issue:44263", "issue:44336|issue:46851" ], - "prepared_review_unit_hash": "4a54af6ffc5b758bcc4896776cb7dc3f7cee831ff505728696c40f16281c5ae1", + "prepared_review_unit_hash": "635ec453bacdeedb0e087b9513560c1bdca623ef49a4704c79b6c352313028c6", "component_budget": { "node_count": 502, "item_count": 502, - "soft_pair_count": 1842, - "serialized_chars": 956182, - "estimated_input_tokens": 239046, - "estimated_eval_tokens": 478348 + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 26, - "serialized_chars": 35520, + "serialized_chars": 35517, "estimated_input_tokens": 8880, "estimated_eval_tokens": 18016 }, @@ -12562,168 +12656,168 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster looks like a broad false-positive similarity bucket: most items are separate bugs across unrelated model families and code paths, so none of the soft pairs should be merged as duplicates.", - "confidence": 0.94, - "canonical_issue_reason": "Issue 45216 is the clearest standalone regression report in the set and the most plausible anchor, but the cluster overall does not form a true duplicate group.", + "summary": "The cluster is heterogeneous: most soft pairs only overlap at a broad subsystem or regression level, but they point to different bugs, models, or code paths. I found no safe duplicate merges.", + "confidence": 0.84, + "canonical_issue_reason": "issue:45278 is the broadest umbrella report in the set (multiple import errors after an upgrade), so it is the best cluster anchor even though it is not a true duplicate of the others.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 45216 is the best representative issue because it is specific, reproducible, and framed as a concrete save_pretrained regression.", + "best_issue_reason": "issue:45278 has the widest blast radius and the clearest version-regression framing, making it the most suitable representative issue in this mixed cluster.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:39692", "right": "issue:43901", "accept": false, - "reason": "Both are documentation-related, but they cover different APIs and failures: SigLIP2 example errors vs. TextClassificationPipeline return_all_scores docs drift." + "reason": "Both are documentation-related, but they describe different docs and different failures (SigLIP2 example errors vs TextClassificationPipeline return_all_scores mismatch)." }, { "left": "issue:43577", "right": "issue:45216", "accept": false, - "reason": "Different subsystems and symptoms: Blip2 dtype propagation vs. Qwen3.5 save_pretrained checkpoint corruption." + "reason": "These are distinct regression reports: one is a Blip2 dtype propagation issue, the other is Qwen3.5 save_pretrained writing incorrect checkpoint keys." }, { "left": "issue:44351", "right": "issue:45278", "accept": false, - "reason": "44351 is a specific missing HybridCache import; 45278 is a broad post-upgrade import error report, not the same bug." + "reason": "45278 is a broad import-regression report, while 44351 is one specific missing export; not enough evidence they are the same underlying defect." }, { - "left": "issue:46082", - "right": "issue:46512", + "left": "issue:42796", + "right": "issue:43650", "accept": false, - "reason": "One is config validation for head_dim; the other is stale config.dtype after from_config(dtype=...), which is a different initialization issue." + "reason": "Completely different problems: tokenizer BPE marker handling vs an unrelated placeholder issue titled 'ADD THE DATA'." }, { - "left": "issue:42796", - "right": "issue:43650", + "left": "issue:46082", + "right": "issue:46512", "accept": false, - "reason": "Completely unrelated: tokenizer BPE marker handling vs. a vague data-related issue." + "reason": "One concerns LlamaConfig head_dim validation, the other stale model.config.dtype after from_config(dtype=...). Different code paths and symptoms." }, { "left": "issue:44779", "right": "issue:45812", "accept": false, - "reason": "Different model families and different failure modes: Deepseek tokenization regression vs. Granite AutoTokenizer wrong IDs." + "reason": "Both are tokenizer regressions, but for different model families and likely different loading logic; not the same concrete bug." }, { "left": "issue:41950", "right": "issue:45020", "accept": false, - "reason": "A video pipeline processor lookup bug is not the same as a generic remote_code compatibility regression." + "reason": "Video-classification processor lookup and remote_code breakage are unrelated failure modes." }, { "left": "issue:43299", "right": "issue:45357", "accept": false, - "reason": "Both involve Qwen3.x, but one is a loading break in v5 RC and the other is incorrect save_pretrained visual encoder keys; different code paths." + "reason": "Different Qwen model/version regressions affecting different behaviors: Moe model loading vs save_pretrained visual encoder key serialization." }, { "left": "issue:44466", "right": "issue:45276", "accept": false, - "reason": "Both touch serialization/embedding weights, but one is device-dependent lm_head serialization and the other is resize_token_embeddings not updating related embeddings." + "reason": "Serialization of tied weights depending on device is not the same bug as resize_token_embeddings failing to update embeddings in Gemma4." }, { "left": "issue:43299", "right": "issue:43854", "accept": false, - "reason": "Unrelated models and bugs: Qwen3VL MoE loading failure vs. GLM-4.7-Flash test load issue." + "reason": "Both mention loading issues, but one is a Qwen3VL MoE regression and the other is a unit-test load failure for GLM-4.7-Flash." }, { "left": "issue:43653", "right": "issue:45593", "accept": false, - "reason": "Tokenizer special-token registration is unrelated to D-FINE auxiliary loss handling." + "reason": "Tokenizer special-token registration and D-FINE auxiliary-loss behavior are unrelated." }, { "left": "issue:41950", "right": "issue:45823", "accept": false, - "reason": "Different pipelines and failures: video-classification processor lookup vs. Gemma4 PLE device mismatch." + "reason": "Video-classification processor lookup and Gemma4 device-mapping mismatch are different bugs." }, { "left": "issue:44242", "right": "issue:45200", "accept": false, - "reason": "MoE load-balancing loss logic is unrelated to Gemma 4 mm_token_type_ids defaults." + "reason": "MoE load-balancing loss gating and Gemma4 mm_token_type_ids defaults are separate model-training issues." }, { "left": "issue:43650", "right": "issue:46209", "accept": false, - "reason": "No shared bug surface; one is a placeholder title, the other is an FP8 inference failure on B200." + "reason": "No shared bug: one is a vague data request issue, the other is an FP8 inference failure on B200 GPUs." }, { "left": "issue:44297", "right": "issue:44462", "accept": false, - "reason": "Both involve tokenizer persistence, but one is wrong tokenizer_class metadata while the other is AutoTokenizer ignoring tokenizer.json; separate issues." + "reason": "One is about save_pretrained writing the wrong tokenizer_class, the other about AutoTokenizer ignoring tokenizer.json when loading." }, { "left": "issue:44683", "right": "issue:45542", "accept": false, - "reason": "Compiled flex_attention on torch 2.9 and tensorboard/TF backend handling are unrelated." + "reason": "Compiled flex_attention on torch>=2.9 and a TensorFlow-backend import issue are unrelated." }, { "left": "issue:43479", "right": "issue:43525", "accept": false, - "reason": "Different model configs and symptoms: Phi4Multimodal default subconfigs vs. Llama4 missing pad_token_id." + "reason": "Different config bugs: default multimodal subconfigs being initialized from None versus missing pad_token_id on Llama4Config." }, { "left": "issue:44016", "right": "issue:44291", "accept": false, - "reason": "Notebook syntax error is unrelated to init_empty_weights passing an unexpected _is_hf_initialized argument." + "reason": "A notebook syntax error is not the same as the init_empty_weights unexpected _is_hf_initialized argument regression." }, { "left": "issue:44361", "right": "issue:45593", "accept": false, - "reason": "MLukeTokenizer task AttributeError is unrelated to D-FINE denoising/auxiliary loss behavior." + "reason": "MLukeTokenizer task failures and D-FINE auxiliary-loss behavior are unrelated." }, { "left": "issue:45081", "right": "issue:45907", "accept": false, - "reason": "Mistral regex patch crashes on tokenizer internals; list_repo_files revision forwarding is a hub API bug." + "reason": "Mistral regex patching crashes and list_repo_files not forwarding revision are separate utility bugs." }, { "left": "issue:33290", "right": "issue:45656", "accept": false, - "reason": "Both mention DeepSpeed, but one is Adafactor OOM and the other is double optimizer.step calls; not the same defect." + "reason": "Both involve DeepSpeed and optimizers, but one is OOM with Adafactor while the other is double optimizer stepping; not the same bug." }, { "left": "issue:43441", "right": "issue:45834", "accept": false, - "reason": "Different models and code paths: Ministral FlashAttention failure vs. Kosmos2.5 long-OCR index error." + "reason": "FlashAttention failure for Ministral-3 and long-ocr index errors in Kosmos2.5 are unrelated." }, { "left": "issue:42994", "right": "issue:43317", "accept": false, - "reason": "Quantized model saving failure and dequantized model device_map offload loading failure are distinct save vs. load problems." + "reason": "Saving a quantized model and loading a dequantized model with device_map=auto are different phases and different failures." }, { "left": "issue:44623", "right": "issue:44749", "accept": false, - "reason": "processor.save_pretrained missing files is not the same as a performance regression after upgrading Transformers." + "reason": "processor.save_pretrained missing files is unrelated to the reported slowdown after upgrading Transformers." }, { "left": "issue:44261", "right": "issue:44263", "accept": false, - "reason": "MLA rms_norm_eps precision drift and GlmMoeDsaIndexer torch.split() behavior are different bugs." + "reason": "These affect different GLM/MoE internals: rms_norm_eps precision behavior versus torch.split return values." }, { "left": "issue:44336", "right": "issue:46851", "accept": false, - "reason": "ANSI code emission in loading_report is unrelated to Qwen3.5 variable-length position_ids handling." + "reason": "ANSI code emission in loading_report is unrelated to Qwen3.5 variable-length position_ids behavior." } ] }, @@ -12827,14 +12921,14 @@ "issue:45103|issue:46399", "issue:43526|issue:45003" ], - "prepared_review_unit_hash": "d6f4792b108c8263a85c0d9d29e005b2d33701bdd7dbc920e4c64a3d8e5c5c61", + "prepared_review_unit_hash": "e70a0a213fd943aa329f9382ae73c4c63ec910090da111e7e5243488ec039a06", "component_budget": { "node_count": 502, "item_count": 502, - "soft_pair_count": 1842, - "serialized_chars": 956182, - "estimated_input_tokens": 239046, - "estimated_eval_tokens": 478348 + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 }, "budget": { "node_count": 47, @@ -12849,192 +12943,192 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "These items are mostly unrelated Transformers issue reports spanning import errors, model-specific regressions, tokenizer/processor bugs, and environment/package compatibility problems. I found no soft-edge pair that clearly shares the same underlying defect, so this cluster does not look like a true duplicate set.", - "confidence": 0.67, - "canonical_issue_reason": "issue:43824 is the strongest representative issue: it is a concrete, user-facing import failure, has the most external signal (inbound references and discussion), and fits the cluster\u2019s broader theme of v5 regression reports.", + "summary": "This cluster is heterogeneous: import/load failures, tokenizer/model-specific regressions, docs issues, and packaging/backend problems. The soft pairs share broad vocabulary or model families, but none look like the same underlying bug or mergeable change.", + "confidence": 0.91, + "canonical_issue_reason": "No single canonical issue: the items are not duplicates of one bug, but a mixed set of unrelated Transformers regressions and maintenance issues.", "canonical_pr_reason": null, - "best_issue_reason": "issue:43824 is the best single anchor for this cluster because it is specific, actionable, and broadly relevant compared with the more niche model- or environment-specific reports.", + "best_issue_reason": "No single best issue to represent the cluster; each candidate is too specific to its own model, backend, or subsystem.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:45103", "right": "issue:45372", "accept": false, - "reason": "Different failures and code paths: auto-docstring annotation handling vs missing ReasoningEffort import in Gemma 4 processor loading." + "reason": "Both are import-time failures, but one is auto-docstring/annotations handling and the other is a missing `ReasoningEffort` dependency for Gemma 4 processor loading." }, { "left": "issue:45600", "right": "issue:45712", "accept": false, - "reason": "Both are repo hygiene problems, but one is stale auto_mappings references and the other is dummy classes leaking into dir(transformers); not the same bug." + "reason": "Both touch repo hygiene, but one is stale `auto_mappings.py` config references and the other is leftover dummy PT classes; different files and fixes." }, { "left": "issue:45859", "right": "issue:46851", "accept": false, - "reason": "Both involve Qwen3.5, but one is CPU offload rotary embedding compatibility and the other is variable-length position_ids behavior; different defects." + "reason": "Same Qwen3.5 family, but one is CPU offload compatibility in rotary embeddings and the other is variable-length `position_ids` handling." }, { "left": "issue:42175", "right": "issue:43976", "accept": false, - "reason": "Unrelated: missing TensorFlow backend in a torch install vs Python version compatibility for Transformers 5.1.0." + "reason": "One is missing TensorFlow from `.[torch]` install extras; the other is a Python 3.9/3.10 compatibility issue. Not the same problem." }, { "left": "issue:44623", "right": "issue:45397", "accept": false, - "reason": "processor.save_pretrained missing files is a different code path from gemma-4 zero3 from_pretrained loading." + "reason": "`processor.save_pretrained()` missing files versus a `from_pretrained()` zero3 loading failure are different code paths and failure modes." }, { "left": "issue:44671", "right": "issue:44779", "accept": false, - "reason": "Different models and symptoms: CamemBERT masked LM predictions vs Deepseek tokenizer output regression." + "reason": "Both are output-correctness bugs, but on different models and tokenization/LM behavior; no shared concrete code path." }, { "left": "issue:43723", "right": "issue:44162", "accept": false, - "reason": "Both are loading problems, but AutoTokenizer loading in v5 and ESM2 breakage are distinct model-specific issues." + "reason": "Generic tokenizer-loading regression versus ESM2 model breakage; they are model-specific and not the same underlying bug." }, { "left": "issue:43901", "right": "issue:44246", "accept": false, - "reason": "Documentation mismatch for return_all_scores is unrelated to intermittent import slowness." + "reason": "Documentation mismatch about `return_all_scores` versus intermittent import slowness are unrelated." }, { "left": "issue:43673", "right": "issue:43761", "accept": false, - "reason": "Different regressions in different subsystems: chunked_prefill cache vs CLIPVisionModel hidden_states handling." + "reason": "Chunked prefill cache regression in generation versus CLIP hidden-states regression; both are v5 issues, but different model/runtime paths." }, { "left": "issue:43824", "right": "issue:43976", "accept": false, - "reason": "An import error for a specific model class is not the same as a Python-version support regression." + "reason": "Missing Qwen2.5-VL class export versus a Python version compatibility break; no shared bug or fix." }, { "left": "issue:45542", "right": "issue:45800", "accept": false, - "reason": "Optional TensorFlow backend installation issue vs torch/Transformers version incompatibility; separate causes." + "reason": "Both mention installation/runtime compatibility, but one is TensorFlow backend detection and the other is a torch/transformers version mismatch." }, { "left": "issue:42796", "right": "issue:43441", "accept": false, - "reason": "Tokenizer BPE marker handling and FlashAttention failure are different Ministral issues." + "reason": "Same Ministral family, but tokenizer BPE-marker handling and FlashAttention failure are different subsystems." }, { "left": "issue:43525", "right": "issue:43881", "accept": false, - "reason": "Llama4Config missing pad_token_id and glm-4v-9b loading failure are distinct model/config bugs." + "reason": "Llama4 config missing `pad_token_id` and glm-4v-9b loading failure are unrelated model-loading bugs." }, { "left": "issue:46291", "right": "issue:46399", "accept": false, - "reason": "Both mention kernels, but one is an import-time version/revision requirement and the other is a missing decorator in Qwen3.5 rotary embedding registration." + "reason": "Both involve kernels, but one is a dependency/version requirement on importing pipelines and the other is a missing decorator registration in Qwen3.5 code." }, { "left": "issue:44188", "right": "issue:44336", "accept": false, - "reason": "Attention-kernel divergence under torch.compile is unrelated to ANSI codes in loading_report." + "reason": "Attention-kernel divergence under `torch.compile` versus ANSI codes in a loading report are unrelated." }, { "left": "issue:44016", "right": "issue:45823", "accept": false, - "reason": "Notebook syntax error and Gemma4 device-mismatch forward failure are unrelated." + "reason": "Notebook syntax error and Gemma4 device-mismatch during forward are completely different issues." }, { "left": "issue:44336", "right": "issue:44373", "accept": false, - "reason": "Terminal color output cleanup is unrelated to a docstring error." + "reason": "One is terminal-color output behavior, the other is a wrong docstring; no shared underlying bug." }, { "left": "issue:44589", "right": "issue:45362", "accept": false, - "reason": "Missing Float8 storage class and Qwen3.5 chat crash are unrelated failures." + "reason": "Float8 storage lookup failure versus Qwen3.5 chat crash are unrelated runtime/model problems." }, { "left": "issue:43525", "right": "issue:43643", "accept": false, - "reason": "Both concern config fields, but one is a specific Llama4Config attribute error and the other is a broader AutoConfig remote-code field loss; not the same concrete bug." + "reason": "Both mention config fields, but one is Llama4-specific `pad_token_id` and the other is `trust_remote_code` returning incomplete AutoConfig objects." }, { "left": "issue:46655", "right": "issue:46726", "accept": false, - "reason": "Video padding math in SmolVLM and AutoProcessor\u2019s missing-backend error message are different processor issues." + "reason": "Both are processor/image-related, but one is a padding-calculation bug and the other is an explicit missing-vision-backend error path." }, { "left": "issue:45325", "right": "issue:46851", "accept": false, - "reason": "Both are Qwen position_ids issues, but one is Qwen2.5-VL rope scaling for still images and the other is Qwen3.5 variable-length input handling." + "reason": "Both concern Qwen position IDs, but one is a Qwen2.5-VL rope scaling bug for still images and the other is Qwen3.5 variable-length input handling." }, { "left": "issue:41628", "right": "issue:43976", "accept": false, - "reason": "AutoImageProcessor import failure is unrelated to the Python 3.9+/3.10+ support regression." + "reason": "`AutoImageProcessor` import failure and Python version incompatibility are unrelated." }, { "left": "issue:44297", "right": "issue:45397", "accept": false, - "reason": "Tokenizer save_pretrained metadata mismatch and gemma-4 zero3 loading failure are separate bugs." + "reason": "Tokenizer save metadata bug versus Gemma4 zero3 loading failure are different subsystems." }, { "left": "issue:43784", "right": "issue:45542", "accept": false, - "reason": "Both involve optional dependencies, but sentence-transformers import breakage and missing TensorFlow backend are different problems." + "reason": "Sentence-transformers import NameError and TensorFlow backend absence are unrelated packaging/runtime issues." }, { "left": "issue:45083", "right": "issue:46851", "accept": false, - "reason": "Different Qwen code paths: feature-extractor length helper vs variable-length position_ids handling." + "reason": "Both are Qwen-related position/length logic, but they affect different models and helper functions." }, { "left": "issue:44246", "right": "issue:44496", "accept": false, - "reason": "Intermittent import slowness is unrelated to an unrecognized model/config parsing error." + "reason": "Intermittent import slowness and unrecognized model config loading are not the same bug." }, { "left": "issue:46399", "right": "issue:46851", "accept": false, - "reason": "Qwen3.5 kernels registration failure and variable-length position_ids are different defects." + "reason": "Both are Qwen3.5 regressions, but one is kernel registration failure and the other is variable-length `position_ids`; distinct code paths." }, { "left": "issue:45561", "right": "issue:45563", "accept": false, - "reason": "A pytest-xdist race in testing utils is unrelated to a stale generation warning." + "reason": "A pytest-xdist file race and a stale generation warning are unrelated test/runtime issues." }, { "left": "issue:45103", "right": "issue:46399", "accept": false, - "reason": "Different subsystems and failure modes: auto-docstring annotation parsing vs Qwen3.5 kernel registration." + "reason": "Auto-docstring annotation handling and Qwen3.5 kernel registration failure have no concrete overlap." }, { "left": "issue:43526", "right": "issue:45003", "accept": false, - "reason": "BeitImageProcessorFast label reduction and unsafe sys.modules access in modeling_utils are unrelated." + "reason": "BEiTImageProcessorFast label reduction and unsafe `sys.modules` access in modeling_utils are unrelated bugs." } ] }, @@ -13136,20 +13230,20 @@ "issue:44295|issue:44749", "issue:44479|issue:46655" ], - "prepared_review_unit_hash": "1926e782cf8f347fecfa9cf8861a1f1ea3e9c9f81025f17e2214e11a1efe82db", + "prepared_review_unit_hash": "6f9ff847ee63b51f42b71a5cf7bfaf5a18f406adc19217819b6d5023ccbba978", "component_budget": { "node_count": 502, "item_count": 502, - "soft_pair_count": 1842, - "serialized_chars": 956182, - "estimated_input_tokens": 239046, - "estimated_eval_tokens": 478348 + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 28, - "serialized_chars": 35501, + "serialized_chars": 35502, "estimated_input_tokens": 8876, "estimated_eval_tokens": 18008 }, @@ -13158,180 +13252,180 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a set of semantically similar but distinct bug reports, not true duplicates. None of the soft pairs look like the same concrete code-path issue; the closest broad anchor is the core-library import/safety bug in `modeling_utils`.", - "confidence": 0.84, - "canonical_issue_reason": "issue:45003 is the broadest framework-level bug in the set and the best anchor for a heterogeneous cluster, but the rest are mostly unrelated model- or feature-specific regressions.", + "summary": "This cluster is mostly heterogeneous issue reports. The best anchor is the broad v5.3.0 video regression report (44479); the soft pairs mostly share a subsystem or model family but describe different concrete failures, so none should be merged as duplicates.", + "confidence": 0.72, + "canonical_issue_reason": "issue:44479 is the strongest representative because it describes a concrete, version-specific regression affecting multiple Qwen video/model variants, with a clear shared code path and broad enough scope to anchor related reports.", "canonical_pr_reason": null, - "best_issue_reason": "issue:45003 stands out as the most central reusable bug report because it targets core `modeling_utils` behavior rather than a single model or edge feature.", + "best_issue_reason": "issue:44479 is the best standalone issue: it is specific, impactful, and frames a regression that plausibly subsumes adjacent video-input complaints better than the narrower model-specific reports.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43540", "right": "issue:46752", "accept": false, - "reason": "Different failures: Qwen3OmniMoe video input processing vs empty-chat-template IndexError." + "reason": "Different failures: Qwen3OmniMoe video input ValueError vs empty-list apply_chat_template IndexError." }, { "left": "issue:46614", "right": "issue:46665", "accept": false, - "reason": "Stale phonemizer backend state is unrelated to `from_pretrained` behavior inside classes." + "reason": "Stale phonemizer backend state and from_pretrained behavior are unrelated code paths." }, { "left": "issue:45072", "right": "issue:46829", "accept": false, - "reason": "One is a bfloat16 dtype mismatch; the other is a wrong output-field assignment in a video classifier." + "reason": "One is a bfloat16 dtype mismatch; the other is a bad hidden_states field in an output object." }, { "left": "issue:44295", "right": "issue:45397", "accept": false, - "reason": "Buffer access error and DeepSpeed `from_pretrained` failure are different code paths." + "reason": "Registered buffer access error and ZeRO3 from_pretrained crash do not point to the same bug." }, { "left": "issue:43531", "right": "issue:45362", "accept": false, - "reason": "Sliding-window handling in Qwen3-MoE is unrelated to chat loading crashes in Qwen3.5-35B." + "reason": "Sliding-window handling and a Qwen3.5 chat crash are different model issues." }, { "left": "issue:42491", "right": "issue:43299", "accept": false, - "reason": "Both involve Qwen3 MoE, but one is hf4.x LoRA compatibility and the other is v5 loading breakage." + "reason": "Both involve Qwen MoE loading, but one is LoRA compatibility across HF versions and the other is Qwen3VL MoE loading; not the same concrete defect." }, { "left": "issue:44246", "right": "issue:44871", "accept": false, - "reason": "MoE load-balancing loss bug is unrelated to Gemma eos-token config inconsistency." + "reason": "Load-balancing loss omission and eos_token_id config inconsistency are unrelated." }, { "left": "issue:44242", "right": "issue:45593", "accept": false, - "reason": "Different auxiliary-loss mechanisms in different models; not the same bug." + "reason": "Both mention auxiliary losses, but they affect different models and triggers; not mergeable as one bug." }, { "left": "issue:44336", "right": "issue:44485", "accept": false, - "reason": "ANSI output formatting and GLM-5 RoPE implementation are unrelated." + "reason": "ANSI output formatting and RoPE implementation are unrelated." }, { "left": "issue:45588", "right": "issue:46851", "accept": false, - "reason": "Flash-attention sink handling is unrelated to variable-length `position_ids` failures." + "reason": "Flash-attention sinkless-model crash and variable-length position_ids failure are different code paths." }, { "left": "issue:45003", "right": "issue:46821", "accept": false, - "reason": "Unsafe `sys.modules` access is not the same as a `window_size` attribute error in TimesFM." + "reason": "Unsafe sys.modules access and TimesFM window_size attribute error are unrelated." }, { "left": "issue:42886", "right": "issue:46132", "accept": false, - "reason": "Offline tokenizer cache loading and attention registration semantics are different bugs." + "reason": "Offline cache loading and AttentionInterface.register behavior changes are different subsystems." }, { "left": "issue:44451", "right": "issue:46752", "accept": false, - "reason": "Model loading for ScandiBERT is unrelated to empty `apply_chat_template()` inputs." + "reason": "Model loading failure and empty-list chat template crash are not the same issue." }, { "left": "issue:43550", "right": "issue:46506", "accept": false, - "reason": "Torch.compile SDPA failure and CPU/CUDA output divergence are different problems." + "reason": "torch.compile/SDPA failure and CPU-vs-CUDA output mismatch are different defects." }, { "left": "issue:44779", "right": "issue:45292", "accept": false, - "reason": "Tokenizer correctness regression is not the same as resizing embeddings not updating output embeddings." + "reason": "Tokenizer correctness regression and resize_token_embeddings output-sync bug are distinct." }, { "left": "issue:36246", "right": "issue:42757", "accept": false, - "reason": "Different missing imports from different dependencies/modules." + "reason": "Import errors for different symbols from different modules; no shared root cause is evident." }, { "left": "issue:36246", "right": "issue:44351", "accept": false, - "reason": "Both are import errors, but they concern different symbols and different root causes." + "reason": "Qwen2_5_VLImageProcessor export failure and HybridCache export failure are separate import problems." }, { "left": "issue:41950", "right": "issue:44220", "accept": false, - "reason": "Video-classification processor lookup and `_torch_extract_fbank_features()` are unrelated." + "reason": "Pipeline image-processor lookup and _torch_extract_fbank_features() are different bugs." }, { "left": "issue:43901", "right": "issue:45003", "accept": false, - "reason": "Docs mismatch for `return_all_scores` is not the same as unsafe module lookup in `modeling_utils`." + "reason": "Docs/behavior mismatch in TextClassificationPipeline and sys.modules access are unrelated." }, { "left": "issue:43257", "right": "issue:45137", "accept": false, - "reason": "Qwen3 MoE weight conversion and DeepSpeed ZeRO3 deque underflow are different failure modes." + "reason": "DeepSpeed weight conversion and ZeRO3 deque underflow are both distributed-training related but not the same bug." }, { "left": "issue:45003", "right": "issue:45561", "accept": false, - "reason": "Core import safety bug and pytest-xdist file-race bug do not share a code path." + "reason": "sys.modules access and pytest-xdist captured_info.txt race are unrelated." }, { "left": "issue:44496", "right": "issue:45003", "accept": false, - "reason": "Unrecognized model config and unsafe `sys.modules` access are unrelated." + "reason": "Unrecognized model config and unsafe sys.modules access are different failure modes." }, { "left": "issue:43540", "right": "issue:46655", "accept": false, - "reason": "Both involve video processing, but the concrete bugs are different models and different failure logic." + "reason": "Both are video-processing reports, but they target different models and distinct processing bugs." }, { "left": "issue:44162", "right": "issue:45200", "accept": false, - "reason": "ESM2 model breakage and Gemma 4 `mm_token_type_ids` defaults are unrelated." + "reason": "ESM2 breakage and Gemma 4 mm_token_type_ids default are unrelated." }, { "left": "issue:45375", "right": "issue:45759", "accept": false, - "reason": "Missing config field handling and `text_config` unwrapping are different composite-config issues." + "reason": "Both touch Qwen 3.5 multimodal config handling, but one is strict-field dropping and the other is from_config text_config unwrapping; not the same concrete change." }, { "left": "issue:43262", "right": "issue:44246", "accept": false, - "reason": "Audio chat-template sampling-rate default and MoE load-balancing loss are unrelated." + "reason": "Audio chat-template sampling-rate default and router-logits load-balancing loss are unrelated." }, { "left": "issue:44295", "right": "issue:44749", "accept": false, - "reason": "A buffer access bug is unrelated to a performance regression after upgrading Transformers." + "reason": "A buffer-related runtime error and a Chinese performance regression report are not duplicates." }, { "left": "issue:44479", "right": "issue:46655", "accept": false, - "reason": "Both are video-related, but one is a Qwen regression and the other is a SmolVLM padding bug." + "reason": "Both concern video regressions, but they affect different model families and likely different padding/processor code paths." } ] }, @@ -13423,192 +13517,186 @@ "issue:42199|issue:46851", "issue:43526|issue:46894", "issue:44060|issue:44077", - "issue:43653|issue:44162", "issue:44368|issue:44977", + "issue:43653|issue:44162", "issue:45412|issue:46123", "issue:44479|issue:46170", "issue:42886|issue:45229", "issue:45397|issue:45656", "issue:43065|issue:43408" ], - "prepared_review_unit_hash": "dd87713157e220797fefb1710f9e658307b2ee398bfac5510915fd80a3304fff", + "prepared_review_unit_hash": "a01bf47f5c52e0ff5a1763246cf7a1c8aaf2f87a49615b50d11aa7eede42da6e", "component_budget": { "node_count": 502, "item_count": 502, - "soft_pair_count": 1842, - "serialized_chars": 956182, - "estimated_input_tokens": 239046, - "estimated_eval_tokens": 478348 + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 25, - "serialized_chars": 35305, - "estimated_input_tokens": 8827, - "estimated_eval_tokens": 17910 + "serialized_chars": 35314, + "estimated_input_tokens": 8829, + "estimated_eval_tokens": 17914 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The items are a heterogeneous set of issue reports with only weak topical similarity in the proposed soft pairs. I don\u2019t see any pair that clearly describes the same underlying bug or change, so I would not merge any soft edges.", - "confidence": 0.94, - "canonical_issue_reason": "Issue 45292 is the cleanest representative bug report here: it names a specific API (`resize_token_embeddings`), a concrete incorrect effect on `output_embeddings`, and has a narrow, actionable scope.", + "summary": "This cluster is a loose collection of unrelated Transformers bug reports. The soft pairs mostly share broad themes (SAM3, Qwen, DeepSpeed, tied embeddings, tokenizer/config issues) but not the same concrete defect or fix path, so none should be merged as duplicates.", + "confidence": 0.97, + "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "Issue 45292 is the strongest standalone issue for triage because it is precise, reproducible, and describes a single code-path defect without extra subsystem noise.", + "best_issue_reason": null, "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43475", "right": "issue:46369", "accept": false, - "reason": "Both are SAM3-related, but one is an attribute error in video vision encoder outputs and the other is a memory overflow in sam3-litetext; different failure modes and code paths." + "reason": "Same SAM3 family, but one is an attribute error in vision encoder outputs and the other is a memory overflow in litetext; different failures and fixes." }, { "left": "issue:45859", "right": "issue:46399", "accept": false, - "reason": "Both involve Qwen3.5-family models, but one is about CPU offload compatibility in rotary embedding and the other is a registration/decorator crash with `use_kernels=True`; not the same bug." + "reason": "Both touch Qwen3.5 rotary/kernel code, but one is CPU offload incompatibility and the other is a missing decorator causing kernel registration failure." }, { "left": "issue:45137", "right": "issue:45600", "accept": false, - "reason": "These are unrelated: one is a DeepSpeed ZeRO3 deque/index error with BertModel, the other is a CI break from stale auto-mapping references." + "reason": "DeepSpeed ZeRO3 deque error vs a broken auto_mappings CI reference; no shared bug path." }, { "left": "issue:36246", "right": "issue:43596", "accept": false, - "reason": "An import error for Qwen2.5-VL image processor and a ZeRO3/BertModel empty-index error are different issues with no shared code path." + "reason": "ImportError for a missing Qwen2.5-VL image processor is unrelated to the BertModel/Zero3 index error." }, { "left": "issue:44877", "right": "issue:45375", "accept": false, - "reason": "Both mention config validation, but they affect different models and different fields; this is too broad to be the same underlying bug." + "reason": "Both involve strict config handling, but they concern different models and different missing fields, so not the same defect." }, { "left": "issue:43295", "right": "issue:45292", "accept": false, - "reason": "One is a regression in custom model code using `processor.tokenizer` and images; the other is `resize_token_embeddings` not updating output embeddings. Different defects." + "reason": "Processor/tokenizer regression and resize_token_embeddings propagation are distinct code paths." }, { "left": "issue:45292", "right": "issue:45584", "accept": false, - "reason": "`resize_token_embeddings` vs Whisper empty transcription after `align_special_tokens` are distinct behaviors and failure modes." + "reason": "Both mention embeddings/tokens, but one is a resize propagation bug while the other is Whisper empty-transcription handling after special-token alignment." }, { "left": "issue:44466", "right": "issue:45335", "accept": false, - "reason": "Both touch tied/resize embedding behavior, but one is inconsistent serialization of `lm_head.weight` by device and the other is `resize_token_embeddings` not affecting `decoder.embed_tokens`; not the same concrete bug." + "reason": "Both are tied-embedding related, but one is inconsistent serialization and the other is resize_token_embeddings not updating decoder embeddings." }, { "left": "issue:44623", "right": "issue:46726", "accept": false, - "reason": "Saving processor files and reporting an unrecognized image processor when a backend is missing are different processor lifecycle issues." + "reason": "Missing saved files and missing vision backend are different save/load problems." }, { "left": "issue:44242", "right": "issue:44857", "accept": false, - "reason": "MoE load-balancing loss omission and a float16/CUDA crash in `LwDetrImageLoss` are unrelated training bugs." + "reason": "Load-balancing loss omission in an MoE model is unrelated to an AMP/CUDA crash in LwDetrImageLoss." }, { "left": "issue:43526", "right": "issue:46897", "accept": false, - "reason": "A BEiT fast image-processor label reduction bug and a Florence2 training-loss double-shift bug are not the same issue." + "reason": "BEiT label reduction and Florence2 training-loss double-shift are different model-specific bugs." }, { "left": "issue:44038", "right": "issue:46506", "accept": false, - "reason": "Qwen3-VL-Moe compatibility and PPDocLayoutV3 CPU/CUDA result differences are unrelated model-specific problems." + "reason": "Qwen3-VL-Moe compatibility with Transformers 5 is unrelated to PPDocLayoutV3 CPU/CUDA result differences." }, { "left": "issue:44987", "right": "issue:45459", "accept": false, - "reason": "Both are loading-related, but one is a version-specific model load failure and the other is tokenizer error handling when protobuf is absent; distinct root causes." + "reason": "Loading failures on transformers>=5.1.0 and tokenizer decode-error masking are separate issues." }, { "left": "issue:29870", "right": "issue:30333", "accept": false, - "reason": "A Bart generation UnboundLocalError and an MLflow status-reporting failure are completely unrelated." + "reason": "BART generate() decoder mask bug and MLFlow job-status reporting are unrelated." }, { "left": "issue:43352", "right": "issue:45250", "accept": false, - "reason": "A model explicitly lacking Flash Attention 2.0 support is not the same as a generic Flash Attention 2.0 topic." + "reason": "One is a specific model not supporting Flash Attention 2.0; the other is only a generic Flash Attention topic." }, { "left": "issue:42199", "right": "issue:46851", "accept": false, - "reason": "DETR-style cardinality handling and Qwen3.5 variable-length `position_ids` behavior are unrelated." - }, - { - "left": "issue:43526", - "right": "issue:46894", - "accept": false, - "reason": "BEiT fast label reduction and a CohereASR training-loss double-shift bug do not share the same code path." + "reason": "DETR cardinality/background-class handling and Qwen3.5 variable-length position_ids are unrelated." }, { "left": "issue:44060", "right": "issue:44077", "accept": false, - "reason": "These involve different models and different problems: incorrect tied-weights warning vs a `post_init` allowance change." + "reason": "Tied-weights warning in Qwen3-Next and patchtsmixer post_init policy are different issues." }, { - "left": "issue:43653", - "right": "issue:44162", + "left": "issue:44368", + "right": "issue:44977", "accept": false, - "reason": "A BigBird tokenizer special-token registration bug and an ESM2 breakage report are not the same underlying defect." + "reason": "One is a tied-embeddings warning during LoRA fine-tuning; the other is a flash-attention generation regression." }, { - "left": "issue:44368", - "right": "issue:44977", + "left": "issue:43653", + "right": "issue:44162", "accept": false, - "reason": "Both mention Qwen3.5 and generation/warnings, but one is a tied-embeddings warning during LoRA fine-tuning and the other is a flash-attention generation failure; not mergeable as one bug." + "reason": "BigBirdTokenizer special-token registration and ESM2 breakage do not share the same code path." }, { "left": "issue:45412", "right": "issue:46123", "accept": false, - "reason": "RT-DETR garbage-collection memory retention and a generation pipeline final-batch flag bug are unrelated." + "reason": "Memory not released on deletion and final partial-batch results being dropped are different lifecycle bugs." }, { "left": "issue:44479", "right": "issue:46170", "accept": false, - "reason": "A Qwen vision video-input regression and a Windows ROCm torch.distributed dependency issue are unrelated." + "reason": "Video-input regression and Windows ROCm torch.distributed import/type errors are unrelated." }, { "left": "issue:42886", "right": "issue:45229", "accept": false, - "reason": "Offline tokenizer cache loading and Gemma4 multi-GPU CUDA OOM are different operational failures." + "reason": "Offline tokenizer cache loading and Gemma4 CUDA OOM are not the same underlying defect." }, { "left": "issue:45397", "right": "issue:45656", "accept": false, - "reason": "Gemma-4 ZeRO3 `from_pretrained` behavior and double optimizer stepping under DeepSpeed are separate DeepSpeed-related issues, not the same bug." + "reason": "Gemma-4 ZeRO3 from_pretrained and a double optimizer step under DeepSpeed are different bugs despite both mentioning DeepSpeed." }, { "left": "issue:43065", "right": "issue:43408", "accept": false, - "reason": "A dummy Conv2d in Sam3PixelDecoder and a model-type mismatch warning for sam3_video vs sam3_tracker are different SAM3 issues." + "reason": "Dummy Conv2d in Sam3PixelDecoder and a sam3_video/sam3_tracker config mismatch are separate SAM3 issues." } ] }, @@ -13691,8 +13779,8 @@ "issue:43931|issue:44368", "issue:44188|issue:46202", "issue:44483|issue:46202", - "issue:45987|issue:46082", "issue:43906|issue:46202", + "issue:45987|issue:46082", "issue:44995|issue:45588", "issue:44188|issue:46421", "issue:45685|issue:46431", @@ -13709,230 +13797,224 @@ "issue:43441|issue:46752", "issue:45375|issue:45478", "issue:42994|issue:43482", - "issue:43475|issue:43643", "issue:43232|issue:43673", + "issue:43475|issue:43643", "issue:44188|issue:46851" ], - "prepared_review_unit_hash": "466ad1c6efec8c25aea7e74ccd92f4c072fdff78dc1e90c4f877f5f0976d4ee8", + "prepared_review_unit_hash": "9b131206d965e0f5605fb4d5c0fa6448ddc18d13c9b12e7f726de968f22a68b8", "component_budget": { "node_count": 502, "item_count": 502, - "soft_pair_count": 1842, - "serialized_chars": 956182, - "estimated_input_tokens": 239046, - "estimated_eval_tokens": 478348 + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 30, - "serialized_chars": 36819, - "estimated_input_tokens": 9205, - "estimated_eval_tokens": 18666 + "serialized_chars": 36815, + "estimated_input_tokens": 9204, + "estimated_eval_tokens": 18664 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a loose grab-bag of unrelated Transformers bugs. The only convincing duplicate-like pair is the Qwen3.5-Moe config loading failure: one issue identifies the missing field/root cause and the other reports the downstream from_pretrained error.", + "summary": "Most pairs are only superficially similar; there is one clear duplicate pair around Qwen3.5 MoE config loading. Everything else appears to be separate bugs or regressions in different code paths.", "confidence": 0.81, - "canonical_issue_reason": "issue:45375 is the clearest root-cause report: it names the missing `deepstack_visual_indexes` field, explains that `@strict` drops it, and matches the Qwen3.5-Moe loading failure described in the paired report.", + "canonical_issue_reason": "issue:45375 is the most specific root-cause report in the only credible duplicate pair: it names the missing `deepstack_visual_indexes` field and the `@strict` drop behavior directly.", "canonical_pr_reason": null, - "best_issue_reason": "issue:45375 is the best representative because it is the most specific and actionable description of the only likely duplicate bug in the set.", + "best_issue_reason": "issue:45375 is the best representative issue for this cluster because it captures the concrete underlying config-loading bug rather than the broader downstream symptom.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43475", "right": "issue:43540", "accept": false, - "reason": "Different model families and failures: SAM3 video encoder attribute error vs Qwen3OmniMoe video-input processing ValueError." + "reason": "Different models and failures: missing `fpn_position_embeddings` in SAM 3 video vs a video-input `ValueError` in Qwen3OmniMoe." }, { "left": "issue:44877", "right": "issue:46170", "accept": false, - "reason": "Unrelated bugs: granite_speech config strict-loading vs Windows/ROCm import/type error from a torch.distributed dependency." + "reason": "Unrelated problems: strict config loading for `granite_speech` vs a Windows/ROCm `torch.distributed` import/type issue." }, { "left": "issue:46170", "right": "issue:46832", "accept": false, - "reason": "Different code paths and symptoms: platform/import failure vs RT-DETR feature-level mismatch in forward." + "reason": "Different subsystems and symptoms: platform import dependency issue vs RT-DETR feature-level mismatch in forward." }, { "left": "issue:43723", "right": "issue:46710", "accept": false, - "reason": "Tokenizer loading failure and an accuracy regression are not the same bug, even if both relate to v5 upgrades." + "reason": "Tokenizer loading failure and an accuracy regression are not the same bug or code path." }, { "left": "issue:44368", "right": "issue:45563", "accept": false, - "reason": "One is a model config warning about `tie_word_embeddings`; the other is a stale warning in paged generation." + "reason": "Both mention warnings, but they are different warnings in different generation/tokenization paths." }, { "left": "issue:44995", "right": "issue:45200", "accept": false, - "reason": "Different models and root causes: stale cache crash on second forward vs Gemma 4 requiring `mm_token_type_ids` defaults." + "reason": "Stale cache crash in GlmMoeDsa is unrelated to Gemma 4 text-only fine-tuning requiring `mm_token_type_ids`." }, { "left": "issue:43931", "right": "issue:44368", "accept": false, - "reason": "Qwen3-VL weight-shape mismatch is unrelated to the Qwen3.5 LoRA warning about `tie_word_embeddings`." + "reason": "Shape-mismatch loading failure for Qwen3-VL is unrelated to a tie-embeddings warning during Qwen3.5 fine-tuning." }, { "left": "issue:44188", "right": "issue:46202", "accept": false, - "reason": "Attention-kernel divergence under `torch.compile` is a different problem from Gemma pipeline roundtrip failure on `\u2581`." + "reason": "Torch.compile attention-kernel divergence is unrelated to Gemma pipeline roundtrip token handling." }, { "left": "issue:44483", "right": "issue:46202", "accept": false, - "reason": "API request rejection and tokenizer roundtrip failure are different layers and different bugs." + "reason": "API request rejection in `/v1/chat/completions` is a different bug from Gemma token roundtripping." }, { - "left": "issue:45987", - "right": "issue:46082", + "left": "issue:43906", + "right": "issue:46202", "accept": false, - "reason": "Cache return-type contract bug vs LlamaConfig validation of `head_dim`; not the same issue." + "reason": "An isolated reproduction of an older issue is not the same as the Gemma pipeline roundtrip failure." }, { - "left": "issue:43906", - "right": "issue:46202", + "left": "issue:45987", + "right": "issue:46082", "accept": false, - "reason": "An isolated reproduction of another issue is too vague here, and it does not clearly match the Gemma pipeline token roundtrip bug." + "reason": "`StaticCache.get_seq_length()` return-type bug and a `LlamaConfig` validation error are distinct issues." }, { "left": "issue:44995", "right": "issue:45588", "accept": false, - "reason": "Stale indexer/cache crash on second forward is unrelated to the flash_attention `s_aux=None` AttributeError." + "reason": "Different failure modes: stale indexer cache crash vs `flash_attention.py` `AttributeError` on `s_aux=None`." }, { "left": "issue:44188", "right": "issue:46421", "accept": false, - "reason": "Both mention cache/compile, but one is attention-kernel branching divergence and the other is lazy cache init causing recompilation; not the same bug." + "reason": "Both involve compile/cache themes, but one is attention-kernel divergence and the other is spurious recompilation from lazy cache init." }, { "left": "issue:45685", "right": "issue:46431", "accept": false, - "reason": "MPS histogram integer error and missing chat template are unrelated." + "reason": "MPS histogram kernel limitation is unrelated to missing chat template support in `apply_chat_template`." }, { "left": "issue:45103", "right": "issue:46858", "accept": false, - "reason": "Docstring-annotation handling and static-cache seq-length tensor arithmetic are different code paths." + "reason": "Docstring annotation handling crash is unrelated to static-cache slice arithmetic failing on non-CUDA backends." }, { "left": "issue:44297", "right": "issue:44623", "accept": false, - "reason": "Both are save_pretrained complaints, but tokenizer config mismatch and processor missing files are distinct failures." + "reason": "Both are save-pretrained problems, but tokenizer config mismatch and missing processor files are different bugs." }, { "left": "issue:43065", "right": "issue:46123", "accept": false, - "reason": "Dummy SAM3 Conv2d and MaskGenerationPipeline partial-batch bookkeeping are unrelated." + "reason": "A dummy Conv2d in SAM 3 pixel decoding is unrelated to a masking pipeline partial-batch result drop." }, { "left": "issue:44361", "right": "issue:44749", "accept": false, - "reason": "Tokenizer AttributeError and a performance regression after upgrade do not look like the same underlying bug." + "reason": "Tokenizer AttributeError and transformer-upgrade slowdown are unrelated." }, { "left": "issue:43994", "right": "issue:46726", "accept": false, - "reason": "Nonsensical model outputs are not the same as an AutoProcessor backend-missing error." + "reason": "Wrong outputs from siglip2 and missing-vision-backend processor detection are different loading/runtime issues." }, { "left": "issue:44749", "right": "issue:45656", "accept": false, - "reason": "Performance slowdown from upgrade and deepspeed double optimizer-step are different problems." + "reason": "Data filtering slowdown after upgrade is unrelated to DeepSpeed double optimizer stepping." }, { "left": "issue:43701", "right": "issue:44749", "accept": false, - "reason": "Checkpoint key mismatch and a data-processing slowdown are unrelated." + "reason": "Checkpoint key mismatch and performance regression are not the same underlying bug." }, { "left": "issue:44871", "right": "issue:46132", "accept": false, - "reason": "Inconsistent `eos_token_id` config and AttentionInterface registration behavior changes are different issues." + "reason": "Gemma eos-token-id config inconsistency is unrelated to AttentionInterface registration behavior changes." }, { "left": "issue:44623", "right": "issue:45656", "accept": false, - "reason": "Processor save_pretrained file omission is unrelated to deepspeed calling optimizer.step twice." + "reason": "Missing files on `processor.save_pretrained` is unrelated to the optimizer being stepped twice under DeepSpeed." }, { "left": "issue:43646", "right": "issue:44162", "accept": false, - "reason": "Both are broad v5 regressions, but one is custom model initialization and the other is an ESM2-specific breakage; not the same bug." + "reason": "Generic v5 custom-model initialization breakage and ESM2 breakage are different reports with different concrete causes." }, { "left": "issue:43441", "right": "issue:46752", "accept": false, - "reason": "FlashAttention failure for Ministral-3 and empty-list chat-template IndexError are unrelated." + "reason": "FlashAttention failure for Ministral-3 and empty-chat-template `IndexError` are unrelated." }, { "left": "issue:45375", "right": "issue:45478", "accept": true, - "reason": "These appear to be the same Qwen3.5-Moe loading bug: 45375 identifies the missing config field/root cause, and 45478 reports the resulting from_pretrained error on newer Transformers." + "reason": "Both describe the same Qwen3.5 MoE config-loading bug: a required field is dropped/ignored under strict loading, producing downstream `from_pretrained` errors." }, { "left": "issue:42994", "right": "issue:43482", "accept": false, - "reason": "Quantized model saving failure and GGUF loading failure are opposite operations on different model paths." - }, - { - "left": "issue:43475", - "right": "issue:43643", - "accept": false, - "reason": "Missing `fpn_position_embeddings` in a SAM3 video output object is not the same as AutoConfig remote-code field loss." + "reason": "Quantized model saving failure and GGUF loading failure are different directions and different model formats." }, { "left": "issue:43232", "right": "issue:43673", "accept": false, - "reason": "Generation kwargs update after `sync_gpus` and missing cache during chunked prefill are separate generation bugs." + "reason": "These are separate generation-related regressions with different triggers: `sync_gpus`/`_update_model_kwargs_for_generation` vs chunked-prefill cache handling." }, { - "left": "issue:43931", - "right": "issue:44368", + "left": "issue:43475", + "right": "issue:43643", "accept": false, - "reason": "Qwen3-VL weight mismatch and a Qwen3.5 tie_word_embeddings warning are different model/config issues." + "reason": "Missing SAM3 vision encoder field and missing fields from `AutoConfig.from_pretrained` are different config-loading issues." }, { "left": "issue:44188", "right": "issue:46851", "accept": false, - "reason": "Torch.compile attention divergence is unrelated to Qwen3.5 variable-length `position_ids` handling." + "reason": "Attention-kernel divergence under compile is unrelated to Qwen3.5 variable-length position-id handling." } ] }, "evaluator_result": { "accept": true, - "feedback": "The summary is grounded in the packet and stays conservative overall. The only accepted soft-edge pair, issue:45375 and issue:45478, is a plausible same-bug match: one reports the missing `deepstack_visual_indexes` field being dropped by `@strict`, and the other reports the downstream Qwen3.5-Moe `from_pretrained` failure. The other pair verdicts are cautious and do not overclaim duplication from weak similarity alone." + "feedback": "The summary is well grounded in the packet metadata. The single accepted pair (45375/45478) is a plausible duplicate rooted in the same Qwen3.5 MoE config-loading failure, and the other soft-edge rejections are conservative based on distinct titles/symptoms." }, "evaluator_used": true, "retried": false, @@ -14005,9 +14087,9 @@ "soft_pairs": [ "issue:28282|issue:30990", "issue:41628|issue:44933", - "issue:43475|issue:46821", - "issue:45310|issue:45375", "issue:45230|issue:45397", + "issue:45310|issue:45375", + "issue:43475|issue:46821", "issue:44964|issue:45137", "issue:43986|issue:46726", "issue:44869|issue:45356", @@ -14029,225 +14111,225 @@ "issue:44038|issue:45310", "issue:44222|issue:45310", "issue:36010|issue:41628", - "issue:46202|issue:46701", "issue:43278|issue:45137", "issue:42994|issue:43824", + "issue:46202|issue:46701", "issue:45081|issue:45103" ], - "prepared_review_unit_hash": "6b3e5b888a53468219b54ced90b8e9a278e0c3439832b81bf37e2f009ffeb6cd", + "prepared_review_unit_hash": "b541748974b90acd393daaaf0e8fb4e0def7fa148c3f1b54b188e6c31c0decfc", "component_budget": { "node_count": 502, "item_count": 502, - "soft_pair_count": 1842, - "serialized_chars": 956182, - "estimated_input_tokens": 239046, - "estimated_eval_tokens": 478348 + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 30, - "serialized_chars": 36415, - "estimated_input_tokens": 9104, - "estimated_eval_tokens": 18464 + "serialized_chars": 36407, + "estimated_input_tokens": 9102, + "estimated_eval_tokens": 18460 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster is mostly a set of unrelated issue reports that happen to share similar wording around imports, model loading, processors, or tokenizer/model regressions. I only found one clear duplicate pair: the repeated Qwen3ForCausalLM VRAM leak report.", - "confidence": 0.89, - "canonical_issue_reason": "issue:42673 is the clearest canonical issue: it exactly matches issue:43704 by title and symptom (Qwen3ForCausalLM VRAM leak with multiple dataloader threads) and has substantially more discussion activity, making it the better anchor.", + "summary": "This cluster is mostly a loose grab-bag of unrelated transformer/model bugs, not a tight duplicate set. The only clear duplicate is the repeated Qwen3ForCausalLM VRAM leak report.", + "confidence": 0.95, + "canonical_issue_reason": "issue:42673 is the earliest and more complete report of the exact same VRAM leak; issue:43704 has the same title and appears to be a duplicate.", "canonical_pr_reason": null, - "best_issue_reason": "issue:42673 is the best representative issue in this cluster because it is an exact duplicate of another item and has the strongest signal/discussion around a concrete reproducible bug.", + "best_issue_reason": "issue:42673 is the cleanest anchor for the cluster: specific, reproducible, and duplicated verbatim by a later report.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:28282", "right": "issue:30990", "accept": false, - "reason": "Both are loading-related, but one is a PyTorch availability ImportError and the other is a hang while loading Sentence Transformers; different failure modes and code paths." + "reason": "Different failures: PyTorch missing import vs a loading hang in Sentence Transformers." }, { "left": "issue:41628", "right": "issue:44933", "accept": false, - "reason": "Both are import errors, but they concern different missing symbols/modules and likely different exports; not the same bug." + "reason": "Both are import-related, but they concern different missing symbols/modules and different code paths." }, { - "left": "issue:43475", - "right": "issue:46821", + "left": "issue:45230", + "right": "issue:45397", "accept": false, - "reason": "Completely different models and errors: SAM 3 Vision encoder attribute vs TimesFM window_size attribute access." + "reason": "Unrelated bugs: a placeholder bug report vs Gemma-4 ZeRO3 from_pretrained failure." }, { "left": "issue:45310", "right": "issue:45375", "accept": false, - "reason": "Same general Qwen3.5 family, but one is a from_pretrained failure and the other is a config field being dropped by @strict; not the same concrete defect." + "reason": "Related model family, but not the same bug: generic Qwen3.5 Moe from_pretrained error vs a missing config field in the vision config." }, { - "left": "issue:45230", - "right": "issue:45397", + "left": "issue:43475", + "right": "issue:46821", "accept": false, - "reason": "Issue:45230 is too generic to tie to the Gemma-4 Zero3 from_pretrained bug; no evidence they are the same underlying problem." + "reason": "Different model-specific attribute errors on different components; no shared underlying bug." }, { "left": "issue:44964", "right": "issue:45137", "accept": false, - "reason": "Different subsystems and symptoms: Phi-4 multimodal loading vs a DeepSpeed ZeRO3 deque error." + "reason": "Different subsystems and symptoms: model load regression vs DeepSpeed ZeRO3 deque underflow." }, { "left": "issue:43986", "right": "issue:46726", "accept": false, - "reason": "Both mention missing vision backend handling, but one is a video-model AutoProcessor crash and the other is an unrecognized image processor message; not enough to treat as the same bug." + "reason": "Both mention missing vision backend handling, but one is a video-model AutoProcessor crash and the other is an unrecognized image processor error." }, { "left": "issue:44869", "right": "issue:45356", "accept": false, - "reason": "Whisper timestamp decoding crash vs Kimi-K2.5 tokenizer codec/regression issue; different pipelines and root causes." + "reason": "Different tokenizer regressions affecting different models and failure modes." }, { "left": "issue:44038", "right": "issue:44623", "accept": false, - "reason": "Qwen3-VL-Moe v5 compatibility error vs processor.save_pretrained missing files; different operations and failure points." + "reason": "Qwen3-VL-Moe loading error is unrelated to processor.save_pretrained missing files." }, { "left": "issue:39692", "right": "issue:46726", "accept": false, - "reason": "Documentation example errors for SigLIP2 versus a missing-vision-backend AutoProcessor error; not the same bug." + "reason": "Doc example errors for SigLIP2 are not the same as the missing-vision-backend processor error." }, { "left": "issue:46463", "right": "issue:46877", "accept": false, - "reason": "T5 decoder-input validation error message vs auto_docstring printing diagnostics at import time; unrelated issues." + "reason": "Incorrect validation error message is unrelated to auto_docstring stdout diagnostics." }, { "left": "issue:44373", "right": "issue:46851", "accept": false, - "reason": "A wrong docstring and a real variable-length position_ids bug are not the same underlying issue." + "reason": "One is a docstring issue; the other is a runtime bug with variable-length position_ids." }, { "left": "issue:42994", "right": "issue:43525", "accept": false, - "reason": "Quantized model saving failure vs missing pad_token_id on Llama4Config; different model behaviors." + "reason": "Quantized model saving failure is unrelated to a missing pad_token_id attribute on Llama4Config." }, { "left": "issue:42371", "right": "issue:45092", "accept": false, - "reason": "TF32 API guidance issue vs legacy InternVL2 meta-init incompatibility; unrelated." + "reason": "TF32 settings guidance is unrelated to InternVL2 meta-initialization incompatibility." }, { "left": "issue:43723", "right": "issue:46326", "accept": false, - "reason": "Both involve tokenizers, but one is an AutoTokenizer v5 loading issue and the other is a mistral_common continue_final_message bug; distinct code paths." + "reason": "Both are tokenizer-related, but they describe different regressions and different code paths." }, { "left": "issue:45406", "right": "issue:46202", "accept": false, - "reason": "Gemma4Processor missing _tokenizer in transformers serve vs a Gemma pipeline roundtrip failure on a specific character; not the same defect." + "reason": "Gemma processor missing _tokenizer is not the same as the Gemma pipeline roundtrip failure." }, { "left": "issue:43873", "right": "issue:44945", "accept": false, - "reason": "Quantization/offloading behavior vs pipeline-parallel output correctness; too broad and different failure modes." + "reason": "Quantization/offloading behavior is unrelated to pipeline-parallel incorrect outputs." }, { "left": "issue:44945", "right": "issue:45753", "accept": false, - "reason": "Pipeline parallelism output corruption vs a Qwen3.5 infinite loop on a specific image; unrelated symptoms." + "reason": "Pipeline-parallel output corruption is not the same as a Qwen3.5 image-specific infinite loop." }, { "left": "issue:41628", "right": "issue:44351", "accept": false, - "reason": "Both are import errors from transformers, but they are for different symbols and likely different export paths." + "reason": "Different missing public imports: AutoImageProcessor vs HybridCache." }, { "left": "issue:43784", "right": "issue:46399", "accept": false, - "reason": "Sentence-transformers import NameError vs a Qwen3.5 kernel registration ValueError; different libraries and code paths." + "reason": "Import-time NameError in sentence-transformers is unrelated to a rotary-embedding registration error." }, { "left": "issue:44857", "right": "issue:46877", "accept": false, - "reason": "AMP/CUDA crash in LwDetrImageLoss vs import-time doc-lint stdout noise; unrelated." + "reason": "Training loss AMP crash is unrelated to auto_docstring emitting diagnostics at import time." }, { "left": "issue:45593", "right": "issue:46877", "accept": false, - "reason": "D-FINE auxiliary-loss bug vs auto_docstring import-time diagnostics; no overlap beyond being bugs." + "reason": "D-FINE auxiliary-loss behavior is unrelated to docstring lint output on import." }, { "left": "issue:42673", "right": "issue:43704", "accept": true, - "reason": "Exact duplicate: identical title and same reported Qwen3ForCausalLM VRAM leak when used in multiple dataloader threads." + "reason": "Exact duplicate: same title and same Qwen3ForCausalLM VRAM leak symptom." }, { "left": "issue:44038", "right": "issue:45310", "accept": false, - "reason": "Qwen3-VL-Moe v5 error and Qwen3.5 Moe from_pretrained error are related by family name only, not the same concrete failure." + "reason": "Different Qwen model families and different loading errors; not the same concrete bug." }, { "left": "issue:44222", "right": "issue:45310", "accept": false, - "reason": "FP8 save_pretrained bug vs from_pretrained error; opposite operations and different symptoms." + "reason": "FP8 save_pretrained failure is distinct from a from_pretrained load error." }, { "left": "issue:36010", "right": "issue:41628", "accept": false, - "reason": "GenerationMixin import failure vs AutoImageProcessor import failure; both imports, but not the same missing object or root cause." - }, - { - "left": "issue:46202", - "right": "issue:46701", - "accept": false, - "reason": "Gemma token roundtrip failure on a specific token vs image-text-to-text pipeline not working; too broad and not enough to unify." + "reason": "Both are import errors, but they involve different missing API symbols and likely different regressions." }, { "left": "issue:43278", "right": "issue:45137", "accept": false, - "reason": "Embedding dtype drift between train/eval vs DeepSpeed ZeRO3 deque underflow; different implementation layers and errors." + "reason": "Dtype drift between train/eval is unrelated to the ZeRO3 deque error." }, { "left": "issue:42994", "right": "issue:43824", "accept": false, - "reason": "Quantized model saving bug vs Qwen2.5-VL import failure; unrelated." + "reason": "Quantized saving failure is unrelated to a missing Qwen2.5-VL import." + }, + { + "left": "issue:46202", + "right": "issue:46701", + "accept": false, + "reason": "Both mention Gemma, but one is a tokenizer roundtrip problem and the other is a pipeline integration issue." }, { "left": "issue:45081", "right": "issue:45103", "accept": false, - "reason": "Mistral regex patch/backend_tokenizer crash vs auto_docstring future-annotations crash; unrelated internals." + "reason": "Mistral tokenizer backend_tokenizer crash is unrelated to the future-annotations docstring crash." } ] }, "evaluator_result": { "accept": true, - "feedback": "The summary is grounded in the packet and stays conservative. The canonical/best issue choice is supported by the identical titles and the higher discussion activity on issue:42673. The only accepted soft edge, issue:42673 \u2194 issue:43704, is a clearly justified exact duplicate. The remaining pair judgments are appropriately cautious and do not overstate shared evidence." + "feedback": "The summary is grounded in the packet and stays conservative. The Qwen3ForCausalLM VRAM leak pair is a clear duplicate, and the remaining soft-edge verdicts are appropriately cautious rather than overstated." }, "evaluator_used": true, "retried": false, @@ -14346,23 +14428,23 @@ "issue:43784|issue:43824", "issue:27957|issue:30333", "issue:44734|issue:45103", - "issue:42222|issue:45200", - "issue:30333|issue:34689" + "issue:30333|issue:34689", + "issue:42222|issue:45200" ], - "prepared_review_unit_hash": "6b451342b5146aa9622ade039434f63e129d8a888605f81af4b73b9d6f4c5768", + "prepared_review_unit_hash": "d7839f6071f88e05676feebc157e07c99618fdbc26ba9a5a74e76ecc68f42597", "component_budget": { "node_count": 502, "item_count": 502, - "soft_pair_count": 1842, - "serialized_chars": 956182, - "estimated_input_tokens": 239046, - "estimated_eval_tokens": 478348 + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 30, - "serialized_chars": 36794, + "serialized_chars": 36795, "estimated_input_tokens": 9199, "estimated_eval_tokens": 18654 }, @@ -14371,192 +14453,192 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster is highly heterogeneous: the items span unrelated bugs in model loading, tokenizers, docs, training, serving, and test infrastructure. The soft pairs mostly share broad subsystems or release timing, but not the same concrete underlying defect.", - "confidence": 0.98, - "canonical_issue_reason": "No issue is a true semantic canonical for this cluster; issue 45561 is only the closest anchor by activity/inbound references, but it does not explain most of the other reports.", + "summary": "This cluster is heterogeneous: the soft edges mostly connect superficially similar issues (tokenizers, multimodal loading, docs, FP8) but not the same underlying bug. No PRs are present.", + "confidence": 0.94, + "canonical_issue_reason": "Issue 45561 is the best representative by triage value: it is open, highly referenced, and has clear reproduction details. That said, the set is not a true duplicate family, so the choice is pragmatic rather than semantic.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 45561 is the best single triage anchor on metadata alone (open, referenced, discussed), but it is not a strong representative of the cluster\u2019s mixed content.", + "best_issue_reason": "Issue 45561 is the strongest standalone issue to surface because it is active, well-scoped, and has multiple inbound references; it is more actionable than the other candidates.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:42831", "right": "issue:44222", "accept": false, - "reason": "Both mention FP8, but one is an accuracy problem and the other is save_pretrained behavior for MoE; different bugs." + "reason": "Both mention FP8, but one is an accuracy regression in FineGrainedFP8 and the other is a save_pretrained/MoE serialization bug; different code paths." }, { "left": "issue:43232", "right": "issue:44792", "accept": false, - "reason": "Generation-state sync bug vs a janus image-generation test failure; no shared code-path." + "reason": "Generation-state update after sync_gpus is unrelated to a Janus image-generation test failure; no shared underlying bug." }, { "left": "issue:42175", "right": "issue:44933", "accept": false, - "reason": "Packaging/backend dependency issue vs a nonexistent import in image_utils; unrelated." + "reason": "Backend packaging/install omission and a missing image_utils import are different failures, even if both are import/setup related." }, { "left": "issue:43504", "right": "issue:45561", "accept": false, - "reason": "Model preset loading for BEiT vs a pytest-xdist file race; different areas and symptoms." + "reason": "Beit legacy-field loading and an xdist captured-info race are unrelated subsystems and symptoms." }, { "left": "issue:45561", "right": "issue:45600", "accept": false, - "reason": "Testing-utils race vs auto_mappings cleanup; not the same defect." + "reason": "A test-file race in patched utils is not the same bug as stale auto_mappings references to removed configs." }, { "left": "issue:43976", "right": "issue:44933", "accept": false, - "reason": "Python version compatibility complaint vs image_utils import typo; unrelated." + "reason": "Python version compatibility and a nonexistent import export are distinct issues." }, { "left": "issue:44496", "right": "issue:45561", "accept": false, - "reason": "Model config/import failure vs test file race; no overlap in root cause." + "reason": "Model config recognition failure and a testing-utils race do not share a code path." }, { "left": "issue:30333", "right": "issue:33290", "accept": false, - "reason": "MLflow reporting on failed jobs vs Deepspeed OOM with Adafactor; different failure modes." + "reason": "MLflow reporting after failed jobs is unrelated to an Adafactor/deepspeed OOM." }, { "left": "issue:45405", "right": "issue:46877", "accept": false, - "reason": "Dependency version bump issue vs import-time doc-lint logging; unrelated." + "reason": "A dependency version pin problem and docstring lint output at import time are unrelated." }, { "left": "issue:30990", "right": "issue:35141", "accept": false, - "reason": "Loading hang vs embedding reinitialization; different subsystems." + "reason": "A SentenceTransformers loading hang is different from output-embedding reinitialization during resize." }, { "left": "issue:42898", "right": "issue:44568", "accept": false, - "reason": "Both are tokenizer regressions in v5, but one is cleanup-space behavior and the other is special-token insertion; not the same bug." + "reason": "Both are tokenizer regressions, but clean_up_tokenization_spaces behavior and add_special_tokens/BOS-EOS handling are different bugs." }, { "left": "issue:42175", "right": "issue:45712", "accept": false, - "reason": "Missing tensorflow backend extra vs leftover dummy classes in repo checks; unrelated." + "reason": "Packaging backend omission and leftover dummy classes leaking into dir() are unrelated maintenance issues." }, { "left": "issue:43479", "right": "issue:46829", "accept": false, - "reason": "Config-default initialization bug vs output field mapping bug; different model paths." + "reason": "Multimodal config default initialization and VideoPrism output field misassignment are different model bugs." }, { "left": "issue:46077", "right": "issue:46123", "accept": false, - "reason": "Attention-mask construction bug vs partial-batch pipeline flag bug; unrelated." + "reason": "Attention-mask construction for Gemma 4 and final-partial-batch handling in MaskGenerationPipeline are separate problems." }, { "left": "issue:44623", "right": "issue:45310", "accept": false, - "reason": "Processor save_pretrained file omission vs Qwen3.5 MoE from_pretrained error; unrelated." + "reason": "Processor.save_pretrained missing files and Qwen3.5 MoE from_pretrained errors are both loading-related but not the same failure." }, { "left": "issue:42673", "right": "issue:46693", "accept": false, - "reason": "VRAM leak in multi-threaded dataloading vs long-context FA2 performance regression; different issues." + "reason": "A VRAM leak in multi-threaded dataloaders is not the same as a long-context FA2 performance regression." }, { "left": "issue:43479", "right": "issue:44869", "accept": false, - "reason": "Config defaults vs Whisper timestamp decode crash; unrelated." + "reason": "Config defaults for Phi4Multimodal and Whisper timestamp decode crashes are unrelated code paths." }, { "left": "issue:43723", "right": "issue:44568", "accept": false, - "reason": "Tokenizer loading failure vs special-token behavior regression; not the same concrete bug." + "reason": "AutoTokenizer.from_pretrained failure and missing BOS/EOS on add_special_tokens are different tokenizer regressions." }, { "left": "issue:42898", "right": "issue:45812", "accept": false, - "reason": "Both tokenizer regressions, but one is cleanup-space handling and the other is wrong token IDs for Granite; different code paths." + "reason": "Both are tokenizer regressions, but one concerns cleanup-space behavior and the other wrong Granite token IDs; not the same bug." }, { "left": "issue:44016", "right": "issue:46231", "accept": false, - "reason": "Notebook syntax error vs documentation parameter-name mismatch; unrelated." + "reason": "A notebook syntax error and a documentation parameter-name mismatch are different doc issues." }, { "left": "issue:44625", "right": "issue:46877", "accept": false, - "reason": "Qwen3.5 num_labels propagation vs docstring lint logging; unrelated." + "reason": "Qwen3.5 num_labels propagation and auto_docstring stdout noise have no shared underlying code path." }, { "left": "issue:42898", "right": "issue:44779", "accept": false, - "reason": "Both are v5 tokenizer regressions, but they affect different models and behaviors; too broad to merge." + "reason": "Both are v5 tokenizer regressions, but they affect different behaviors: cleanup spacing versus DeepSeek token output correctness." }, { "left": "issue:39692", "right": "issue:46909", "accept": false, - "reason": "Two broken docs examples, but for different models and failures; not the same underlying change." + "reason": "These are both broken docs examples, but for different models/examples and different failure modes." }, { "left": "issue:43479", "right": "issue:44079", "accept": false, - "reason": "Config initialization bug vs ModelOutput key assignment bug; unrelated." + "reason": "Multimodal config initialization and ModelOutput key assignment are separate core issues." }, { "left": "issue:43874", "right": "issue:45092", "accept": false, - "reason": "Different multimodal loading failures in different model families; no shared concrete root cause." + "reason": "A missing GLM46V processor method and an InternVL2 meta-init incompatibility are different multimodal-loading bugs." }, { "left": "issue:43784", "right": "issue:43824", "accept": false, - "reason": "Both are import failures, but they affect different libraries/models and failure points." + "reason": "Both are import-related, but one is a NameError in sentence-transformers and the other is a missing Qwen2.5-VL class export." }, { "left": "issue:27957", "right": "issue:30333", "accept": false, - "reason": "Flash Attention model support vs MLflow reporting; unrelated." + "reason": "Flash Attention 2 support for XLMRoberta is unrelated to MLflow reporting after failed training." }, { "left": "issue:44734", "right": "issue:45103", "accept": false, - "reason": "Serving KV-cache indexing crash vs auto-docstring annotation crash; unrelated." + "reason": "A serving KV-cache indexing crash and an auto_docstring annotations crash are different subsystems." }, { - "left": "issue:42222", - "right": "issue:45200", + "left": "issue:30333", + "right": "issue:34689", "accept": false, - "reason": "VitPose model breakage vs Gemma4 mm_token_type_ids defaulting; different models and bugs." + "reason": "MLflow reporting and Llama 3.2 Vision model loading regressions are unrelated." }, { - "left": "issue:30333", - "right": "issue:34689", + "left": "issue:42222", + "right": "issue:45200", "accept": false, - "reason": "Failed-job MLflow reporting vs Llama 3.2 Vision model-loading regression; unrelated." + "reason": "VitPose model breakage and Gemma 4 mm_token_type_ids defaulting are different model-family bugs." } ] }, @@ -14634,8 +14716,8 @@ "issue:44295|issue:45656", "issue:43116|issue:44190", "issue:30333|issue:35141", - "issue:45405|issue:45712", "issue:44222|issue:44297", + "issue:45405|issue:45712", "issue:22355|issue:28282", "issue:44060|issue:45125", "issue:44075|issue:46231", @@ -14660,14 +14742,14 @@ "issue:44263|issue:45092", "issue:44190|issue:46082" ], - "prepared_review_unit_hash": "10a153dfefeeef79c2024c3ca87c740b6c80ef0e0eb9342d1f6b72a481964742", + "prepared_review_unit_hash": "ee1cbba10a4aa09f4dce9177175f13c7a8b36b092248bf142f5faced20f071ba", "component_budget": { "node_count": 502, "item_count": 502, - "soft_pair_count": 1842, - "serialized_chars": 956182, - "estimated_input_tokens": 239046, - "estimated_eval_tokens": 478348 + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 }, "budget": { "node_count": 47, @@ -14682,192 +14764,192 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is a grab-bag of unrelated Transformers issues spanning imports, tokenizer regressions, config validation, training/runtime bugs, serialization, and model-specific accuracy problems. None of the soft pairs look like true duplicates or mergeable PR-equivalents.", - "confidence": 0.98, - "canonical_issue_reason": "No single canonical issue stands out: the items are heterogeneous and do not share one underlying bug or code path.", + "summary": "These soft pairs are all false positives: they share broad themes like tokenization, imports, or save_pretrained, but each points to a different concrete bug or code path.", + "confidence": 0.97, + "canonical_issue_reason": "No canonical issue: the items span unrelated failures, so there is no duplicate hub to collapse into one issue.", "canonical_pr_reason": null, - "best_issue_reason": "No global best issue for deduplication: the cluster is too diverse, so any single issue would be a poor representative.", + "best_issue_reason": "No single best issue overall; if a representative were forced, 42898 is one of the broadest regression reports, but it does not subsume the rest.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:42898", "right": "issue:43295", "accept": false, - "reason": "Both are regressions, but one is about `clean_up_tokenization_spaces` behavior in v5 while the other is about `processor.tokenizer`/image passthrough; different code paths." + "reason": "Both are v5 regressions, but one is about clean_up_tokenization_spaces and the other about processor.tokenizer/image passing; different bugs." }, { "left": "issue:42898", "right": "issue:45356", "accept": false, - "reason": "Both involve tokenizer behavior, but one is a generic v5 cleanup change and the other is a Kimi-K2.5 codec/fix_mistral_regex regression; not the same bug." + "reason": "Tokenizer regressions in general, but 45356 is a Kimi-K2.5 codec/regex warning issue, not the same behavior change as 42898." }, { "left": "issue:44295", "right": "issue:45656", "accept": false, - "reason": "`position_ids` buffer access and optimizer stepping twice under deepspeed are unrelated failures." + "reason": "Registering position_ids as a buffer causing read errors is unrelated to DeepSpeed making optimizer.step run twice." }, { "left": "issue:43116", "right": "issue:44190", "accept": false, - "reason": "Both touch example scripts, but one is empty multi-label outputs and the other is local dataset loading; different underlying issues." + "reason": "Both hit example scripts, but one is multi-label classification returning empty results and the other is local dataset loading failure in a different script." }, { "left": "issue:30333", "right": "issue:35141", "accept": false, - "reason": "MLflow reporting on failed training and embedding reinitialization after resize are unrelated." + "reason": "MLflow not reporting failed training is unrelated to output embeddings being reinitialized after resize." }, { - "left": "issue:45405", - "right": "issue:45712", + "left": "issue:44222", + "right": "issue:44297", "accept": false, - "reason": "A dependency version bump and leftover dummy classes leaking into `dir(transformers)` are different problems." + "reason": "FP8 MoE save_pretrained problems and tokenizer_class mismatch in tokenizer_config.json are different save paths and failures." }, { - "left": "issue:44222", - "right": "issue:44297", + "left": "issue:45405", + "right": "issue:45712", "accept": false, - "reason": "Both mention `save_pretrained`, but one is FP8/moe model saving and the other is tokenizer metadata mismatch; not the same concrete bug." + "reason": "A dependency minimum-version bump is unrelated to leftover dummy classes leaking into the package namespace." }, { "left": "issue:22355", "right": "issue:28282", "accept": false, - "reason": "Missing `transformers.onnx` and missing PyTorch for `AutoModel` are separate import/dependency issues." + "reason": "Both are import errors, but one is missing transformers.onnx and the other is missing PyTorch for AutoModel; different causes." }, { "left": "issue:44060", "right": "issue:45125", "accept": false, - "reason": "Both are Qwen MoE-related, but one is a tied-weights warning bug and the other is a missing tensor-parallel plan; different code paths." + "reason": "An incorrect tied-weights warning is not the same as a missing _tp_plan for tensor parallelism." }, { "left": "issue:44075", "right": "issue:46231", "accept": false, - "reason": "An optimizer argument bug and a documentation parameter-name typo are not the same underlying issue." + "reason": "Unused SGD arguments in the optimizer are a code bug, while the set_training issue is just a documentation parameter-name mismatch." }, { "left": "issue:44933", "right": "issue:45712", "accept": false, - "reason": "Nonexistent `image_utils` import and dummy class leakage are unrelated import/export problems." + "reason": "A nonexistent image_utils import and dummy classes leaking into dir(transformers) are separate packaging issues." }, { "left": "issue:44360", "right": "issue:46762", "accept": false, - "reason": "A DSA indexer activation issue and Minimax attention numerical drift are different subsystems and symptoms." + "reason": "The DSA indexer missing a ReLU is an architecture/design discussion, not the same as Minimax attention numerical differences." }, { "left": "issue:42831", "right": "issue:44623", "accept": false, - "reason": "FP8 accuracy problems and missing files in `processor.save_pretrained` are unrelated." + "reason": "FineGrainedFP8 accuracy loss is unrelated to processor.save_pretrained omitting files." }, { "left": "issue:44297", "right": "issue:45656", "accept": false, - "reason": "Tokenizer config mismatch and double optimizer stepping have no shared code path." + "reason": "Tokenizer metadata mismatch on save is unrelated to an optimizer stepping twice under DeepSpeed." }, { "left": "issue:42796", "right": "issue:43723", "accept": false, - "reason": "Both are tokenizer-related, but one is BPE marker handling and the other is `AutoTokenizer.from_pretrained` loading in v5; not the same bug." + "reason": "Ministral BPE-marker handling and AutoTokenizer.from_pretrained loading failures are different tokenizer bugs." }, { "left": "issue:44222", "right": "issue:44623", "accept": false, - "reason": "Both involve saving, but one is FP8 model serialization and the other is missing processor files; different artifacts and fixes." + "reason": "FP8 save_pretrained for MoE and processor save_pretrained missing files are different components and failure modes." }, { "left": "issue:44448", "right": "issue:46132", "accept": false, - "reason": "A Pegasus output regression and `AttentionInterface.register` behavior change are different issues." + "reason": "Pegasus output regression across versions is unrelated to AttentionInterface.register changing callback behavior." }, { "left": "issue:45245", "right": "issue:46132", "accept": false, - "reason": "Category-count overflow and attention registration behavior are unrelated." + "reason": "Category-count runtime limits and AttentionInterface.register semantics are unrelated." }, { "left": "issue:43479", "right": "issue:43644", "accept": false, - "reason": "Default config initialization with `None` and non-persistent buffers being filled with junk are different bugs." + "reason": "Bad default vision/audio config initialization is unrelated to non-persistent buffers being filled with junk." }, { "left": "issue:41628", "right": "issue:42757", "accept": false, - "reason": "They are both import errors, but for different symbols and modules; no shared root cause is evident." + "reason": "Both are missing-import reports, but for different symbols in different packages with different fixes." }, { "left": "issue:44373", "right": "issue:44485", "accept": false, - "reason": "A wrong docstring and a RoPE implementation discussion are not duplicates." + "reason": "A wrong docstring for position_ids is not the same as a GLM-5 RoPE implementation bug." }, { "left": "issue:44297", "right": "issue:45310", "accept": false, - "reason": "Tokenizer class metadata mismatch and Qwen3.5 MoE `from_pretrained` failure are different problems." + "reason": "Tokenizer save metadata mismatch and Qwen3.5 MoE from_pretrained failure are different issues." }, { "left": "issue:46132", "right": "issue:46710", "accept": false, - "reason": "Behavior change in `AttentionInterface.register` and an accuracy regression after upgrade are unrelated." + "reason": "AttentionInterface.register behavior changes do not match a model accuracy regression after upgrading Transformers." }, { "left": "issue:45071", "right": "issue:46082", "accept": false, - "reason": "Both concern config validation, but one is `PretrainedConfig` type checking and the other is `LlamaConfig` head_dim constraints; not the same bug." - }, - { - "left": "issue:44263", - "right": "issue:45092", - "accept": false, - "reason": "A `torch.split` return-value issue in GlmMoeDsaIndexer and an InternVL2 meta-init incompatibility are unrelated." - }, - { - "left": "issue:44190", - "right": "issue:46082", - "accept": false, - "reason": "Example-script dataset loading and LlamaConfig head_dim validation are unrelated." + "reason": "PretrainedConfig type-checking regression is unrelated to LlamaConfig rejecting explicit head_dim under divisibility constraints." }, { "left": "issue:44222", "right": "issue:44484", "accept": false, - "reason": "FP8/MoE save-pretrained bug and the default 50GB shard-size question are unrelated." + "reason": "FP8 save_pretrained bugs and the default max_shard_size question are not the same problem." }, { "left": "issue:43723", "right": "issue:44977", "accept": false, - "reason": "Tokenizer loading in v5 and Qwen3.5 flash-attention generation failure are different issues." + "reason": "Tokenizer loading issues in v5 are unrelated to Qwen3.5 flash-attention generation failures." }, { "left": "issue:46614", "right": "issue:46932", "accept": false, - "reason": "Stale phonemizer language state and Gemma4 quantized embedder dtype casting are unrelated." + "reason": "Stale phonemizer language backend state is unrelated to Gemma4 quantized vision/audio embedder dtype casting." }, { "left": "issue:43122", "right": "issue:46209", "accept": false, - "reason": "Tokenization changes across versions and FP8 inference on B200 are different bugs with different failure modes." + "reason": "Tokenization differences across versions and FP8 inference failing on B200 are unrelated bugs." + }, + { + "left": "issue:44263", + "right": "issue:45092", + "accept": false, + "reason": "torch.split return handling in GlmMoeDsaIndexer is unrelated to remote-code checkpoint incompatibility with v5 meta initialization." + }, + { + "left": "issue:44190", + "right": "issue:46082", + "accept": false, + "reason": "Local dataset loading in an example script is unrelated to LlamaConfig head_dim validation." } ] }, @@ -14953,15 +15035,15 @@ "issue:42831|issue:45310", "issue:45092|issue:45357", "issue:46369|issue:46909", - "issue:44279|issue:46328", "issue:42907|issue:46050", + "issue:44279|issue:46328", "issue:45381|issue:46305", "issue:44261|issue:45092", - "issue:44483|issue:45406", "issue:22355|issue:27957", - "issue:44829|issue:46002", + "issue:44483|issue:45406", "issue:43644|issue:46305", "issue:42831|issue:44038", + "issue:44829|issue:46002", "issue:41628|issue:43824", "issue:43408|issue:46402", "issue:42831|issue:45753", @@ -14972,20 +15054,20 @@ "issue:42757|issue:45278", "issue:27957|issue:30990" ], - "prepared_review_unit_hash": "d78a24dec8c955ace178cf42f3b54dcf83be067edf2d845713e349792aca21f4", + "prepared_review_unit_hash": "5e4b41f9f37573d035a6069c0268053acae84c75c85fe86d9e91021d52943b75", "component_budget": { "node_count": 502, "item_count": 502, - "soft_pair_count": 1842, - "serialized_chars": 956182, - "estimated_input_tokens": 239046, - "estimated_eval_tokens": 478348 + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 30, - "serialized_chars": 36163, + "serialized_chars": 36161, "estimated_input_tokens": 9041, "estimated_eval_tokens": 18338 }, @@ -14994,192 +15076,192 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly heterogeneous bug reports and regression reports across unrelated Transformers subsystems. The soft-similarity links are mostly superficial keyword matches; none look like true duplicates of the same concrete bug/change.", - "confidence": 0.84, - "canonical_issue_reason": "Issue 45278 is the broadest umbrella-style report in the set, covering multiple import regressions after upgrading Transformers; however, it is still not a true duplicate of the more specific issues.", + "summary": "This set is not a single duplicate cluster; it spans many unrelated bugs across tokenization, imports, quantization, SAM3/Qwen models, cache/offline behavior, and docs. None of the soft pairs are the same concrete underlying issue.", + "confidence": 0.97, + "canonical_issue_reason": "No clear canonical issue: the items are heterogeneous and do not share one underlying bug or change to consolidate around.", "canonical_pr_reason": null, - "best_issue_reason": "45278 is the best single issue to anchor the cluster because it is the most general and likely to attract related import-regression reports, but the cluster is too diverse for a strong canonical duplicate.", + "best_issue_reason": "No issue is a strong global canonical representative because the set is too diverse; any single issue would be overly narrow and misleading.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43232", "right": "issue:46424", "accept": false, - "reason": "Different bugs: generation kwargs sync logic vs cache length handling." + "reason": "Different code paths: generation kwargs after sync_gpus vs cache max length handling." }, { "left": "issue:43232", "right": "issue:45834", "accept": false, - "reason": "Different code paths and symptoms: generation/sync_gpus vs OCR index error." + "reason": "Unrelated generation sync bug vs Kosmos2.5 OCR index error." }, { "left": "issue:44295", "right": "issue:44623", "accept": false, - "reason": "Unrelated: buffer read error for position_ids vs missing files on processor save." + "reason": "Position_ids buffer access error is unrelated to missing processor save files." }, { "left": "issue:43475", "right": "issue:46305", "accept": false, - "reason": "Different models and defects: missing SAM3 attribute vs Moonshine off-by-one mask bug." + "reason": "SAM3 attribute error and MoonshineStreaming mask off-by-one are different models and bugs." }, { "left": "issue:46209", "right": "issue:46431", "accept": false, - "reason": "FP8 inference failure on B200 is unrelated to missing chat template support." + "reason": "FP8 B200 inference failure is unrelated to processor chat-template/save_pretrained behavior." }, { "left": "issue:44263", "right": "issue:44360", "accept": false, - "reason": "Same subsystem but different concrete defects in DSA indexer; not the same bug." + "reason": "Both touch DSA indexer, but one is split output behavior and the other is missing ReLU; not the same bug." }, { "left": "issue:46305", "right": "issue:46655", "accept": false, - "reason": "Both are video/streaming edge cases, but the actual bugs differ: mask off-by-one vs padding calculation." + "reason": "Different processors/models: streaming mask logic vs video padding calculation." }, { "left": "issue:44263", "right": "issue:44373", "accept": false, - "reason": "Indexer torch.split behavior vs a docstring issue for position_ids; not duplicates." + "reason": "Indexer tensor split bug vs docstring issue for position_ids." }, { "left": "issue:30990", "right": "issue:38175", "accept": false, - "reason": "Sentence Transformers loading hang and zero probabilities in SigLIP2 are unrelated." + "reason": "Sentence Transformers loading hang is unrelated to SigLIP2 zero probabilities." }, { "left": "issue:42831", "right": "issue:45310", "accept": false, - "reason": "Different failures: FineGrainedFP8 accuracy regression vs Qwen3.5 MoE from_pretrained error." + "reason": "FineGrainedFP8 accuracy issue is unrelated to Qwen3.5 MoE from_pretrained failure." }, { "left": "issue:45092", "right": "issue:45357", "accept": false, - "reason": "Both concern Qwen3.5/InternVL2-style loading, but one is meta-init compatibility and the other is save_pretrained key mapping." + "reason": "Different models and failures: InternVL2 meta-init incompatibility vs Qwen3.5 save_pretrained key regression." }, { "left": "issue:46369", "right": "issue:46909", "accept": false, - "reason": "Memory overflow in sam3-litetext is unrelated to a broken documentation example." + "reason": "SAM3 LiteText memory overflow is unrelated to a broken DIA documentation example." }, { - "left": "issue:44279", - "right": "issue:46328", + "left": "issue:42907", + "right": "issue:46050", "accept": false, - "reason": "Dependency issue vs SequenceFeatureExtractor.pad performance inefficiency; not the same bug." + "reason": "Saving dequantized Ministrals/Devstrals differs from sinq quantized save/load corruption." }, { - "left": "issue:42907", - "right": "issue:46050", + "left": "issue:44279", + "right": "issue:46328", "accept": false, - "reason": "Both involve quantized model persistence, but they are different quantization paths and failure modes." + "reason": "General dependency issue is unrelated to SequenceFeatureExtractor.pad performance regression." }, { "left": "issue:45381", "right": "issue:46305", "accept": false, - "reason": "Different models and logic: Qwen2.5-VL vision_position_ids vs MoonshineStreaming mask windowing." + "reason": "Qwen2.5-VL video position_ids bug is unrelated to MoonshineStreaming window masking." }, { "left": "issue:44261", "right": "issue:45092", "accept": false, - "reason": "MLA norm epsilon mismatch is unrelated to InternVL2 meta initialization incompatibility." - }, - { - "left": "issue:44483", - "right": "issue:45406", - "accept": false, - "reason": "API request rejection and Gemma4Processor _tokenizer attribute error are unrelated." + "reason": "Layernorm epsilon precision mismatch is unrelated to remote-code checkpoint meta initialization." }, { "left": "issue:22355", "right": "issue:27957", "accept": false, - "reason": "Transformers.onnx import failure and XLMRoberta Flash Attention 2 are unrelated." + "reason": "ONNX import failure and Flash Attention 2 behavior are unrelated." }, { - "left": "issue:44829", - "right": "issue:46002", + "left": "issue:44483", + "right": "issue:45406", "accept": false, - "reason": "Both mention training/initialization problems, but they are different failure modes and models." + "reason": "v5.3 chat/completions request rejection is unrelated to Gemma4Processor missing _tokenizer." }, { "left": "issue:43644", "right": "issue:46305", "accept": false, - "reason": "Non-persistent buffer junk and Moonshine window masking are unrelated bugs." + "reason": "Non-persistent buffer junk in v5.0 is unrelated to streaming mask off-by-one." }, { "left": "issue:42831", "right": "issue:44038", "accept": false, - "reason": "FineGrainedFP8 accuracy regression is unrelated to the Qwen3-VL-Moe from_pretrained bug." + "reason": "FP8 accuracy issue is unrelated to Qwen3-VL-Moe from_pretrained errors." + }, + { + "left": "issue:44829", + "right": "issue:46002", + "accept": false, + "reason": "Flash Attention 3 training degeneration is unrelated to NaNs on checkout/init." }, { "left": "issue:41628", "right": "issue:43824", "accept": false, - "reason": "Both are import errors, but for different symbols and likely different missing exports." + "reason": "Both are import errors, but different missing symbols and likely different causes." }, { "left": "issue:43408", "right": "issue:46402", "accept": false, - "reason": "Model type warning during instantiation vs missing text encoder weights loaded in SAM3." + "reason": "SAM3 tracker/model-type warning is unrelated to missing text encoder weights in SAM3 model loading." }, { "left": "issue:42831", "right": "issue:45753", "accept": false, - "reason": "Different problems: FP8 accuracy issue vs Qwen3_5 infinite loop on a specific image." + "reason": "FP8 accuracy issue and Qwen3_5 infinite loop are different symptoms and code paths." }, { "left": "issue:22355", "right": "issue:28218", "accept": false, - "reason": "No module transformers.onnx and tokenizer spacing after added token are unrelated." + "reason": "ONNX module import failure is unrelated to tokenizer added-token spacing." }, { "left": "issue:39401", "right": "issue:44568", "accept": false, - "reason": "Both are tokenizer bugs, but they affect different models and different mechanics." + "reason": "Both are tokenizer bugs, but offset_mapping and add_special_tokens/BOS-EOS are different defects." }, { "left": "issue:42886", "right": "issue:43066", "accept": false, - "reason": "Offline cache loading bug is unrelated to wrong tokenizer decoder type in v5." + "reason": "Offline cache loading failure is unrelated to wrong tokenizer decoder type in v5." }, { "left": "issue:42886", "right": "issue:42994", "accept": false, - "reason": "Tokenizer offline cache loading and quantized model saving are unrelated." + "reason": "Tokenizer cache loading and quantized model saving are separate issues." }, { "left": "issue:42757", "right": "issue:45278", "accept": false, - "reason": "A specific huggingface_hub import error is not the same as a broader Transformers import-regression report." + "reason": "A single huggingface_hub import error is not the same as many import errors after a transformers upgrade." }, { "left": "issue:27957", "right": "issue:30990", "accept": false, - "reason": "Flash Attention 2 on XLMRoberta and Sentence Transformers loading hang are unrelated." + "reason": "XLMRoberta Flash Attention 2 issue is unrelated to Sentence Transformers loading hang." } ] }, @@ -15284,220 +15366,220 @@ "issue:42907|issue:43452", "issue:44038|issue:44295" ], - "prepared_review_unit_hash": "8539f268ba0bb827f390c4d1efa3e5857a365f7d78ee0d5bef0ac2fb3f1d5917", + "prepared_review_unit_hash": "6b6e8c69bc4451e4095ad1266da23e4e1f739e25e3be204bce4d66542fc05b45", "component_budget": { "node_count": 502, "item_count": 502, - "soft_pair_count": 1842, - "serialized_chars": 956182, - "estimated_input_tokens": 239046, - "estimated_eval_tokens": 478348 + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 31, - "serialized_chars": 35938, - "estimated_input_tokens": 8985, - "estimated_eval_tokens": 18226 + "serialized_chars": 35942, + "estimated_input_tokens": 8986, + "estimated_eval_tokens": 18228 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Most items are distinct regressions across tokenization, loading, generation, multimodal chat templates, and model-specific bugs. The only loose thematic overlap is among version-upgrade import failures, but the concrete failure modes still differ, so I rejected all soft edges.", - "confidence": 0.64, - "canonical_issue_reason": "issue:45278 is the broadest umbrella report in this set for upgrade-related import failures, making it the best anchor issue among the title-only evidence.", + "summary": "This cluster is broad and mostly heterogeneous: a few issues relate to v5 upgrade/import regressions, but most are model-specific tokenizer, generation, or multimodal bugs. None of the soft pairs look like the same underlying bug or change.", + "confidence": 0.42, + "canonical_issue_reason": "issue:45278 is the broadest representative of the cluster\u2019s main theme: upgrade regressions in v5 causing import failures and API breakage.", "canonical_pr_reason": null, - "best_issue_reason": "issue:45278 has the most general framing, active discussion, and the highest chance of representing the shared import-regression family better than the narrower symptom-specific reports.", + "best_issue_reason": "issue:45278 is the best global representative because it captures the widest, most reusable symptom in the cluster (many import errors after upgrading).", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:44188", "right": "issue:44373", "accept": false, - "reason": "Different bug classes: attention-kernel divergence vs a docstring issue." + "reason": "Different bugs: torch.compile attention-kernel divergence vs a docstring issue for position_ids." }, { "left": "issue:43408", "right": "issue:46369", "accept": false, - "reason": "Both mention SAM3-related areas, but one is a model-type warning and the other is a memory overflow on a different model path." + "reason": "Different SAM 3 problems: model-type warning on import/config vs memory overflow in sam3-litetext." }, { "left": "issue:43232", "right": "issue:45593", "accept": false, - "reason": "Generation kwargs handling and D-FINE auxiliary-loss behavior are unrelated code paths." + "reason": "Unrelated: generation kwargs after sync_gpus vs D-FINE auxiliary-loss behavior." }, { "left": "issue:42796", "right": "issue:43122", "accept": false, - "reason": "Both concern tokenization, but one is a specific BPE-marker bug and the other is a version-to-version tokenization change report." + "reason": "Both tokenizer-related, but one is BPE marker handling for a specific model and the other is a cross-version tokenization drift report; not the same bug." }, { "left": "issue:43646", "right": "issue:46002", "accept": false, - "reason": "Custom model initialization breakage and NaNs in classification heads are different failures." + "reason": "Custom model initialization breakage is distinct from NaNs in classification heads after checkout/init." }, { "left": "issue:45381", "right": "issue:46909", "accept": false, - "reason": "Qwen2.5-VL video position IDs and a DIA documentation example are unrelated." + "reason": "Multimodal video position-id bug vs broken documentation example; unrelated." }, { "left": "issue:43540", "right": "issue:46909", "accept": false, - "reason": "Video-input ValueError in Qwen3OmniMoe is not the same as a broken docs example." + "reason": "Qwen3OmniMoe video-processing ValueError is unrelated to a DIA docs example." }, { "left": "issue:30990", "right": "issue:33357", "accept": false, - "reason": "Both are loading failures, but they affect different models/platforms and describe different symptoms." + "reason": "Different loading failures: Sentence Transformers getting stuck vs MacOS bus error with a CLIP model." }, { "left": "issue:42907", "right": "issue:44945", "accept": false, - "reason": "Dequantized save failure and pipeline-parallel output corruption are different bugs." + "reason": "Dequantized Ministrals save failure is unrelated to incorrect pipeline-parallel LLM outputs." }, { "left": "issue:44295", "right": "issue:44297", "accept": false, - "reason": "Reading a registered buffer and tokenizer-class metadata mismatch are separate tokenizer/model serialization issues." + "reason": "Position_ids buffer access error and tokenizer_class mismatch are different tokenizer serialization bugs." }, { "left": "issue:43576", "right": "issue:43784", "accept": false, - "reason": "Broken env command and a NameError during sentence-transformers import do not share the same failure." + "reason": "Both are v5 breakages, but one is the env command and the other is an import NameError; different code paths." }, { "left": "issue:42371", "right": "issue:44263", "accept": false, - "reason": "TF32 API guidance and a torch.split return-value issue are unrelated." + "reason": "TF32 API guidance and torch.split return semantics in GlmMoeDsaIndexer are unrelated." }, { "left": "issue:39401", "right": "issue:44488", "accept": false, - "reason": "Wrong offset mappings and a model load failure are different tokenizer/model-loading problems." + "reason": "Qwen3 offset_mapping bug vs inability to load a separate model; not the same issue." }, { "left": "issue:42907", "right": "issue:44964", "accept": false, - "reason": "Saving dequantized models and loading Phi-4 multimodal instruct are not the same issue." + "reason": "Saving dequantized Ministrals/Devstrals is unrelated to loading Phi-4 multimodal." }, { "left": "issue:43976", "right": "issue:44945", "accept": false, - "reason": "Python-version support regression and pipeline-parallel incorrect output are different problems." + "reason": "Python version support regression vs pipeline-parallel output corruption; unrelated." }, { "left": "issue:46172", "right": "issue:46762", "accept": false, - "reason": "DeepSeek tensor transpose logic and Minimax attention numerical differences are different model-specific bugs." + "reason": "Both involve numerical behavior, but they are different models and different attention implementations; not mergeable as one fix." }, { "left": "issue:42907", "right": "issue:43937", "accept": false, - "reason": "Model save/dequantization failure and invalid GenerationConfig are unrelated." + "reason": "Dequantized model saving and invalid GenerationConfig for GLM-5 are separate problems." }, { "left": "issue:43756", "right": "issue:44060", "accept": false, - "reason": "RoPE-layer dropping and tied-weights warning logic are different model configuration bugs." + "reason": "RoPE layer selection in Smollm3 is unrelated to tied-weights warnings in Qwen3-Next." }, { "left": "issue:41628", "right": "issue:45278", "accept": false, - "reason": "The latter is a broad umbrella import-regression report, but this pair does not clearly show the same concrete import failure." + "reason": "Both mention import failures, but one is a specific missing AutoImageProcessor symbol and the other is broad post-upgrade import breakage; not enough evidence of the same root cause." }, { "left": "issue:43784", "right": "issue:45600", "accept": false, - "reason": "Sentence-transformers import NameError and stale auto_mappings references are different breakages." + "reason": "Import NameError in sentence-transformers is unrelated to stale auto_mappings references breaking CI." }, { "left": "issue:44117", "right": "issue:44704", "accept": false, - "reason": "Tokenizer mapping lookup and AutoProcessor kwargs forwarding are separate API bugs." + "reason": "TOKENIZER_MAPPING_NAMES returning None and AutoProcessor not forwarding kwargs are different from_pretrained plumbing bugs." }, { "left": "issue:39401", "right": "issue:44521", "accept": false, - "reason": "Offset-mapping correctness and multimodal assistant-mask generation are unrelated tokenizer behaviors." + "reason": "Tokenizer offset mapping and multimodal assistant masks are distinct tokenizer/template issues." }, { "left": "issue:44704", "right": "issue:45290", "accept": false, - "reason": "Both involve apply_chat_template, but one is cached-file kwarg forwarding while the other is a crash on tool-call assistant messages." + "reason": "AutoProcessor cached_file kwargs and apply_chat_template tool-call crash are unrelated APIs." }, { "left": "issue:44625", "right": "issue:44829", "accept": false, - "reason": "Config propagation for num_labels and flash-attention training degeneration are unrelated." + "reason": "Qwen3.5 num_labels propagation and flash_attention_3 training degeneration are unrelated." }, { "left": "issue:44360", "right": "issue:44485", "accept": false, - "reason": "DSA indexer activation shape/behavior and GLM-5 RoPE implementation are different issues." + "reason": "DSA indexer ReLU discussion and GLM-5 RoPE implementation are different model internals." }, { "left": "issue:43408", "right": "issue:43475", "accept": false, - "reason": "Both are SAM3-related, but the warning about model types is not the same as the missing fpn_position_embeddings attribute." + "reason": "Both are SAM 3 Video, but one is a config/type warning and the other is a missing attribute in the encoder output; different bugs." }, { "left": "issue:43122", "right": "issue:44521", "accept": false, - "reason": "Tokenizer version drift and multimodal assistant-mask output are different behaviors." + "reason": "Cross-version tokenization drift and all-zero assistant masks are different tokenizer/template failures." }, { "left": "issue:44038", "right": "issue:44297", "accept": false, - "reason": "Qwen3-VL-Moe compatibility and tokenizer save metadata mismatch are unrelated." + "reason": "Qwen3-VL-Moe bug and tokenizer_class mismatch are unrelated." }, { "left": "issue:43122", "right": "issue:46326", "accept": false, - "reason": "Different tokenization reports with different tokenizer stacks and failure modes." + "reason": "Tokenization differences across versions and mistral_common continue_final_message behavior are not the same underlying bug." }, { "left": "issue:42907", "right": "issue:43452", "accept": false, - "reason": "Dequantized save failure and gguf_file broken from_pretrained loading are separate model I/O bugs." + "reason": "Dequantized model saving and gguf_file loading for AutoTokenizer/AutoModelForCausalLM are unrelated." }, { "left": "issue:44038", "right": "issue:44295", "accept": false, - "reason": "Qwen3-VL-Moe issues and registered-buffer position_ids reading are different code paths." + "reason": "Qwen3-VL-Moe issue and position_ids buffer read problem are different code paths." } ] }, @@ -15571,8 +15653,8 @@ ], "soft_pairs": [ "issue:45216|issue:45478", - "issue:39401|issue:46614", "issue:22355|issue:29159", + "issue:39401|issue:46614", "issue:44038|issue:44222", "issue:43824|issue:44351", "issue:43116|issue:45200", @@ -15589,8 +15671,8 @@ "issue:38617|issue:45278", "issue:42898|issue:46614", "issue:43066|issue:44779", - "issue:43611|issue:46002", "issue:44038|issue:45478", + "issue:43611|issue:46002", "issue:43976|issue:45446", "issue:44355|issue:46231", "issue:43116|issue:43493", @@ -15601,14 +15683,14 @@ "issue:45083|issue:45446", "issue:44610|issue:46909" ], - "prepared_review_unit_hash": "dabea3befca64c8474c997d63eeb64e818f3ef43bc3e56af9ae17985afac432f", + "prepared_review_unit_hash": "3787419c5da0287b38cdab3c4836b075d1067a326b1419041edbe5174afb59a3", "component_budget": { "node_count": 502, "item_count": 502, - "soft_pair_count": 1842, - "serialized_chars": 956182, - "estimated_input_tokens": 239046, - "estimated_eval_tokens": 478348 + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 }, "budget": { "node_count": 47, @@ -15623,186 +15705,186 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Most pairs are clearly unrelated. The Qwen3.5 save/load pair is suggestive but still too weak to treat as the same root bug without direct linkage in the issue bodies, so it should be kept separate or treated as tentative at most.", - "confidence": 0.73, - "canonical_issue_reason": "issue:45216 is the best anchor for the Qwen3.5 serialization problem because it is the more direct report of the save_pretrained regression, but the connection to the from_pretrained failure in issue:45478 is still only suggestive.", + "summary": "The cluster is mostly heterogeneous and does not contain a clear duplicate set. A few items revolve around Qwen3.5 save/load/tokenizer regressions, but they still look like separate bugs rather than one mergeable fix.", + "confidence": 0.87, + "canonical_issue_reason": "Issue 45216 is the strongest anchor among the related reports: it is a concrete, version-scoped Qwen3.5 save_pretrained regression with inbound references and a precise failure mode.", "canonical_pr_reason": null, - "best_issue_reason": "issue:45216 is the strongest issue candidate for the Qwen3.5 save/load cluster because it describes the upstream checkpoint-writing problem more directly than the downstream load error.", + "best_issue_reason": "Issue 45216 is the best representative issue if one must be chosen, because it is specific, actionable, and closest to the small Qwen3.5 serialization subgroup.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:45216", "right": "issue:45478", "accept": false, - "reason": "Related area, but the packet only shows a bad save_pretrained checkpoint on one side and a from_pretrained error on the other; that is not enough to confirm the same root bug." + "reason": "Same model family and release window, but one is a save_pretrained regression and the other is a from_pretrained import/load error; too different to treat as the same bug." }, { - "left": "issue:39401", - "right": "issue:46614", + "left": "issue:22355", + "right": "issue:29159", "accept": false, - "reason": "Different model families and failure modes: Qwen3 tokenizer offset mapping vs Wav2Vec2 phonemizer backend state." + "reason": "Unrelated topics: ONNX module import failure vs tokenizer slow/fast inconsistency." }, { - "left": "issue:22355", - "right": "issue:29159", + "left": "issue:39401", + "right": "issue:46614", "accept": false, - "reason": "Missing transformers.onnx module vs tokenizer fast/slow inconsistency are unrelated bugs." + "reason": "Different subsystems and symptoms: tokenizer offset mapping vs phonemizer language backend state." }, { "left": "issue:44038", "right": "issue:44222", "accept": false, - "reason": "Both mention Qwen/MoE, but the concrete problems differ: Qwen3-VL-Moe bug vs FP8 save_pretrained issue." + "reason": "Both mention Qwen/MoE, but one is a general transformers 5.0 compatibility bug and the other is FP8 save_pretrained behavior; not the same concrete failure." }, { "left": "issue:43824", "right": "issue:44351", "accept": false, - "reason": "Both are import errors, but they involve different missing symbols and likely different root causes." + "reason": "Both are import errors, but for different missing symbols and likely different fixes." }, { "left": "issue:43116", "right": "issue:45200", "accept": false, - "reason": "Different features and symptoms: example-script multi-label output vs Gemma 4 mm_token_type_ids defaulting." + "reason": "Classification example script returns empty results vs Gemma 4 multimodal token type IDs defaulting; unrelated bugs." }, { "left": "issue:43122", "right": "issue:44779", "accept": false, - "reason": "Both are tokenizer regressions, but they affect different model families and different behaviors." + "reason": "Both are tokenizer regressions, but they affect different model families and different observed failures." }, { "left": "issue:42757", "right": "issue:44351", "accept": false, - "reason": "Different packages and imports: huggingface_hub vs transformers." + "reason": "Different import sources and symbols: huggingface_hub import error vs transformers HybridCache import error." }, { "left": "issue:36296", "right": "issue:43116", "accept": false, - "reason": "Tensor-parallel training bug is unrelated to the classification example script issue." + "reason": "Tensor parallel training bug is unrelated to the multi-label classification example issue." }, { "left": "issue:43278", "right": "issue:44945", "accept": false, - "reason": "Different code paths and symptoms: train/eval dtype drift vs incorrect output under pipeline parallelism." + "reason": "Embedding dtype drift in eval is not the same as incorrect output under pipeline parallelism." }, { "left": "issue:42175", "right": "issue:44938", "accept": false, - "reason": "Package/backend install issue is unrelated to Python 3.14 load failure." + "reason": "Package backend dependency issue vs Python 3.14 load failure; no shared bug." }, { "left": "issue:22355", "right": "issue:29127", "accept": false, - "reason": "Missing module import vs LayoutLMv3 error-message clarity are not the same bug." + "reason": "ONNX import failure and LayoutLMv3 error-message clarity are unrelated." }, { "left": "issue:43493", "right": "issue:45753", "accept": false, - "reason": "SigLIP2 HF/JAX discrepancy and Qwen3_5 image infinite loop are different model-specific bugs." + "reason": "Both touch Qwen3_5 vision, but one is a HF-vs-JAX implementation discrepancy and the other is an infinite loop on one image; too different to merge." }, { "left": "issue:46231", "right": "issue:46877", "accept": false, - "reason": "Documentation parameter-name mismatch is unrelated to import-time stdout diagnostics." + "reason": "Documentation parameter name mismatch vs auto_docstring printing diagnostics at import time are unrelated." }, { "left": "issue:44188", "right": "issue:46328", "accept": false, - "reason": "Torch.compile attention-kernel divergence is unrelated to feature-extractor padding inefficiency." + "reason": "Attention-kernel divergence under torch.compile is unrelated to numpy/list padding performance." }, { "left": "issue:43937", "right": "issue:44485", "accept": false, - "reason": "Same model family, but different concrete bugs: invalid GenerationConfig vs RoPE implementation discussion." + "reason": "GenerationConfig validation failure and RoPE implementation discussion are different GLM-5 issues." }, { "left": "issue:38617", "right": "issue:45278", "accept": false, - "reason": "Both involve imports, but one is a specific missing symbol and the other is a broad upgrade regression; not enough to merge." + "reason": "Specific missing import vs broad post-upgrade import breakage; too coarse to call duplicate." }, { "left": "issue:42898", "right": "issue:46614", "accept": false, - "reason": "Tokenization cleanup behavior change is unrelated to stale phonemizer language state." + "reason": "Tokenization cleanup behavior changes and stale phonemizer language backend are unrelated." }, { "left": "issue:43066", "right": "issue:44779", "accept": false, - "reason": "Both are v5 tokenizer regressions, but they concern different tokenizer internals and different models." + "reason": "Both are v5 tokenizer regressions, but on different models and likely different code paths." }, { - "left": "issue:43611", - "right": "issue:46002", + "left": "issue:44038", + "right": "issue:45478", "accept": false, - "reason": "Loading via base_model_prefix and NaNs in classification heads are unrelated issues." + "reason": "Qwen3-VL-Moe compatibility bug vs Qwen3.5 Moe from_pretrained error; related area, but not the same underlying bug." }, { - "left": "issue:44038", - "right": "issue:45478", + "left": "issue:43611", + "right": "issue:46002", "accept": false, - "reason": "Related Qwen variants, but one is a Qwen3-VL-Moe issue and the other is a Qwen3.5 MoE from_pretrained failure; not the same concrete defect." + "reason": "Base model prefix loading regression is unrelated to NaNs in classification heads after init/checkout." }, { "left": "issue:43976", "right": "issue:45446", "accept": false, - "reason": "Python version compatibility issue is unrelated to the flex_attention version-check bug." + "reason": "Python version support regression is unrelated to the flex_attention AuxRequest version check." }, { "left": "issue:44355", "right": "issue:46231", "accept": false, - "reason": "Running compiled Python files and a documentation parameter-name mismatch are unrelated." + "reason": "Compiled Python file import errors are unrelated to a documentation typo." }, { "left": "issue:43116", "right": "issue:43493", "accept": false, - "reason": "Example-script output bug is unrelated to SigLIP2 implementation parity." + "reason": "Classification example bug is unrelated to a SigLIP2 implementation discrepancy." }, { "left": "issue:42673", "right": "issue:45072", "accept": false, - "reason": "VRAM leak in Qwen3ForCausalLM is unrelated to dtype mismatches in inference tests." + "reason": "VRAM leak in multi-threaded dataloader use is unrelated to dtype mismatch in bfloat16 inference." }, { "left": "issue:45357", "right": "issue:45478", "accept": false, - "reason": "Both are Qwen3.5-related, but one is a visual-encoder key serialization bug and the other is a from_pretrained load error; not clearly the same defect." + "reason": "Both are Qwen3.5 save/load regressions, but one is incorrect visual encoder key saving and the other is a from_pretrained error; similar area, distinct bug." }, { "left": "issue:44625", "right": "issue:46002", "accept": false, - "reason": "Qwen3.5 num_labels propagation and NaNs in classification heads are different bugs." + "reason": "Qwen3.5 num_labels propagation and NaNs in classification heads are not the same failure." }, { "left": "issue:42371", "right": "issue:43874", "accept": false, - "reason": "TF32 backend-setting guidance issue is unrelated to the missing image-processor method error." + "reason": "TF32 behavior guidance and missing image-patch API are unrelated." }, { "left": "issue:45083", "right": "issue:45446", "accept": false, - "reason": "Different problems in different areas: helper length calculation vs PyTorch version gate for AuxRequest import." + "reason": "Feat-extract output length helper behavior in qwen3_omni_moe is unrelated to flex_attention PyTorch version checking." }, { "left": "issue:44610", @@ -15812,12 +15894,9 @@ } ] }, - "evaluator_result": { - "accept": false, - "feedback": "The analysis is mostly conservative on the unrelated pairs, but it overstates the Qwen3.5 save/load linkage. From the packet we only know one issue is about a bad `save_pretrained` checkpoint and the other is a `from_pretrained` error on Qwen3.5 MoE; that suggests a possible relation, but not enough to confidently call them the same root bug. The canonical-issue claim should stay tentative, and the accepted soft-edge verdict should be downgraded to uncertain/rejected unless there is direct linkage in the issue bodies." - }, - "evaluator_used": true, - "retried": true, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, "accepted_nontrivial_soft_edge": false, "error_kind": null, "error_message": null @@ -15914,202 +15993,202 @@ "issue:43937|issue:44977", "issue:43450|issue:43986" ], - "prepared_review_unit_hash": "b6162c723373f2ac2f6984e245c2762eff80183f2f46edbe3dfa473e5f4b360a", + "prepared_review_unit_hash": "6fdcd21c6b4eb6bf474436b9b764152d93a510dd3cf0232af0c2cfd709da0b6f", "component_budget": { "node_count": 502, "item_count": 502, - "soft_pair_count": 1842, - "serialized_chars": 956182, - "estimated_input_tokens": 239046, - "estimated_eval_tokens": 478348 + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 28, - "serialized_chars": 35840, - "estimated_input_tokens": 8960, - "estimated_eval_tokens": 18176 + "serialized_chars": 35842, + "estimated_input_tokens": 8961, + "estimated_eval_tokens": 18178 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "All suggested edges are broad topical similarities, not the same underlying bug/change. The set spans unrelated tokenizer, generation, vision, attention, and training issues, so every soft pair is rejected.", - "confidence": 0.97, - "canonical_issue_reason": "No single issue cleanly canonicalizes this set; the candidates are separate defects that only share broad subsystem keywords.", + "summary": "This cluster is highly heterogeneous: most pairs are only loosely similar by subsystem or wording and do not look like the same underlying bug. No PRs are present; the only plausible anchor issue is a broad tokenizer-regression report, but even that is a weak fit for the whole cluster.", + "confidence": 0.43, + "canonical_issue_reason": "issue:33453 is the broadest and most generic report in the set (tokenizer loading regression), so it is the least-specific issue to use as an anchor. That said, the cluster spans many unrelated bugs, so no issue is a strong canonical fit.", "canonical_pr_reason": null, - "best_issue_reason": "No issue is a strong global duplicate hub. If forced, issue 43937 is the most concrete v5-generation regression, but it still does not represent the rest of the cluster.", + "best_issue_reason": "issue:33453 is the most umbrella-like issue and could subsume nearby tokenizer regressions better than the other, more specialized reports. It is still not a great global representative because the cluster is not a true duplicate set.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43992", "right": "issue:44704", "accept": false, - "reason": "Different code paths: UMT5Encoder missing `embed_tokens.weight` vs AutoProcessor not forwarding kwargs to `cached_file`." + "reason": "Different code paths: encoder weight loading vs AutoProcessor/cached_file argument passing." }, { "left": "issue:44075", "right": "issue:44936", "accept": false, - "reason": "Optimizer SGD args not used is unrelated to `trainer.evaluate()` failing after `trainer.train()`." + "reason": "Unrelated bugs: SGD argument handling vs trainer.evaluate() failure after train()." }, { "left": "issue:43618", "right": "issue:46829", "accept": false, - "reason": "Both concern outputs, but one is missing CLIP attentions and the other maps video classifier hidden states incorrectly." + "reason": "Both are output-field issues, but they affect different model classes and different dataclass fields." }, { "left": "issue:43116", "right": "issue:43825", "accept": false, - "reason": "Example-script multi-label classification failure is unrelated to a pipeline() translation-task error message regression." + "reason": "Example-script multi-label output bug vs pipeline() error-message/docs mismatch; not the same failure." }, { "left": "issue:43066", "right": "issue:46710", "accept": false, - "reason": "Tokenizer decoder type bug and DeepSeek accuracy regression are different failures in different paths." + "reason": "Tokenizer decoder typing issue vs model accuracy regression; too far apart to be the same bug." }, { "left": "issue:28348", "right": "issue:46640", "accept": false, - "reason": "One is a Flash Attention 2 support request for GPT-2; the other is a specific flash-attention crash in a sink-less model." + "reason": "Enhancement request for GPT2 FlashAttention 2 support vs a separate Flash Attention bug report." }, { "left": "issue:44263", "right": "issue:46762", "accept": false, - "reason": "GLM-MoE `torch.split` return handling and Minimax attention numerical differences are unrelated bugs." + "reason": "GlmMoeDsa indexing bug vs Minimax attention numerical mismatch; different models and symptoms." }, { "left": "issue:46614", "right": "issue:46752", "accept": false, - "reason": "Stale phonemizer language/backend state is unrelated to `apply_chat_template` crashing on an empty list." + "reason": "Tokenizer phonemizer language-state bug vs empty-chat-template IndexError; unrelated." }, { "left": "issue:44995", "right": "issue:46697", "accept": false, - "reason": "Stale indexer cache causing a second-forward crash is unrelated to Swin2SR over-padding images." + "reason": "Second-forward stale cache crash vs image over-padding; different subsystems entirely." }, { "left": "issue:45092", "right": "issue:46762", "accept": false, - "reason": "Meta-initialization incompatibility for old InternVL2 checkpoints is unrelated to Minimax attention numerical drift." + "reason": "Remote-code/meta-init incompatibility vs attention numerical differences; no shared code-path." }, { "left": "issue:42222", "right": "issue:43441", "accept": false, - "reason": "VitPose model breakage and Ministral flash-attention failures are different model-specific issues." + "reason": "VitPose model breakage and Ministral FlashAttention failure are unrelated model-specific issues." }, { "left": "issue:43827", "right": "issue:45446", "accept": false, - "reason": "Docs still mentioning pipeline() is unrelated to the flex_attention PyTorch version check." + "reason": "Docs still mention pipeline() vs PyTorch version check in flex_attention; not the same bug." }, { "left": "issue:33453", "right": "issue:39401", "accept": false, - "reason": "Tokenizer loading regression and Qwen3 offset_mapping errors are not the same defect." + "reason": "Both are tokenizer-related, but one is loading regression and the other is offset_mapping correctness for a specific model." }, { "left": "issue:43064", "right": "issue:43673", "accept": false, - "reason": "Wrong optimizer state/params under FSDP2+PEFT is unrelated to missing GenerationMixin cache during chunked prefill." + "reason": "Training/FSDP optimizer-state corruption vs chunked_prefill cache missing during generation." }, { "left": "issue:42673", "right": "issue:42796", "accept": false, - "reason": "Qwen3 VRAM leak in dataloader threads is unrelated to Ministral tokenizer BPE marker handling." + "reason": "VRAM leak in threaded dataloader use vs tokenizer BPE-marker handling; different layers." }, { "left": "issue:43937", "right": "issue:45563", "accept": false, - "reason": "Invalid GenerationConfig and a stale num_return_sequences warning are different generation issues." + "reason": "GenerationConfig validation error vs stale warning in paged generate(); not the same defect." }, { "left": "issue:44380", "right": "issue:46077", "accept": false, - "reason": "GPT-2 attention scaling ignored under SDPA/FlashAttention is unrelated to Gemma4 building causal masks in bidirectional mode." + "reason": "Different attention bugs: ignored scaling config vs causal mask construction for bidirectional attention." }, { "left": "issue:43673", "right": "issue:46328", "accept": false, - "reason": "Generation cache missing during chunked prefill is unrelated to a SequenceFeatureExtractor.pad performance issue." + "reason": "Generation cache bug vs feature extractor padding performance issue; unrelated." }, { "left": "issue:42673", "right": "issue:46655", "accept": false, - "reason": "VRAM leakage in Qwen3ForCausalLM does not match SmolVLM video padding miscalculation." + "reason": "Threaded VRAM leak and video padding calculation are unrelated." }, { "left": "issue:42222", "right": "issue:46752", "accept": false, - "reason": "VitPose model breakage and empty-list chat template crashes are unrelated." + "reason": "VitPose breakage vs apply_chat_template empty-input crash; different components." }, { "left": "issue:44291", "right": "issue:45071", "accept": false, - "reason": "Unexpected `_is_hf_initialized` argument and PretrainedConfig type-checking are both v5-related but not the same code-path failure." + "reason": "Meta-init argument TypeError vs PretrainedConfig type-checking regression; similar release area, but not the same bug." }, { "left": "issue:45405", "right": "issue:45446", "accept": false, - "reason": "A PEFT minimum-version bump is unrelated to an AuxRequest import version check in flex_attention." + "reason": "Dependency version bump problem vs flex_attention import version check; unrelated." }, { "left": "issue:44117", "right": "issue:45588", "accept": false, - "reason": "Tokenizer mapping assumptions and an `s_aux=None` flash-attention crash are unrelated." + "reason": "Tokenizer mapping None-handling vs flash_attention sinkless-model AttributeError; different paths." }, { "left": "issue:39401", "right": "issue:44779", "accept": false, - "reason": "Both are tokenizer bugs, but they affect different models and different mechanics (`offset_mapping` vs overall generation/tokenization correctness)." + "reason": "Both are tokenizer regressions, but model-specific offset mapping and DeepSeek output errors are not shown to share a root cause." }, { "left": "issue:44410", "right": "issue:45310", "accept": false, - "reason": "Missing qwen3next projections and Qwen3.5 MoE from_pretrained errors are not the same defect." + "reason": "Qwen3Next missing projections vs Qwen3.5 MoE from_pretrained failure; related family, but insufficient evidence of same underlying bug." }, { "left": "issue:44466", "right": "issue:46002", "accept": false, - "reason": "`lm_head.weight` serialization inconsistency is unrelated to NaNs in classification heads after checkout/init." + "reason": "Serialization inconsistency of tied weights vs NaNs after checkout/init; different failure modes." }, { "left": "issue:43937", "right": "issue:44977", "accept": false, - "reason": "Invalid GenerationConfig is unrelated to Qwen3.5 generation failing under flash-attention." + "reason": "GenerationConfig invalid vs flash-attention generation failure; no clear shared code-path." }, { "left": "issue:43450", "right": "issue:43986", "accept": false, - "reason": "Batched video processor shape bugs are unrelated to AutoProcessor crashing without torchvision." + "reason": "Batched video processor shape bug vs AutoProcessor crash without torchvision; different causes." } ] }, @@ -16202,8 +16281,8 @@ "issue:44410|issue:45478", "issue:42371|issue:46617", "issue:44261|issue:46762", - "issue:43208|issue:46894", "issue:44933|issue:45278", + "issue:43208|issue:46894", "issue:38617|issue:44220", "issue:44075|issue:45656", "issue:43066|issue:43122", @@ -16213,222 +16292,229 @@ "issue:43673|issue:45341", "issue:41628|issue:45071", "issue:42222|issue:45542", - "issue:45071|issue:45542" + "issue:45071|issue:45542", + "issue:44484|issue:45478" ], - "prepared_review_unit_hash": "db1b7acc8220a3a082a4d0eeb80bd5bdfb0fc1e2d5096ba970e0c90c8e7166af", + "prepared_review_unit_hash": "eb626806da11ff3737f8d444d6d529bdeeb6593e4b2760445d923a93af103cd2", "component_budget": { "node_count": 502, "item_count": 502, - "soft_pair_count": 1842, - "serialized_chars": 956182, - "estimated_input_tokens": 239046, - "estimated_eval_tokens": 478348 + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 }, "budget": { "node_count": 48, "item_count": 48, - "soft_pair_count": 31, - "serialized_chars": 36567, - "estimated_input_tokens": 9142, - "estimated_eval_tokens": 18540 + "soft_pair_count": 32, + "serialized_chars": 36865, + "estimated_input_tokens": 9217, + "estimated_eval_tokens": 18690 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a grab-bag of unrelated Transformers regressions. The only clear duplicate-looking overlap is the Qwen3.5 MoE from_pretrained report pair; the rest should stay separate.", - "confidence": 0.39, - "canonical_issue_reason": "issue:45310 is the clearest duplicate target: it reports the same Qwen3.5 MoE from_pretrained regression as issue:45478, with a precise version-scoped failure description.", + "summary": "This cluster is a loose collection of distinct Transformers regression/bug reports, mostly around imports, tokenizers, model loading, and training behavior. I do not see any true duplicate pair among the provided soft edges.", + "confidence": 0.87, + "canonical_issue_reason": "issue:45278 is the broadest and most representative post-upgrade regression report; it best serves as an umbrella for the cluster\u2019s import/API-breakage theme, even though the other items are mostly separate bugs.", "canonical_pr_reason": null, - "best_issue_reason": "It is the most concrete and representative report among the items, and it matches an exact duplicate issue already present in the set.", + "best_issue_reason": "issue:45278 has the clearest cross-cutting impact and a general regression framing, making it the most useful anchor issue in an otherwise heterogeneous set.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:45341", "right": "issue:46328", "accept": false, - "reason": "Different bug classes: a testing_utils bug versus a SequenceFeatureExtractor.pad performance issue." + "reason": "One is a small testing_utils bug, the other is a SequenceFeatureExtractor.pad performance issue; different code paths and symptoms." }, { "left": "issue:45341", "right": "issue:45561", "accept": false, - "reason": "Both mention testing utils, but one is a captured_info.txt race under xdist and the other is a different testing_utils bug; not the same failure." + "reason": "Both mention testing utils, but one is a generic bug report and the other is a specific pytest-xdist race on captured_info.txt; not the same defect." }, { "left": "issue:44829", "right": "issue:45216", "accept": false, - "reason": "Flash-attention training instability and a Qwen3.5 save_pretrained regression are separate code paths." + "reason": "Flash-attention training degeneration and save_pretrained checkpoint corruption are different failure modes in different paths." }, { "left": "issue:45071", "right": "issue:46512", "accept": false, - "reason": "PretrainedConfig type-checking breakage is not the same as from_config(dtype=...) leaving config.dtype stale." + "reason": "PretrainedConfig type-checking breakage is unrelated to from_config(dtype) leaving config.dtype stale." }, { "left": "issue:44117", "right": "issue:45081", "accept": false, - "reason": "Tokenizer mapping None handling and a Mistral regex patch crash are different tokenizer bugs." + "reason": "A TOKENIZER_MAPPING_NAMES None-handling bug is not the same as the Mistral regex patch crash in tokenizer backend access." }, { "left": "issue:38617", "right": "issue:41628", "accept": false, - "reason": "Both are import errors, but for different symbols and modules; not one underlying bug." + "reason": "Both are import errors, but they target different symbols and modules; no evidence of a shared root cause." }, { "left": "issue:43408", "right": "issue:43742", "accept": false, - "reason": "These are unrelated model-loading failures for different model families." + "reason": "Different model-loading failures for different models, with unrelated error surfaces." }, { "left": "issue:43122", "right": "issue:46298", "accept": false, - "reason": "A tokenizer behavior change and a noisy logging message are unrelated." + "reason": "Tokenizer regression versus a noisy logging message; unrelated issues." }, { "left": "issue:42617", "right": "issue:45341", "accept": false, - "reason": "3d_parallel.py failing to run is unrelated to a small testing_utils bug." + "reason": "Training script failure and a testing_utils bug are separate concerns." }, { "left": "issue:45362", "right": "issue:46328", "accept": false, - "reason": "Qwen3.5 chat crashing and a pad conversion performance issue are separate problems." + "reason": "Chat crashes for Qwen3.5-35B and a padding performance issue are unrelated." }, { "left": "issue:42222", "right": "issue:46178", "accept": false, - "reason": "Vitpose breakage and a tensorflow_text backend mapping error are different issues." + "reason": "VitPose model breakage and tensorflow_text backend mapping error affect different subsystems." }, { "left": "issue:44484", "right": "issue:45310", "accept": false, - "reason": "save_pretrained shard-size policy is unrelated to the Qwen3.5 MoE from_pretrained error." + "reason": "A question about shard size defaults is not the same as a Qwen3.5 MoE from_pretrained error." }, { "left": "issue:42371", "right": "issue:44261", "accept": false, - "reason": "TF32 API guidance and an MLA layernorm epsilon precision bug are not the same bug." + "reason": "TF32 API guidance and MLA rms_norm_eps precision drift are unrelated." }, { "left": "issue:44877", "right": "issue:46231", "accept": false, - "reason": "Strict config loading and a docs typo about set_training parameters are unrelated." + "reason": "Strict config loading for granite_speech is unrelated to a documentation parameter-name mismatch." }, { "left": "issue:43937", "right": "issue:44496", "accept": false, - "reason": "A GenerationConfig validation failure is different from an unrecognized-model config error." + "reason": "GenerationConfig validation on GLM-5 and missing model_type in config.json are different loading problems." }, { "left": "issue:39401", "right": "issue:46209", "accept": false, - "reason": "Qwen3 tokenizer offset mapping and FP8 inference on B200 are unrelated." + "reason": "Qwen3 offset_mapping bug and FP8 inference failure on B200 are unrelated." }, { "left": "issue:44410", "right": "issue:45478", "accept": false, - "reason": "Missing Qwen3next projections and the Qwen3.5 MoE from_pretrained regression are different model/code-path failures." + "reason": "Missing qwen3next projections and Qwen3.5 MoE from_pretrained errors are separate model-specific defects." }, { "left": "issue:42371", "right": "issue:46617", "accept": false, - "reason": "TF32 behavior settings and trust_remote_code cache loading are unrelated." + "reason": "TF32 configuration advice is unrelated to a cache FileNotFoundError for trust_remote_code loading." }, { "left": "issue:44261", "right": "issue:46762", "accept": false, - "reason": "An epsilon mismatch in MLA q_a_layernorm is not the same as Minimax MSA numerical disagreement." + "reason": "Both involve numerical differences, but they are different models and different attention implementations; not the same bug." }, { - "left": "issue:43208", - "right": "issue:46894", + "left": "issue:44933", + "right": "issue:45278", "accept": false, - "reason": "xLSTM training blockers and CohereASR double-shift loss bug are different model-specific issues." + "reason": "A specific missing image_utils import is not enough to equate it with the broader post-upgrade import-error report." }, { - "left": "issue:44933", - "right": "issue:45278", + "left": "issue:43208", + "right": "issue:46894", "accept": false, - "reason": "A missing import from image_utils is too specific to be the same as a broader many-import-errors report." + "reason": "xLSTM training blockers and CohereASR double-shift loss bug are distinct training issues." }, { "left": "issue:38617", "right": "issue:44220", "accept": false, - "reason": "ImportError for layer_type_validation and a _torch_extract_fbank_features issue are different symbols/code paths." + "reason": "An import name error and an audio feature extraction bug are unrelated." }, { "left": "issue:44075", "right": "issue:45656", "accept": false, - "reason": "SGD args being ignored and deepspeed calling optimizer.step() twice are different optimizer bugs." + "reason": "SGD args being ignored and optimizer stepping twice under deepspeed are different optimizer bugs." }, { "left": "issue:43066", "right": "issue:43122", "accept": false, - "reason": "Wrong decoder type and changed tokenization output are related to tokenizers but not the same concrete defect." + "reason": "Tokenizer decoder type regression and tokenization differences across versions are related thematically but not shown to be the same underlying defect." }, { "left": "issue:44263", "right": "issue:44485", "accept": false, - "reason": "A torch.split return-value bug in GlmMoeDsaIndexer is unrelated to GLM-5 RoPE implementation." + "reason": "A torch.split return-value issue and GLM-5 RoPE implementation concern different code paths." }, { "left": "issue:43097", "right": "issue:44493", "accept": false, - "reason": "Removed tie_embeddings_and_encoder_decoder and unexpected position-id keys are different regressions." + "reason": "Removal of tie_embeddings_and_encoder_decoder and unexpected position-id keys are distinct API/serialization issues." }, { "left": "issue:43931", "right": "issue:44410", "accept": false, - "reason": "MobileLLM weight-shape mismatch and Qwen3next missing projections are unrelated model-loading failures." + "reason": "Weight-shape mismatch for Qwen3-VL and missing projections in qwen3next are unrelated model-loading problems." }, { "left": "issue:43673", "right": "issue:45341", "accept": false, - "reason": "chunked_prefill cache regression and a testing_utils bug are unrelated." + "reason": "GenerationMixin cache handling during chunked_prefill is unrelated to the testing_utils bug." }, { "left": "issue:41628", "right": "issue:45071", "accept": false, - "reason": "AutoImageProcessor import failure and PretrainedConfig type-checking breakage are separate API regressions." + "reason": "Missing AutoImageProcessor import and PretrainedConfig type-checking regression are separate API breakages." }, { "left": "issue:42222", "right": "issue:45542", "accept": false, - "reason": "Vitpose model breakage and the TensorBoard-without-TensorFlow backend error are unrelated." + "reason": "VitPose model breakage and the tensorflow/tensorboard backend issue do not share a concrete root cause." }, { "left": "issue:45071", "right": "issue:45542", "accept": false, - "reason": "PretrainedConfig type-checking and a missing TensorFlow backend definition are different issues." + "reason": "PretrainedConfig type checking and the undefined tf backend error are unrelated." + }, + { + "left": "issue:44484", + "right": "issue:45478", + "accept": false, + "reason": "max_shard_size default behavior has nothing to do with the Qwen3.5 MoE from_pretrained error." } ] }, @@ -16488,7 +16574,6 @@ "issue:45375", "issue:45440", "issue:45464", - "issue:45478", "issue:45588", "issue:45753", "issue:46050", @@ -16499,11 +16584,11 @@ "issue:46655", "issue:46664", "issue:46710", + "issue:46752", "issue:46897" ], "soft_pairs": [ "issue:44483|issue:45464", - "issue:44484|issue:45478", "issue:43874|issue:44261", "issue:44263|issue:46172", "issue:44360|issue:46172", @@ -16522,239 +16607,240 @@ "issue:43531|issue:45341", "issue:43673|issue:46123", "issue:43097|issue:45588", - "issue:44410|issue:45375", - "issue:44038|issue:44410", "issue:38617|issue:42757", + "issue:44038|issue:44410", + "issue:44410|issue:45375", "issue:43208|issue:46897", "issue:43064|issue:44315", "issue:43986|issue:45372", "issue:43097|issue:46050", "issue:43986|issue:44484", - "issue:42890|issue:46298", "issue:46655|issue:46664", "issue:46172|issue:46710", - "issue:38617|issue:44351" + "issue:38617|issue:44351", + "issue:42890|issue:46298", + "issue:46298|issue:46752" ], - "prepared_review_unit_hash": "bcce7bd4a2683ec747bfbfc8c803394297d6f675a61ba494c2ad7977ce32c975", + "prepared_review_unit_hash": "ec4a09d90f14b6e4e5f4657d2ff147216ab28d5af594a777d72a14f601585488", "component_budget": { "node_count": 502, "item_count": 502, - "soft_pair_count": 1842, - "serialized_chars": 956182, - "estimated_input_tokens": 239046, - "estimated_eval_tokens": 478348 + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 32, - "serialized_chars": 37119, - "estimated_input_tokens": 9280, - "estimated_eval_tokens": 18816 + "serialized_chars": 37139, + "estimated_input_tokens": 9285, + "estimated_eval_tokens": 18826 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "These items are largely heterogeneous and do not form a strong duplicate cluster. No soft-edge pair looks like the same concrete bug or change, so I would not merge any of them as duplicates.", - "confidence": 0.83, - "canonical_issue_reason": "issue:43097 is the broadest, most reusable regression candidate: a v5.0.0 compatibility/API-removal problem with inbound references and a clear user-facing impact. It is the best anchor only in a loose sense; the rest of the set is not a single duplicate family.", + "summary": "This set is mostly a grab-bag of unrelated issue reports across different models and subsystems. No pull requests are present. None of the soft pairs look like the same underlying defect, so all are rejected.", + "confidence": 0.66, + "canonical_issue_reason": "If one issue has to anchor the cluster, issue 43097 is the broadest and most systemic regression report here; the others are mostly model-specific or unrelated. That said, this is not a true duplicate cluster.", "canonical_pr_reason": null, - "best_issue_reason": "issue:43097 is the strongest global issue candidate because it represents a general versioning regression rather than a narrow model-specific symptom.", + "best_issue_reason": "Issue 43097 is the strongest single representative because it describes a clear, reusable v5 regression with wider impact than the other entries, and is more central than the highly model-specific reports.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:44483", "right": "issue:45464", "accept": false, - "reason": "Both mention chat/completions, but one is a generic v5.3 request rejection while the other is a model-specific streaming failure; different bugs." - }, - { - "left": "issue:44484", - "right": "issue:45478", - "accept": false, - "reason": "Max shard size behavior and a Qwen3.5-Moe from_pretrained error are unrelated code paths." + "reason": "Same API area, but different failures: a generic /v1/chat/completions rejection vs a Qwen3.5 streaming inference problem." }, { "left": "issue:43874", "right": "issue:44261", "accept": false, - "reason": "Image-patch counting in GLM46V differs from an MLA RMS norm epsilon precision issue; not the same defect." + "reason": "Different models and different bugs: missing image-patch method vs missing rms_norm_eps config." }, { "left": "issue:44263", "right": "issue:46172", "accept": false, - "reason": "DSA torch.split return handling and a DeepSeek-V4 transpose/cast issue are different model internals." + "reason": "Both are DeepSeek/GLM-adjacent math bugs, but torch.split return values and comb.to(dtype).transpose(-1, -2) are different code paths." }, { "left": "issue:44360", "right": "issue:46172", "accept": false, - "reason": "A missing ReLU in the DSA indexer is not the same as the DeepSeek-V4 dtype/transpose bug." + "reason": "DSA indexer lacking ReLU is unrelated to the DeepSeek-v4 dtype/transpose issue." }, { "left": "issue:27957", "right": "issue:43352", "accept": false, - "reason": "Both are Flash Attention 2 topics, but they concern different models and different support/failure modes." + "reason": "Both mention Flash Attention 2 support, but they are separate model-specific compatibility gaps, not one shared bug." }, { "left": "issue:43931", "right": "issue:44038", "accept": false, - "reason": "Both are Qwen3-VL family loading issues, but one is a shape mismatch and the other is a v5/Qwen3-VL-Moe failure; too different to be duplicates." + "reason": "Both are Qwen3-VL loading failures, but shape mismatch and VL-Moe/version behavior are distinct defects." }, { "left": "issue:43012", "right": "issue:43577", "accept": false, - "reason": "A PyTorch precision warning on bfloat16 compilation is not the same as Blip2 dtype not propagating." + "reason": "One is a compile-time warning about bf16 precision; the other is a dtype propagation/loading bug in BLIP2 submodules." }, { "left": "issue:43441", "right": "issue:44829", "accept": false, - "reason": "Both involve Flash Attention, but one is a model load/support failure and the other is degenerate training with flash_attention_3; different concrete bugs." + "reason": "Both involve FlashAttention, but one is a Ministral-3 compatibility crash and the other is a FlashAttention-3 training regression." }, { "left": "issue:43097", "right": "issue:43452", "accept": false, - "reason": "A removed config flag and a gguf_file tokenizer/model loading break are unrelated." + "reason": "Removed config behavior and gguf_file loading breakage are different issues and would not be fixed by one PR." }, { "left": "issue:45003", "right": "issue:45341", "accept": false, - "reason": "Unsafe sys.modules access in modeling_utils is unrelated to a testing_utils bug." + "reason": "Unrelated modules and bug classes: unsafe sys.modules access vs a testing_utils bug." }, { "left": "issue:43986", "right": "issue:45341", "accept": false, - "reason": "Missing torchvision handling for video models is unrelated to a testing_utils issue." + "reason": "Video AutoProcessor crash without torchvision is unrelated to the testing_utils issue." }, { "left": "issue:44261", "right": "issue:44485", "accept": false, - "reason": "RMS norm epsilon precision mismatch and GLM-5 RoPE implementation are different model/math issues." + "reason": "Both are GLM-family reports, but missing rms_norm_eps and RoPE implementation are separate defects." }, { "left": "issue:45753", "right": "issue:46369", "accept": false, - "reason": "An infinite loop on a specific Qwen3_5 image is not the same as SAM3-litetext memory overflow." + "reason": "Different multimodal models and failure modes: one infinite loop, the other memory overflow." }, { "left": "issue:43874", "right": "issue:44263", "accept": false, - "reason": "Missing get_number_of_image_patches and torch.split return-value handling are different bugs in different GLM components." + "reason": "Both touch GLM internals, but the missing image-patch method and torch.split contract issue are not the same bug." }, { "left": "issue:44117", "right": "issue:44987", "accept": false, - "reason": "A tokenizer mapping None assumption and a specific physical-intelligence/fast load failure are not clearly the same root cause." + "reason": "Generic tokenizer mapping None handling is not the same as a specific physical-intelligence/fast load failure." }, { "left": "issue:45440", "right": "issue:46172", "accept": false, - "reason": "DeepSeekV3 divergence from remote implementation and a DeepSeek-V4 transpose/cast bug are different generations and symptoms." + "reason": "Different DeepSeek generations and different tensor logic; not the same underlying bug." }, { "left": "issue:43531", "right": "issue:45341", "accept": false, - "reason": "Qwen3-MoE sliding_window behavior is unrelated to a testing_utils bug." + "reason": "Qwen3-MoE sliding_window behavior is unrelated to a testing utility bug." }, { "left": "issue:43673", "right": "issue:46123", "accept": false, - "reason": "Chunked-prefill cache handling and MaskGenerationPipeline final-batch handling are different generation/pipeline bugs." + "reason": "Chunked-prefill cache handling and MaskGenerationPipeline partial-batch logic are different code paths." }, { "left": "issue:43097", "right": "issue:45588", "accept": false, - "reason": "Config removal in v5.0.0 and an AttributeError in flash_attention.py are unrelated." + "reason": "Tie-embeddings removal and the s_aux=None flash_attention crash are unrelated regressions." }, { - "left": "issue:44410", - "right": "issue:45375", + "left": "issue:38617", + "right": "issue:42757", "accept": false, - "reason": "Missing attn projections in qwen3next and a silently dropped config field in Qwen3_5MoeVisionConfig are different issues." + "reason": "Both are import errors, but they involve different symbols, modules, and likely different dependency breakages." }, { "left": "issue:44038", "right": "issue:44410", "accept": false, - "reason": "A Qwen3-VL-Moe load regression and missing qkv/gate projections in qwen3next are not the same bug." + "reason": "Different model families and different missing pieces: Qwen3-VL-Moe vs qwen3next layer projections." }, { - "left": "issue:38617", - "right": "issue:42757", + "left": "issue:44410", + "right": "issue:45375", "accept": false, - "reason": "Both are import errors, but for different symbols and different modules; no shared root cause is evident." + "reason": "Both are Qwen3-family issues, but missing projections and a silently dropped config field are not the same defect." }, { "left": "issue:43208", "right": "issue:46897", "accept": false, - "reason": "xLSTM training blockers and Florence2 double-shift loss bug are distinct model/training defects." + "reason": "xLSTM training blockers are unrelated to Florence2's training-loss double-shift bug." }, { "left": "issue:43064", "right": "issue:44315", "accept": false, - "reason": "Wrong optimizer state placement under FSDP2/PEFT is unrelated to Liger Kernel not being applied with model_init." + "reason": "FSDP2/PEFT optimizer-state corruption is not the same as Liger Kernel not being applied under model_init." }, { "left": "issue:43986", "right": "issue:45372", "accept": false, - "reason": "Missing torchvision for video processors and a mistral_common import error for Gemma 4 loading are different failure causes." + "reason": "Both are processor-loading ImportErrors, but they stem from different missing optional symbols/dependencies." }, { "left": "issue:43097", "right": "issue:46050", "accept": false, - "reason": "A removed embedding-tie config and a broken sinq quantization save/load path are not the same defect." + "reason": "Quantized save/load breakage is unrelated to the tie_embeddings_and_encoder_decoder removal." }, { "left": "issue:43986", "right": "issue:44484", "accept": false, - "reason": "Video-model torchvision dependency handling is unrelated to save_pretrained shard sizing." - }, - { - "left": "issue:42890", - "right": "issue:46298", - "accept": false, - "reason": "Flaky SamHQ integration tests due to missing seeds and noisy logging are unrelated." + "reason": "A torchvision-dependent video-model crash is not the same as a max_shard_size policy question." }, { "left": "issue:46655", "right": "issue:46664", "accept": false, - "reason": "Both are SmolVLM video-processor issues, but one is padding computation and the other is num_frames semantics; different fixes." + "reason": "Same component area, but padding calculation and num_frames semantics are separate video-processor bugs." }, { "left": "issue:46172", "right": "issue:46710", "accept": false, - "reason": "DeepSeek-V4 tensor handling and a broad accuracy regression after upgrade are not the same concrete bug." + "reason": "The DeepSeek-v4 tensor bug is not the same as a broader accuracy regression after upgrading Transformers." }, { "left": "issue:38617", "right": "issue:44351", "accept": false, - "reason": "Different missing imports from transformers; same symptom class, but not the same underlying issue." + "reason": "Both are import-name failures, but they are for different symbols in different modules." + }, + { + "left": "issue:42890", + "right": "issue:46298", + "accept": false, + "reason": "A missing set_seed in tests is unrelated to a noisy logging message." + }, + { + "left": "issue:46298", + "right": "issue:46752", + "accept": false, + "reason": "Noisy logging and empty-list chat template IndexError are unrelated bugs." } ] }, @@ -16790,6 +16876,7 @@ "issue:43208", "issue:43352", "issue:43502", + "issue:43526", "issue:43550", "issue:43582", "issue:43824", @@ -16824,11 +16911,9 @@ "issue:46365", "issue:46421", "issue:46701", - "issue:46752", "issue:46821" ], "soft_pairs": [ - "issue:46298|issue:46752", "issue:44062|issue:44117", "issue:43097|issue:43873", "issue:43012|issue:46821", @@ -16861,240 +16946,241 @@ "issue:42175|issue:46365", "issue:42757|issue:43824", "issue:43064|issue:44380", - "issue:42490|issue:43012" + "issue:42490|issue:43012", + "issue:43208|issue:43526" ], - "prepared_review_unit_hash": "482a2204af70f4e5f74e1bb1e3fb0f07a888c29ad06c36f3331fdd44cd2d2fce", + "prepared_review_unit_hash": "3d929e79e4817e941f47c463a9d983ad819593ed05ef418b5cf163a1ec5a1b54", "component_budget": { "node_count": 502, "item_count": 502, - "soft_pair_count": 1842, - "serialized_chars": 956182, - "estimated_input_tokens": 239046, - "estimated_eval_tokens": 478348 + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 34, - "serialized_chars": 37698, - "estimated_input_tokens": 9425, - "estimated_eval_tokens": 19106 + "serialized_chars": 37723, + "estimated_input_tokens": 9431, + "estimated_eval_tokens": 19118 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The items are a grab-bag of unrelated bugs and regressions; none of the soft pairs look like true duplicates, and there are no PRs in the set.", - "confidence": 0.82, - "canonical_issue_reason": "issue:43824 is a concrete, well-scoped import failure with clear impact and inbound references, so it is the best anchor if one issue must represent the set.", - "canonical_pr_reason": null, - "best_issue_reason": "issue:43873 has the strongest engagement and broad user impact, making it the best standalone issue even though the cluster is heterogeneous.", + "summary": "These items are largely unrelated transformer bugs and regressions; none of the soft pairs looks like the same underlying defect, so I would not merge them into duplicate clusters.", + "confidence": 0.9, + "canonical_issue_reason": "Issue 43824 is the best single representative only because it is a concrete, well-documented runtime import failure with strong discussion and inbound references; the overall set is otherwise heterogeneous.", + "canonical_pr_reason": "No PRs are present in this cluster.", + "best_issue_reason": "Issue 43824 is the strongest anchor if one issue must represent the set: clear repro, specific failure mode, and high signal compared with the rest.", "best_pr_reason": null, "soft_edge_verdicts": [ - { - "left": "issue:46298", - "right": "issue:46752", - "accept": false, - "reason": "Noisy logging vs empty-list IndexError; different symptoms and code paths." - }, { "left": "issue:44062", "right": "issue:44117", "accept": false, - "reason": "AddedToken special-arg collision vs tokenizer-mapping None; unrelated tokenizer bugs." + "reason": "Both involve tokenizers, but one is an AddedToken constructor bug and the other is a mapping/lookup assumption bug." }, { "left": "issue:43097", "right": "issue:43873", "accept": false, - "reason": "Removed-API warning vs quantization offloading behavior; not the same failure mode." + "reason": "Unrelated areas: v5 API removal vs quantization/offloading behavior." }, { "left": "issue:43012", "right": "issue:46821", "accept": false, - "reason": "bfloat16 compile warning vs TimesFM window_size AttributeError; different models and issues." + "reason": "Different subsystems and symptoms: PyTorch precision warning vs TimesFM config AttributeError." }, { "left": "issue:44117", "right": "issue:46701", "accept": false, - "reason": "Tokenizer mapping None vs pipeline failure on image-text generation; unrelated loader paths." + "reason": "Tokenizer mapping failure and diffusion pipeline failure are separate code paths." }, { "left": "issue:43502", "right": "issue:43992", "accept": false, - "reason": "local_files_only network leak vs missing embed_tokens weight; different pretraining/loading bugs." + "reason": "Local-files-only network leak vs missing checkpoint weight are distinct bugs." }, { "left": "issue:44936", "right": "issue:46365", "accept": false, - "reason": "trainer.evaluate() failure vs Trainer import blocked by datasets; separate trainer problems." + "reason": "Training/evaluation state issue vs import-time dependency conflict." }, { "left": "issue:44704", "right": "issue:45103", "accept": false, - "reason": "cached_file kwargs passthrough vs auto_docstring annotations crash; different subsystems." + "reason": "Pretrained kwargs forwarding and docstring annotation parsing are unrelated." }, { "left": "issue:44380", "right": "issue:46421", "accept": false, - "reason": "GPT2 attention scaling ignored vs cache lazy initialization recompilation; unrelated backend behavior." + "reason": "Attention scaling/backend dispatch vs chunked-prefill cache initialization." }, { "left": "issue:44938", "right": "issue:46178", "accept": false, - "reason": "Python 3.14 load failure vs tensorflow_text backend mapping error; different import stacks." + "reason": "Python version load failure is unrelated to a tensorflow_text backend mapping error." }, { "left": "issue:43986", "right": "issue:44117", "accept": false, - "reason": "AutoProcessor torchvision crash vs tokenizer mapping None; unrelated failure points." + "reason": "Video-model torchvision fallback issue vs tokenizer mapping None issue." }, { "left": "issue:42907", "right": "issue:44655", "accept": false, - "reason": "Dequantized-save failure vs pipeline save_pretrained support; different save paths." + "reason": "Different save_pretrained paths: dequantized model save vs pipeline object serialization." }, { "left": "issue:42907", "right": "issue:45753", "accept": false, - "reason": "Dequantized-save failure vs Qwen3_5 image infinite loop; unrelated model behaviors." + "reason": "One is a save failure; the other is an inference infinite loop on a specific image." }, { "left": "issue:43825", "right": "issue:44655", "accept": false, - "reason": "pipeline translation messaging vs pipeline save_pretrained; distinct user-facing issues." + "reason": "Pipeline task messaging bug is not the same as pipeline save_pretrained failure." }, { "left": "issue:43012", "right": "issue:43550", "accept": false, - "reason": "bfloat16 warning vs Bamba SDPA compile failure; different model/backend defects." + "reason": "Precision warning during compilation is not the same as Bamba SDPA compile failure." }, { "left": "issue:45712", "right": "issue:46298", "accept": false, - "reason": "dummy-class cleanup vs noisy logging; repo hygiene vs logging." + "reason": "Repository cleanup of dummy classes vs a noisy log message." }, { "left": "issue:43208", "right": "issue:43550", "accept": false, - "reason": "xLSTM training bugs vs Bamba compile SDPA failure; different models and root causes." + "reason": "xLSTM training blockers are unrelated to Bamba torch.compile/SDPA behavior." }, { "left": "issue:43012", "right": "issue:44368", "accept": false, - "reason": "bfloat16 precision warning vs tie_word_embeddings warning; separate warning conditions." + "reason": "General precision warning vs tied-embedding config warning; different triggers and fixes." }, { "left": "issue:43992", "right": "issue:44484", "accept": false, - "reason": "Missing embed_tokens weight vs max_shard_size question; unrelated save/load concerns." + "reason": "Missing encoder weights is unrelated to shard-size defaults." }, { "left": "issue:43352", "right": "issue:43976", "accept": false, - "reason": "FlashAttention support gap vs Python version compatibility; unrelated issues." + "reason": "Model-specific FlashAttention support issue vs Python compatibility issue." }, { "left": "issue:43352", "right": "issue:44380", "accept": false, - "reason": "FlashAttention support gap vs GPT2 attention scaling bug; different models/backends." + "reason": "Nemotron FlashAttention support and GPT2 attention scaling are different bugs." }, { "left": "issue:43066", "right": "issue:43906", "accept": false, - "reason": "Wrong decoder type vs isolated reproduction of another issue; not the same bug." + "reason": "Tokenizer decoder typing regression is not the same as an isolated reproduction issue." }, { "left": "issue:44162", "right": "issue:46172", "accept": false, - "reason": "ESM2 broken vs deepseek matmul/transpose bug; different architectures and failures." + "reason": "ESM2 model breakage and a deepseek tensor operation error are unrelated." }, { "left": "issue:43097", "right": "issue:46298", "accept": false, - "reason": "API removal warning vs noisy logging; unrelated compatibility and logging issues." + "reason": "API removal and a noisy logging message do not share a code path." }, { "left": "issue:44380", "right": "issue:46132", "accept": false, - "reason": "GPT2 attention scaling ignored vs AttentionInterface.register behavior change; different APIs." + "reason": "Both are attention-related, but one is backend-specific scaling and the other is register-side behavior change." }, { "left": "issue:45584", "right": "issue:46365", "accept": false, - "reason": "Whisper empty transcription crash vs Trainer import conflict; unrelated code paths." + "reason": "Whisper generation edge case and Trainer import conflict are separate problems." }, { "left": "issue:44077", "right": "issue:46365", "accept": false, - "reason": "patchtsmixer post_init issue vs Trainer import conflict; unrelated failures." + "reason": "Model initialization policy issue vs import dependency conflict." }, { "left": "issue:44222", "right": "issue:45478", "accept": false, - "reason": "Same model family, but save_pretrained FP8 issue vs from_pretrained error are different pathways and not clearly one root cause." + "reason": "FP8 save_pretrained failure and Qwen3.5 MoE from_pretrained error are different directions of failure." }, { "left": "issue:36683", "right": "issue:43582", "accept": false, - "reason": "Gemma3Config vocab_size AttributeError vs Apple Silicon warmup TypeError; unrelated." + "reason": "Gemma config attribute error and Apple Silicon allocator TypeError are unrelated." }, { "left": "issue:44380", "right": "issue:45479", "accept": false, - "reason": "GPT2 attention scaling bug vs zero-loss classification bug; unrelated model logic." + "reason": "Attention backend config bug vs classification loss degeneracy." }, { "left": "issue:42175", "right": "issue:46365", "accept": false, - "reason": "TensorFlow extra-install packaging issue vs Trainer import blocked by datasets; unrelated import/packaging problems." + "reason": "Backend packaging issue is not the same as Trainer import failure." }, { "left": "issue:42757", "right": "issue:43824", "accept": false, - "reason": "Missing is_offline_mode import vs Qwen2.5 VL symbol import; different missing symbols and packages." + "reason": "Both are import errors, but for different symbols and different packages." }, { "left": "issue:43064", "right": "issue:44380", "accept": false, - "reason": "FSDP2 optimizer-state bug vs GPT2 attention scaling bug; different training vs attention problems." + "reason": "Distributed optimizer-state corruption and GPT2 attention scaling are unrelated." }, { "left": "issue:42490", "right": "issue:43012", "accept": false, - "reason": "Trailing-slash local model warning vs bfloat16 compile warning; different import and compile behaviors." + "reason": "Trailing-slash local model warning and bf16 compile warning are separate behaviors." + }, + { + "left": "issue:43208", + "right": "issue:43526", + "accept": false, + "reason": "xLSTM training blockers and BeitImageProcessorFast label reduction bug do not match." } ] }, @@ -17133,7 +17219,6 @@ "issue:43475", "issue:43502", "issue:43525", - "issue:43526", "issue:43611", "issue:43618", "issue:43756", @@ -17159,22 +17244,21 @@ "issue:45216", "issue:45276", "issue:45278", + "issue:45290", "issue:45310", "issue:46077", "issue:46082", "issue:46365", - "issue:46369", "issue:46829", "issue:46877" ], "soft_pairs": [ - "issue:43208|issue:43526", "issue:44483|issue:44514", "issue:43618|issue:44380", "issue:38617|issue:43502", "issue:44038|issue:44230", - "issue:42994|issue:44655", "issue:43756|issue:44485", + "issue:42994|issue:44655", "issue:43618|issue:43761", "issue:43299|issue:43867", "issue:36010|issue:42757", @@ -17182,12 +17266,12 @@ "issue:43352|issue:43937", "issue:44355|issue:46877", "issue:45276|issue:46077", - "issue:44483|issue:46369", + "issue:44483|issue:45290", "issue:36683|issue:43475", "issue:43611|issue:44655", "issue:43901|issue:44655", - "issue:43976|issue:44938", "issue:43867|issue:45216", + "issue:43976|issue:44938", "issue:36683|issue:44077", "issue:44625|issue:46077", "issue:46077|issue:46082", @@ -17201,226 +17285,220 @@ "issue:43010|issue:45278", "issue:44075|issue:46365" ], - "prepared_review_unit_hash": "9fed6bb94a2243749b80fac1e10e9de55c3ca22031a70e2c90ae9a97b34426a0", + "prepared_review_unit_hash": "3f1b467e74265e0a250735c115ff120b3113afb80ff2bf35763bd696bfe7bd42", "component_budget": { "node_count": 502, "item_count": 502, - "soft_pair_count": 1842, - "serialized_chars": 956182, - "estimated_input_tokens": 239046, - "estimated_eval_tokens": 478348 + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 }, "budget": { - "node_count": 48, - "item_count": 48, - "soft_pair_count": 32, - "serialized_chars": 36923, - "estimated_input_tokens": 9231, - "estimated_eval_tokens": 18718 + "node_count": 47, + "item_count": 47, + "soft_pair_count": 31, + "serialized_chars": 36073, + "estimated_input_tokens": 9019, + "estimated_eval_tokens": 18294 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster is a grab bag of unrelated Transformers issues across models, configs, saving/loading, and import regressions. None of the soft-edge pairs look like the same underlying bug/change, so I would not merge any of them.", - "confidence": 0.91, - "canonical_issue_reason": "No single canonical issue really subsumes this set because the items are heterogeneous. If an anchor is needed, issue:36683 is the most active open, concrete regression report, but it does not represent the rest of the cluster.", + "summary": "The cluster is highly heterogeneous: most pairs are only loosely related by subsystem keywords (imports, loading, saving, attention, chat templates) but describe different concrete bugs. I would not merge any of the soft edges as duplicates.", + "confidence": 0.21, + "canonical_issue_reason": "issue:45278 is the broadest umbrella-style report in the set ('many import errors after update'), so it is the closest thing to a cluster anchor, though it is still not a true duplicate target for most others.", "canonical_pr_reason": null, - "best_issue_reason": "Issue:36683 is the most actionable single issue in the set: open, heavily discussed, and framed as a concrete reproducible bug. It is only a weak representative because the cluster is not a true duplicate group.", + "best_issue_reason": "issue:45278 is the best global issue candidate only because it is the most general regression report; however, the set is too mixed for a strong canonical duplicate choice.", "best_pr_reason": null, "soft_edge_verdicts": [ - { - "left": "issue:43208", - "right": "issue:43526", - "accept": false, - "reason": "Different models and failures: xLSTM training bugs vs BeitImageProcessorFast label reduction." - }, { "left": "issue:44483", "right": "issue:44514", "accept": false, - "reason": "Both mention request handling, but one is the chat-completions API and the other is Qwen2_5_VLProcessor batched chat templating; different code paths." + "reason": "Both involve chat/completions behavior, but one is the v5.3 /v1/chat/completions request acceptance path and the other is apply_chat_template batching with padding=False; different code paths and failure modes." }, { "left": "issue:43618", "right": "issue:44380", "accept": false, - "reason": "Both are transformer regressions, but CLIP attentions missing and GPT-2 attention scaling ignored are separate bugs." + "reason": "Both mention attention behavior, but one is CLIPOutput losing attentions and the other is GPT-2 attention scaling being ignored under SDPA/FlashAttention; not the same bug." }, { "left": "issue:38617", "right": "issue:43502", "accept": false, - "reason": "Import error for a missing symbol vs local_files_only still triggering API calls; unrelated issues." + "reason": "One is an import symbol breakage from transformers.configuration_utils; the other is offline/local_files_only still triggering API calls. Different root causes." }, { "left": "issue:44038", "right": "issue:44230", "accept": false, - "reason": "Same model family, but one is a loading regression and the other is fp8 inference support; not the same fix." + "reason": "Both reference Qwen3-VL-Moe, but one is a loading regression in transformers 5.0 and the other is fp8 inference support; not mergeable as one fix." }, { - "left": "issue:42994", - "right": "issue:44655", + "left": "issue:43756", + "right": "issue:44485", "accept": false, - "reason": "Quantized model save failure and pipeline save_pretrained failure are different serialization problems." + "reason": "Both mention RoPE, but one is a SmolLM3 layer-dropping bug and the other is a GLM-5 RoPE implementation discussion; different model-specific issues." }, { - "left": "issue:43756", - "right": "issue:44485", + "left": "issue:42994", + "right": "issue:44655", "accept": false, - "reason": "Both involve RoPE, but they target different models and different implementation bugs." + "reason": "Saving is the only overlap: quantized model serialization failure vs Pipeline.save_pretrained support. Different objects and failure modes." }, { "left": "issue:43618", "right": "issue:43761", "accept": false, - "reason": "Both are CLIP output regressions, but one is missing attentions and the other is hidden_states not returned; too specific to assume one bug." + "reason": "Both are CLIP v5 regressions, but one is attentions missing from output objects and the other is hidden_states not returned; separate return-field bugs." }, { "left": "issue:43299", "right": "issue:43867", "accept": false, - "reason": "Both are loading errors, but one is Qwen3VL-Moe regression and the other is a generic state_dict sort problem." + "reason": "Both are loading problems, but one is Qwen3-VL-Moe model loading and the other is a sorted state_dict load error; too different to treat as the same bug." }, { "left": "issue:36010", "right": "issue:42757", "accept": false, - "reason": "Different import failures from different packages and symbols; no shared underlying bug." + "reason": "Both are ImportErrors, but for different symbols and different packages (transformers.generation vs huggingface_hub)." }, { "left": "issue:44625", "right": "issue:46365", "accept": false, - "reason": "Qwen3.5 num_labels propagation vs inability to import Trainer due to datasets; unrelated." + "reason": "One is Qwen3.5 config propagation, the other is an import-time Trainer problem tied to datasets; unrelated." }, { "left": "issue:43352", "right": "issue:43937", "accept": false, - "reason": "Flash Attention support gating for Nemotron vs invalid GenerationConfig for GLM-5; unrelated." + "reason": "Different models and different failures: Flash Attention 2 unsupported vs invalid GenerationConfig." }, { "left": "issue:44355", "right": "issue:46877", "accept": false, - "reason": "Compiled Python file errors and import-time docstring lint output are different problems." + "reason": "Running compiled Python files vs auto_docstring stdout noise at import time are unrelated issues." }, { "left": "issue:45276", "right": "issue:46077", "accept": false, - "reason": "Both are Gemma4 bugs, but one is embedding resize propagation and the other is attention mask construction." + "reason": "Both are Gemma4-related, but one is token embedding/output resizing and the other is bidirectional attention mask construction; different code paths." }, { "left": "issue:44483", - "right": "issue:46369", + "right": "issue:45290", "accept": false, - "reason": "API request acceptance and SAM3 litetext memory overflow are unrelated." + "reason": "Both touch chat template/completions, but one is API request acceptance and the other is assistant tool-call templating with empty content; not the same concrete bug." }, { "left": "issue:36683", "right": "issue:43475", "accept": false, - "reason": "Different missing-attribute regressions in different model families: Gemma3 config vs SAM3 vision output." + "reason": "Both are missing-attribute errors, but in different model families and unrelated fields (vocab_size vs fpn_position_embeddings)." }, { "left": "issue:43611", "right": "issue:44655", "accept": false, - "reason": "Model loading via base_model_prefix vs saving Pipeline objects; different subsystems." + "reason": "One is loading models with base_model_prefix under v5; the other is saving Pipeline objects. Different operations and fixes." }, { "left": "issue:43901", "right": "issue:44655", "accept": false, - "reason": "Pipeline docs/behavior mismatch vs pipeline serialization; not the same bug." + "reason": "Docs mismatch for TextClassificationPipeline vs inability to save Pipeline objects; unrelated." }, { - "left": "issue:43976", - "right": "issue:44938", + "left": "issue:43867", + "right": "issue:45216", "accept": false, - "reason": "Platform/Python compatibility issues, but they concern different Python versions and failure modes." + "reason": "Both involve model serialization/loading, but one is a sorted state_dict load failure and the other is a bad saved checkpoint for Qwen3.5; not clearly the same root cause." }, { - "left": "issue:43867", - "right": "issue:45216", + "left": "issue:43976", + "right": "issue:44938", "accept": false, - "reason": "Both involve loading/saving state, but the titles point to different regressions and model-specific checkpoint issues." + "reason": "Both are Python-version compatibility reports, but for different versions and likely different breakpoints; not the same bug." }, { "left": "issue:36683", "right": "issue:44077", "accept": false, - "reason": "Different config bugs in different models: missing vocab_size vs an optional post_init allowance change." + "reason": "Gemma3 config attribute missing vs patchtsmixer post_init policy; no substantive overlap." }, { "left": "issue:44625", "right": "issue:46077", "accept": false, - "reason": "Different Gemma4/Qwen3.5 config issues with no shared code-path indicated." + "reason": "Qwen3.5 num_labels propagation vs Gemma4 bidirectional attention masking are unrelated model-specific bugs." }, { "left": "issue:46077", "right": "issue:46082", "accept": false, - "reason": "Different model configs and validation problems; same broad area only." + "reason": "Both are config/behavior issues, but for different models and different constraints (Gemma4 attention masking vs LlamaConfig head_dim validation)." }, { "left": "issue:43618", "right": "issue:45216", "accept": false, - "reason": "CLIP output regression vs Qwen3.5 saved checkpoint regression; unrelated." + "reason": "Attention-output regression in CLIP vs corrupted Qwen3.5 save_pretrained checkpoint; different components." }, { "left": "issue:43208", "right": "issue:46829", "accept": false, - "reason": "xLSTM training bugs and VideoPrism output misassignment are unrelated." + "reason": "xLSTM training bugs and VideoPrism output-field confusion are unrelated." }, { "left": "issue:36683", "right": "issue:43525", "accept": false, - "reason": "Missing Gemma3 vocab_size and LlamaConfig head_dim validation are separate configuration issues." + "reason": "Both are config attribute errors, but for different attributes and different models; no shared code-path bug is evident." }, { "left": "issue:36683", "right": "issue:44625", "accept": false, - "reason": "Different model families and different config propagation bugs." + "reason": "Gemma3 missing vocab_size and Qwen3.5 num_labels propagation are different configuration issues." }, { "left": "issue:29870", "right": "issue:36296", "accept": false, - "reason": "Bart generate UnboundLocalError and a tensor-parallel training bug are unrelated." + "reason": "BART generate() decoder_attention_mask initialization bug vs tensor parallel training bug; unrelated." }, { "left": "issue:43992", "right": "issue:45310", "accept": false, - "reason": "Both are from_pretrained/load failures, but on different models and with different failure mechanisms." + "reason": "Both are from_pretrained/loading regressions, but they affect different models and appear to stem from different missing keys or version issues." }, { "left": "issue:43064", "right": "issue:44492", "accept": false, - "reason": "Distributed optimizer-state bug vs a cache-strategy typo; no overlap." + "reason": "FSDP2/PEFT optimizer-state corruption vs a cache-strategy typo; no overlap." }, { "left": "issue:43010", "right": "issue:45278", "accept": false, - "reason": "Cache update no_grad bug and broad import errors after upgrade are unrelated." + "reason": "Cache/update no_grad behavior vs broad import errors after upgrade; entirely different bugs." }, { "left": "issue:44075", "right": "issue:46365", "accept": false, - "reason": "Unused SGD args and Trainer import failure due to datasets are different issues." + "reason": "SGD optimizer args ignored vs Trainer import failure tied to datasets installation; unrelated." } ] }, @@ -17525,14 +17603,14 @@ "issue:43502|issue:44485", "issue:43874|issue:46617" ], - "prepared_review_unit_hash": "aafc5c9a1191b1d4d95aa6518bcffbe210cefc57f714b98c3b82553fe59fdec4", + "prepared_review_unit_hash": "798ae6b9e0c779257a5d0d0c57f9e46754f90f74cca85060bcbbdf7c055dd67f", "component_budget": { "node_count": 502, "item_count": 502, - "soft_pair_count": 1842, - "serialized_chars": 956182, - "estimated_input_tokens": 239046, - "estimated_eval_tokens": 478348 + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 }, "budget": { "node_count": 48, @@ -17547,192 +17625,198 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Mostly unrelated closed issues spanning import performance, model/config regressions, tensor-parallelism, tokenizer loading, and test flakiness. Only the import-slowdown pair looks like the same underlying bug; the rest are too far apart to merge as duplicates.", - "confidence": 0.68, - "canonical_issue_reason": "issue:41669 is the strongest canonical issue in this set because it is the most concrete root-cause report: it identifies a specific import-time bottleneck (`import *` usage in models) rather than just a vague symptom.", + "summary": "Mostly unrelated issues grouped by superficial similarity (v5 regressions, loading problems, test flakiness). Only one pair looks like a likely true duplicate around slow `transformers` import time.", + "confidence": 0.78, + "canonical_issue_reason": "issue:41669 is the clearest, most specific root-cause report in the set: it identifies the import-time slowdown and points to the `import *` usage in models as the likely cause.", "canonical_pr_reason": null, - "best_issue_reason": "issue:41669 is the best issue candidate for dedupe handling: it is specific, actionable, and can subsume the generic import-performance complaint in issue:44246.", + "best_issue_reason": "issue:41669 is the best representative issue because it is actionable, has a concrete symptom/cause pairing, and appears to subsume the broader import-slowdown report.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:44908", "right": "issue:46851", "accept": false, - "reason": "Different bugs: inverse-sqrt scheduler kwargs handling vs Qwen3.5 variable-length `position_ids`." + "reason": "Different bugs: scheduler kwargs propagation vs Qwen3.5 variable-length position_ids handling." }, { "left": "issue:29870", "right": "issue:33453", "accept": false, - "reason": "Different surfaces: BART `generate()` decoder mask bug vs tokenizer loading regression." + "reason": "Bart generate-time decoder attention mask error and tokenizer-loading regression are unrelated symptoms/code paths." }, { "left": "issue:44938", "right": "issue:45071", "accept": false, - "reason": "One is a Python 3.14 import failure; the other is a `PretrainedConfig` type-checking regression." + "reason": "Both mention version compatibility, but one is Python 3.14 import failure and the other is `PretrainedConfig` type checking; not the same bug." }, { "left": "issue:44355", "right": "issue:44855", "accept": false, - "reason": "Compiled-file import errors and a Python 3.13 `@torch.jit.script` parsing issue are distinct code paths." + "reason": "One is a generic compiled-file import error; the other is a specific Python 3.13 `torch.jit.script` parsing issue in DebertaV2Model." }, { "left": "issue:43867", "right": "issue:46202", "accept": false, - "reason": "State-dict loading order issue vs Gemma pipeline roundtrip tokenization issue; not the same bug." + "reason": "A sorted-state_dict loading error is not the same as Gemma pipeline roundtrip failure on the underline token." }, { "left": "issue:36683", "right": "issue:43064", "accept": false, - "reason": "Gemma3 config missing `vocab_size` is unrelated to the FSDP2/PEFT optimizer-state problem." + "reason": "Gemma3Config missing `vocab_size` and FSDP2/PEFT optimizer state corruption are different failures." }, { "left": "issue:43867", "right": "issue:45375", "accept": false, - "reason": "Model load failure from sorted state_dicts is unrelated to the Qwen3_5MoeVision config field omission." + "reason": "Different models and different failure modes: state_dict loading vs missing vision config field." }, { "left": "issue:42890", "right": "issue:45561", "accept": false, - "reason": "Test flakiness from missing `set_seed()` is different from the xdist file-race in patched testing utils." + "reason": "Both are flaky-test related, but one is missing `set_seed()` in SAM HQ tests and the other is an xdist file race." }, { "left": "issue:43010", "right": "issue:45850", "accept": false, - "reason": "A cache/layer `@torch.no_grad` fix is not the same as a broad report that v5 breaks many models." + "reason": "Decoratoring cache `update()` with `torch.no_grad` is a specific fix; \u201cbreaks too many models\u201d is a broad complaint, not the same issue." }, { "left": "issue:41669", "right": "issue:44246", "accept": true, - "reason": "Both point to the same import-time performance regression in `transformers`; 41669 gives a likely root cause (`import *` in models) and 44246 reports the user-facing slowdown." + "reason": "Both point to the same underlying problem: `transformers` import is unexpectedly slow, and 41669 gives the likely root cause in model-side `import *` usage." }, { "left": "issue:36296", "right": "issue:43575", "accept": false, - "reason": "Tensor-parallel training bug and Qwen2-57B load OOM are different failures with different triggers." + "reason": "Tensor-parallel training bug and Qwen2-57B-A14B-Instruct OOM under tp are related only at a broad subsystem level, not the same concrete bug." }, { "left": "issue:43010", "right": "issue:43976", "accept": false, - "reason": "`update()` missing `no_grad` is unrelated to Python 3.9/3.10 compatibility." + "reason": "A no-grad cache update change and a Python version support issue are unrelated." }, { "left": "issue:43618", "right": "issue:44077", "accept": false, - "reason": "CLIP attentions regression and `patchtsmixer` post-init policy are unrelated." + "reason": "CLIP attentions assignment and patchtsmixer `post_init` policy are different model-specific issues." }, { "left": "issue:43992", "right": "issue:45478", "accept": false, - "reason": "UMT5 encoder weight-loading bug is unrelated to Qwen3.5 MoE from-pretrained failure." + "reason": "UMT5Encoder missing weights and Qwen3.5 Moe `from_pretrained` error are different model-loading problems." }, { "left": "issue:43867", "right": "issue:45357", "accept": false, - "reason": "Sorted-state_dict loading and Qwen3.5 visual encoder save-pretrained keys are different serialization problems." + "reason": "State-dict loading failure and incorrect visual encoder keys on save are different sides of different bugs." }, { "left": "issue:28282", "right": "issue:33453", "accept": false, - "reason": "PyTorch-missing ImportError and tokenizer-loading regression are unrelated." + "reason": "PyTorch missing ImportError and tokenizer-loading regression are unrelated." }, { "left": "issue:43618", "right": "issue:46285", "accept": false, - "reason": "Different CLIP regressions: missing attentions vs structural inconsistency of text models." + "reason": "CLIP attentions assignment and CLIP text-model structure inconsistency are distinct changes." }, { "left": "issue:44360", "right": "issue:45440", "accept": false, - "reason": "DSA indexer activation concern and DeepseekV3MoE divergence are unrelated model-implementation issues." + "reason": "DSA indexer ReLU discussion and DeepSeekV3 MoE divergence are unrelated." }, { "left": "issue:42890", "right": "issue:44964", "accept": false, - "reason": "Seed-related CI flakiness is unrelated to loading Phi-4 multimodal with latest transformers." + "reason": "SAM HQ test seeding flakiness is unrelated to Phi-4 multimodal model loading." }, { "left": "issue:42890", "right": "issue:44112", "accept": false, - "reason": "Both are CI/test issues, but one is seed flakiness and the other is a stale device-override test failure." + "reason": "Different CI/test failures: missing seeds vs stale device override in GraniteSpeech." }, { "left": "issue:42890", "right": "issue:45600", "accept": false, - "reason": "Missing seeds in SAM HQ tests is unrelated to broken auto-mappings for removed configs." + "reason": "Test seeding flakiness and auto-mappings referencing removed configs are not the same bug." }, { "left": "issue:42898", "right": "issue:46664", "accept": false, - "reason": "Tokenization-space behavior change and SmolVLM frame-count semantics are different feature regressions." + "reason": "Tokenization cleanup behavior changes and SmolVLM `num_frames` semantics are unrelated." }, { "left": "issue:43502", "right": "issue:44062", "accept": false, - "reason": "Offline-loading network calls and `AddedToken(... special=...)` argument duplication are unrelated." + "reason": "Local-files-only network leakage and AddedToken keyword duplication are unrelated tokenizer/loading issues." }, { "left": "issue:44734", "right": "issue:46617", "accept": false, - "reason": "KV-cache continuation indexing bug in serve is unrelated to trust-remote-code cache file lookup." + "reason": "KV-cache continuation indexing crash and trust_remote_code cache FileNotFoundError are different server/loading failures." }, { "left": "issue:43723", "right": "issue:44556", "accept": false, - "reason": "Tokenizer loading regression and checkpoint reload incompatibility in v5 are distinct failure modes." + "reason": "Tokenizer `from_pretrained` regression and checkpoint reload incompatibility across v4/v5 are not the same concrete bug." }, { "left": "issue:44222", "right": "issue:45125", "accept": false, - "reason": "FP8 `save_pretrained` behavior and missing `_tp_plan` for tensor parallelism are different bugs." + "reason": "FP8 save_pretrained for MoE and missing tensor-parallel plan are separate features/bugs." }, { "left": "issue:44387", "right": "issue:44556", "accept": false, - "reason": "Int4 CUDA memory growth/OOM is unrelated to v4.57 checkpoint reload incompatibility." + "reason": "Int4 quantization reserved-memory/OOM and checkpoint reload incompatibility are different regression classes." }, { "left": "issue:43824", "right": "issue:46617", "accept": false, - "reason": "Missing Qwen2_5_VL import and FileNotFoundError when loading from cache are not the same issue." + "reason": "Missing exported Qwen2_5_VL class and cache FileNotFoundError are unrelated." + }, + { + "left": "issue:43502", + "right": "issue:44485", + "accept": false, + "reason": "API requests despite `local_files_only=True` and GLM-5 RoPE implementation are unrelated." }, { "left": "issue:43874", "right": "issue:46617", "accept": false, - "reason": "GLM46V image-patch counting bug is unrelated to trust-remote-code cache loading failure." + "reason": "Missing image-patch helper and cache FileNotFoundError are different issues." } ] }, "evaluator_result": { "accept": true, - "feedback": "The summary is grounded in the packet and stays conservative overall. The canonical pairing of issue:41669 with issue:44246 is well supported by the titles and the shared import-slowdown theme. The other soft-edge verdicts are phrased cautiously enough and do not overclaim duplicate status from the limited evidence provided." + "feedback": "The summary is well grounded in the packet and stays conservative: it identifies only one likely duplicate pair (41669 vs 44246) and the rest of the soft-edge rejections are appropriately cautious. I don\u2019t see unsupported escalation or overstatement." }, "evaluator_used": true, "retried": false, @@ -17839,14 +17923,14 @@ "issue:44485|issue:46664", "issue:43673|issue:44492" ], - "prepared_review_unit_hash": "2c72da6667398d0d7c05bf65de51f671d2438af652f7fe155c376fabe1c2b089", + "prepared_review_unit_hash": "822bff15e8b00b9e0d044cd004a7bd90be32f93dc4a24fe92e6f1954d491f7c1", "component_budget": { "node_count": 502, "item_count": 502, - "soft_pair_count": 1842, - "serialized_chars": 956182, - "estimated_input_tokens": 239046, - "estimated_eval_tokens": 478348 + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 }, "budget": { "node_count": 47, @@ -17861,180 +17945,180 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This set is mostly a grab-bag of unrelated Transformers issues (imports, model loading, docs, cache/generation, FP8, RoFormer, etc.), so I do not see any clear duplicate PR/issue merges among the soft pairs. The most central issue in the list is the broad v5 regression around the `transformers env` command.", - "confidence": 0.77, - "canonical_issue_reason": "issue:43576 is the broadest and most central user-facing regression in the set, with inbound references and multiple soft links; it is a better anchor than the many narrow model-specific or docs-only issues.", + "summary": "This cluster is not a true duplicate set; it mixes unrelated Transformers v5-era regressions, import errors, generation bugs, docs issues, and model-specific loading failures. None of the soft pairs look like the same underlying bug/change closely enough to merge.", + "confidence": 0.79, + "canonical_issue_reason": "Issue 45478 is the most concrete, actionable representative: it describes a specific versioned `from_pretrained` regression for Qwen3.5 MoE with a clear failure mode and affected path.", "canonical_pr_reason": null, - "best_issue_reason": "issue:43576 is the best standalone triage representative because it is a concrete, widely relevant v5 breakage rather than a narrow model-specific symptom.", + "best_issue_reason": "45478 is the strongest stand-alone triage target in the set because it has the clearest repro surface, explicit version boundary, and a narrow model-loading code path.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43576", "right": "issue:46132", "accept": false, - "reason": "Different problems: a broken `transformers env` command vs changed behavior in `AttentionInterface.register`." + "reason": "CLI/env command breakage and AttentionInterface.register behavior are different subsystems and different bugs." }, { "left": "issue:43010", "right": "issue:43065", "accept": false, - "reason": "Cache update no-grad bug and SAM3 pixel decoder dummy Conv2d are unrelated code paths." + "reason": "Cache/layer `update()` no_grad decoration and SAM3 pixel decoder dummy conv are unrelated code paths." }, { "left": "issue:45125", "right": "issue:45478", "accept": false, - "reason": "Both involve Qwen3.5 MoE, but one is missing `_tp_plan` and the other is a broader `from_pretrained` failure; not clearly the same bug." + "reason": "Both involve Qwen3.5 MoE, but one is tensor-parallel `_tp_plan` support and the other is a `from_pretrained` loading regression." }, { "left": "issue:30333", "right": "issue:33453", "accept": false, - "reason": "MLflow reporting on training failure is unrelated to tokenizer loading regression." + "reason": "MLflow job-state reporting and tokenizer loading are unrelated failures." }, { "left": "issue:44355", "right": "issue:44749", "accept": false, - "reason": "Compiled-file loading errors and a post-upgrade data filtering slowdown are different issues." + "reason": "Compiled Python file errors and upgrade-related data slowdown are different problems." }, { "left": "issue:42548", "right": "issue:44351", "accept": false, - "reason": "Both are import errors, but for different symbols (`PreTrainedModel` vs `HybridCache`) and likely different fixes." + "reason": "Missing `PreTrainedModel` export and missing `HybridCache` export are separate import regressions." }, { "left": "issue:45125", "right": "issue:45375", "accept": false, - "reason": "Tensor-parallel `_tp_plan` issue is unrelated to the missing `deepstack_visual_indexes` config field." + "reason": "Tensor-parallel plan support and a missing vision config field are not the same bug." }, { "left": "issue:46664", "right": "issue:46726", "accept": false, - "reason": "Video `num_frames` semantics and missing vision backend handling are distinct bugs." + "reason": "Video `num_frames` semantics and missing vision-backend handling are distinct issues." }, { "left": "issue:43502", "right": "issue:44961", "accept": false, - "reason": "`local_files_only=True` API leakage is unrelated to the unrelated `racoon` report." + "reason": "`local_files_only=True` network leakage and an unrelated placeholder issue do not match." }, { "left": "issue:43502", "right": "issue:44222", "accept": false, - "reason": "Offline-loading API calls and FP8 `save_pretrained` behavior are different code paths." + "reason": "Offline-loading behavior and FP8 `save_pretrained` are unrelated." }, { "left": "issue:43010", "right": "issue:43232", "accept": false, - "reason": "Both touch generation internals, but one is about `Cache.update` decoration and the other about `_update_model_kwargs_for_generation` after GPU sync." + "reason": "Both touch generation internals, but the cache `update()` no_grad issue and sync_gpus kwargs handling are different root causes." }, { "left": "issue:30990", "right": "issue:31795", "accept": false, - "reason": "A loading hang in Sentence Transformers is not the same as a documentation confusion issue." + "reason": "Loading hang and documentation confusion about `input_ids`/`past_key_values` are not duplicates." }, { "left": "issue:43827", "right": "issue:44961", "accept": false, - "reason": "Docs still referencing `pipeline()` and the unrelated `racoon` issue do not match." + "reason": "Pipeline-removal docs breakage and an unrelated placeholder issue are not the same change." }, { "left": "issue:30333", "right": "issue:36296", "accept": false, - "reason": "MLflow job reporting and tensor-parallel training are unrelated." + "reason": "MLflow reporting and tensor-parallel training are unrelated." }, { "left": "issue:43299", "right": "issue:44556", "accept": false, - "reason": "Both are upgrade/loading regressions, but one is Qwen3VL MoE-specific and the other is a general checkpoint reload compatibility issue." + "reason": "Both are version-upgrade load regressions, but on different models and with different symptoms; too broad to merge." }, { "left": "issue:29870", "right": "issue:31795", "accept": false, - "reason": "BART generate-time UnboundLocalError is unrelated to the forward-docs confusion issue." + "reason": "A concrete `generate()` runtime error and a documentation wording issue are different classes of bug." }, { "left": "issue:44355", "right": "issue:45071", "accept": false, - "reason": "Compiled Python file errors and `PretrainedConfig` type-checking breakage are different failures." + "reason": "Compiled-file execution errors and `PretrainedConfig` type-checking are unrelated." }, { "left": "issue:33453", "right": "issue:34689", "accept": false, - "reason": "Tokenizer loading regression and Llama 3.2 vision model-loading regression are different symptoms and likely different root causes." + "reason": "Tokenizer-loading regression and Llama 3.2 Vision model-loading breakage are different model/load paths." }, { "left": "issue:42757", "right": "issue:43576", "accept": false, - "reason": "Missing `is_offline_mode` import from `huggingface_hub` is not the same as the broken `transformers env` command." + "reason": "Missing `is_offline_mode` import from `huggingface_hub` and a broken `transformers env` command are unrelated." }, { "left": "issue:43576", "right": "issue:45071", "accept": false, - "reason": "CLI breakage and `PretrainedConfig` type-checking are unrelated." + "reason": "CLI breakage and config type-checking regression are different problems." }, { "left": "issue:44315", "right": "issue:44360", "accept": false, - "reason": "Liger Kernel not applied via `model_init` and a DSA indexer/ReLU discussion are unrelated." + "reason": "Liger Kernel application in `model_init` and a DSA indexer discussion are not the same issue." }, { "left": "issue:31795", "right": "issue:38617", "accept": false, - "reason": "Docs confusion around `input_ids`/`past_key_values` is unrelated to the `layer_type_validation` import error." + "reason": "Docs confusion and missing `layer_type_validation` import are unrelated." }, { "left": "issue:44961", "right": "issue:45468", "accept": false, - "reason": "`racoon` and Gemma-4 audio positional encoding are unrelated." + "reason": "Placeholder text and Gemma-4 audio positional encoding are unrelated." }, { "left": "issue:43576", "right": "issue:43824", "accept": false, - "reason": "Broken `transformers env` command and missing `Qwen2_5_VLForConditionalGeneration` import are unrelated." + "reason": "Broken CLI env command and missing `Qwen2_5_VLForConditionalGeneration` export are different failures." }, { "left": "issue:44485", "right": "issue:44961", "accept": false, - "reason": "GLM-5 RoPE implementation discussion is unrelated to the `racoon` issue." + "reason": "GLM-5 RoPE implementation and an unrelated placeholder issue do not match." }, { "left": "issue:44351", "right": "issue:46617", "accept": false, - "reason": "`HybridCache` import failure and trust-remote-code cache `FileNotFoundError` are different problems." + "reason": "Missing `HybridCache` export and FileNotFoundError when loading cached trust_remote_code models are different bug classes." }, { "left": "issue:44908", "right": "issue:44961", "accept": false, - "reason": "Scheduler kwargs bug is unrelated to the `racoon` report." + "reason": "Scheduler kwargs handling and an unrelated placeholder issue are not duplicates." }, { "left": "issue:43576", "right": "issue:44938", "accept": false, - "reason": "`transformers env` breakage and Python 3.14 loading failure are distinct regressions." + "reason": "Broken `transformers env` command and Python 3.14 load failure are different issues." }, { "left": "issue:36296", @@ -18046,43 +18130,43 @@ "left": "issue:46143", "right": "issue:46144", "accept": false, - "reason": "One is about missing `**kwargs` forwarding; the other is about RoFormer not using the attention interface. Same area, but not the same concrete bug." + "reason": "Both are RoFormer-related, but one is missing `**kwargs` passthrough and the other is attention-interface integration; not the same concrete bug." }, { "left": "issue:43475", "right": "issue:46617", "accept": false, - "reason": "SAM3 vision encoder attribute error and trust-remote-code cache loading failure are unrelated." + "reason": "SAM3 vision output attribute error and cached trust_remote_code FileNotFoundError are different failures." }, { "left": "issue:44222", "right": "issue:44230", "accept": false, - "reason": "FP8 MoE save-pretrained bug and FP8 MoE inference support request are adjacent but not the same code-path issue." + "reason": "FP8 `save_pretrained` bug and FP8 inference support request are not the same change." }, { "left": "issue:41628", "right": "issue:42548", "accept": false, - "reason": "Different missing imports: `AutoImageProcessor` vs `PreTrainedModel`." + "reason": "Missing `AutoImageProcessor` and missing `PreTrainedModel` imports are separate exports/regressions." }, { "left": "issue:44315", "right": "issue:46209", "accept": false, - "reason": "Liger Kernel application and FP8 inference on B200 are unrelated." + "reason": "Liger Kernel/model-init application and FP8 inference on B200 are unrelated." }, { "left": "issue:44485", "right": "issue:46664", "accept": false, - "reason": "GLM-5 RoPE implementation and SmolVLM `num_frames` semantics are unrelated." + "reason": "RoPE implementation details and video processor frame semantics are unrelated." }, { "left": "issue:43673", "right": "issue:44492", "accept": false, - "reason": "A v5 generation-cache bug and a typo in cache strategy docs/discussion are not the same issue." + "reason": "Generation cache missing in v5 and a typo in cache strategies are not the same bug." } ] }, @@ -18195,268 +18279,268 @@ "issue:44315|issue:44829", "issue:44360|issue:45468" ], - "prepared_review_unit_hash": "4c07036f99d549ce9fe54fb2b584b4eaeceb656d9ed1a6d74cac520cb8dd25d1", + "prepared_review_unit_hash": "a9c656f1fa3c4b3a826725c123107e7692fed4f07ec1cad0c25334ae23f836c8", "component_budget": { "node_count": 502, "item_count": 502, - "soft_pair_count": 1842, - "serialized_chars": 956182, - "estimated_input_tokens": 239046, - "estimated_eval_tokens": 478348 + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 39, - "serialized_chars": 38224, - "estimated_input_tokens": 9556, - "estimated_eval_tokens": 19368 + "serialized_chars": 38227, + "estimated_input_tokens": 9557, + "estimated_eval_tokens": 19370 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The set is mostly a grab-bag of unrelated bugs and regressions. There is some loose overlap around v5/import/lazy-loading issues, but no soft pair is a safe duplicate merge.", - "confidence": 0.57, - "canonical_issue_reason": "issue:45850 is the broadest umbrella-style regression report in the set, but it is too vague to be a precise canonical duplicate target.", + "summary": "This cluster is mostly a loose collection of unrelated Transformers v5 regressions, model-specific incompatibilities, and a few documentation/performance issues. None of the soft pairs looks like the same concrete underlying bug or change.", + "confidence": 0.96, + "canonical_issue_reason": "issue:45850 is the broadest representative of the cluster\u2019s recurring theme of v5 breakage complaints, but it is still only a loose umbrella, not a true duplicate hub.", "canonical_pr_reason": null, - "best_issue_reason": "issue:45850 is the most suitable general reference issue if one must pick from this list, because it covers the widest regression surface even though it is not specific.", + "best_issue_reason": "issue:45850 is the best global issue candidate because it most directly captures the cluster\u2019s dominant user-visible regression pattern across many models and upgrade paths.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:45362", "right": "issue:45850", "accept": false, - "reason": "Generic v5 breakage vs a specific Qwen3.5 chat crash; same theme, not the same bug." + "reason": "Different bugs: Qwen3.5 chat crash vs generic v5 model-breakage complaint." }, { "left": "issue:44273", "right": "issue:45542", "accept": false, - "reason": "Lazy-loading/import behavior is unrelated to the TensorBoard-without-TF backend error." + "reason": "Lazy loading failure and tf backend error are unrelated code paths." }, { "left": "issue:44623", "right": "issue:45230", "accept": false, - "reason": "A concrete save_pretrained file bug vs an unnamed generic bug report." + "reason": "Specific processor save_pretrained file bug vs a generic bug report with no concrete overlap." }, { "left": "issue:44360", "right": "issue:46664", "accept": false, - "reason": "DSA/ReLU discussion is unrelated to SmolVLM video num_frames semantics." + "reason": "DSA indexer ReLU discussion and video num_frames semantics are unrelated." }, { "left": "issue:44038", "right": "issue:45125", "accept": false, - "reason": "Both mention Qwen3 MoE, but one is a model support/report issue and the other is tensor-parallel metadata; different bugs." + "reason": "Same model family, but different failures: unsupported Qwen3-VL-Moe path vs missing tensor-parallel plan." }, { "left": "issue:44273", "right": "issue:45003", "accept": false, - "reason": "Lazy loading/import performance is not the same as unsafe sys.modules access." + "reason": "Both involve import/lazy-loading concerns, but the actual bugs and fixes are different." }, { "left": "issue:44393", "right": "issue:45071", "accept": false, - "reason": "Qwen3-VL output quality bug vs PretrainedConfig type-checking regression." + "reason": "Qwen3-VL output bug and PretrainedConfig type-checking regression are unrelated." }, { "left": "issue:44393", "right": "issue:45753", "accept": false, - "reason": "Both are Qwen3-VL related, but 2D bbox hallucination and infinite-loop-on-image are distinct failures." + "reason": "Both are Qwen3.5-VL related, but one is bbox output failure and the other is an infinite-loop image-specific bug." }, { "left": "issue:43576", "right": "issue:44556", "accept": false, - "reason": "Broken env command vs checkpoint reload regression; same version era, different code paths." + "reason": "Broken env command and checkpoint reload incompatibility are different upgrade regressions." }, { "left": "issue:41669", "right": "issue:43673", "accept": false, - "reason": "Import-time explosion from import * cleanup is unrelated to chunked_prefill cache missing." + "reason": "Import-* / lazy-loading cleanup vs chunked_prefill cache missing are distinct issues." }, { "left": "issue:43525", "right": "issue:45850", "accept": false, - "reason": "Missing pad_token_id on Llama4Config is a specific config bug, not a broad v5 breakage report." + "reason": "Llama4Config pad_token_id attribute bug is model-specific and not the same as broad v5 breakage." }, { "left": "issue:44556", "right": "issue:45092", "accept": false, - "reason": "Both are v5 compatibility complaints, but one is checkpoint reload and the other is remote-code/meta-init incompatibility." + "reason": "Both are upgrade-related, but one is checkpoint reload failure and the other is remote-code/meta-init incompatibility." }, { "left": "issue:43572", "right": "issue:45850", "accept": false, - "reason": "StableLmConfig missing pad_token_idx is a specific config regression, not the same as generic model breakage." + "reason": "StableLmConfig pad token attribute regression is separate from the generic v5 breakage complaint." }, { "left": "issue:44492", "right": "issue:45850", "accept": false, - "reason": "A typo in cache-strategy text does not match a broad model-breaking regression." + "reason": "Cache-strategy typo and broad model breakage are not the same underlying bug." }, { "left": "issue:45278", "right": "issue:45850", "accept": false, - "reason": "Many import errors after upgrade overlaps only loosely with generic model breakage; not the same underlying issue." + "reason": "Broad upgrade complaints overlap only thematically; one is import errors, the other generic model breakage." }, { "left": "issue:44230", "right": "issue:46095", "accept": false, - "reason": "FP8 inference support request vs a weight-conversion-script question are different asks." + "reason": "FP8 support request for Qwen3-VL is unrelated to a DeepSeek weight-conversion question." }, { "left": "issue:42371", "right": "issue:44492", "accept": false, - "reason": "TF32 API guidance is unrelated to a cache-strategy typo." + "reason": "TF32 settings API and cache-strategy typo are unrelated." }, { "left": "issue:41669", "right": "issue:46218", "accept": false, - "reason": "Import performance regression and missing Qwen3_5 symbol are different import failures." + "reason": "Import-* lazy-loading issue and missing Qwen3_5 module import are different failures." }, { "left": "issue:43408", "right": "issue:44315", "accept": false, - "reason": "Model-type mismatch warning is unrelated to Liger Kernel not being applied in model_init." + "reason": "Sam3 tracker/model mismatch and Liger Kernel not applied with model_init are unrelated." }, { "left": "issue:42890", "right": "issue:44908", "accept": false, - "reason": "Missing set_seed in tests and inverse_sqrt kwargs handling are unrelated." + "reason": "Missing test seeds and inverse_sqrt scheduler kwargs are different bugs." }, { "left": "issue:44246", "right": "issue:44273", "accept": false, - "reason": "Both touch import/lazy loading, but one is intermittent slow imports and the other is lazy-loading correctness; not the same bug." + "reason": "Slow import sometimes and lazy loading broken are related in topic only, not the same defect." }, { "left": "issue:43576", "right": "issue:44908", "accept": false, - "reason": "Broken env command and scheduler kwargs bug are unrelated." + "reason": "Environment command regression and scheduler kwargs bug are unrelated." }, { "left": "issue:43352", "right": "issue:44393", "accept": false, - "reason": "Flash Attention 2 support gating is unrelated to Qwen3-VL bbox hallucination/output errors." + "reason": "Flash Attention 2 support error for Nemotron and Qwen3-VL bbox hallucination are different model issues." }, { "left": "issue:43976", "right": "issue:44393", "accept": false, - "reason": "Python-version compatibility regression is unrelated to Qwen3-VL output bug." + "reason": "Python version compatibility regression and Qwen3-VL output bug do not share a concrete code path." }, { "left": "issue:44908", "right": "issue:45071", "accept": false, - "reason": "Scheduler kwargs and config type checking affect different subsystems." + "reason": "Scheduler kwargs handling and PretrainedConfig type checking are unrelated." }, { "left": "issue:44295", "right": "issue:45230", "accept": false, - "reason": "position_ids buffer access error is a concrete modeling bug, unlike the generic report." + "reason": "Position_ids buffer access error and a generic bug report are not duplicates." }, { "left": "issue:44297", "right": "issue:45230", "accept": false, - "reason": "tokenizer_class mismatch is a specific save_pretrained issue, not a generic bug report." + "reason": "Tokenizer save_pretrained mismatch is a concrete tokenizer bug, unlike the generic bug report." }, { "left": "issue:44704", "right": "issue:44908", "accept": false, - "reason": "AutoProcessor cached_file kwargs and inverse_sqrt scheduler kwargs are unrelated kwargs bugs." + "reason": "AutoProcessor cached_file kwargs forwarding and scheduler kwargs are different subsystems." }, { "left": "issue:44492", "right": "issue:46328", "accept": false, - "reason": "Cache-strategy typo and SequenceFeatureExtractor.pad performance regression are different issues." + "reason": "Cache-strategy typo and SequenceFeatureExtractor.pad performance issue are unrelated." }, { "left": "issue:41669", "right": "issue:44351", "accept": false, - "reason": "Import-star cleanup/import-time regression is not the same as missing HybridCache export." + "reason": "Both are import-related, but one is import-* cleanup and the other is HybridCache export/import failure." }, { "left": "issue:41669", "right": "issue:45341", "accept": false, - "reason": "Import-related refactor issue vs a small testing_utils bug; no shared underlying defect." + "reason": "Import cleanup regression and a testing_utils bug are not the same issue." }, { "left": "issue:43976", "right": "issue:45850", "accept": false, - "reason": "Python support regression and broad model breakage are only loosely related by version, not by bug." + "reason": "Python-version incompatibility and generic v5 model breakage are only broadly related." }, { "left": "issue:44393", "right": "issue:45478", "accept": false, - "reason": "Qwen3-VL bbox issue and Qwen3.5 Moe from_pretrained error are different model-path failures." + "reason": "Qwen3-VL bbox error and Qwen3.5 MoE from_pretrained failure are different model-specific bugs." }, { "left": "issue:43976", "right": "issue:46218", "accept": false, - "reason": "Python-version compatibility and missing Qwen3_5 import are unrelated." + "reason": "Python support regression and missing Qwen3_5ForCausalLM import are unrelated." }, { "left": "issue:44077", "right": "issue:44861", "accept": false, - "reason": "Optional post_init handling and tied-weight-key crash are different internals." + "reason": "Patchtsmixer post_init validation and tied-weight-key crash are different code paths." }, { "left": "issue:31795", "right": "issue:33453", "accept": false, - "reason": "Documentation confusion and tokenizer loading regression are unrelated." + "reason": "Documentation confusion and tokenizer-loading regression are not the same underlying bug." }, { "left": "issue:43576", "right": "issue:44273", "accept": false, - "reason": "Env-command regression and lazy-loading issue do not share the same code path." + "reason": "Broken env command and lazy loading malfunction are separate v5 infra issues." }, { "left": "issue:44315", "right": "issue:44829", "accept": false, - "reason": "Liger Kernel application in model_init and flash_attention_3 training degeneration are separate problems." + "reason": "Liger Kernel application problem and flash_attention_3 degenerate training are distinct bugs." }, { "left": "issue:44360", "right": "issue:45468", "accept": false, - "reason": "DSA/ReLU discussion is unrelated to Gemma-4 audio positional encoding." + "reason": "DSA indexer discussion and Gemma-4 audio positional encoding bug are unrelated." } ] }, @@ -18532,8 +18616,8 @@ "soft_pairs": [ "issue:43519|issue:46655", "issue:44273|issue:44908", - "issue:44655|issue:44861", "issue:45230|issue:45310", + "issue:44655|issue:44861", "issue:43576|issue:46218", "issue:43519|issue:44485", "issue:45468|issue:46894", @@ -18563,8 +18647,8 @@ "issue:31515|issue:38617", "issue:43572|issue:44861", "issue:43525|issue:44861", - "issue:28348|issue:43352", "issue:44230|issue:44623", + "issue:28348|issue:43352", "issue:29870|issue:31515", "issue:30333|issue:31515", "issue:31515|issue:36296", @@ -18573,20 +18657,20 @@ "issue:43298|issue:44230", "issue:43717|issue:46144" ], - "prepared_review_unit_hash": "31deabcf6495ad419a6bdf37459cf12927d2d69c8a16f6469a1365a81fd5097c", + "prepared_review_unit_hash": "f0dfca480ae299ca72a894fa08fc0b7c656ea92ed69bdbacb69a4dd52d27921b", "component_budget": { "node_count": 502, "item_count": 502, - "soft_pair_count": 1842, - "serialized_chars": 956182, - "estimated_input_tokens": 239046, - "estimated_eval_tokens": 478348 + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 42, - "serialized_chars": 39366, + "serialized_chars": 39367, "estimated_input_tokens": 9842, "estimated_eval_tokens": 19940 }, @@ -18595,114 +18679,114 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is highly heterogeneous. A few items cluster around v5/model-loading/tokenizer regressions, but most are unrelated subsystem bugs or broad requests. I did not find any soft-edge pair that clearly describes the same underlying bug/change.", - "confidence": 0.66, - "canonical_issue_reason": "Issue 45850 is the broadest umbrella for the v5 regression-style reports, so it is the most reasonable anchor for this mixed set, even though the cluster is not truly duplicate-coherent.", + "summary": "No soft pair looks like the same underlying bug/change closely enough to merge. Most pairs are only superficially similar by shared model family, a broad subsystem, or a generic regression theme; the concrete symptoms and code paths differ.", + "confidence": 0.92, + "canonical_issue_reason": "issue:43525 is the clearest self-contained bug report in the set: a specific model, a concrete missing attribute, and a narrow actionable failure path.", "canonical_pr_reason": null, - "best_issue_reason": "45850 is the most representative issue here because it captures the broad 'v5 breaks many models' theme that overlaps with several nearby reports more than any other single issue.", + "best_issue_reason": "issue:43525 is the strongest canonical candidate because it is specific, reproducible, and points to a well-scoped fix surface.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43519", "right": "issue:46655", "accept": false, - "reason": "Different model families and failures: Qwen3VL timestamp math vs SmolVLM video padding calculation." + "reason": "Different models and different processors: timestamp calculation in Qwen3VL vs padding calculation in SmolVLM video processing." }, { "left": "issue:44273", "right": "issue:44908", "accept": false, - "reason": "Lazy loading failure and inverse_sqrt scheduler kwargs are unrelated code paths." + "reason": "Unrelated subsystems: lazy loading vs learning-rate scheduler kwargs." }, { - "left": "issue:44655", - "right": "issue:44861", + "left": "issue:45230", + "right": "issue:45310", "accept": false, - "reason": "Saving Pipeline objects and tied-weight key handling are different bugs." + "reason": "One is a generic bug placeholder; the other is a specific Qwen3.5 MoE from_pretrained regression." }, { - "left": "issue:45230", - "right": "issue:45310", + "left": "issue:44655", + "right": "issue:44861", "accept": false, - "reason": "Generic bug report vs a specific Qwen3.5 MoE from_pretrained error; not the same issue." + "reason": "Saving pipeline objects and tied-weight-key handling are different code paths with different failures." }, { "left": "issue:43576", "right": "issue:46218", "accept": false, - "reason": "Broken env command and missing Qwen3_5ForCausalLM import are different regressions." + "reason": "CLI/env command breakage is unrelated to an import failure for a specific model class." }, { "left": "issue:43519", "right": "issue:44485", "accept": false, - "reason": "Timestamp calculation in Qwen3VL is unrelated to GLM-5 RoPE implementation." + "reason": "Different model-specific bugs: processor timestamp logic vs GLM-5 RoPE implementation." }, { "left": "issue:45468", "right": "issue:46894", "accept": false, - "reason": "Gemma-4 audio positional encoding and CohereASR training-loss shifting are distinct problems." + "reason": "Different models and symptoms: Gemma-4 audio positional encoding vs CohereASR training-loss shifting." }, { "left": "issue:43519", "right": "issue:44908", "accept": false, - "reason": "Video timestamp math does not match scheduler keyword-argument handling." + "reason": "Processor timestamp logic and scheduler kwargs are unrelated." }, { "left": "issue:41669", "right": "issue:43704", "accept": false, - "reason": "Import-star cleanup and a Qwen3ForCausalLM VRAM leak are not the same bug." + "reason": "Import-star cleanup/perf issue is unrelated to a VRAM leak in multi-threaded dataloading." }, { "left": "issue:31795", "right": "issue:34689", "accept": false, - "reason": "Documentation confusion for forward inputs/past_key_values is unrelated to a Llama 3.2 Vision loading breakage." + "reason": "Documentation confusion about forward inputs is not the same as a model-loading regression." }, { "left": "issue:46018", "right": "issue:46172", "accept": false, - "reason": "DeepSeek-V4 shared-expert gating and a comb.transpose dtype issue are different model-path bugs." + "reason": "Both mention DeepSeek-V4, but the reported bugs are different: gating logic vs a tensor transpose/cast issue." }, { "left": "issue:43704", "right": "issue:44485", "accept": false, - "reason": "VRAM leakage in threaded use and GLM-5 RoPE implementation are unrelated." + "reason": "VRAM leakage in threading and RoPE implementation are distinct bugs." }, { "left": "issue:44485", "right": "issue:45468", "accept": false, - "reason": "Two different model architectures and failure modes; no shared code-path evidence." + "reason": "Different models and different implementation areas; no shared failing code path." }, { "left": "issue:43502", "right": "issue:43519", "accept": false, - "reason": "Local-files-only network calls and Qwen3VL timestamp calculation are separate defects." + "reason": "local_files_only network leakage is unrelated to timestamp calculation in a video processor." }, { "left": "issue:42371", "right": "issue:43704", "accept": false, - "reason": "TF32 API guidance is unrelated to a Qwen3ForCausalLM memory leak." + "reason": "TF32 settings guidance and a VRAM leak are unrelated issues." }, { "left": "issue:45850", "right": "issue:46218", "accept": false, - "reason": "Broad v5 breakage report and a specific missing import for Qwen3_5ForCausalLM are not the same underlying bug." + "reason": "Broad 'many models break' report is too vague to match a specific Qwen3_5 import failure." }, { "left": "issue:43519", "right": "issue:43704", "accept": false, - "reason": "Timestamp calculation and multi-thread VRAM leakage are different issues." + "reason": "Different failure modes in different model families; no evidence of one underlying bug." }, { "left": "issue:44263", @@ -18714,145 +18798,145 @@ "left": "issue:44261", "right": "issue:45468", "accept": false, - "reason": "Missing rms_norm_eps in MLA q_a_layernorm is a different bug from Gemma-4 audio positional encoding." + "reason": "Different models and different numerical issues; not the same bug." }, { "left": "issue:43298", "right": "issue:46095", "accept": false, - "reason": "Continuous batching support for audio models and DeepSeek-V4 weight conversion are unrelated requests." + "reason": "Continuous batching support for audio models is not a weight-conversion-script request." }, { "left": "issue:44351", "right": "issue:46218", "accept": false, - "reason": "HybridCache import failure and Qwen3_5ForCausalLM import failure are similar in shape but different missing symbols/code paths." + "reason": "Two different import failures with no shared target or code path." }, { "left": "issue:43352", "right": "issue:46218", "accept": false, - "reason": "Flash Attention 2.0 support for NemotronHForCausalLM is unrelated to a Qwen3_5ForCausalLM import error." + "reason": "Flash-Attention support for Nemotron and an import error for Qwen3_5 are unrelated." }, { "left": "issue:42913", "right": "issue:43066", "accept": false, - "reason": "Both mention v5 tokenizer changes, but they point to different symptoms and likely different root causes." + "reason": "Both are tokenizer-v5 complaints, but the symptoms are different enough that they do not clearly describe one concrete bug." }, { "left": "issue:28348", "right": "issue:45250", "accept": false, - "reason": "One is GPT2-specific Flash Attention 2.0 support; the other is a generic Flash Attention 2.0 request." + "reason": "GPT2LMHeadModel flash-attention support is more specific than the generic Flash Attention 2.0 topic." }, { "left": "issue:44908", "right": "issue:44998", "accept": false, - "reason": "Scheduler kwargs handling and an off-topic issue are unrelated." + "reason": "Scheduler kwargs bug and an unrelated generic issue title do not match." }, { "left": "issue:44492", "right": "issue:44998", "accept": false, - "reason": "A cache-strategy typo and an off-topic issue do not match." + "reason": "A typo in cache-strategy naming is unrelated to the generic issue." }, { "left": "issue:43618", "right": "issue:44998", "accept": false, - "reason": "CLIP attentions assignment regression is unrelated to the off-topic issue." + "reason": "CLIPOutput attentions assignment regression is unrelated to the generic issue title." }, { "left": "issue:44998", "right": "issue:45850", "accept": false, - "reason": "The off-topic issue does not share a concrete bug/change with the broad v5 regression report." + "reason": "A generic 'too many models break' complaint does not identify the same concrete bug as the generic placeholder issue." }, { "left": "issue:44938", "right": "issue:44998", "accept": false, - "reason": "Python 3.14 loading failure is unrelated to the off-topic issue." + "reason": "Python 3.14 import/load failure is unrelated to the generic issue." }, { "left": "issue:44861", "right": "issue:45290", "accept": false, - "reason": "Tied-weights key crash and chat-template assistant tool-call crash are different code paths." + "reason": "Tied-weight key handling and chat-template tool-call crashes are different features and failures." }, { "left": "issue:31515", "right": "issue:38617", "accept": false, - "reason": "Slow checkpoint loading and a missing import error are unrelated." + "reason": "Slow from_pretrained checkpoint loading and an import error for layer_type_validation are different problems." }, { "left": "issue:43572", "right": "issue:44861", "accept": false, - "reason": "Missing pad_token_idx in StableLmConfig is unrelated to tied-weight key crashes." + "reason": "Missing pad_token_idx in config is not the same as tied-weight key list handling." }, { "left": "issue:43525", "right": "issue:44861", "accept": false, - "reason": "Missing pad_token_id in Llama4Config is a different config regression." + "reason": "Llama4Config missing pad_token_id and tied-weight key handling are distinct code paths." }, { - "left": "issue:28348", - "right": "issue:43352", + "left": "issue:44230", + "right": "issue:44623", "accept": false, - "reason": "Both mention Flash Attention 2.0, but they target different models and support requests." + "reason": "FP8 inference support request and missing processor save files are unrelated." }, { - "left": "issue:44230", - "right": "issue:44623", + "left": "issue:28348", + "right": "issue:43352", "accept": false, - "reason": "FP8 Qwen3-VL/Qwen3.5 support and missing processor.save_pretrained files are unrelated." + "reason": "Both mention Flash Attention 2.0, but one is GPT2 support and the other is Nemotron unsupported status; not the same concrete bug or change." }, { "left": "issue:29870", "right": "issue:31515", "accept": false, - "reason": "Bart generate decoder_attention_mask crash and slow from_pretrained loading are unrelated." + "reason": "BART generate() decoder_attention_mask bug and slow from_pretrained loading are unrelated." }, { "left": "issue:30333", "right": "issue:31515", "accept": false, - "reason": "MLFlow reporting on failed jobs has no overlap with checkpoint loading performance." + "reason": "MLFlow reporting on training failure is not the same as checkpoint loading performance." }, { "left": "issue:31515", "right": "issue:36296", "accept": false, - "reason": "Slow model loading and tensor-parallel training bugs are different problems." + "reason": "from_pretrained slowness and tensor-parallel training bugs are different failure modes." }, { "left": "issue:31515", "right": "issue:31795", "accept": false, - "reason": "Checkpoint loading performance and forward-pass documentation confusion do not match." + "reason": "Checkpoint loading performance and forward-API documentation confusion are unrelated." }, { "left": "issue:44230", "right": "issue:45310", "accept": false, - "reason": "FP8 support request for Qwen3-VL/Qwen3.5 MoE and a from_pretrained error are not the same underlying bug." + "reason": "FP8 inference support request is unrelated to a Qwen3.5 MoE from_pretrained error." }, { "left": "issue:43298", "right": "issue:44230", "accept": false, - "reason": "Continuous batching for audio models and FP8 support for Qwen3-VL/Qwen3.5 MoE are unrelated." + "reason": "Audio continuous batching support and FP8 inference support are different feature requests." }, { "left": "issue:43717", "right": "issue:46144", "accept": false, - "reason": "Mamba init_weights usage and RoFormer attention interface usage are different implementation issues." + "reason": "Mamba init_weights behavior and RoFormer attention interface usage are unrelated implementation issues." } ] }, @@ -18898,20 +18982,20 @@ "issue:37934|issue:43668", "issue:42936|issue:43668" ], - "prepared_review_unit_hash": "1a6ef8f5167360fb91eab445479447148e5f9d6eca0fa57773254a67761046e3", + "prepared_review_unit_hash": "4cf3715f25058c3c5bf72a5e9b627bdbd4d63ccc254992c82ea27136d4aa4094", "component_budget": { "node_count": 502, "item_count": 502, - "soft_pair_count": 1842, - "serialized_chars": 956182, - "estimated_input_tokens": 239046, - "estimated_eval_tokens": 478348 + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 }, "budget": { "node_count": 8, "item_count": 8, "soft_pair_count": 11, - "serialized_chars": 7773, + "serialized_chars": 7776, "estimated_input_tokens": 1944, "estimated_eval_tokens": 4144 }, @@ -18920,78 +19004,78 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster is heterogeneous: model-specific bugs, a tokenizer regression, a typo, and a config type-hint issue. None of the soft pairs look like the same underlying bug, so all should stay separate. The most representative issue is the generic functional bug about kwargs not reaching the loss function.", - "confidence": 0.9, - "canonical_issue_reason": "issue:43240 is the broadest, most actionable functional bug in the set: it describes a concrete runtime failure in loss computation and is more representative than the typo/type-hint/narrow model issues.", + "summary": "These issues cover unrelated bugs across different models/components (norms, loss kwargs, tokenizer warnings, attention interfaces, and a typo). No duplicate cluster is evident, so all soft-edge pairs should be rejected.", + "confidence": 0.93, + "canonical_issue_reason": "issue:43240 is the most concrete and actionable report in the set: it describes a specific loss-kwargs propagation bug with a clear code-path implication.", "canonical_pr_reason": null, - "best_issue_reason": "issue:43240 is the strongest standalone issue to anchor the cluster because it has a clear code-path defect and likely wider impact than the more model-specific or cosmetic reports.", + "best_issue_reason": "issue:43240 is the best representative issue because it is narrowly scoped, implementation-specific, and likely to map to a single fix.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:42936", "right": "issue:46018", "accept": false, - "reason": "Different model families and different failure modes: Mask2Former ignore_value handling vs DeepSeek-V4 gating logic." + "reason": "Different models and different failure modes: Mask2Former ignore_value handling vs DeepSeek-V4 shared expert gating." }, { "left": "issue:43668", "right": "issue:44031", "accept": false, - "reason": "One is a ModernBERT config type-hint problem; the other is a tokenizer warning regression. Not the same bug." + "reason": "A config type-hint mistake is unrelated to a tokenizer regex warning regression." }, { "left": "issue:42936", "right": "issue:43240", "accept": false, - "reason": "Both are bugs, but in unrelated code paths: Mask2Former ignore_value vs passing kwargs into loss calculation." + "reason": "One is about Mask2Former ignore_value usage; the other is about loss kwargs not being forwarded." }, { "left": "issue:43240", "right": "issue:46144", "accept": false, - "reason": "Generic kwargs-to-loss bug vs RoFormer attention interface implementation; no concrete overlap." + "reason": "Loss-function kwargs propagation and RoFormer attention interface usage are distinct code-path problems." }, { "left": "issue:42936", "right": "issue:44031", "accept": false, - "reason": "Mask2Former loss handling and tokenizer regex warnings are unrelated subsystems and defects." + "reason": "Mask2Former ignore_value behavior is unrelated to tokenizer warning behavior." }, { "left": "issue:42936", "right": "issue:46144", "accept": false, - "reason": "Different models and different bugs; no evidence they stem from the same underlying change." + "reason": "These concern different subsystems and different bugs; no shared underlying fix is apparent." }, { "left": "issue:42936", "right": "issue:43813", "accept": false, - "reason": "Mask2Former ignore_value logic is unrelated to a typo in a quantization-related string." + "reason": "Ignore_value handling in a model head is unrelated to a typo in a quantization attribute name." }, { "left": "issue:43240", "right": "issue:43668", "accept": false, - "reason": "Loss-function kwargs propagation is unrelated to a config annotation/type-hint error." + "reason": "Forwarding kwargs to a loss function is unrelated to a ModernBERTConfig type annotation issue." }, { "left": "issue:43668", "right": "issue:43813", "accept": false, - "reason": "ModernBERT config typing and a quantization typo are distinct cosmetic/code-quality issues." + "reason": "A type-hint error and a string/attribute typo are different defect types with no common code path." }, { "left": "issue:37934", "right": "issue:43668", "accept": false, - "reason": "Llama4TextL2Norm behavior question vs ModernBERT norm_eps type hint; different models and defect types." + "reason": "RMS-norm behavior question for Llama4TextL2Norm is unrelated to a ModernBERT config type hint." }, { "left": "issue:42936", "right": "issue:43668", "accept": false, - "reason": "Mask2Former loss behavior and ModernBERT config typing are unrelated." + "reason": "Mask2Former loss/ignore_value logic and ModernBERT config typing do not describe the same bug." } ] }, @@ -19024,52 +19108,52 @@ "issue:10519|issue:11005", "issue:11005|issue:11307" ], - "prepared_review_unit_hash": "f5511004e7561dd868d3f620dc5fa49ad4e9b59c9def4a608d45268575ccba10", + "prepared_review_unit_hash": "ada96ec7f012a44d1948ab1aff1fb32cf36cee0143dfa255eb191c2367c5de8f", "component_budget": { "node_count": 3, "item_count": 3, "soft_pair_count": 3, - "serialized_chars": 2877, - "estimated_input_tokens": 720, - "estimated_eval_tokens": 1696 + "serialized_chars": 2875, + "estimated_input_tokens": 719, + "estimated_eval_tokens": 1694 }, "budget": { "node_count": 3, "item_count": 3, "soft_pair_count": 3, - "serialized_chars": 2877, - "estimated_input_tokens": 720, - "estimated_eval_tokens": 1696 + "serialized_chars": 2875, + "estimated_input_tokens": 719, + "estimated_eval_tokens": 1694 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These three issues are not duplicates: they cover unrelated feature requests in token truncation, learning-rate scheduling, and Wav2Vec2 word timing offsets. No soft pair looks like the same underlying bug/change.", + "summary": "These three issues are not duplicates: one is about token truncation direction, one about ReduceLROnPlateau-like scheduling, and one about Wav2Vec2 word time offsets. The cluster looks like a false-positive similarity grouping rather than a single underlying bug.", "confidence": 0.98, - "canonical_issue_reason": "No clear canonical duplicate exists because the issues are about different subsystems and different user goals. If one must be chosen as a representative, #10519 is the most concrete and narrowly scoped.", + "canonical_issue_reason": "No clear canonical issue exists because the topics are unrelated and do not describe the same bug or feature request.", "canonical_pr_reason": null, - "best_issue_reason": "#10519 is the best representative issue only in the sense that it is the most specific and actionable request; the other two are broader or different in domain.", + "best_issue_reason": "None of the issues is a good global canonical representative for the cluster; 10519 is the most self-contained, but it still has no relationship to the other two.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:10519", "right": "issue:11307", "accept": false, - "reason": "Different features in different areas: tokenizer truncation direction vs Wav2Vec2 word boundary timing. No shared code-path or same underlying bug." + "reason": "Different domains and requests: tokenizer truncation behavior vs Wav2Vec2 word boundary timing." }, { "left": "issue:10519", "right": "issue:11005", "accept": false, - "reason": "Unrelated requests: truncation behavior vs ReduceLROnPlateau-like scheduler functionality. Similarity is only superficial." + "reason": "Unrelated functionality: truncation option vs learning-rate scheduler behavior." }, { "left": "issue:11005", "right": "issue:11307", "accept": false, - "reason": "Different problem domains and expected changes; one is about training scheduler behavior, the other about ASR word timestamp extraction." + "reason": "No shared underlying bug or change; scheduler feature request is unrelated to Wav2Vec2 timestamp extraction." } ] }, @@ -19101,12 +19185,12 @@ "issue:13244|issue:16618", "issue:15980|issue:16618" ], - "prepared_review_unit_hash": "abc58debed1564a07304eb0feddf92750bedeb72ebcd7f183bab20ef15570423", + "prepared_review_unit_hash": "2fef424e1aa6db6c0457ae8a812a119b6e9c418513283024be98a7e9192bda07", "component_budget": { "node_count": 3, "item_count": 3, "soft_pair_count": 2, - "serialized_chars": 2477, + "serialized_chars": 2479, "estimated_input_tokens": 620, "estimated_eval_tokens": 1496 }, @@ -19114,7 +19198,7 @@ "node_count": 3, "item_count": 3, "soft_pair_count": 2, - "serialized_chars": 2477, + "serialized_chars": 2479, "estimated_input_tokens": 620, "estimated_eval_tokens": 1496 }, @@ -19123,24 +19207,24 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "This cluster is mostly mixed/unrelated: one issue is about Tapas tokenization, one about a private-model download/login error, and one about a general model-loading failure. The shared wording is too broad to treat them as the same bug.", - "confidence": 0.86, - "canonical_issue_reason": "issue:16618 is the broadest and most central of the three, since it is the only one directly about a model-loading failure; the other two describe different problems in different paths.", + "summary": "These are three unrelated issue reports. The only potentially general model-loading failure is #16618, while #15980 is specifically about an unauthenticated private-model download error and #13244 is about Tapas tokenization behavior, so neither soft pair should be merged.", + "confidence": 0.93, + "canonical_issue_reason": "#16618 is the clearest standalone bug report about model loading failure and is more central than the other two, which describe different subsystems and failure modes.", "canonical_pr_reason": null, - "best_issue_reason": "issue:16618 is the best representative issue for this cluster because it matches the only model-loading themed report and is the least specific of the set.", + "best_issue_reason": "#16618 is the best representative issue because it names the concrete user-facing failure, whereas #15980 is an auth/error-message problem and #13244 is a tokenizer parity issue.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:13244", "right": "issue:16618", "accept": false, - "reason": "Tapas tokenization mismatch is a different subsystem and symptom from a 'can't load model' failure; no evidence they share the same underlying bug." + "reason": "These describe different problems: Tapas tokenization mismatch versus inability to load bert-base-uncased. They are not the same underlying bug." }, { "left": "issue:15980", "right": "issue:16618", "accept": false, - "reason": "Private-model download/login error is auth-specific, while the other issue is a general public-model loading failure; these are not the same code-path problem." + "reason": "#15980 is about the error shown when downloading a private model without login, not the same failure as #16618's model-loading issue." } ] }, @@ -19170,7 +19254,7 @@ "soft_pairs": [ "issue:26350|issue:33260" ], - "prepared_review_unit_hash": "581d5a69ba8cc63c35cf554ebf469c96299bcc32e29916e5e596a52736732855", + "prepared_review_unit_hash": "8f76a7e6011a335b5c884e5ecc4eb93de67f787cfaa6167a6cae6da475c12cf8", "component_budget": { "node_count": 2, "item_count": 2, @@ -19192,18 +19276,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These two issues are not duplicates: one requests Flash Attention 2 support for more architectures, while the other requests GGUF support for more architectures. They are separate feature areas and should not be clustered as the same underlying request.", - "confidence": 0.97, - "canonical_issue_reason": "No canonical issue: the topics are different feature requests (Flash Attention 2 vs GGUF support) and do not describe the same bug or change.", + "summary": "The two issues are about different feature requests: Flash Attention 2 support for more architectures vs GGUF support for more architectures. They are not duplicates and should not be clustered together.", + "confidence": 0.98, + "canonical_issue_reason": "No canonical issue: the issues describe distinct enhancement requests for different code paths and formats.", "canonical_pr_reason": null, - "best_issue_reason": "No single best issue for duplication purposes; both are distinct community contribution requests with different implementation targets.", + "best_issue_reason": "If a single representative is required, issue 26350 is slightly better due to being older and having more discussion, but it is not a duplicate of issue 33260.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:26350", "right": "issue:33260", "accept": false, - "reason": "Different features and code paths: Flash Attention 2 support is unrelated to GGUF support, so they are not the same underlying request." + "reason": "Different features with different technical goals (Flash Attention 2 vs GGUF support); same broad theme of architecture support is too weak for duplicate/merge triage." } ] }, @@ -19268,218 +19352,215 @@ "issue:46222|issue:46931", "issue:46226|issue:46931" ], - "prepared_review_unit_hash": "32c9ab7725ac219a03a546b4baa6720b1a52b169b9e474521c8b6914995ef2fe", + "prepared_review_unit_hash": "6d6d4edb052383ac3ec27e42ce44ac422e53244189b2489bcec37e39cb5c1046", "component_budget": { "node_count": 9, "item_count": 9, "soft_pair_count": 29, - "serialized_chars": 15971, - "estimated_input_tokens": 3993, - "estimated_eval_tokens": 8242 + "serialized_chars": 15982, + "estimated_input_tokens": 3996, + "estimated_eval_tokens": 8248 }, "budget": { "node_count": 9, "item_count": 9, "soft_pair_count": 29, - "serialized_chars": 15971, - "estimated_input_tokens": 3993, - "estimated_eval_tokens": 8242 + "serialized_chars": 15982, + "estimated_input_tokens": 3996, + "estimated_eval_tokens": 8248 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "The cluster is not a single duplicate set; it contains multiple separate exact-title duplicate-looking subgroups. In particular, the placeholder i18n issues 43834/44365/44370 form one repeated-title group, and 46222/46226 form another exact-title Thai pair, while the language-specific translation requests remain distinct.", - "confidence": 0.77, - "canonical_issue_reason": "issue:43834 is a reasonable canonical anchor for the repeated exact-title placeholder subgroup because it is the earliest of that trio, but the cluster also contains a separate exact-title Thai duplicate pair, so it is not the only duplicate-looking issue here.", + "summary": "These are not true duplicates: most are separate docs-translation requests for different languages, while a few are generic template instances. The cluster is driven by template similarity rather than the same underlying issue.", + "confidence": 0.97, + "canonical_issue_reason": "No single issue is a real duplicate canonical because each concrete issue targets a different language translation request; issue 43834 is only the closest generic template match.", "canonical_pr_reason": null, - "best_issue_reason": "issue:43834 is the strongest issue representative within the placeholder-title subgroup, but there are multiple exact-title duplicate-looking subgroups in the cluster, so it should be treated as a local representative rather than a unique cluster-wide canonical.", + "best_issue_reason": "Issue 43834 is the best representative only in the narrow sense that it is the generic placeholder form, but it is not a good global canonical because the cluster contains multiple independent language-specific requests.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:27088", "right": "issue:44370", "accept": false, - "reason": "Different translation requests; the shared i18n boilerplate is not enough to treat them as the same issue." + "reason": "Different language translation requests; same template wording but not the same underlying task." }, { "left": "issue:27088", "right": "issue:44365", "accept": false, - "reason": "Different language/request; only the docs-translation template is similar." + "reason": "Different language translation requests; they are structurally similar but not duplicates." }, { "left": "issue:27088", "right": "issue:43834", "accept": false, - "reason": "Different translation requests; no evidence they target the same underlying issue." + "reason": "One is a concrete Turkish request and the other is a generic placeholder template, so not the same issue." }, { "left": "issue:27088", "right": "issue:38435", "accept": false, - "reason": "Turkish and Romanian translation requests are separate issues despite similar template wording." + "reason": "Different target languages (Turkish vs Romanian), so these are separate requests." }, { "left": "issue:27088", "right": "issue:28936", "accept": false, - "reason": "Different target language and separate translation request." + "reason": "Different target languages (Turkish vs Spanish), so not duplicates." }, { "left": "issue:28936", "right": "issue:43834", "accept": false, - "reason": "Spanish request versus a placeholder template issue; similarity is boilerplate only." + "reason": "Generic template vs a concrete Spanish translation request; not the same underlying issue." }, { "left": "issue:28936", "right": "issue:44365", "accept": false, - "reason": "Different translation requests; shared template text does not make them duplicates." + "reason": "Different translation requests despite similar template text." }, { "left": "issue:28936", "right": "issue:38435", "accept": false, - "reason": "Spanish and Romanian docs requests are distinct." + "reason": "Different target languages, so separate issues." }, { "left": "issue:28936", "right": "issue:44370", "accept": false, - "reason": "Only the surrounding i18n template matches; no evidence of the same request." + "reason": "Different translation requests; similarity is only in format." }, { "left": "issue:38435", "right": "issue:44370", "accept": false, - "reason": "Romanian request and generic template issue are not the same underlying task." + "reason": "Different instances of the translation template and different actual requests." }, { "left": "issue:38435", "right": "issue:44365", "accept": false, - "reason": "Romanian translation request versus a different placeholder issue; keep separate." + "reason": "Separate language-translation requests, not one underlying bug/change." }, { "left": "issue:38435", "right": "issue:43834", "accept": false, - "reason": "Same broad translation theme, but different request content." + "reason": "Romanian request vs generic placeholder template; not a duplicate." }, { "left": "issue:38435", "right": "issue:46222", "accept": false, - "reason": "Different request content; no sign these are the same translation issue." + "reason": "Different issues; only the wording/template is similar." }, { "left": "issue:38435", "right": "issue:46226", "accept": false, - "reason": "Different request content; no evidence of the same underlying issue." + "reason": "Different issues; these are not the same concrete request." }, { "left": "issue:43834", "right": "issue:44370", - "accept": true, - "reason": "Same exact placeholder title and nearly identical body; looks like a repeated template issue." + "accept": false, + "reason": "Both are template-like, but they still represent separate translation requests rather than one shared bug/change." }, { "left": "issue:43834", "right": "issue:44365", - "accept": true, - "reason": "Same exact placeholder title and near-identical content; strong duplicate signal." + "accept": false, + "reason": "Generic placeholders only; not enough to conclude the same underlying issue." }, { "left": "issue:43834", "right": "issue:46222", "accept": false, - "reason": "Different title/content; one is a placeholder template issue and the other is a distinct Thai-titled request." + "reason": "Different concrete issues even if the template is identical." }, { "left": "issue:43834", "right": "issue:46226", "accept": false, - "reason": "Different title/content; no evidence they are the same request." + "reason": "Different concrete issues even if the template is identical." }, { "left": "issue:43834", "right": "issue:46931", "accept": false, - "reason": "Generic i18n template issue and Thai-titled issue are not clearly the same request." + "reason": "The titles are unrelated beyond matching the template pattern; not a duplicate." }, { "left": "issue:44365", "right": "issue:44370", - "accept": true, - "reason": "Same exact placeholder title and nearly identical body; clear duplicate-looking pair." + "accept": false, + "reason": "Two separate template-based translation requests; no evidence they are the same request." }, { "left": "issue:44365", "right": "issue:46222", "accept": false, - "reason": "Different title/content; only broad translation similarity." + "reason": "Same template pattern but not the same underlying issue." }, { "left": "issue:44365", "right": "issue:46226", "accept": false, - "reason": "Different title/content; no sign of the same underlying issue." + "reason": "Same template pattern but not the same underlying issue." }, { "left": "issue:44365", "right": "issue:46931", "accept": false, - "reason": "Different issue content despite similar i18n structure." + "reason": "Titles do not indicate the same concrete task or bug." }, { "left": "issue:44370", "right": "issue:46222", "accept": false, - "reason": "Different title/content; not the same translation request." + "reason": "Same generic translation template, but separate issues." }, { "left": "issue:44370", "right": "issue:46226", "accept": false, - "reason": "Different title/content; no evidence they are duplicates." + "reason": "Same generic translation template, but separate issues." }, { "left": "issue:44370", "right": "issue:46931", "accept": false, - "reason": "Template similarity only; not enough to merge." + "reason": "No clear shared underlying issue beyond superficial similarity." }, { "left": "issue:46222", "right": "issue:46226", - "accept": true, - "reason": "Same exact Thai title and identical payload; clear duplicate." + "accept": false, + "reason": "Even though the text is nearly identical, they appear to be distinct issue entries rather than evidence of one concrete duplicate." }, { "left": "issue:46222", "right": "issue:46931", "accept": false, - "reason": "Different issue content and no sign of the same request." + "reason": "Only superficial textual similarity; not enough to treat as the same issue." }, { "left": "issue:46226", "right": "issue:46931", "accept": false, - "reason": "Different issue content; not the same underlying request." + "reason": "Only superficial textual similarity; not enough to treat as the same issue." } ] }, - "evaluator_result": { - "accept": false, - "feedback": "The output is too strong on the duplicate classification. In the packet, there are multiple exact-title matches among the templated i18n issues (43834, 44365, and 44370 all share the same title), so 46222/46226 are not the only exact-title duplicate-looking pair. The canonical/best-issue rationale should be more conservative and not imply uniqueness here." - }, - "evaluator_used": true, - "retried": true, - "accepted_nontrivial_soft_edge": true, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, "error_kind": null, "error_message": null }, @@ -19558,274 +19639,274 @@ "issue:42503|issue:46653", "issue:45027|issue:46653" ], - "prepared_review_unit_hash": "e502188847f2fb5a3e896ce261c0d0db15b98bdcd01bce28f4559cc189156af0", + "prepared_review_unit_hash": "e546ac0256f0a3d3b3d0e3c357719e087b83d4a6fcb13bf75ae2bbc33dc7a0d0", "component_budget": { "node_count": 19, "item_count": 19, "soft_pair_count": 40, - "serialized_chars": 24676, - "estimated_input_tokens": 6169, - "estimated_eval_tokens": 12594 + "serialized_chars": 24672, + "estimated_input_tokens": 6168, + "estimated_eval_tokens": 12592 }, "budget": { "node_count": 19, "item_count": 19, "soft_pair_count": 40, - "serialized_chars": 24676, - "estimated_input_tokens": 6169, - "estimated_eval_tokens": 12594 + "serialized_chars": 24672, + "estimated_input_tokens": 6168, + "estimated_eval_tokens": 12592 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "This cluster is a loose mix of model-addition requests, not duplicate reports. The titles often share template wording (\u201cAdd \u2026 model/support\u201d) and some are in adjacent families, but each item targets a different model or capability, so no pair looks like the same underlying change.", - "confidence": 0.9, - "canonical_issue_reason": "No true duplicate anchor: the issues are separate requests for different models/features, so there is no single canonical issue for deduping the cluster.", + "summary": "This cluster is a loose set of model-add/support requests, not a duplicate set. The titles share a broad pattern ('Add X model/support'), but each pair targets a different model, modality, or feature, so no soft edge looks like the same underlying change. No PRs are present.", + "confidence": 0.78, + "canonical_issue_reason": "issue:36879 is the clearest concrete model-addition request and a good representative of the cluster\u2019s common pattern, even though the cluster itself is not made of true duplicates.", "canonical_pr_reason": null, - "best_issue_reason": "If a representative is needed, #36879 is the cleanest generic model-add request, but it still represents only one specific model and is not a duplicate of the others.", + "best_issue_reason": "issue:36879 is the most representative issue overall: specific, well-scoped, and central to the repeated 'add model' theme across the cluster.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:36718", "right": "issue:41115", "accept": false, - "reason": "Different model requests: RoMa keypoint matcher vs MiniCPM3 architecture." + "reason": "Different artifacts: keypoint matcher vs MiniCPM3 architecture request." }, { "left": "issue:36817", "right": "issue:41211", "accept": false, - "reason": "EuroBert config entry vs DEIMv2 model add; unrelated changes." + "reason": "EuroBert config request and DEIMv2 are unrelated model additions." }, { "left": "issue:36817", "right": "issue:43671", "accept": false, - "reason": "EuroBert model config vs Qwen3-TTS support; different tasks." + "reason": "BERT config change vs Qwen3-TTS support; different model families and code paths." }, { "left": "issue:36817", "right": "issue:36879", "accept": false, - "reason": "EuroBert vs RF-DETR are distinct models with no shared code change implied." + "reason": "EuroBert and RF-DETR are separate model requests, not one change." }, { "left": "issue:36879", "right": "issue:41211", "accept": false, - "reason": "RF-DETR and DEIMv2 are different model additions." + "reason": "RF-DETR and DEIMv2 are distinct models with different integration work." }, { "left": "issue:36879", "right": "issue:43909", "accept": false, - "reason": "RF-DETR detection model vs LFM2.5 Audio 1.5B; different domains." + "reason": "Object detection model vs audio model; no shared underlying bug or feature." }, { "left": "issue:36879", "right": "issue:44121", "accept": false, - "reason": "RF-DETR model add vs weight-sparse transformer request; not the same feature." + "reason": "RF-DETR model addition and weight-sparse transformer request are unrelated." }, { "left": "issue:36879", "right": "issue:43837", "accept": false, - "reason": "RF-DETR vs Qwen3-ASR support are different model integrations." + "reason": "RF-DETR and Qwen3-ASR are different models and different tasks." }, { "left": "issue:36879", "right": "issue:43671", "accept": false, - "reason": "RF-DETR detection model vs Qwen3-TTS support; no same code path." + "reason": "Different model families and modalities; not the same implementation." }, { "left": "issue:37015", "right": "issue:43837", "accept": false, - "reason": "NeoBERT model add vs Qwen3-ASR support; unrelated model families." + "reason": "NeoBERT addition and Qwen3-ASR support are unrelated requests." }, { "left": "issue:37015", "right": "issue:44121", "accept": false, - "reason": "NeoBERT vs weight-sparse transformer request; different targets." + "reason": "NeoBERT and circuit-sparsity transformer changes do not share a concrete code-path fix." }, { "left": "issue:37015", "right": "issue:43909", "accept": false, - "reason": "NeoBERT text model vs LFM2.5 Audio 1.5B; different capabilities." + "reason": "BERT-style model request vs audio model support; different targets." }, { "left": "issue:36879", "right": "issue:37015", "accept": false, - "reason": "RF-DETR and NeoBERT are distinct model requests." + "reason": "RF-DETR and NeoBERT are distinct model additions." }, { "left": "issue:37015", "right": "issue:42503", "accept": false, - "reason": "NeoBERT vs ModernVBERT models; similar naming but different model additions." + "reason": "NeoBERT and ModernVBERT are separate models; same broad area only." }, { "left": "issue:41115", "right": "issue:43837", "accept": false, - "reason": "MiniCPM3 architecture request vs Qwen3-ASR support; different model work." + "reason": "MiniCPM3 architecture and Qwen3-ASR support are different model implementations." }, { "left": "issue:41115", "right": "issue:44121", "accept": false, - "reason": "MiniCPM3 architecture vs weight-sparse transformer; not the same underlying change." + "reason": "MiniCPM3 model support vs weight-sparse transformer request are not the same change." }, { "left": "issue:41115", "right": "issue:42503", "accept": false, - "reason": "MiniCPM3 vs ModernVBERT are separate model requests." + "reason": "MiniCPM3 and ModernVBERT are different model requests." }, { "left": "issue:41115", "right": "issue:45027", "accept": false, - "reason": "MiniCPM3 architecture vs Voxtral TTS support; different model paths." + "reason": "MiniCPM3 architecture and Voxtral TTS support are unrelated." }, { "left": "issue:41211", "right": "issue:43671", "accept": false, - "reason": "DEIMv2 model add vs Qwen3-TTS support; different models." + "reason": "DEIMv2 and Qwen3-TTS are different models and modalities." }, { "left": "issue:42503", "right": "issue:43837", "accept": false, - "reason": "ModernVBERT models vs Qwen3-ASR support; unrelated features." + "reason": "ModernVBERT models and Qwen3-ASR support are separate requests." }, { "left": "issue:42503", "right": "issue:43671", "accept": false, - "reason": "ModernVBERT vs Qwen3-TTS support; different model families." + "reason": "ModernVBERT and Qwen3-TTS are unrelated model additions." }, { "left": "issue:42503", "right": "issue:42738", "accept": false, - "reason": "ModernVBERT models vs BERT-like models with RoPE; one is a concrete model add, the other is a broad architecture request." + "reason": "ModernVBERT models vs a general RoPE-related BERT-like proposal; not the same concrete change." }, { "left": "issue:42738", "right": "issue:45027", "accept": false, - "reason": "BERT-like RoPE architecture request vs Voxtral TTS support; not the same code change." + "reason": "General RoPE support for BERT-like models and Voxtral TTS support are unrelated." }, { "left": "issue:42738", "right": "issue:46653", "accept": false, - "reason": "Model architecture request vs a person-name issue; clearly unrelated." + "reason": "BERT/RoPE model request vs an unrelated personal-name issue; no overlap." }, { "left": "issue:43671", "right": "issue:43837", "accept": false, - "reason": "Qwen3-TTS support and Qwen3-ASR support are adjacent but distinct tasks; strict duplicate criteria not met." + "reason": "Qwen3-TTS and Qwen3-ASR are related by name only; distinct support paths." }, { "left": "issue:43671", "right": "issue:45027", "accept": false, - "reason": "Qwen3-TTS support vs Voxtral TTS support; different models and implementations." + "reason": "Both audio-related, but different models and tasks; not one mergeable fix." }, { "left": "issue:43671", "right": "issue:45177", "accept": false, - "reason": "Qwen3-TTS support vs DeepSeek-OCR2; different model domains." + "reason": "TTS support and DeepSeek-OCR2 are different model integrations." }, { "left": "issue:43671", "right": "issue:44121", "accept": false, - "reason": "Qwen3-TTS support vs weight-sparse transformer request; no same underlying bug/change." + "reason": "Qwen3-TTS support and weight-sparse transformer request are unrelated." }, { "left": "issue:43837", "right": "issue:43909", "accept": false, - "reason": "Qwen3-ASR support vs LFM2.5 Audio 1.5B; both audio-related but different models." + "reason": "Qwen3-ASR and LFM2.5 Audio are different audio models with separate work." }, { "left": "issue:43837", "right": "issue:45322", "accept": false, - "reason": "Qwen3-ASR support vs EUPE encoder; different architectures and goals." + "reason": "ASR support vs EUPE encoder request; different model additions." }, { "left": "issue:43837", "right": "issue:44121", "accept": false, - "reason": "Qwen3-ASR support vs weight-sparse transformer request; unrelated." + "reason": "Qwen3-ASR support and circuit-sparsity transformer request are not the same feature." }, { "left": "issue:43909", "right": "issue:44205", "accept": false, - "reason": "LFM2.5 Audio 1.5B vs SAM3-LiteText; different model categories." + "reason": "Audio model support and SAM3-LiteText are different integration targets." }, { "left": "issue:43909", "right": "issue:46645", "accept": false, - "reason": "Audio model request vs Granite SWA+sink option; one is a model add, the other a model feature toggle." + "reason": "LFM2.5 Audio support and Granite SWA+sink option are different changes." }, { "left": "issue:44121", "right": "issue:45322", "accept": false, - "reason": "Weight-sparse transformer request vs EUPE encoder; different model architectures." + "reason": "Weight-sparse transformer request and EUPE model request are unrelated." }, { "left": "issue:43837", "right": "issue:45027", "accept": false, - "reason": "Qwen3-ASR support vs Voxtral TTS support; different models even though both are speech-related." + "reason": "Qwen3-ASR and Voxtral TTS are separate speech-model support requests." }, { "left": "issue:37015", "right": "issue:45027", "accept": false, - "reason": "NeoBERT vs Voxtral TTS support; unrelated requests." + "reason": "NeoBERT and Voxtral TTS are different model requests." }, { "left": "issue:43837", "right": "issue:45177", "accept": false, - "reason": "Qwen3-ASR support vs DeepSeek-OCR2; different model targets." + "reason": "Qwen3-ASR support and DeepSeek-OCR2 are distinct model additions." }, { "left": "issue:45322", "right": "issue:46133", "accept": false, - "reason": "EUPE encoder vs TIPSv2 vision-language pretraining; distinct model additions." + "reason": "EUPE and TIPSv2 are different vision-language models; same broad domain only." }, { "left": "issue:42503", "right": "issue:46653", "accept": false, - "reason": "ModernVBERT models vs a person-name issue; clearly unrelated." + "reason": "ModernVBERT model request and unrelated personal-name issue share no underlying change." }, { "left": "issue:45027", "right": "issue:46653", "accept": false, - "reason": "Voxtral TTS support vs a person-name issue; unrelated." + "reason": "Voxtral TTS support and unrelated personal-name issue are not duplicates." } ] }, @@ -19855,7 +19936,7 @@ "soft_pairs": [ "issue:38740|issue:46443" ], - "prepared_review_unit_hash": "813a6873233ab2eb9609a2395f4d2ea7cfdb67c23f762d1f2e58f7337974287c", + "prepared_review_unit_hash": "af5bd9264c2ade167cea9d8286bf3cfc54e3cba527c4ffb63c671c77f5a949a5", "component_budget": { "node_count": 2, "item_count": 2, @@ -19877,18 +19958,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These two issues are unrelated: one is a documentation request about adding Pruna as an optimization framework, while the other is a vision RoPE refactor. No duplicate relationship is supported.", + "summary": "These two issues are not duplicates: one is a docs request to add Pruna as an optimization framework, while the other is a technical Vision RoPE refactor. They do not describe the same underlying problem or change.", "confidence": 0.98, - "canonical_issue_reason": null, + "canonical_issue_reason": "No canonical issue: the topics are unrelated and there is no shared bug or feature request to collapse onto one issue.", "canonical_pr_reason": null, - "best_issue_reason": "Neither issue is a good canonical representative for the other; they cover different topics and scopes.", + "best_issue_reason": "Issue #46443 is the more substantive and specific technical work item; it is the better representative issue in this pair, though not a duplicate target for #38740.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:38740", "right": "issue:46443", "accept": false, - "reason": "Different subject areas and intentions: docs/vendor inclusion vs. a vision RoPE code refactor. They do not describe the same underlying bug or change." + "reason": "Different scopes and intent: documentation/framework listing request vs a Vision RoPE code refactor. No evidence they refer to the same underlying change." } ] }, @@ -19918,7 +19999,7 @@ "soft_pairs": [ "issue:41084|issue:43037" ], - "prepared_review_unit_hash": "4ceba559fa8cb8f8545fa19b31c92e0a34535279bf61b97b8056c240edfa5cb3", + "prepared_review_unit_hash": "e2ee96cba9396022e1a3c9926aa435bc1f40fa5bb33042bda070729ba0de732b", "component_budget": { "node_count": 2, "item_count": 2, @@ -19940,18 +20021,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These two issues are not duplicates: one requests \"Set Block Decoding\" and the other asks for \"DeepSeek v3.2 support.\" They describe different problems and different likely code paths.", - "confidence": 0.98, - "canonical_issue_reason": "No canonical duplicate here; the two issues are about different requests rather than the same underlying bug or feature.", + "summary": "The two issues do not look like duplicates: one is about set block decoding, the other about DeepSeek v3.2 support. The soft similarity signal is too weak to justify merging them.", + "confidence": 0.97, + "canonical_issue_reason": "Issue #41084 is the better canonical representative because it is the more general technical topic, while #43037 is a specific model-support request.", "canonical_pr_reason": null, - "best_issue_reason": "issue:43037 is the clearer standalone request because it names a specific model support target and is more actionable.", + "best_issue_reason": "#41084 is the best issue to anchor the cluster since it is broader and more likely to describe the underlying implementation area rather than a single model request.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:41084", "right": "issue:43037", "accept": false, - "reason": "Different concerns: a decoding-mode change versus adding support for a specific model. No evidence they are the same underlying issue." + "reason": "Reject: the titles and likely intents are unrelated; 'Set Block Decoding' concerns a decoding mode/implementation detail, while 'DeepSeek v3.2 support' is a model compatibility request. No shared files, targets, or explicit linkage." } ] }, @@ -19981,40 +20062,40 @@ "soft_pairs": [ "issue:41306|issue:41826" ], - "prepared_review_unit_hash": "4ed425b1f1bf198d8b2e7577cef0e8a2d7599a60617c434701a13bbcc9d14d5d", + "prepared_review_unit_hash": "fc07ad48c73dd5a0e92986297fae7aaf6f7ad52cc2cab0ba11a17939862034fb", "component_budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 1584, - "estimated_input_tokens": 396, - "estimated_eval_tokens": 1048 + "serialized_chars": 1585, + "estimated_input_tokens": 397, + "estimated_eval_tokens": 1050 }, "budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 1584, - "estimated_input_tokens": 396, - "estimated_eval_tokens": 1048 + "serialized_chars": 1585, + "estimated_input_tokens": 397, + "estimated_eval_tokens": 1050 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "The two issues are unrelated: one is about adding a `TiledFusedLogitsLoss` / FLCE flag in `from_pretrained()`, while the other is about integrating `TiledMLP` for lower memory usage.", - "confidence": 0.98, + "summary": "These two issues are both about memory-efficient/tiled model components, but they describe different features in different code paths: one is about exposing a `TiledFusedLogitsLoss` / FLCE flag in `from_pretrained()`, while the other requests integrating `TiledMLP`. They should not be clustered as duplicates.", + "confidence": 0.95, "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": null, + "best_issue_reason": "If a representative must be chosen, issue 41306 is slightly more concrete and narrowly scoped, but it is not a true duplicate of 41826.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:41306", "right": "issue:41826", "accept": false, - "reason": "Different features and codepaths with no clear overlap beyond both being model-optimization related; they do not look like the same underlying bug/change." + "reason": "Different underlying requests and likely different code paths: FLCE/from_pretrained configuration vs. adding TiledMLP for lower memory use. No explicit linkage or shared filenames/targets." } ] }, @@ -20044,7 +20125,7 @@ "soft_pairs": [ "issue:43598|issue:43600" ], - "prepared_review_unit_hash": "a5530f8bb9f3d92dfc67b8245cb29802a98d4ee0ed28621caf24ccf9113d4602", + "prepared_review_unit_hash": "9c9e8f6882ae04cf1beaaf9732aa144760b8007df33d63f82f1f1a0f667ca023", "component_budget": { "node_count": 2, "item_count": 2, @@ -20066,18 +20147,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "The two issues are related to Trainer customization, but they describe different problems: one is about `remove_unused_column`, the other about FLOPS estimation for custom trainers. They are not the same underlying bug.", + "summary": "These are two distinct Trainer customization issues: one about `remove_unused_column` behavior, the other about FLOPS estimation with custom trainers. They share the same broad area but not the same underlying bug/change.", "confidence": 0.91, - "canonical_issue_reason": "Issue #43598 is the cleaner canonical candidate if forced, because it is the more specific Trainer customization problem and has a concrete setting (`remove_unused_column`) rather than a broader metrics-estimation concern.", + "canonical_issue_reason": "Issue #43598 is the more specific and actionable bug report, centered on a concrete Trainer behavior (`remove_unused_column`) and its customizability.", "canonical_pr_reason": null, - "best_issue_reason": "Issue #43598 is slightly better as a standalone representative because it names a specific Trainer behavior and likely points to a focused code-path change.", + "best_issue_reason": "Issue #43598 is the better representative for the cluster because it is narrowly scoped and directly describes a single Trainer customization problem.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43598", "right": "issue:43600", "accept": false, - "reason": "Both mention custom Trainers, but they target different functionality: unused-column handling versus FLOPS counting. Same subsystem, not the same bug or change." + "reason": "Both mention custom trainers, but they address different code paths and outcomes: column removal customization vs FLOPS estimation. They are not the same underlying bug or change." } ] }, @@ -20107,12 +20188,12 @@ "soft_pairs": [ "issue:43684|issue:44405" ], - "prepared_review_unit_hash": "788292a3409d84d3814679b4f67b8d6ba0a17e3745a7d26775ea9043497eaa7a", + "prepared_review_unit_hash": "7ad541e71e04a019b7c43d5911ed2e0400342429971c56c5012bc1b227d42a28", "component_budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 1611, + "serialized_chars": 1612, "estimated_input_tokens": 403, "estimated_eval_tokens": 1062 }, @@ -20120,7 +20201,7 @@ "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 1611, + "serialized_chars": 1612, "estimated_input_tokens": 403, "estimated_eval_tokens": 1062 }, @@ -20129,18 +20210,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These are related only at a high level (both request AutoModel registration for Qwen-family configs), but they target different models and different auto classes, so they are not duplicates.", - "confidence": 0.95, - "canonical_issue_reason": "Issue 43684 is the broader request: add Qwen3-Omni registration to AutoModel and AutoModelForConditionalGeneration. If choosing one issue as the cluster anchor, it better represents the generic auto-registration gap, but it does not subsume the sequence-classification request.", + "summary": "The two issues are about different model families and different AutoModel APIs: one requests Qwen3-Omni registration for AutoModel/AutoModelForConditionalGeneration, while the other asks for AutoModelForSequenceClassification support for Qwen3.5. They are related only at a broad \u201cadd model support\u201d level, not the same underlying bug or change.", + "confidence": 0.97, + "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "Issue 43684 is the better representative because it asks for core auto-class registration across more central classes and is phrased as a general model-registration gap rather than a single head-specific support request.", + "best_issue_reason": "Issue #43684 is the broader and more concrete API-registration request, but it is not a duplicate of #44405; the topics and target classes differ.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43684", "right": "issue:44405", "accept": false, - "reason": "Different underlying tasks: one is Qwen3-Omni registration for AutoModel/AutoModelForConditionalGeneration, the other is AutoModelForSequenceClassification support for Qwen3.5 (Qwen3_5Config). They involve different configs, different auto classes, and are not the same bug/change." + "reason": "Different models (Qwen3-Omni vs Qwen3.5) and different auto classes (generation registration vs sequence classification support); not the same underlying change." } ] }, @@ -20177,70 +20258,70 @@ "issue:43845|issue:43859", "issue:43846|issue:43859" ], - "prepared_review_unit_hash": "a84ffa797ffa6db614f7687a34244e1e769e3ae47acedfd80cce69d3be8859de", + "prepared_review_unit_hash": "e429079caea6c4cc196581fef0609df9cab7b3386b02971e593d613b18d003db", "component_budget": { "node_count": 4, "item_count": 4, "soft_pair_count": 6, - "serialized_chars": 4266, - "estimated_input_tokens": 1067, - "estimated_eval_tokens": 2390 + "serialized_chars": 4262, + "estimated_input_tokens": 1066, + "estimated_eval_tokens": 2388 }, "budget": { "node_count": 4, "item_count": 4, "soft_pair_count": 6, - "serialized_chars": 4266, - "estimated_input_tokens": 1067, - "estimated_eval_tokens": 2390 + "serialized_chars": 4262, + "estimated_input_tokens": 1066, + "estimated_eval_tokens": 2388 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These four issues look like low-information duplicates/spam, with three near-identical \"huggingface \" titles and one unrelated \"Fett\" issue. The strongest textual match is between 43845 and 43846, but there is still not enough substance to confidently treat any pair as the same real bug report.", - "confidence": 0.91, - "canonical_issue_reason": "Issue 43845 is the earliest of the near-identical \"huggingface \" reports and is the best representative of that mini-cluster, despite the lack of useful body detail.", + "summary": "All four issues are too sparse to establish real duplicate identity; the three identical-title \"huggingface\" issues are the closest candidates, but there is not enough substantive evidence to confidently merge any pair.", + "confidence": 0.42, + "canonical_issue_reason": "issue:43859 is the best placeholder canonical because it matches the repeated title and has the only discussion activity, but the cluster lacks body/detail evidence.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 43845 is the best global issue choice because it is the earliest clearly duplicate-looking report among the only substantive title match set.", + "best_issue_reason": "issue:43859 is the most suitable cluster representative: it is one of the repeated-title issues and has slightly more activity than the others.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43835", "right": "issue:43845", "accept": false, - "reason": "Titles are unrelated and there is no body evidence tying \"Fett\" to \"huggingface\"." + "reason": "Only weak title similarity; no body details, links, or shared targets to show the same underlying issue." }, { "left": "issue:43835", "right": "issue:43846", "accept": false, - "reason": "Titles are unrelated and the issues contain no descriptive detail showing a shared bug." + "reason": "Only weak title similarity; no substantive evidence of the same bug/problem." }, { "left": "issue:43835", "right": "issue:43859", "accept": false, - "reason": "No meaningful overlap beyond both being low-information closed issues." + "reason": "Different title from the repeated huggingface issues, and there is no content evidence tying it to the others." }, { "left": "issue:43845", "right": "issue:43846", "accept": false, - "reason": "They have the same bland title and look duplicate-like, but there is still no concrete bug description to confirm they are the same underlying issue." + "reason": "Identical short title, but the issues are otherwise empty; that is not enough to safely treat them as the same underlying report." }, { "left": "issue:43845", "right": "issue:43859", "accept": false, - "reason": "High title similarity, but the reports are too empty to verify they describe the same problem." + "reason": "Same generic title, but there is no description or issue-target evidence to confirm a duplicate." }, { "left": "issue:43846", "right": "issue:43859", "accept": false, - "reason": "Same generic title pattern, but insufficient evidence of a shared underlying bug or change." + "reason": "Same generic title, but insufficient evidence to conclude they refer to the same concrete issue." } ] }, @@ -20281,88 +20362,88 @@ "issue:44572|issue:44937", "issue:44573|issue:44937" ], - "prepared_review_unit_hash": "2b54c81c46ab33f223699f1ede47b47afa96665b1776fbd57ce182d2299cbedc", + "prepared_review_unit_hash": "edd89155e9f2dd29dde7e59e0880390fcb7d12af902bf26da389f92a3bc01cda", "component_budget": { "node_count": 5, "item_count": 5, "soft_pair_count": 9, - "serialized_chars": 5794, - "estimated_input_tokens": 1449, - "estimated_eval_tokens": 3154 + "serialized_chars": 5791, + "estimated_input_tokens": 1448, + "estimated_eval_tokens": 3152 }, "budget": { "node_count": 5, "item_count": 5, "soft_pair_count": 9, - "serialized_chars": 5794, - "estimated_input_tokens": 1449, - "estimated_eval_tokens": 3154 + "serialized_chars": 5791, + "estimated_input_tokens": 1448, + "estimated_eval_tokens": 3152 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "This cluster looks like unrelated low-information/spam issues, not a duplicate set for the same underlying bug or request. No soft pair is convincing enough to merge.", + "summary": "These are low-information, likely spam or placeholder issues with no evidence of a shared underlying bug. The similar titles are not enough to treat them as duplicates, so the soft links should all be rejected.", "confidence": 0.97, - "canonical_issue_reason": null, + "canonical_issue_reason": "No clear canonical issue exists: the items are unrelated spam/placeholder reports with too little substantive detail to identify a common bug.", "canonical_pr_reason": null, - "best_issue_reason": null, + "best_issue_reason": "If one must be chosen as a representative, issue 44418 is the least spam-like title, but it still does not represent a real shared issue cluster.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:44371", "right": "issue:44572", "accept": false, - "reason": "Both are just '' with no actionable content; same placeholder text is not enough to establish the same underlying issue." + "reason": "Both are placeholder spam titles, but there is no substantive body, target, or code-path evidence showing the same underlying issue." }, { "left": "issue:44371", "right": "issue:44573", "accept": false, - "reason": "Both are just '' and provide no shared bug/report details to deduplicate." + "reason": "Identical spam-style titles are not enough; there is no concrete bug description tying them together." }, { "left": "issue:44371", "right": "issue:44418", "accept": false, - "reason": "One is '' and the other is a generic documentation enhancement title; no evidence they describe the same problem." + "reason": "One is spam and the other is a generic documentation suggestion; no shared underlying issue is shown." }, { "left": "issue:44371", "right": "issue:44937", "accept": false, - "reason": "'' and 'Check out \"Google LLC\"' are different spam-like titles with no shared concrete issue." + "reason": "Completely different titles and no supporting evidence of the same problem." }, { "left": "issue:44418", "right": "issue:44572", "accept": false, - "reason": "Generic documentation enhancement vs. spam placeholder; not the same underlying request." + "reason": "Generic documentation suggestion versus spam placeholder; no concrete common issue." }, { "left": "issue:44418", "right": "issue:44573", "accept": false, - "reason": "Generic documentation enhancement vs. spam placeholder; no concrete overlap." + "reason": "Generic documentation suggestion versus spam placeholder; no evidence they describe the same bug or change." }, { "left": "issue:44572", "right": "issue:44573", "accept": false, - "reason": "Both are '' placeholders, but there is no actual issue description to show they are the same bug or change." + "reason": "Although the titles match, both are bare spam placeholders with no details to verify a shared underlying bug." }, { "left": "issue:44572", "right": "issue:44937", "accept": false, - "reason": "Spam placeholder vs. unrelated titled issue; insufficient evidence of duplication." + "reason": "Spam placeholder versus unrelated promotional-looking title; no shared issue." }, { "left": "issue:44573", "right": "issue:44937", "accept": false, - "reason": "Spam placeholder vs. unrelated titled issue; not a credible duplicate match." + "reason": "Spam placeholder versus unrelated promotional-looking title; no shared issue." } ] }, @@ -20392,12 +20473,12 @@ "soft_pairs": [ "issue:45145|issue:45151" ], - "prepared_review_unit_hash": "049a2ff7de1507e4f89376d4a066d10cb224710534cc861cbd39974e98267cad", + "prepared_review_unit_hash": "679779e3e7a577e4947fdfe3b6c6a87fe88d9f92348332d8e4fde54ea9ce6e7e", "component_budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 1645, + "serialized_chars": 1646, "estimated_input_tokens": 412, "estimated_eval_tokens": 1080 }, @@ -20405,7 +20486,7 @@ "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 1645, + "serialized_chars": 1646, "estimated_input_tokens": 412, "estimated_eval_tokens": 1080 }, @@ -20414,28 +20495,28 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These two closed issues appear to be duplicates of the same report: identical title, very similar body length/content, and close creation times. Issue #45145 is the better canonical candidate because it was opened earlier and already has an inbound reference.", - "confidence": 0.98, - "canonical_issue_reason": "#45145 is the stronger canonical issue: it was created first and has an inbound reference, while #45151 has none. The titles are identical and the bodies appear to describe the same request/problem.", + "summary": "These two closed issues have identical titles and strong soft-similarity signals, but the available evidence is still limited to similarity rather than a deterministic duplicate link. Issue 45145 remains the better canonical choice because it was created first and already has one inbound reference.", + "confidence": 0.87, + "canonical_issue_reason": "Issue 45145 is the earliest instance and has an inbound reference, so it is the cleaner canonical representative for this likely duplicate report.", "canonical_pr_reason": null, - "best_issue_reason": "#45145 is the best issue to keep as the representative record because it likely attracted the existing reference and is the earliest instance of this duplicate report.", + "best_issue_reason": "Issue 45145 is the best representative because it is first chronologically and slightly better connected, making it the most suitable target for consolidation.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:45145", "right": "issue:45151", - "accept": true, - "reason": "Same title and evidently the same underlying energy-reduction report; close timestamps and near-identical body size support a duplicate relationship." + "accept": false, + "reason": "The titles match and the issues look highly similar, but the evidence shown is only soft similarity and deterministic_accept is false. This is best treated as a likely duplicate rather than a confirmed duplicate pair." } ] }, "evaluator_result": { - "accept": true, - "feedback": "The summary is grounded in the packet: both issues have the same title, very similar body lengths, close timestamps, and the packet\u2019s soft similarity signal supports a duplicate relationship. The canonical choice for #45145 is also justified by its earlier creation time and inbound reference. The wording stays reasonably conservative overall." + "accept": false, + "feedback": "The canonical-choice reasoning is grounded in the packet (same title, earlier creation time, and one inbound reference vs. none). However, the soft-edge verdict is too strong for the available evidence: the packet only shows identical titles plus soft-similarity signals, and deterministic_accept is false. This should be phrased as a possible/likely duplicate, not a confirmed valid duplicate pair." }, "evaluator_used": true, - "retried": false, - "accepted_nontrivial_soft_edge": true, + "retried": true, + "accepted_nontrivial_soft_edge": false, "error_kind": null, "error_message": null }, @@ -20512,7 +20593,7 @@ "soft_pairs": [ "issue:45762|issue:45763" ], - "prepared_review_unit_hash": "37541174eb025705a5b2230f1e51cafd59336ff90131f96b0de6f3de6fac62f2", + "prepared_review_unit_hash": "b177fd47360863275875b048f94a02f4a7eef8e4135f8540d58a8a745590cb71", "component_budget": { "node_count": 2, "item_count": 2, @@ -20534,18 +20615,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Two closed issues with only name/date titles and no body content; they do not appear to describe the same underlying bug or request, so this looks unrelated rather than duplicate.", + "summary": "These two closed issues do not look like duplicates of the same technical problem; they are just distinct personal-name/date entries with no body or code-related context. There is no meaningful canonical issue or PR in this cluster.", "confidence": 0.98, - "canonical_issue_reason": "No canonical issue: the two issues are unrelated by title/content and do not share a concrete problem statement.", + "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "If a representative must be chosen, issue 45762 is only marginally more readable, but neither issue is a meaningful canonical duplicate target.", + "best_issue_reason": null, "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:45762", "right": "issue:45763", "accept": false, - "reason": "The titles are just different personal names with different dates and there is no evidence of the same bug, feature, or code path." + "reason": "Titles only share a surname and both are non-technical personal-name/date strings; there is no evidence they describe the same underlying bug or request." } ] }, @@ -20575,7 +20656,7 @@ "soft_pairs": [ "issue:45764|issue:45765" ], - "prepared_review_unit_hash": "3588f12d7c84a32ddd1de67e2c301df2704ed9dd1f4a0700b50133b5dedd6220", + "prepared_review_unit_hash": "148c902d520cd5a41eff0457cd1b90042167eb4a80d340a4d92ae38baada3ab7", "component_budget": { "node_count": 2, "item_count": 2, @@ -20597,18 +20678,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "The two issues are not meaningfully related: both are empty, user-name/date-style titles with no body, links, or shared technical context. The soft similarity is not enough to treat them as duplicates.", - "confidence": 0.96, - "canonical_issue_reason": null, + "summary": "These two closed issues are title-only stubs with unrelated personal-name/date strings and no bug or feature description. They do not appear to be duplicates or part of the same underlying problem.", + "confidence": 0.98, + "canonical_issue_reason": "No clear canonical issue: neither item describes an actionable problem, and the titles do not indicate the same subject matter beyond both being low-signal stubs.", "canonical_pr_reason": null, - "best_issue_reason": null, + "best_issue_reason": "If a representative must be chosen, issue 45764 is only marginally preferable because it was created slightly earlier, but it is not a meaningful canonical target.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:45764", "right": "issue:45765", "accept": false, - "reason": "Reject: the titles look like unrelated personal records, and there is no issue body or shared context indicating the same bug or request." + "reason": "Reject: the titles are just different names/dates with no shared bug, code path, or explicit linkage; they do not look like the same underlying issue." } ] }, @@ -20638,7 +20719,7 @@ "soft_pairs": [ "issue:45810|issue:46295" ], - "prepared_review_unit_hash": "bf7b98e6d65741e01c1fc5c0024ae56a4687e07d8934eded6f4f56c07898eb5d", + "prepared_review_unit_hash": "aa54513823ba01c4bf26b057029044fc24f3df324bc0923041cf039da12e85e1", "component_budget": { "node_count": 2, "item_count": 2, @@ -20660,18 +20741,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "The cluster contains two unrelated open issues: one requests a new Qwen3_5 token-classification/value-model class, and the other requests FP8 training support. They do not describe the same bug or change.", + "summary": "These two issues are unrelated: one requests a new Qwen3_5 token-classification/value-model class, while the other asks for FP8 training support. No duplicate triage relationship is supported.", "confidence": 0.97, - "canonical_issue_reason": null, + "canonical_issue_reason": "No clear canonical issue for this cluster; the reports describe different features in different areas of the stack.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 45810 is the more concrete, narrowly scoped feature request; issue 46295 is broader and about a separate training capability.", + "best_issue_reason": "Neither issue is a good cluster representative because they are not duplicates of each other.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:45810", "right": "issue:46295", "accept": false, - "reason": "Different feature areas and code paths: model class addition for token classification/value modeling vs FP8 training support. They are not the same underlying request." + "reason": "Different underlying requests: model/class support for token classification versus FP8 training support. They do not describe the same bug or change." } ] }, @@ -20711,12 +20792,12 @@ "issue:45871|issue:45873", "issue:45872|issue:45873" ], - "prepared_review_unit_hash": "cce65d871d448bd92f6568ab9c585da647ef9c5203d6a0c76d47fdb09436f593", + "prepared_review_unit_hash": "26f6de163a91cb4a8c33b7aca5fa45bcaad73bf9990f4529dac448291feebddc", "component_budget": { "node_count": 5, "item_count": 5, "soft_pair_count": 8, - "serialized_chars": 5725, + "serialized_chars": 5727, "estimated_input_tokens": 1432, "estimated_eval_tokens": 3120 }, @@ -20724,7 +20805,7 @@ "node_count": 5, "item_count": 5, "soft_pair_count": 8, - "serialized_chars": 5725, + "serialized_chars": 5727, "estimated_input_tokens": 1432, "estimated_eval_tokens": 3120 }, @@ -20733,66 +20814,66 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "There are two clear duplicate issue pairs: 45870/45871 and 45872/45873. Issue 45869 is related to the same financial-data theme but describes a different dataset/sample request, not the same underlying task.", - "confidence": 0.94, - "canonical_issue_reason": "Issue 45870 is a strong representative for its duplicate pair because it matches 45871 essentially exactly and has slightly more discussion activity, making it the cleanest anchor among the issues shown.", + "summary": "This cluster contains two separate exact-duplicate issue pairs: 45870/45871 and 45872/45873. Issue 45869 is only loosely related by financial-data theme, but it describes a different dataset/sample offering rather than the same request or bug.", + "confidence": 0.95, + "canonical_issue_reason": "45870 is the earliest and clearest representative of the exact-duplicate pair 45870/45871, with the same title and apparent content; 45871 is redundant.", "canonical_pr_reason": null, - "best_issue_reason": "45870 is the best single issue to keep as the representative example of the duplicate pair 'Example: Financial time series prediction with DataForge.' The other identical pair (45872/45873) appears to be a separate issue cluster, so no one issue spans everything here.", + "best_issue_reason": "45870 is the best overall issue representative because it is a clean exact-duplicate anchor with the earliest timestamp and a concrete, self-contained topic. The other issues split into a separate duplicate pair or are only thematically related.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:45869", "right": "issue:45870", "accept": false, - "reason": "Related financial-data wording, but one is a dataset/sample offer and the other is an example request; not the same underlying issue." + "reason": "Related financial-data theme, but one is a dataset/sample listing and the other is an example prediction use case; not the same underlying issue." }, { "left": "issue:45869", "right": "issue:45871", "accept": false, - "reason": "Related financial-data wording, but one is a dataset/sample offer and the other is an example request; not the same underlying issue." + "reason": "Same as 45869/45870: broad theme overlap only, not the same request or change." }, { "left": "issue:45870", "right": "issue:45871", "accept": true, - "reason": "Same title and effectively the same issue content; this is a duplicate." + "reason": "Exact duplicate titles and highly similar content; same example issue, with 45870 as the earlier copy." }, { "left": "issue:45870", "right": "issue:45872", "accept": false, - "reason": "Both mention financial time series, but one is an example request and the other is about a scraping service; different deliverables." + "reason": "Both mention financial time series/DataForge, but one is an example issue and the other is a scraping-service topic; different underlying subject." }, { "left": "issue:45870", "right": "issue:45873", "accept": false, - "reason": "Both mention financial time series, but one is an example request and the other is about a scraping service; different deliverables." + "reason": "Different concrete topic from 45870; similar financial/DataForge wording is too broad to treat as a duplicate." }, { "left": "issue:45871", "right": "issue:45872", "accept": false, - "reason": "Same broad topic, but the requested work differs: example request versus scraping-service-oriented item." + "reason": "Different issue themes: example prediction use case versus scraping-service integration." }, { "left": "issue:45871", "right": "issue:45873", "accept": false, - "reason": "Same broad topic, but the requested work differs: example request versus scraping-service-oriented item." + "reason": "Different concrete subject; shared keywords are not enough to merge them." }, { "left": "issue:45872", "right": "issue:45873", "accept": true, - "reason": "Same title and effectively the same issue content; this is a duplicate." + "reason": "Exact duplicate titles and matching apparent content; same scraping-service issue, with 45872 as the earlier copy." } ] }, "evaluator_result": { "accept": true, - "feedback": "The summary is well grounded in the packet: 45870/45871 and 45872/45873 have identical titles and matching body lengths, so treating them as duplicate pairs is a conservative inference. The 45869 rejections are also reasonable because its title clearly describes a different dataset/sample offer rather than the example-request issues. No soft-edge verdict appears overstated." + "feedback": "The summary is grounded in the packet: 45870/45871 share an identical title and body length, and 45872/45873 do as well, while 45869 is only loosely thematically related by title. The canonical choice of the earliest duplicate in each exact pair is conservative, and the soft-edge rejections do not overclaim beyond the available evidence." }, "evaluator_used": true, "retried": false, @@ -20819,40 +20900,40 @@ "soft_pairs": [ "issue:46439|issue:46441" ], - "prepared_review_unit_hash": "dcabd2621324ed631d5adbcf56cf5d55227078f3a83a926b17e20627f4f74b19", + "prepared_review_unit_hash": "1e10978e612d0667297c0bd0b45091859ac8f4e6cb4ce33f8043c9b2ab0d3ecd", "component_budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 1636, - "estimated_input_tokens": 409, - "estimated_eval_tokens": 1074 + "serialized_chars": 1637, + "estimated_input_tokens": 410, + "estimated_eval_tokens": 1076 }, "budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 1636, - "estimated_input_tokens": 409, - "estimated_eval_tokens": 1074 + "serialized_chars": 1637, + "estimated_input_tokens": 410, + "estimated_eval_tokens": 1076 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These two issues both involve `cache_implementation=\"static\"` and `linear_attention`, but they describe different failure modes: shape corruption/runtime error vs `KeyError: 'linear_attention'`. They are related, but not clearly the same underlying bug, so I would not merge them as duplicates.", - "confidence": 0.67, - "canonical_issue_reason": "Issue 46439 is the better canonical because it states a clearer root cause (`early_initialization` corrupting linear-attention layers) and describes the broader failure mode.", + "summary": "Both issues point at static-cache failures around `linear_attention`, but they describe different concrete breakages: one is shape corruption during `early_initialization`, the other is a `KeyError` in Qwen3.5 with `cache_implementation=\"static\"`. They are related, but not clearly the same underlying bug.", + "confidence": 0.66, + "canonical_issue_reason": "Issue 46439 is the better canonical issue because it states the likely root cause more directly (`early_initialization` corrupting linear-attention layers) and is broader than a model-specific crash report.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 46439 is the strongest representative for the cluster since it is more diagnostic and likely explains the underlying mechanism behind the static-cache failure.", + "best_issue_reason": "Issue 46439 is the strongest representative for the cluster since it captures the underlying static-cache/linear-attention failure mode rather than a single downstream symptom.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:46439", "right": "issue:46441", "accept": false, - "reason": "Both mention static cache and linear-attention, but one is a shape-corruption/runtime-error bug and the other is a `KeyError` in Qwen3.5; the evidence is too weak to treat them as the same concrete issue." + "reason": "Same broad area, but the reported failures differ materially: wrong-shape RuntimeError during early initialization vs. a `KeyError: 'linear_attention'` in Qwen3.5. Not enough evidence that they are the same bug or patch target." } ] }, @@ -20882,12 +20963,12 @@ "soft_pairs": [ "issue:46551|issue:46552" ], - "prepared_review_unit_hash": "d83e1597d54e34ba55cf1c9dc35878cb5c304c126811f543961b6be92fceb9a0", + "prepared_review_unit_hash": "13ad7501bc622533ac19da1691560dad2d8be28776d0d4634544ffa9ee4e80cd", "component_budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 1706, + "serialized_chars": 1707, "estimated_input_tokens": 427, "estimated_eval_tokens": 1110 }, @@ -20895,7 +20976,7 @@ "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 1706, + "serialized_chars": 1707, "estimated_input_tokens": 427, "estimated_eval_tokens": 1110 }, @@ -20904,24 +20985,24 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "The two issues are near-perfect duplicates: same title, same reported tokenizer bug, and the same underlying v5 Wav2Vec2CTCTokenizer special-token leakage behavior.", + "summary": "These two issues are effectively duplicates: they have the same title, same underlying Wav2Vec2CTCTokenizer v5 bug description, and near-identical evidence. The open issue is the better canonical tracker.", "confidence": 0.99, - "canonical_issue_reason": "issue #46552 is the better canonical issue because it is still open and already has inbound references, making it the active tracker for the bug.", + "canonical_issue_reason": "Issue #46552 is open and appears to be the live tracker for the exact same tokenizer bug, so it is the better canonical issue.", "canonical_pr_reason": null, - "best_issue_reason": "issue #46552 is the best issue to keep as the primary duplicate target: it is open, active, and identical in scope to #46551.", + "best_issue_reason": "#46552 is the best issue to keep because it is open, current, and describes the same concrete bug as #46551.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:46551", "right": "issue:46552", "accept": true, - "reason": "Same title and same bug description; both report the Wav2Vec2CTCTokenizer v5 special-word-delimiter leak into get_special_tokens_mask and convert_ids_to_tokens, so they are duplicates." + "reason": "Same title and same reported Wav2Vec2CTCTokenizer v5 behavior; these are the same underlying bug report." } ] }, "evaluator_result": { "accept": true, - "feedback": "The conclusion is grounded in the packet: the two issues have identical titles, equal body length, and max soft similarity (Jaccard 1.0). Choosing the open issue #46552 as the canonical tracker is conservative and supported by its inbound references. No overstatement detected." + "feedback": "The conclusion is well grounded: the two issues have identical titles, identical body length, and a soft-similarity score of 1.0, so treating them as duplicates and choosing the open issue as canonical is conservative and supported by the packet." }, "evaluator_used": true, "retried": false, @@ -20948,7 +21029,7 @@ "soft_pairs": [ "issue:46657|issue:46692" ], - "prepared_review_unit_hash": "a6c4b29da050503a8cf0dd6bd85498698d4602255bad8cf004f3c654df344267", + "prepared_review_unit_hash": "28ff32d86f1243ec80bdfb66a679d82433386e7bf3d8558ef54533424b3da81a", "component_budget": { "node_count": 2, "item_count": 2, @@ -20970,18 +21051,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These two issues are not the same underlying bug: one reports a tensor shape mismatch during object detection training, while the other reports an out-of-memory failure after several epochs on RT-DETR v2. The similarity is only superficial.", - "confidence": 0.92, - "canonical_issue_reason": "Issue 46657 is the better anchor because it describes a specific training-time tensor mismatch error, while 46692 is a different failure mode (OOM) and model-specific.", + "summary": "These two issues do not describe the same bug: one is a tensor shape mismatch during object-detection training, the other is a training-time out-of-memory problem in RT-DETR v2 after several epochs. The soft similarity is only at the broad \u201cobject detection training\u201d level, so it should not be merged.", + "confidence": 0.95, + "canonical_issue_reason": "Issue #46657 is the better canonical representative because it describes a concrete, specific failure mode (tensor size mismatch) rather than a broad resource symptom.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 46657 is the most representative issue in this pair since it captures a concrete code-path failure; 46692 does not look like a duplicate or a close variant.", + "best_issue_reason": "Issue #46657 is the most suitable issue to anchor on because it is more precise and actionable, with a clearer error signature than the OOM report.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:46657", "right": "issue:46692", "accept": false, - "reason": "Different symptoms and likely different root causes: tensor size mismatch vs. memory exhaustion after epochs, with one tied to RT-DETR v2 specifically." + "reason": "Different underlying problems: one is a shape mismatch error, the other is memory exhaustion during later epochs. They share the training context but not the same code-path bug." } ] }, @@ -21011,12 +21092,12 @@ "soft_pairs": [ "issue:46822|issue:46823" ], - "prepared_review_unit_hash": "a2e23d19728650d819dce5fc5268f5163be3b51926b2a978377e4c85e366afb6", + "prepared_review_unit_hash": "15a541f8b122cce2b4d36be096da720e8bdba98fe301b5f6f7d7749d4cac7186", "component_budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 1671, + "serialized_chars": 1672, "estimated_input_tokens": 418, "estimated_eval_tokens": 1092 }, @@ -21024,7 +21105,7 @@ "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 1671, + "serialized_chars": 1672, "estimated_input_tokens": 418, "estimated_eval_tokens": 1092 }, @@ -21033,18 +21114,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These two issues are not duplicates: one is about ZeRO-3 zero.Init failing to partition a composite language submodule and causing OOM during multi-GPU load, while the other is about 4-bit loading with device_map='auto' and max_memory incorrectly dispatching to CPU and raising ValueError. Different code paths, symptoms, and fixes.", - "confidence": 0.96, - "canonical_issue_reason": "issue:46822 is the better canonical issue because it has more discussion/inbound references and a more specific, actionable reproduction of the load-time partitioning failure.", + "summary": "These are two separate model-loading bugs, not duplicates: one is a ZeRO-3/zero.Init partitioning failure causing OOM, the other is a 4-bit device_map/max_memory dispatch error causing CPU placement and ValueError.", + "confidence": 0.97, + "canonical_issue_reason": "issue:46822 is the better canonical representative because it has the richer reproduction context and more discussion, and it is the more specific ZeRO-3/zero.Init bug in this pair.", "canonical_pr_reason": null, - "best_issue_reason": "issue:46822 is the strongest representative for the cluster since it has the clearest bug description and more evidence of impact.", + "best_issue_reason": "issue:46822 is the most representative issue in the cluster, but only because it is the more detailed report; it should not be treated as the same bug as issue:46823.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:46822", "right": "issue:46823", "accept": false, - "reason": "Different underlying bugs: ZeRO-3 partitioning/OOM versus 4-bit device_map/max_memory CPU dispatch. They do not look like the same fixable code-path problem." + "reason": "Reject: these fail in different code paths and with different mechanisms. 46822 is about ZeRO-3 zero.Init not partitioning a composite submodule and OOMing on load; 46823 is about 4-bit load_in_4bit with device_map='auto' and max_memory still dispatching to CPU and raising ValueError." } ] }, @@ -21074,12 +21155,12 @@ "soft_pairs": [ "pull_request:40546|pull_request:42781" ], - "prepared_review_unit_hash": "5ad765a1c0391ab50702fcba2217ed907621e504374a5eb627aa7eab2da4c45b", + "prepared_review_unit_hash": "54467223b2ac28cf6ba9bd41777c79f701d7688ba4fbdcde66ed1bf62e9ee252", "component_budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 7223, + "serialized_chars": 7224, "estimated_input_tokens": 1806, "estimated_eval_tokens": 3868 }, @@ -21087,7 +21168,7 @@ "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 7223, + "serialized_chars": 7224, "estimated_input_tokens": 1806, "estimated_eval_tokens": 3868 }, @@ -21096,18 +21177,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These PRs share a VibeVoice codebase, but they are not the same change: 40546 is the base VibeVoice implementation, while 42781 is a draft for VibeVoice Realtime plus acoustic tokenizer/generation updates.", - "confidence": 0.93, + "summary": "These are related VibeVoice PRs, but they do not look like duplicates: one is the main VibeVoice implementation, the other adds VibeVoice Realtime plus a separate acoustic tokenizer and generation changes. Overlap is mostly in shared model/docs scaffolding, not the same underlying change.", + "confidence": 0.91, "canonical_issue_reason": null, - "canonical_pr_reason": "40546 is the cleaner canonical PR for the core VibeVoice implementation; 42781 is a separate realtime/tokenizer follow-up rather than a duplicate of the same fix/change.", + "canonical_pr_reason": "PR 40546 is the main, more established VibeVoice implementation and is better suited as the canonical representative of the feature work; PR 42781 is a narrower realtime extension in draft state.", "best_issue_reason": null, - "best_pr_reason": "40546 is the best representative PR in this cluster because it is the main, non-draft base implementation and has the broader completed core integration work.", + "best_pr_reason": "PR 40546 is the best overall PR candidate because it appears to be the primary implementation with substantial review activity and broader feature coverage, whereas PR 42781 is an add-on realtime follow-up.", "soft_edge_verdicts": [ { "left": "pull_request:40546", "right": "pull_request:42781", "accept": false, - "reason": "Reject: they target related but different deliverables. 40546 implements VibeVoice itself; 42781 adds VibeVoice Realtime and acoustic tokenizer changes, so they are not the same underlying change and would not plausibly merge into one PR." + "reason": "Reject: despite shared VibeVoice filenames, 42781 adds realtime support and a new acoustic tokenizer with additional generation/modeling changes. This is an extension to the same family, not the same concrete bug or one mergeable fix." } ] }, @@ -21137,7 +21218,7 @@ "soft_pairs": [ "pull_request:40755|pull_request:41224" ], - "prepared_review_unit_hash": "3a203fd41beeb3bdaf3a7d0c56d0e2750a64966a106926932545612d4511bbc9", + "prepared_review_unit_hash": "5a18f7cf0d18702711ecf80318d296faae214f3ba27955c5b77c2d471c919a23", "component_budget": { "node_count": 2, "item_count": 2, @@ -21159,18 +21240,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "The two pull requests are unrelated feature additions in different model families, so they should not be clustered as duplicates or merged together.", - "confidence": 0.99, + "summary": "These two PRs are unrelated feature additions: one adds TimesFM covariate forecasting, the other adds DINOv3 image-classification support. They do not describe the same bug/change, so the soft duplicate edge should be rejected.", + "confidence": 0.98, "canonical_issue_reason": null, - "canonical_pr_reason": "No canonical PR: one adds TimesFM covariate forecasting support, the other adds DINOv3 image classification support; they address different code paths and use cases.", + "canonical_pr_reason": "No single canonical PR for a duplicate cluster here; if forced to pick a representative, PR 40755 is the more self-contained and substantial change set, while PR 41224 is a separate model feature.", "best_issue_reason": null, - "best_pr_reason": "Neither PR is a strong canonical representative for the other because they are unrelated changes in different subsystems.", + "best_pr_reason": "PR 40755 is the better cluster representative only by breadth and coherence: it implements one feature across code/docs/tests. PR 41224 is an entirely different model capability and does not fit the same underlying change.", "soft_edge_verdicts": [ { "left": "pull_request:40755", "right": "pull_request:41224", "accept": false, - "reason": "Reject: TimesFM covariate forecasting and DINOv3 image classification are different model features with no shared underlying bug or concrete code-path fix." + "reason": "Different models and code paths: TimesFM covariate forecasting vs. DINOv3 image-classification support. Shared \"model support\" theme is too broad to indicate the same underlying change." } ] }, @@ -21301,214 +21382,196 @@ "pull_request:43823|pull_request:46911", "pull_request:43251|pull_request:43254" ], - "prepared_review_unit_hash": "388784c572a56acfa57ee0c07c735b6cad3646e589fddbdab36ef962ce83fc72", + "prepared_review_unit_hash": "abde52afea45308b81feceace91f09d88d8d1da1054a09ee42310decaf762274", "component_budget": { "node_count": 15, "item_count": 15, "soft_pair_count": 26, - "serialized_chars": 42244, - "estimated_input_tokens": 10561, - "estimated_eval_tokens": 21378 + "serialized_chars": 42250, + "estimated_input_tokens": 10563, + "estimated_eval_tokens": 21382 }, "budget": { "node_count": 15, "item_count": 15, "soft_pair_count": 26, - "serialized_chars": 42244, - "estimated_input_tokens": 10561, - "estimated_eval_tokens": 21378 + "serialized_chars": 42250, + "estimated_input_tokens": 10563, + "estimated_eval_tokens": 21382 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "This cluster is mostly unrelated PRs that only share broad subsystem similarity. The only clear duplicate pair is the two `fixed_cross_entropy` changes, which both address passing supported kwargs through to `nn.functional.cross_entropy`.", - "confidence": 0.92, + "summary": "These PRs are overwhelmingly unrelated and should not be clustered as duplicates. The only true overlap is the pair of fixed_cross_entropy changes (43251 and 43254), which both address the same cross-entropy kwargs bug; the rest are distinct features, refactors, or tests in different areas.", + "confidence": 0.98, "canonical_issue_reason": null, - "canonical_pr_reason": "`pull_request:43251` is the strongest canonical PR: it has an explicit issue target (`43240`) and directly fixes the `fixed_cross_entropy` kwargs passthrough bug in the exact code path.", + "canonical_pr_reason": "No single canonical PR exists for the whole set because the items cover unrelated changes. If one must be chosen as the representative duplicate-like change, 43251 is the better canonical PR: it is the broader, issue-linked implementation of the same fixed_cross_entropy kwargs fix that 43254 narrows to.", "best_issue_reason": null, - "best_pr_reason": "`pull_request:43251` is the best representative PR because it most clearly captures the underlying fix, is explicitly issue-linked, and overlaps substantively with the near-duplicate `43254`.", + "best_pr_reason": "43251 is the strongest standalone PR in this set because it targets a concrete bug, is explicitly linked to an issue, and subsumes the same underlying fix addressed more narrowly by 43254.", "soft_edge_verdicts": [ { "left": "pull_request:41855", "right": "pull_request:43823", "accept": false, - "reason": "Different model areas and different changes: Mistral tokenizer methods vs adding MobileLLM. Shared tokenizer/model work is too broad to be the same change." + "reason": "Unrelated changes: Mistral tokenizer API additions vs adding a new MobileLLM model." }, { "left": "pull_request:41855", "right": "pull_request:43424", "accept": false, - "reason": "Tokenizer API addition vs an Executorch export test. No shared concrete bug or mergeable code path." + "reason": "Different code paths and goals: tokenizer methods vs Executorch dynamic-shape export tests." }, { "left": "pull_request:42668", "right": "pull_request:43823", "accept": false, - "reason": "Both concern model/processor registration, but they are for different models and different fixes. Too broad to be duplicates." + "reason": "Processor auto-loading robustness is unrelated to MobileLLM model addition." }, { "left": "pull_request:43251", "right": "pull_request:43823", "accept": false, - "reason": "`fixed_cross_entropy` kwargs passthrough is unrelated to adding a new model implementation." + "reason": "Cross-entropy kwargs fix is unrelated to MobileLLM model scaffolding." }, { "left": "pull_request:43251", "right": "pull_request:44827", "accept": false, - "reason": "Loss-function kwargs support and Mistral4 test/model changes are unrelated code paths." + "reason": "Same broad loss/modeling repo area, but different problems: loss helper kwargs vs Mistral4 test fixes." }, { "left": "pull_request:43424", "right": "pull_request:46911", "accept": false, - "reason": "Executorch export dynamic-shape testing is unrelated to Olmo3 RoPE configuration changes." + "reason": "Executorch dynamic export testing is unrelated to Olmo3 RoPE-per-layer behavior." }, { "left": "pull_request:43424", "right": "pull_request:46707", "accept": false, - "reason": "Exportability tests and FSDP2 migration are different concerns and do not fix the same bug." + "reason": "Executorch export tests and native FSDP2 migration are distinct changes." }, { "left": "pull_request:43424", "right": "pull_request:44827", "accept": false, - "reason": "Dynamic export testing does not match Mistral4 test/model changes; only both are test-adjacent." + "reason": "Different concrete issues: export dynamic shapes vs Mistral4 test fixes." }, { "left": "pull_request:43424", "right": "pull_request:43823", "accept": false, - "reason": "Different model areas and different purposes; no shared underlying bug/change." + "reason": "No shared bug or change: export testing vs adding MobileLLM." }, { "left": "pull_request:43823", "right": "pull_request:45655", "accept": false, - "reason": "Adding MobileLLM and fixing `cls.config` resolution are unrelated changes." - }, - { - "left": "pull_request:43823", - "right": "pull_request:43995", - "accept": false, - "reason": "New model addition vs Falcon output-interface refactor; not the same concrete problem." - }, - { - "left": "pull_request:43995", - "right": "pull_request:45213", - "accept": false, - "reason": "Falcon refactor and model-creation skill are completely different changes." - }, - { - "left": "pull_request:43995", - "right": "pull_request:44827", - "accept": false, - "reason": "Falcon output-collection refactor and Mistral4 fixes are unrelated model-specific changes." + "reason": "New model addition and config-resolution fix are unrelated." }, { - "left": "pull_request:44827", - "right": "pull_request:46911", + "left": "pull_request:43823", + "right": "pull_request:43995", "accept": false, - "reason": "Mistral4 test fixes and Olmo3 per-layer RoPE support are distinct model behaviors." + "reason": "MobileLLM model files vs Falcon output-collection refactor are different changes." }, { - "left": "pull_request:44827", - "right": "pull_request:46707", + "left": "pull_request:43823", + "right": "pull_request:45213", "accept": false, - "reason": "Mistral4 test/model work is unrelated to FSDP2 migration." + "reason": "Adding MobileLLM is not the same as adding the model-creation skill tooling." }, { - "left": "pull_request:44827", - "right": "pull_request:45170", + "left": "pull_request:43823", + "right": "pull_request:46707", "accept": false, - "reason": "Mistral4 test fixes and `layrnorm` typo cleanup span different model families and different issues." + "reason": "Model addition vs FSDP2 module migration are unrelated." }, { - "left": "pull_request:44827", - "right": "pull_request:46267", + "left": "pull_request:43823", + "right": "pull_request:46911", "accept": false, - "reason": "Mistral4-specific work is not the same as updating serve chat parsing." + "reason": "MobileLLM model addition and Olmo3 RoPE changes are unrelated." }, { - "left": "pull_request:44827", - "right": "pull_request:46911", + "left": "pull_request:43995", + "right": "pull_request:44827", "accept": false, - "reason": "Mistral4 test/model changes and Olmo3 RoPE logic are unrelated." + "reason": "Falcon output-interface refactor is not the same as Mistral4 test fixes." }, { - "left": "pull_request:45170", - "right": "pull_request:46707", + "left": "pull_request:43995", + "right": "pull_request:45213", "accept": false, - "reason": "A weight-name typo fix for CLIP-like models is not the same as native FSDP2 migration." + "reason": "Falcon refactor and model-creation skill are entirely different concerns." }, { - "left": "pull_request:46048", - "right": "pull_request:46707", + "left": "pull_request:44827", + "right": "pull_request:45170", "accept": false, - "reason": "TDT loss kernel integration and FSDP2 module migration are unrelated code paths." + "reason": "Both touch model code, but one fixes Mistral4 tests while the other corrects a CLIP-family spelling rename; not the same bug." }, { "left": "pull_request:44827", - "right": "pull_request:46048", + "right": "pull_request:45213", "accept": false, - "reason": "Mistral4 fixes and TDT loss kernel work do not address the same underlying problem." + "reason": "Mistral4 tests are unrelated to the model-creation skill PR." }, { "left": "pull_request:44827", - "right": "pull_request:45213", + "right": "pull_request:45655", "accept": false, - "reason": "Model tests/fixes vs model-creation skill scaffolding are not the same change." + "reason": "Different issues: Mistral4 test fixes vs config-resolution ordering." }, { - "left": "pull_request:43823", - "right": "pull_request:45213", + "left": "pull_request:44827", + "right": "pull_request:46048", "accept": false, - "reason": "Adding MobileLLM is unrelated to the model-creation skill content." + "reason": "Mistral4 tests are unrelated to adding the TDT loss kernel." }, { "left": "pull_request:44827", - "right": "pull_request:45655", + "right": "pull_request:46267", "accept": false, - "reason": "Mistral4 fixes and `cls.config` resolution order are unrelated." + "reason": "Different subsystems and goals: model tests vs chat parsing updates." }, { - "left": "pull_request:46048", + "left": "pull_request:44827", "right": "pull_request:46707", "accept": false, - "reason": "Kernel-loading code for TDT loss and FSDP2 migration are different bugs/features." + "reason": "Mistral4 test fixes and FSDP2 migration are unrelated." }, { "left": "pull_request:44827", - "right": "pull_request:46048", + "right": "pull_request:46911", "accept": false, - "reason": "Mistral4 test fixes and TDT loss kernel integration are unrelated." + "reason": "Mistral4 test fixes do not overlap with Olmo3 RoPE changes." }, { - "left": "pull_request:43823", + "left": "pull_request:45170", "right": "pull_request:46707", "accept": false, - "reason": "Adding MobileLLM and FSDP2 migration are separate broad features." + "reason": "Typos in CLIP-like weight renaming are unrelated to native FSDP2 migration." }, { - "left": "pull_request:43823", - "right": "pull_request:46911", + "left": "pull_request:46048", + "right": "pull_request:46707", "accept": false, - "reason": "MobileLLM addition and Olmo3 RoPE changes are unrelated model-specific work." + "reason": "TDT loss kernel support and FSDP2 migration are separate feature additions." }, { "left": "pull_request:43251", "right": "pull_request:43254", "accept": true, - "reason": "Both PRs fix the same `fixed_cross_entropy` bug: allow supported kwargs to reach `nn.functional.cross_entropy` in `loss_utils.py`. The implementations differ slightly, but the underlying code-path fix is the same." + "reason": "Same underlying bug: both fix fixed_cross_entropy to pass supported kwargs through to nn.functional.cross_entropy, and they could plausibly be merged into one PR." } ] }, "evaluator_result": { "accept": true, - "feedback": "The summary is grounded in the packet and stays conservative about the soft edges. The only accepted pair, `43251`/`43254`, does look mergeable into one PR for the same `fixed_cross_entropy` kwargs passthrough fix in `loss_utils.py`; the rest are reasonably treated as unrelated or only broadly similar." + "feedback": "The summary is grounded in the packet and stays conservative. The only supported overlap is 43251/43254, which both touch the same fixed_cross_entropy kwargs bug and look mergeable into one bugfix PR. The rest of the pairwise rejections are consistent with the diffs and titles, and I don\u2019t see any overstatement of evidence." }, "evaluator_used": true, "retried": false, @@ -21535,12 +21598,12 @@ "soft_pairs": [ "pull_request:42975|pull_request:42976" ], - "prepared_review_unit_hash": "243670c5bd8647dc3ac759ad3a2e6cd648f862498c8c676f884a0d2aa5c59315", + "prepared_review_unit_hash": "90def3c23cf3a673dbce63eee60f32c958723072c9d9e4bb028dd7dacf086a1b", "component_budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 5935, + "serialized_chars": 5934, "estimated_input_tokens": 1484, "estimated_eval_tokens": 3224 }, @@ -21548,7 +21611,7 @@ "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 5935, + "serialized_chars": 5934, "estimated_input_tokens": 1484, "estimated_eval_tokens": 3224 }, @@ -21557,18 +21620,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Both items are broad GitHub Actions maintenance PRs, but they address different upgrade sets: one targets Node 24-compatible action versions, the other bumps docker/build-push-action across Docker image workflows. They overlap in some workflow files, yet they do not look like the same concrete fix and would not naturally collapse into one PR.", - "confidence": 0.89, + "summary": "Two workflow-maintenance PRs overlap on GitHub Actions files, but they are different upgrade efforts: one is specifically about Node 24 compatibility for core actions, while the other updates Docker/build-related actions to newer versions. They are related in area, not the same underlying change.", + "confidence": 0.96, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 42975 is the better canonical PR because its title and diff are more specific about the actual maintenance goal (Node 24 compatibility) and it touches a wider set of workflow action upgrades.", + "canonical_pr_reason": "PR 42975 is the better canonical representative because it is the more broadly scoped and clearly motivated GitHub Actions compatibility upgrade, touching many of the shared workflow files in this cluster.", "best_issue_reason": null, - "best_pr_reason": "PR 42975 is the strongest representative of the cluster since it is the clearer umbrella workflow-upgrade change; PR 42976 is a separate, narrower docker action version bump.", + "best_pr_reason": "PR 42975 is the best single representative of the cluster: it covers the common workflow-action upgrade theme more comprehensively and has the clearest stated purpose.", "soft_edge_verdicts": [ { "left": "pull_request:42975", "right": "pull_request:42976", "accept": false, - "reason": "Reject: both are workflow dependency upgrade PRs, but they change different actions and target different problems (Node 24 compatibility vs docker/build-push-action v6). Shared workflow filenames alone is not enough to treat them as the same underlying change." + "reason": "They both update workflow dependencies, but they target different action families and different concrete goals (Node 24 compatibility vs latest Docker/build action versions). This is not the same underlying change and would not plausibly be merged into one PR." } ] }, @@ -21609,82 +21672,82 @@ "pull_request:45490|pull_request:46562", "pull_request:46562|pull_request:46658" ], - "prepared_review_unit_hash": "f5fde6bdfe22b8bb531773315ebbeba5dc4dbc687338757974577b96acb66cef", + "prepared_review_unit_hash": "52437740d2b97d939f913d551bbce2952845ce1cfb0c58bdf77bf7b6294b96ac", "component_budget": { "node_count": 6, "item_count": 6, "soft_pair_count": 8, - "serialized_chars": 21198, - "estimated_input_tokens": 5300, - "estimated_eval_tokens": 10856 + "serialized_chars": 21192, + "estimated_input_tokens": 5298, + "estimated_eval_tokens": 10852 }, "budget": { "node_count": 6, "item_count": 6, "soft_pair_count": 8, - "serialized_chars": 21198, - "estimated_input_tokens": 5300, - "estimated_eval_tokens": 10856 + "serialized_chars": 21192, + "estimated_input_tokens": 5298, + "estimated_eval_tokens": 10852 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These PRs are all separate model-addition or model-support efforts. They share broad Transformers/model-doc infrastructure, but the only close pair (Step3p7 vs. Step3.7 flash scripts) still targets different codepaths and artifacts, so none should be merged as duplicates.", - "confidence": 0.92, + "summary": "These are mostly unrelated model-introduction PRs that only overlap on shared boilerplate files. The lone high-similarity pair (Step 3.7 Flash vs a mini-model conversion/debug script) is not a duplicate and should not be merged.", + "confidence": 0.93, "canonical_issue_reason": null, - "canonical_pr_reason": "PR #46562 is the strongest representative of the cluster: it is a full core-library model addition with source code, auto mappings, docs, and tests, whereas the others are either different model families or auxiliary/script-only work.", + "canonical_pr_reason": "PR 46562 is the best representative of the cluster: it is the substantive library integration for Step 3.7 Flash, while the other Step 3.7 item is a standalone script/debug artifact rather than the same change.", "best_issue_reason": null, - "best_pr_reason": "PR #46562 is the best standalone PR in this set because it contains the most complete end-to-end model integration and the clearest core code-path change.", + "best_pr_reason": "PR 46562 is the strongest single PR in the set because it contains the actual model implementation, docs, auto mappings, and tests; the others are separate model-addition PRs or auxiliary scripts.", "soft_edge_verdicts": [ { "left": "pull_request:42978", "right": "pull_request:43448", "accept": false, - "reason": "Different model families (ViT NEPA vs Molmo) with separate model code; shared docs/auto-registry files are just standard integration plumbing." + "reason": "Different model families (ViT NEPA vs Molmo); shared docs/auto-registration files are boilerplate and do not indicate the same change." }, { "left": "pull_request:42978", "right": "pull_request:43451", "accept": false, - "reason": "ViT NEPA and Molmo2 are distinct model additions with different implementations; overlap is only in shared registration/docs scaffolding." + "reason": "Different model families (ViT NEPA vs Molmo2) with distinct codepaths and test coverage; overlap is only in shared integration files." }, { "left": "pull_request:42978", "right": "pull_request:45490", "accept": false, - "reason": "Completely different model additions (ViT NEPA vs CTSM); same repository plumbing does not indicate the same bug/change." + "reason": "ViT NEPA and CTSM are unrelated model additions; same broad repo scaffolding, not the same bug or feature." }, { "left": "pull_request:43448", "right": "pull_request:43451", "accept": false, - "reason": "Molmo and Molmo2 are related by name only; they add different model families and different codepaths." + "reason": "Molmo and Molmo2 are separate model integrations with different modules, tests, and conversion logic; naming similarity is not enough." }, { "left": "pull_request:43448", "right": "pull_request:45490", "accept": false, - "reason": "Molmo and CTSM are unrelated model integrations; shared auto/docs files are not enough to treat them as duplicates." + "reason": "Molmo and CTSM target different architectures and codepaths; shared registry/docs files are incidental." }, { "left": "pull_request:43451", "right": "pull_request:45490", "accept": false, - "reason": "Molmo2 and CTSM are separate model additions with different modeling/processing code and different targets." + "reason": "Molmo2 and CTSM are distinct model implementations, not the same underlying fix or feature." }, { "left": "pull_request:45490", "right": "pull_request:46562", "accept": false, - "reason": "CTSM and Step3p7 are different model families; they only overlap in generic auto-mapping/docs updates." + "reason": "CTSM and Step 3.7 Flash are unrelated models; overlap is limited to auto-mapping/boilerplate updates." }, { "left": "pull_request:46562", "right": "pull_request:46658", "accept": false, - "reason": "Both mention Step3.7 Flash, but #46562 adds the core transformers model implementation while #46658 is a WIP scripts/mini-model conversion workspace; they are not the same concrete code-path change and would not naturally collapse into one PR." + "reason": "The Step 3.7 Flash PR adds the actual library model, while 46658 is a standalone mini-model/build/debug script for conversion work; they are not the same concrete code-path change and would not be merged as one PR." } ] }, @@ -21733,7 +21796,7 @@ "pull_request:46539|pull_request:46922", "pull_request:46886|pull_request:46922" ], - "prepared_review_unit_hash": "1f95e45eda1ee33c7e6333085dc4abadd17182238bdfc731dbfc73c31465ebd2", + "prepared_review_unit_hash": "3cb60424b34a3f1457eb1f7e196966860dc2960eb886b7669414316de01e2ad8", "component_budget": { "node_count": 7, "item_count": 7, @@ -21755,108 +21818,108 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Most items are unrelated PRs that only weakly resemble each other by touching tests or the same broad model code. The only true duplicate-looking pair is the YOLOS device-mismatch fixes: PR 46886 and PR 46922 both target issue 46869 and change the same code path in `modeling_yolos.py`.", - "confidence": 0.95, + "summary": "Only the two YOLOS pull requests look like the same underlying bug fix; the rest target unrelated models, CI, generation validation, or tests. 46886 is the stronger canonical PR because it has the explicit issue target and the clearer, more focused title.", + "confidence": 0.98, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 46922 is the best canonical PR for the duplicate YOLOS fix: it directly patches the failing forward path with `.to(hidden_states.device)` and also includes the split-module guard, making it the more complete fix for the same underlying device-mismatch bug.", + "canonical_pr_reason": "PR 46886 is the cleanest representative: it explicitly targets issue 46869, has a precise title about the YOLOS device_map='auto' mismatch, and the diff directly addresses that bug path.", "best_issue_reason": null, - "best_pr_reason": "PR 46922 is the strongest representative of the cluster\u2019s only real duplicate pair because it addresses the concrete YOLOS device placement failure, matches the explicit issue target 46869, and is more complete than PR 46886.", + "best_pr_reason": "PR 46886 is the best overall representative of the cluster because it is the most specific and complete fix for the only true duplicate pair in the set.", "soft_edge_verdicts": [ { "left": "pull_request:43291", "right": "pull_request:43665", "accept": false, - "reason": "Both touch tests, but they are for different models and different problems; 43665 is just placeholder failing assertions, not the same underlying bug as the Whisper tokenizer/test fix in 43291." + "reason": "Different models and different concerns: Whisper tokenizer/test cleanup versus CLIP/ViT test edits." }, { "left": "pull_request:43291", "right": "pull_request:44815", "accept": false, - "reason": "Whisper decoding changes and FP8 dequant/model-loading changes are unrelated code paths and unrelated bugs." + "reason": "Whisper decoding changes are unrelated to FP8 dequantization and model loading logic." }, { "left": "pull_request:43291", "right": "pull_request:46380", "accept": false, - "reason": "Whisper test/model code and generation watermarking validation are unrelated; no shared concrete bug or change." + "reason": "Whisper tests/tokenization versus generation watermarking validation; different code paths and bugs." }, { "left": "pull_request:43291", "right": "pull_request:46922", "accept": false, - "reason": "Whisper tokenizer/test fixes do not match the YOLOS device-mismatch bug; same PR category is not enough to treat them as duplicates." + "reason": "Whisper tokenizer/test work is unrelated to the YOLOS device-mismatch fix." }, { "left": "pull_request:43665", "right": "pull_request:46539", "accept": false, - "reason": "Different subsystems and goals: 43665 is test noise across CLIP/ViT, while 46539 is a Phimoe CI integration fix." + "reason": "Model test assertions in CLIP/ViT are unrelated to Phimoe CI integration test infrastructure." }, { "left": "pull_request:43665", "right": "pull_request:44815", "accept": false, - "reason": "No common bug: placeholder test failures are unrelated to FP8 dequant/model loading." + "reason": "Different subsystems and failure modes: vision-model tests versus FP8 dequant/model-loading changes." }, { "left": "pull_request:43665", "right": "pull_request:46380", "accept": false, - "reason": "Test-only edits in unrelated model suites are not the same as a generation-config validation bug." + "reason": "Both touch tests, but they are for different bugs and different models; not the same change." }, { "left": "pull_request:43665", "right": "pull_request:46922", "accept": false, - "reason": "Both touch model tests/code only superficially; the underlying issues are unrelated (placeholder test failures vs YOLOS device mismatch)." + "reason": "CLIP/ViT test edits do not match the YOLOS runtime device-placement bug." }, { "left": "pull_request:44815", "right": "pull_request:46539", "accept": false, - "reason": "FP8 dequant/model-loading changes are unrelated to CI runner-memory handling for Phimoe integration tests." + "reason": "FP8 quantization/dequantization fixes are unrelated to CI memory/device-map setup." }, { "left": "pull_request:44815", "right": "pull_request:46380", "accept": false, - "reason": "Different bugs and code paths: FP8 padding/dequant vs generation watermarking validation." + "reason": "Different core problems: FP8 dequantization versus generation-config validation." }, { "left": "pull_request:44815", "right": "pull_request:46922", "accept": false, - "reason": "They both touch model code, but they fix different problems in different models; not mergeable as one PR." + "reason": "No shared bug path; FP8 kernel/model-loading work is unrelated to YOLOS device mismatch." }, { "left": "pull_request:46380", "right": "pull_request:46539", "accept": false, - "reason": "Generation configuration validation and Phimoe CI/test environment fixes are unrelated; no shared code-path bug." + "reason": "Generation configuration validation is unrelated to Phimoe integration test CI behavior." }, { "left": "pull_request:46380", "right": "pull_request:46922", "accept": false, - "reason": "Watermarking-config validation is unrelated to YOLOS device placement; same broad repo area only." + "reason": "Different bug domains and code paths: generation config versus YOLOS forward/device placement." }, { "left": "pull_request:46539", "right": "pull_request:46922", "accept": false, - "reason": "CI memory patching for integration tests does not describe the same underlying bug as the YOLOS model device mismatch." + "reason": "CI instance-sharing memory capping does not overlap with the YOLOS device_map='auto' model bug." }, { "left": "pull_request:46886", "right": "pull_request:46922", "accept": true, - "reason": "Both PRs target the same YOLOS device-mismatch issue on the same file and same explicit issue 46869; 46922 is a fuller version of the same fix." + "reason": "Both PRs fix the same YOLOS device-mismatch issue under device_map='auto' in the same file and share the same issue target; they are plausible alternatives for one merged fix." } ] }, "evaluator_result": { "accept": true, - "feedback": "The summary is grounded in the packet and stays conservative on the soft edges. The only accepted duplicate pair, PR 46886 and PR 46922, is well supported by the shared YOLOS file and explicit issue 46869, and the note that 46922 is the fuller fix is consistent with the diff. The other pair rejections do not overstate similarity and are reasonably cautious." + "feedback": "The analysis is grounded in the packet: the only clearly related pair is PR 46886 and PR 46922, which both target the YOLOS device-map/device-mismatch bug in the same file and are plausibly mergeable into one fix. The rest of the pairwise rejections are conservative and consistent with the differing models, code paths, and issue contexts shown in the packet. The canonical PR choice is also supported by the explicit issue target and the more focused title." }, "evaluator_used": true, "retried": false, @@ -21886,12 +21949,12 @@ "pull_request:43488|pull_request:43532", "pull_request:43532|pull_request:46747" ], - "prepared_review_unit_hash": "575f519a53c995a84164884ecb79945c7c3456d126c8e7cb76b77701c47a7ea1", + "prepared_review_unit_hash": "b6b254c5cf42c063af6e9a0bf85f6e7d3aa8e2b8f74d64e8f7aa6ed5cea5c6c2", "component_budget": { "node_count": 3, "item_count": 3, "soft_pair_count": 3, - "serialized_chars": 6349, + "serialized_chars": 6350, "estimated_input_tokens": 1588, "estimated_eval_tokens": 3432 }, @@ -21899,7 +21962,7 @@ "node_count": 3, "item_count": 3, "soft_pair_count": 3, - "serialized_chars": 6349, + "serialized_chars": 6350, "estimated_input_tokens": 1588, "estimated_eval_tokens": 3432 }, @@ -21908,30 +21971,30 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "All three items are unrelated synthetic PRs (\u201cdon't merge\u201d checks) touching different code paths: packaging/versioning, CI/comment workflows, and a security-gate workflow. They share only superficial repo-bot/CI similarity, not the same underlying change.", - "confidence": 0.95, + "summary": "All three PRs are distinct \"don\u2019t merge\" test changes with only superficial overlap in CI/testing areas; they should not be deduplicated.", + "confidence": 0.88, "canonical_issue_reason": null, - "canonical_pr_reason": "None of these PRs looks like a duplicate of the others; if one must be chosen, PR #43488 is the most self-contained and clearly scoped to a single repo-bot/version-format test change, but it is not a good canonical for the cluster because the others are different workflow edits.", + "canonical_pr_reason": "PR 46747 is the narrowest and most self-contained change, but the three PRs are not duplicates of one another.", "best_issue_reason": null, - "best_pr_reason": "PR #46747 is the smallest and most focused change, but it is not a global best representative for the cluster since the three PRs do not fix the same concrete problem and should not be consolidated.", + "best_pr_reason": "46747 is the best representative of the cluster because it is the smallest, clearest workflow-only change.", "soft_edge_verdicts": [ { "left": "pull_request:43488", "right": "pull_request:46747", "accept": false, - "reason": "Both are synthetic workflow/repo-bot style PRs, but the actual changes are unrelated: 43488 tweaks setup/versioning and test dependencies, while 46747 switches a security-gate workflow trigger. Different code paths and different purposes." + "reason": "Different changes: 43488 touches packaging/dependency versioning and a test file, while 46747 only switches a workflow trigger for the security gate. Not the same underlying bug or fix." }, { "left": "pull_request:43488", "right": "pull_request:43532", "accept": false, - "reason": "43488 is a packaging/version-format test PR; 43532 rewrites several CI workflows and test utilities. They only overlap on a shared test file, not on the same underlying bug or change." + "reason": "Unrelated edits: 43488 is setup/dependency/test scaffolding, while 43532 rewires several CI workflows and helper code. Shared Vit test file is not enough to make them the same change." }, { "left": "pull_request:43532", "right": "pull_request:46747", "accept": false, - "reason": "Both touch CI/workflow files, but they address different workflow behaviors in different files and are not plausibly one merged PR for the same fix." + "reason": "Both are workflow-related, but they fix different code paths: 43532 changes comment/scheduled/slack CI plumbing, while 46747 only changes the security-gate workflow trigger." } ] }, @@ -21961,7 +22024,7 @@ "soft_pairs": [ "pull_request:43612|pull_request:43613" ], - "prepared_review_unit_hash": "e12130a12d4e1e89f81fe212de1ecf572f77ffcda333084b57b2b8f3afb5181c", + "prepared_review_unit_hash": "1e7b33a5c2d2c2d4c38d9e46f7caf25093af41176d7f68c0dbd762e58ee133a6", "component_budget": { "node_count": 2, "item_count": 2, @@ -21983,18 +22046,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These are two separate PRs adding different segmentation pipelines: one for promptable concept segmentation and one for promptable visual segmentation. They share scaffolding and docs updates, but the underlying models, filenames, and tests are different, so they should not be treated as duplicates.", - "confidence": 0.97, + "summary": "These are two related but distinct PRs: one adds a Promptable Concept Segmentation pipeline, the other adds a Promptable Visual Segmentation pipeline. They share scaffolding/docs, but target different tasks and model integrations, so they are not duplicates.", + "confidence": 0.98, "canonical_issue_reason": null, - "canonical_pr_reason": "No single canonical PR: 43612 and 43613 implement different pipeline features and touch different model/test code paths.", + "canonical_pr_reason": null, "best_issue_reason": null, - "best_pr_reason": "If forced to pick one representative, 43613 is the broader PR, but it is still a distinct feature from 43612 and not a duplicate.", + "best_pr_reason": "PR 43613 is the broader, more representative pipeline addition, with more model integrations and tests; however, it is still a separate change from 43612.", "soft_edge_verdicts": [ { - "left": "pull_request:43612", - "right": "pull_request:43613", + "left": "Add Promptable Concept Segmentation pipeline", + "right": "Add Promptable Visual Segmentation pipeline", "accept": false, - "reason": "Reject: they add different pipelines with different underlying functionality and model integrations (concept segmentation vs visual segmentation). Shared docs/registry changes are not enough to merge them as one change." + "reason": "Different segmentation tasks and different implementation/test surfaces; shared docs and pipeline boilerplate are not enough to treat them as the same change." } ] }, @@ -22024,7 +22087,7 @@ "soft_pairs": [ "pull_request:43651|pull_request:43663" ], - "prepared_review_unit_hash": "08ee6ab859b6703008923f3789fc249439e908ffe9d98da20c7d51e22727cddf", + "prepared_review_unit_hash": "1d0cca5ed35ad72887c3e8bdf665f24ef757a0f4030598386502040ff5e008c6", "component_budget": { "node_count": 2, "item_count": 2, @@ -22046,18 +22109,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These are two unrelated trainer.py pull requests: one changes gradient-accumulation loss scaling behavior, the other refactors signature-column filtering for custom trainers. They are not duplicates and should stay separate.", - "confidence": 0.97, + "summary": "The two pull requests both touch `trainer.py` but address different trainer extension points: one changes gradient-accumulation loss scaling, the other refactors signature-column filtering for custom trainers/PEFT models. They are not duplicates of the same underlying change.", + "confidence": 0.96, "canonical_issue_reason": null, "canonical_pr_reason": null, "best_issue_reason": null, - "best_pr_reason": "PR #43651 is the more substantive standalone change because it alters training_step behavior and adds a new control point for custom trainers.", + "best_pr_reason": "PR #43651 is the more concrete fix to a specific training-path bug (loss scaling during gradient accumulation), while PR #43663 is a broader refactor/override hook for column filtering.", "soft_edge_verdicts": [ { "left": "pull_request:43651", "right": "pull_request:43663", "accept": false, - "reason": "Same file and similar extension-hook framing, but they address different code paths: GA loss scaling vs input column/signature filtering. They could not plausibly be merged as one PR." + "reason": "Same file, but different code paths and problems: one fixes loss normalization in `training_step`, the other adds `_get_signature_columns` for input column filtering. They are not the same underlying change and would not be merged into one PR." } ] }, @@ -22087,12 +22150,12 @@ "soft_pairs": [ "pull_request:43656|pull_request:43842" ], - "prepared_review_unit_hash": "0ca2f20b1fb671b0ac00897f54beee51f2ed6a4ce7e78058820a3554c0f0932a", + "prepared_review_unit_hash": "a0bedeeca104af34c1872f05231bb327abaead2218cd1b73e4ebeff2e77eea5d", "component_budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 3663, + "serialized_chars": 3664, "estimated_input_tokens": 916, "estimated_eval_tokens": 2088 }, @@ -22100,7 +22163,7 @@ "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 3663, + "serialized_chars": 3664, "estimated_input_tokens": 916, "estimated_eval_tokens": 2088 }, @@ -22109,18 +22172,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These are related by file and typing-related CLI fixes, but they address different NameError root causes in `src/transformers/cli/serve.py` and should not be deduplicated.", - "confidence": 0.87, + "summary": "Two open PRs in the same CLI file both touch TypeAdapter-related typing, but they appear to address different runtime/import issues rather than the same exact fix.", + "confidence": 0.77, "canonical_issue_reason": null, - "canonical_pr_reason": "43842 is the more self-contained fix: it directly prevents runtime evaluation of `TypeAdapter` when pydantic is absent, matching its issue target and minimal patch.", + "canonical_pr_reason": "PR #43842 is the more specific and direct fix: it targets the TypeAdapter NameError when pydantic is not installed, with a minimal TYPE_CHECKING import/quoted annotation change.", "best_issue_reason": null, - "best_pr_reason": "43842 is the best representative of the cluster because it has a clear, concrete failure mode, a tight change set, and an explicit issue link for the pydantic-missing path.", + "best_pr_reason": "PR #43842 is the best representative of the cluster because its scope is narrowly aligned to the concrete TypeAdapter failure mode, whereas #43656 is a broader typing/postponed-annotations cleanup.", "soft_edge_verdicts": [ { "left": "pull_request:43656", "right": "pull_request:43842", "accept": false, - "reason": "Both touch `serve.py` and typing hints, but 43656 appears to fix a different annotation/runtime typing problem than 43842\u2019s pydantic-absent `TypeAdapter` NameError; they are not the same underlying bug." + "reason": "Both modify src/transformers/cli/serve.py and mention TypeAdapter, but they do not look like the same concrete change: #43656 adds postponed annotations and broader typing cleanup, while #43842 specifically fixes optional pydantic/TypeAdapter import handling. The issue targets also differ." } ] }, @@ -22268,8 +22331,8 @@ "pull_request:44085|pull_request:44086", "pull_request:44059|pull_request:44086", "pull_request:44098|pull_request:44101", - "pull_request:44059|pull_request:44722", "pull_request:44018|pull_request:44071", + "pull_request:44059|pull_request:44722", "pull_request:44017|pull_request:44161", "pull_request:44017|pull_request:44722", "pull_request:44017|pull_request:44073", @@ -22313,12 +22376,12 @@ "pull_request:44044|pull_request:44154", "pull_request:44024|pull_request:44076" ], - "prepared_review_unit_hash": "ad58c4d62195b23f7a5a7ca2dc8886d98121ff877cd0000af175efd7cebecf9c", + "prepared_review_unit_hash": "35749bac72ee7c14cf4a9ba5997130760252b49bbadf68095381606dcbab60ee", "component_budget": { "node_count": 28, "item_count": 28, "soft_pair_count": 77, - "serialized_chars": 84509, + "serialized_chars": 84511, "estimated_input_tokens": 21128, "estimated_eval_tokens": 42512 }, @@ -22326,7 +22389,7 @@ "node_count": 28, "item_count": 28, "soft_pair_count": 72, - "serialized_chars": 76941, + "serialized_chars": 76943, "estimated_input_tokens": 19236, "estimated_eval_tokens": 38728 }, @@ -22335,360 +22398,378 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a set of model-by-model output-tracing refactors around the same broader initiative (#43979). Most pairs are not true duplicates because they target different architectures and code paths. The only clear duplicate-like groups are the GPT-Neo pair and the GPT-J lineage, which all refactor the same underlying block/attention output handling.", - "confidence": 0.84, + "summary": "This cluster is mostly a broad wave of similar output-tracing refactors across many different model implementations, not one duplicate bug. The only convincing duplicates are within the GPT-Neo and GPT-J families, where multiple PRs touch the same concrete code paths and appear to be overlapping revisions of the same refactor.", + "confidence": 0.87, "canonical_issue_reason": null, - "canonical_pr_reason": null, + "canonical_pr_reason": "PR 44722 is the strongest representative of the true duplicate subgroup: it covers the GPT-J output-tracing refactor and also extends it to CodeGen, making it the most complete version of that concrete change family.", "best_issue_reason": null, - "best_pr_reason": "PR 44066 is the clearest representative: it is well-scoped, explicitly linked to the tracking issue, and covers the core GPT-J/codegen output-tracing refactor in a coherent implementation.", + "best_pr_reason": "PR 44722 is the best global representative because it is the latest and broadest concrete implementation of the overlapping GPT-J/CodeGen tracing refactor, with explicit issue linkage and the clearest code-path overlap.", "soft_edge_verdicts": [ { "left": "pull_request:43996", "right": "pull_request:44085", "accept": false, - "reason": "Different models and code paths (CVT/FNet vs GPT-J); same general refactor theme is not enough." + "reason": "Different model families and different code paths; same tracking issue is only a shared refactor theme." }, { "left": "pull_request:43996", "right": "pull_request:44044", "accept": false, - "reason": "Unrelated architectures (CVT/FNet vs DeBERTa v2); no shared concrete bug/change." + "reason": "Different models and files; both are output-tracing refactors but not the same concrete bug/change." }, { "left": "pull_request:44066", "right": "pull_request:44085", "accept": true, - "reason": "Same GPT-J output-tracing refactor for the same underlying block/attention path; 44085 appears to be another iteration of the same change." + "reason": "Same GPT-J output-tracing code path in the same file; 44085 looks like a narrower overlap of the broader GPT-J/CodeGen refactor." }, { "left": "pull_request:44007", "right": "pull_request:44072", "accept": false, - "reason": "ResNet and EfficientNet are separate model implementations; these are parallel refactors, not duplicates." + "reason": "RegNet/ResNet-family vs EfficientNet: same umbrella refactor pattern, but not the same underlying change." }, { "left": "pull_request:44072", "right": "pull_request:44722", "accept": false, - "reason": "EfficientNet vs GPT-J/CodeGen are unrelated code paths despite both being output-tracing work." + "reason": "EfficientNet and GPT-J are unrelated code paths; this is only similarity in refactor style." }, { "left": "pull_request:44066", "right": "pull_request:44072", "accept": false, - "reason": "GPT-J/CodeGen and EfficientNet are different model families; no same-code-path overlap." + "reason": "GPT-J/CodeGen and EfficientNet are different implementations; no concrete overlap beyond the generic tracing refactor." }, { "left": "pull_request:44013", "right": "pull_request:44044", "accept": false, - "reason": "MobileNetV2 and DeBERTa v2 are unrelated implementations." + "reason": "MobileNetV2 vs DeBERTa are unrelated models; shared issue target is not enough to merge them." }, { "left": "pull_request:44066", "right": "pull_request:44086", "accept": false, - "reason": "GPT-J/CodeGen vs MGP-STR are different models and different output flows." + "reason": "Different model code paths (GPT-J/CodeGen vs MGP-STR); only the refactor pattern is similar." }, { "left": "pull_request:44018", "right": "pull_request:44068", "accept": true, - "reason": "Both are GPT-Neo output-tracing refactors touching the same model code and tests; clearly the same underlying change." + "reason": "Both are GPT-Neo tracing refactors in the same file/test area and look like overlapping revisions of the same change." }, { "left": "pull_request:44066", "right": "pull_request:44071", "accept": false, - "reason": "GPT-J/CodeGen vs MPT are unrelated model code paths." + "reason": "GPT-J/CodeGen and MPT are unrelated implementations; no same bug/change." }, { "left": "pull_request:44066", "right": "pull_request:44068", "accept": false, - "reason": "GPT-J/CodeGen vs GPT-Neo are different models; similar refactor pattern only." + "reason": "GPT-J/CodeGen vs GPT-Neo: same theme, different model and different code path." }, { "left": "pull_request:44018", "right": "pull_request:44066", "accept": false, - "reason": "GPT-Neo and GPT-J are separate model implementations; not the same concrete fix." + "reason": "GPT-Neo vs GPT-J/CodeGen are different model families; not the same concrete refactor." }, { "left": "pull_request:44019", "right": "pull_request:44722", "accept": false, - "reason": "ResNet vs GPT-J/CodeGen are unrelated architectures." + "reason": "ResNet and GPT-J are unrelated code paths; shared issue label does not imply duplication." }, { "left": "pull_request:44086", "right": "pull_request:44722", "accept": false, - "reason": "MGP-STR and GPT-J/CodeGen are different code paths." + "reason": "MGP-STR vs GPT-J are different models and different output paths." }, { "left": "pull_request:44019", "right": "pull_request:44085", "accept": false, - "reason": "ResNet and GPT-J are unrelated model implementations." + "reason": "ResNet and GPT-J are unrelated implementations." }, { "left": "pull_request:44018", "right": "pull_request:44019", "accept": false, - "reason": "GPT-Neo and ResNet are different models; no duplicate bug/change." + "reason": "GPT-Neo and ResNet are different model families; only the refactor theme matches." }, { "left": "pull_request:44019", "right": "pull_request:44071", "accept": false, - "reason": "ResNet vs MPT are unrelated code paths." + "reason": "ResNet and MPT are unrelated model code paths." }, { "left": "pull_request:44068", "right": "pull_request:44722", "accept": false, - "reason": "GPT-Neo vs GPT-J/CodeGen are different models; only the refactor style is similar." + "reason": "GPT-Neo and GPT-J are different model implementations; not the same fix." }, { "left": "pull_request:44068", "right": "pull_request:44116", "accept": false, - "reason": "GPT-Neo and Flaubert are separate models with different output logic." + "reason": "GPT-Neo and Flaubert are unrelated models." }, { "left": "pull_request:44018", - "right": "pull_request:44116", - "accept": false, - "reason": "GPT-Neo and Flaubert are unrelated implementations." - }, - { - "left": "pull_request:44071", - "right": "pull_request:44722", + "right": "pull_request:44071", "accept": false, - "reason": "MPT and GPT-J/CodeGen are different code paths." + "reason": "GPT-Neo and MPT are unrelated code paths." }, { "left": "pull_request:44018", "right": "pull_request:44085", "accept": false, - "reason": "GPT-Neo vs GPT-J are different models; not a duplicate." + "reason": "GPT-Neo vs GPT-J/CodeGen: same refactor pattern, different model code." }, { "left": "pull_request:44018", "right": "pull_request:44086", "accept": false, - "reason": "GPT-Neo and MGP-STR are unrelated model code paths." + "reason": "GPT-Neo and MGP-STR are unrelated implementations." }, { - "left": "pull_request:44059", + "left": "pull_request:44068", "right": "pull_request:44086", "accept": false, - "reason": "GPT-2 and MGP-STR are different models with different output tracing." + "reason": "GPT-Neo and MGP-STR are unrelated code paths." }, { - "left": "pull_request:44098", - "right": "pull_request:44101", + "left": "pull_request:44071", + "right": "pull_request:44722", "accept": false, - "reason": "ViLT and Flaubert/XLM are unrelated model families." + "reason": "MPT and GPT-J are different model families; no shared concrete bug." + }, + { + "left": "pull_request:44018", + "right": "pull_request:44059", + "accept": false, + "reason": "GPT-Neo and GPT-2 are different models; only the tracing refactor theme overlaps." }, { "left": "pull_request:44059", "right": "pull_request:44722", "accept": false, - "reason": "GPT-2 and GPT-J/CodeGen are separate model implementations." + "reason": "GPT-2 and GPT-J are different code paths." }, { - "left": "pull_request:44018", - "right": "pull_request:44071", + "left": "pull_request:44017", + "right": "pull_request:44161", "accept": false, - "reason": "GPT-Neo and MPT are different models; same refactor theme only." + "reason": "Segformer and LongT5 are unrelated models." }, { "left": "pull_request:44017", - "right": "pull_request:44161", + "right": "pull_request:44722", "accept": false, - "reason": "SegFormer and LongT5 are unrelated architectures." + "reason": "Segformer and GPT-J are unrelated code paths." }, { - "left": "pull_request:43997", - "right": "pull_request:44017", + "left": "pull_request:44017", + "right": "pull_request:44073", "accept": false, - "reason": "RegNet and SegFormer are different vision models; no same fix." + "reason": "Segformer and VisualBERT are different implementations." }, { - "left": "pull_request:43997", - "right": "pull_request:44056", + "left": "pull_request:44068", + "right": "pull_request:44071", "accept": false, - "reason": "RegNet and MPNet are unrelated code paths." + "reason": "GPT-Neo and MPT are different model families." }, { - "left": "pull_request:44071", - "right": "pull_request:44076", + "left": "pull_request:43997", + "right": "pull_request:44073", "accept": false, - "reason": "MPT and ImageGPT are separate model implementations." + "reason": "RegNet and VisualBERT are unrelated models." }, { - "left": "pull_request:44074", + "left": "pull_request:44073", "right": "pull_request:44101", "accept": false, - "reason": "TextNet and Flaubert/XLM are different architectures." + "reason": "VisualBERT and Flaubert/XLM are different code paths." }, { - "left": "pull_request:43997", - "right": "pull_request:44018", + "left": "pull_request:44074", + "right": "pull_request:44101", "accept": false, - "reason": "RegNet and GPT-Neo are unrelated models." + "reason": "TextNet and Flaubert/XLM are unrelated implementations." }, { "left": "pull_request:44056", "right": "pull_request:44161", "accept": false, - "reason": "MPNet and LongT5 are different model code paths." + "reason": "MPNet and LongT5 are unrelated models." }, { - "left": "pull_request:43997", - "right": "pull_request:44161", + "left": "pull_request:43996", + "right": "pull_request:44017", "accept": false, - "reason": "RegNet and LongT5 are unrelated." + "reason": "Cvt/FNet and Segformer are different model families." }, { - "left": "pull_request:44074", - "right": "pull_request:44161", + "left": "pull_request:44010", + "right": "pull_request:44071", "accept": false, - "reason": "TextNet and LongT5 are different models." + "reason": "SqueezeBERT and MPT are unrelated code paths." }, { - "left": "pull_request:44018", - "right": "pull_request:44161", + "left": "pull_request:44010", + "right": "pull_request:44073", "accept": false, - "reason": "GPT-Neo and LongT5 are unrelated implementations." + "reason": "SqueezeBERT and VisualBERT are different models." }, { "left": "pull_request:44010", - "right": "pull_request:44071", + "right": "pull_request:44018", "accept": false, - "reason": "SqueezeBert and MPT are unrelated code paths." + "reason": "SqueezeBERT and GPT-Neo are unrelated implementations." }, { "left": "pull_request:44010", - "right": "pull_request:44073", + "right": "pull_request:44044", "accept": false, - "reason": "SqueezeBert and VisualBert are different model families." + "reason": "SqueezeBERT and DeBERTa are different model code paths." }, { - "left": "pull_request:44018", - "right": "pull_request:44129", + "left": "pull_request:44044", + "right": "pull_request:44074", "accept": false, - "reason": "GPT-Neo and SpeechT5 are unrelated models." + "reason": "DeBERTa and TextNet are unrelated models." }, { - "left": "pull_request:43997", - "right": "pull_request:44074", + "left": "pull_request:44024", + "right": "pull_request:44076", "accept": false, - "reason": "RegNet and TextNet are different architectures." + "reason": "FocalNet and ImageGPT are different code paths; same generic refactor style only." }, { - "left": "pull_request:44010", - "right": "pull_request:44018", + "left": "pull_request:44059", + "right": "pull_request:44071", + "accept": false, + "reason": "GPT-2 and MPT are unrelated model families." + }, + { + "left": "pull_request:44059", + "right": "pull_request:44068", "accept": false, - "reason": "SqueezeBert and GPT-Neo are unrelated models." + "reason": "GPT-2 and GPT-Neo are different implementations." + }, + { + "left": "pull_request:44074", + "right": "pull_request:44101", + "accept": false, + "reason": "TextNet and Flaubert/XLM are unrelated models." }, { "left": "pull_request:43997", - "right": "pull_request:44129", + "right": "pull_request:44056", + "accept": false, + "reason": "RegNet and MPNet are unrelated code paths." + }, + { + "left": "pull_request:43997", + "right": "pull_request:44074", "accept": false, - "reason": "RegNet and SpeechT5 are unrelated code paths." + "reason": "RegNet and TextNet are different model families." }, { "left": "pull_request:44071", - "right": "pull_request:44129", + "right": "pull_request:44076", "accept": false, - "reason": "MPT and SpeechT5 are different models." + "reason": "MPT and ImageGPT are unrelated implementations." }, { - "left": "pull_request:43996", + "left": "pull_request:44074", "right": "pull_request:44161", "accept": false, - "reason": "CVT/FNet and LongT5 are unrelated." + "reason": "TextNet and LongT5 are different model code paths." }, { - "left": "pull_request:43996", - "right": "pull_request:44017", + "left": "pull_request:44018", + "right": "pull_request:44129", "accept": false, - "reason": "CVT/FNet and SegFormer are separate model implementations." + "reason": "GPT-Neo and SpeechT5 are unrelated models." }, { - "left": "pull_request:44010", - "right": "pull_request:44044", + "left": "pull_request:44018", + "right": "pull_request:44059", "accept": false, - "reason": "SqueezeBert and DeBERTa v2 are different code paths." + "reason": "GPT-Neo and GPT-2 are different model families." }, { - "left": "pull_request:44044", - "right": "pull_request:44074", + "left": "pull_request:44056", + "right": "pull_request:44129", "accept": false, - "reason": "DeBERTa v2 and TextNet are unrelated models." + "reason": "MPNet and SpeechT5 are unrelated code paths." }, { "left": "pull_request:44066", "right": "pull_request:44722", "accept": true, - "reason": "Same GPT-J/CodeGen output-tracing change; 44722 looks like a later rework of the same refactor." + "reason": "Both touch the same GPT-J tracing refactor; 44722 is the broader follow-up that subsumes the GPT-J part of 44066." }, { "left": "pull_request:44085", "right": "pull_request:44722", "accept": true, - "reason": "Both target the same GPT-J output-tracing refactor on the same underlying code path." + "reason": "Same GPT-J file and same output-tracing migration; 44722 is a broader version of the same concrete change." }, { "left": "pull_request:44085", "right": "pull_request:44129", "accept": false, - "reason": "GPT-J and SpeechT5 are unrelated models." + "reason": "GPT-J and SpeechT5 are unrelated models; shared issue target is only thematic." }, { "left": "pull_request:43996", "right": "pull_request:44007", "accept": false, - "reason": "FNet/CVT and ResNet are unrelated model families." + "reason": "Cvt/FNet and RegNet/ResNet-family are unrelated model implementations." }, { "left": "pull_request:43996", "right": "pull_request:44129", "accept": false, - "reason": "FNet/CVT and SpeechT5 are unrelated code paths." + "reason": "Cvt/FNet and SpeechT5 are unrelated code paths." }, { "left": "pull_request:44044", "right": "pull_request:44129", "accept": false, - "reason": "DeBERTa v2 and SpeechT5 are different models." + "reason": "DeBERTa and SpeechT5 are unrelated models." }, { "left": "pull_request:44044", "right": "pull_request:44154", "accept": false, - "reason": "DeBERTa v2 and ViTS are unrelated implementations." + "reason": "DeBERTa and VITS are different model families." }, { - "left": "pull_request:44066", - "right": "pull_request:44129", + "left": "pull_request:44024", + "right": "pull_request:44076", "accept": false, - "reason": "GPT-J/CodeGen and SpeechT5 are unrelated models." + "reason": "FocalNet and ImageGPT are unrelated model code paths." }, { - "left": "pull_request:44024", - "right": "pull_request:44076", + "left": "pull_request:44018", + "right": "pull_request:44161", "accept": false, - "reason": "FocalNet and ImageGPT are different architectures; same refactor theme only." + "reason": "GPT-Neo and LongT5 are unrelated implementations." } ] }, "evaluator_result": { "accept": true, - "feedback": "The summary is grounded in the packet: it correctly frames the cluster as mostly separate model-specific output-tracing refactors under #43979, and it limits duplicate-like claims to the GPT-Neo pair and the GPT-J lineage. The accepted PR pairs are conservative and look mergeable as the same bugfix within overlapping code paths. No clear overstatement stands out." + "feedback": "The summary is well grounded in the packet and stays conservative: it distinguishes the broad refactor wave from the smaller duplicate subgroups, and the accepted soft-edge pairs are supported by shared files/code paths or clear overlap in the same model family. The choice of PR 44722 as the best representative is also justified by its later, broader GPT-J+CodeGen scope and explicit issue linkage." }, "evaluator_used": true, "retried": false, @@ -22723,12 +22804,12 @@ "pull_request:43997|pull_request:44010", "pull_request:44010|pull_request:44129" ], - "prepared_review_unit_hash": "8f777c6416f36ac773d24fb19ddb194eb0d17da9eb65875973585ba5f0da7c43", + "prepared_review_unit_hash": "1e3a85c8c78cea756f8f59f5b69370761eb47e4f6e37f76851b63cd583734e3e", "component_budget": { "node_count": 28, "item_count": 28, "soft_pair_count": 77, - "serialized_chars": 84509, + "serialized_chars": 84511, "estimated_input_tokens": 21128, "estimated_eval_tokens": 42512 }, @@ -22745,42 +22826,42 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "All items are model-specific output-tracing refactors, not duplicates of the same underlying bug/change. The cluster is thematically similar, but each PR targets a different model implementation and code path.", - "confidence": 0.86, + "summary": "All six PRs are related by the same general refactor pattern: migrating model-specific output tracing to standardized capture/decorator utilities. However, they touch different model implementations and concrete code paths, so they are not duplicates or mergeable into a single PR.", + "confidence": 0.94, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 44129 is the best representative of the group: it has an explicit issue target and a concrete, self-contained output-tracing refactor in one model, whereas the others are parallel model-by-model migrations.", + "canonical_pr_reason": "PR #44129 is the best representative of the cluster because it has the clearest concrete scope, an explicit issue target, and shows a substantive standardized output-capture refactor in one model implementation.", "best_issue_reason": null, - "best_pr_reason": "PR 44129 is the strongest anchor for the cluster because it is tied to a tracked issue and clearly describes a specific standardized output-capture change rather than a broad generic migration.", + "best_pr_reason": "PR #44129 is the strongest standalone representative: it is directly tied to an issue, has a clear refactor goal, and is more fully specified than the others. That said, it is still not a duplicate of the other PRs.", "soft_edge_verdicts": [ { "left": "pull_request:44073", "right": "pull_request:44074", "accept": false, - "reason": "Both are output-tracing refactors, but they touch different models (VisualBert vs TextNet) and different forward/capture plumbing. Similar pattern, not the same underlying change or mergeable into one PR." + "reason": "Both are output-tracing refactors, but they modify different model files and different code paths (VisualBert vs TextNet). Same theme, not the same underlying change." }, { "left": "pull_request:44056", "right": "pull_request:44073", "accept": false, - "reason": "MPNet and VisualBert are separate model implementations with distinct output-recording setup. The shared theme is only the refactor style, not the same bug or code path." + "reason": "These are separate model refactors in different implementations (MPNet vs VisualBert). The similarity is architectural pattern, not a shared bug or change." }, { "left": "pull_request:44056", "right": "pull_request:44074", "accept": false, - "reason": "These are parallel standardized-output migrations in different models. No shared concrete failure mode or implementation target that would justify treating them as duplicates." + "reason": "Different models and different forward/output-capture logic. They are related refactors, but not duplicates and not safely mergeable into one PR." }, { "left": "pull_request:43997", "right": "pull_request:44010", "accept": false, - "reason": "RegNet and SqueezeBert refactors are model-specific and affect different forward signatures and helper wiring. They are not the same change, just the same migration pattern." + "reason": "RegNet and SqueezeBert each migrate their own output handling; same modernization theme, but no shared code path or concrete duplicate fix." }, { "left": "pull_request:44010", "right": "pull_request:44129", "accept": false, - "reason": "SqueezeBert and SpeechT5 touch different code paths and different output-capture mechanisms. The overlap is only in moving to standardized tracing, which is too broad to count as a duplicate." + "reason": "SqueezeBert and SpeechT5 are unrelated model implementations. The overlap is only in standardized output tracing, not in the same underlying change." } ] }, @@ -22860,274 +22941,274 @@ "pull_request:44028|pull_request:44030", "pull_request:44029|pull_request:44030" ], - "prepared_review_unit_hash": "78f7a072c4394a6176476c69cb46c025e2f5d97b7bbf667ca373f201ae03aadc", + "prepared_review_unit_hash": "d51c0716fe10b3e4d8639b6423034c198f36517369eff1236a2a965285787597", "component_budget": { "node_count": 13, "item_count": 13, "soft_pair_count": 40, - "serialized_chars": 40180, - "estimated_input_tokens": 10045, - "estimated_eval_tokens": 20346 + "serialized_chars": 40170, + "estimated_input_tokens": 10043, + "estimated_eval_tokens": 20342 }, "budget": { "node_count": 13, "item_count": 13, "soft_pair_count": 40, - "serialized_chars": 40180, - "estimated_input_tokens": 10045, - "estimated_eval_tokens": 20346 + "serialized_chars": 40170, + "estimated_input_tokens": 10043, + "estimated_eval_tokens": 20342 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These PRs share a broad \"output tracing\" refactor theme, but they target different model implementations and code paths, so they are not duplicates of one another.", - "confidence": 0.96, + "summary": "These PRs share a generic \"output tracing\" refactor theme, but they touch different model implementations and code paths, so they are related cleanups rather than duplicates.", + "confidence": 0.95, "canonical_issue_reason": null, - "canonical_pr_reason": "If one PR must anchor the cluster, 44025 is the most central representative because it spans two related depth-anything files and sits in the middle of the shared output-tracing pattern; however, it is still a model-specific refactor, not the same change as the others.", + "canonical_pr_reason": "No canonical PR: each change is scoped to a different model family, with distinct forward signatures and implementation details.", "best_issue_reason": null, - "best_pr_reason": "44025 is the best overall representative for this cluster for the same reason: it is substantial, model-specific, and closest to the common refactor theme without being the same underlying fix as the rest.", + "best_pr_reason": "44025 is the broadest exemplar of the recurring tracing refactor pattern, but it still does not subsume the others or serve as a true duplicate target.", "soft_edge_verdicts": [ { "left": "pull_request:43998", "right": "pull_request:44002", "accept": false, - "reason": "Different models and code paths; same refactor pattern, but not the same underlying change." + "reason": "Different model families (`timm_backbone` vs `upernet`); same refactor theme, different code paths." }, { "left": "pull_request:43998", "right": "pull_request:44000", "accept": false, - "reason": "Different models and code paths; same refactor pattern, but not the same underlying change." + "reason": "`timm_backbone` and `vision_text_dual_encoder` are unrelated implementations despite similar tracing cleanup." }, { "left": "pull_request:43998", "right": "pull_request:44001", "accept": false, - "reason": "Different models and code paths; same refactor pattern, but not the same underlying change." + "reason": "Image backbone refactor vs audio generator refactor; not the same bug/change." }, { "left": "pull_request:43998", "right": "pull_request:44003", "accept": false, - "reason": "Different models and code paths; same refactor pattern, but not the same underlying change." + "reason": "`timm_backbone` cleanup is separate from the `mamba`/`falcon_mamba` tracing changes." }, { "left": "pull_request:43998", "right": "pull_request:44029", "accept": false, - "reason": "Different models and code paths; same refactor pattern, but not the same underlying change." + "reason": "Different architectures (`timm_backbone` vs `rwkv`); only the helper pattern is shared." }, { "left": "pull_request:43999", "right": "pull_request:44001", "accept": false, - "reason": "Different models and code paths; same refactor pattern, but not the same underlying change." + "reason": "MobileNetV1 and UnivNet are different domains and forward paths; not duplicates." }, { "left": "pull_request:43999", "right": "pull_request:44002", "accept": false, - "reason": "Different models and code paths; same refactor pattern, but not the same underlying change." + "reason": "CNN tracing refactor vs segmentation-model tracing refactor; distinct model code." }, { "left": "pull_request:43999", "right": "pull_request:44025", "accept": false, - "reason": "Different models and code paths; same refactor pattern, but not the same underlying change." + "reason": "MobileNetV1 and DepthAnything are separate models; shared helper usage is too broad to merge." }, { "left": "pull_request:43999", "right": "pull_request:44026", "accept": false, - "reason": "Different models and code paths; same refactor pattern, but not the same underlying change." + "reason": "MobileNetV1 refactor is unrelated to the vision-encoder/decoder wrapper change." }, { "left": "pull_request:43999", "right": "pull_request:44027", "accept": false, - "reason": "Different models and code paths; same refactor pattern, but not the same underlying change." + "reason": "Different model stacks (`mobilenet_v1` vs `speech_encoder_decoder`); not the same underlying change." }, { "left": "pull_request:44000", "right": "pull_request:44001", "accept": false, - "reason": "Different models and code paths; same refactor pattern, but not the same underlying change." + "reason": "Multimodal dual encoder vs UnivNet audio model; same pattern, different implementation." }, { "left": "pull_request:44000", "right": "pull_request:44002", "accept": false, - "reason": "Different models and code paths; same refactor pattern, but not the same underlying change." + "reason": "These touch different model families and forward contracts, so they are not duplicates." }, { "left": "pull_request:44000", "right": "pull_request:44003", "accept": false, - "reason": "Different models and code paths; same refactor pattern, but not the same underlying change." + "reason": "`vision_text_dual_encoder` is unrelated to the `mamba` refactor beyond shared helper cleanup." }, { "left": "pull_request:44000", "right": "pull_request:44025", "accept": false, - "reason": "Different models and code paths; same refactor pattern, but not the same underlying change." + "reason": "Different vision model families; no shared concrete bug or change." }, { "left": "pull_request:44001", "right": "pull_request:44002", "accept": false, - "reason": "Different models and code paths; same refactor pattern, but not the same underlying change." + "reason": "UnivNet and UperNet are unrelated architectures; only the refactor style matches." }, { "left": "pull_request:44001", "right": "pull_request:44003", "accept": false, - "reason": "Different models and code paths; same refactor pattern, but not the same underlying change." + "reason": "Audio generation model vs Mamba variant; not the same code path." }, { "left": "pull_request:44001", "right": "pull_request:44025", "accept": false, - "reason": "Different models and code paths; same refactor pattern, but not the same underlying change." + "reason": "DepthAnything and UnivNet are separate domains; no duplicate evidence." }, { "left": "pull_request:44002", "right": "pull_request:44029", "accept": false, - "reason": "Different models and code paths; same refactor pattern, but not the same underlying change." + "reason": "Segmentation model refactor vs RWKV refactor; different subsystems." }, { "left": "pull_request:44002", "right": "pull_request:44003", "accept": false, - "reason": "Different models and code paths; same refactor pattern, but not the same underlying change." + "reason": "UperNet cleanup does not overlap with the Mamba/Falcon-Mamba change set." }, { "left": "pull_request:44003", "right": "pull_request:44028", "accept": false, - "reason": "Different models and code paths; same refactor pattern, but not the same underlying change." + "reason": "Falcon/Mamba stateful model refactor and SuperPoint refactor are unrelated." }, { "left": "pull_request:44003", "right": "pull_request:44004", "accept": false, - "reason": "Different models and code paths; same refactor pattern, but not the same underlying change." + "reason": "Different model code paths (`mamba` vs `codegen`); not one concrete fix." }, { "left": "pull_request:43998", "right": "pull_request:44004", "accept": false, - "reason": "Different models and code paths; same refactor pattern, but not the same underlying change." + "reason": "Timm backbone cleanup and CodeGen cache/signature refactor are distinct changes." }, { "left": "pull_request:44001", "right": "pull_request:44004", "accept": false, - "reason": "Different models and code paths; same refactor pattern, but not the same underlying change." + "reason": "UnivNet audio output tracing and CodeGen forward signature changes are not the same issue." }, { "left": "pull_request:44002", "right": "pull_request:44004", "accept": false, - "reason": "Different models and code paths; same refactor pattern, but not the same underlying change." + "reason": "UperNet segmentation refactor vs CodeGen refactor; different code paths." }, { "left": "pull_request:44000", "right": "pull_request:44004", "accept": false, - "reason": "Different models and code paths; same refactor pattern, but not the same underlying change." + "reason": "Multimodal encoder/decoder refactor does not match CodeGen's cache/attention API cleanup." }, { "left": "pull_request:44025", "right": "pull_request:44026", "accept": false, - "reason": "Different model families and code paths; same tracing refactor theme, not the same fix." + "reason": "DepthAnything and VisionEncoderDecoder are different model families; same helper pattern only." }, { "left": "pull_request:44025", "right": "pull_request:44027", "accept": false, - "reason": "Different model families and code paths; same tracing refactor theme, not the same fix." + "reason": "DepthAnything refactor is unrelated to SpeechEncoderDecoder beyond generic tracing cleanup." }, { "left": "pull_request:44025", "right": "pull_request:44028", "accept": false, - "reason": "Different model families and code paths; same tracing refactor theme, not the same fix." + "reason": "DepthAnything and SuperPoint are separate vision models with different forward logic." }, { "left": "pull_request:44025", "right": "pull_request:44029", "accept": false, - "reason": "Different model families and code paths; same tracing refactor theme, not the same fix." + "reason": "DepthAnything and RWKV touch unrelated architectures and code paths." }, { "left": "pull_request:44025", "right": "pull_request:44030", "accept": false, - "reason": "Different model families and code paths; same tracing refactor theme, not the same fix." + "reason": "DepthAnything refactor is not the same as DPR encoder wrapper cleanup." }, { "left": "pull_request:44026", "right": "pull_request:44029", "accept": false, - "reason": "Different models and code paths; same refactor pattern, but not the same underlying change." + "reason": "VisionEncoderDecoder wrapper and RWKV model changes are unrelated." }, { "left": "pull_request:44026", "right": "pull_request:44027", "accept": false, - "reason": "Different models and code paths; same refactor pattern, but not the same underlying change." + "reason": "VisionEncoderDecoder vs SpeechEncoderDecoder are different wrapper implementations; not duplicates." }, { "left": "pull_request:44026", "right": "pull_request:44028", "accept": false, - "reason": "Different models and code paths; same refactor pattern, but not the same underlying change." + "reason": "Different model families (`vision_encoder_decoder` vs `superpoint`)." }, { "left": "pull_request:44026", "right": "pull_request:44030", "accept": false, - "reason": "Different models and code paths; same refactor pattern, but not the same underlying change." + "reason": "Encoder-decoder wrapper cleanup and DPR refactor are separate model changes." }, { "left": "pull_request:44027", "right": "pull_request:44028", "accept": false, - "reason": "Different models and code paths; same refactor pattern, but not the same underlying change." + "reason": "Speech encoder-decoder wrapper and SuperPoint model are unrelated." }, { "left": "pull_request:44027", "right": "pull_request:44029", "accept": false, - "reason": "Different models and code paths; same refactor pattern, but not the same underlying change." + "reason": "SpeechEncoderDecoder and RWKV differ in architecture and forward path." }, { "left": "pull_request:44027", "right": "pull_request:44030", "accept": false, - "reason": "Different models and code paths; same refactor pattern, but not the same underlying change." + "reason": "SpeechEncoderDecoder refactor does not match DPR's wrapper change." }, { "left": "pull_request:44028", "right": "pull_request:44029", "accept": false, - "reason": "Different models and code paths; same refactor pattern, but not the same underlying change." + "reason": "SuperPoint and RWKV are unrelated codebases/architectures." }, { "left": "pull_request:44028", "right": "pull_request:44030", "accept": false, - "reason": "Different models and code paths; same refactor pattern, but not the same underlying change." + "reason": "SuperPoint vision model refactor is separate from DPR's encoder wrapper cleanup." }, { "left": "pull_request:44029", "right": "pull_request:44030", "accept": false, - "reason": "Different models and code paths; same refactor pattern, but not the same underlying change." + "reason": "RWKV and DPR changes touch different model families and are not mergeable as one fix." } ] }, @@ -23157,7 +23238,7 @@ "soft_pairs": [ "pull_request:44662|pull_request:45401" ], - "prepared_review_unit_hash": "58fdc30825f05a463fcf31dcefff9806f1c40f9eb6089237e6b625d68573d055", + "prepared_review_unit_hash": "b0ca5f39923acbe07b5f023b2800e6d34f3c4362a359c2264496bc737c73dbdf", "component_budget": { "node_count": 2, "item_count": 2, @@ -23179,18 +23260,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These are two separate model-integration PRs that overlap only on generic Transformers plumbing and docs registration, not on the underlying change.", - "confidence": 0.98, + "summary": "These are two separate model-addition PRs (PenguinVL vs Voxtral TTS). They share framework registration/docs boilerplate, but they do not fix the same underlying code-path problem.", + "confidence": 0.97, "canonical_issue_reason": null, - "canonical_pr_reason": "No canonical PR: #44662 adds PenguinVL, while #45401 adds Voxtral TTS; they are distinct implementations for different models.", + "canonical_pr_reason": "No single canonical PR: 44662 introduces PenguinVL, while 45401 introduces Voxtral TTS; they target different models and independent implementation paths.", "best_issue_reason": null, - "best_pr_reason": "If a representative is needed, #45401 is slightly more self-contained because it explicitly targets issue #45027, but it is still not a duplicate of #44662.", + "best_pr_reason": "45401 is the cleaner anchor for its own change because it is draft, explicitly tied to issue #45027, and has a narrower, well-scoped addition.", "soft_edge_verdicts": [ { "left": "pull_request:44662", "right": "pull_request:45401", "accept": false, - "reason": "Reject: the PRs implement different model families (PenguinVL vs. Voxtral TTS). Shared files are just common auto-registration/docs scaffolding, not the same code-path fix." + "reason": "Different model integrations with distinct configs, processors, and model code; the shared auto-registry/docs edits are generic boilerplate, not evidence of the same bug or change." } ] }, @@ -23220,12 +23301,12 @@ "soft_pairs": [ "pull_request:44965|pull_request:46845" ], - "prepared_review_unit_hash": "9e6b2be094eb4dca39f99247d452c8acedf027e35af4ab6dc3bc80c54092eee7", + "prepared_review_unit_hash": "3587ac714a751f2d2319d48f0d70c536540f6ec8799d6e46c6e8769c114b8abe", "component_budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 4218, + "serialized_chars": 4219, "estimated_input_tokens": 1055, "estimated_eval_tokens": 2366 }, @@ -23233,7 +23314,7 @@ "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 4218, + "serialized_chars": 4219, "estimated_input_tokens": 1055, "estimated_eval_tokens": 2366 }, @@ -23242,18 +23323,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "The two PRs are not duplicates: one changes a PR-documentation/draft workflow, while the other adjusts CI routing and includes unrelated code edits. The soft similarity is too broad to indicate the same underlying change.", - "confidence": 0.95, + "summary": "The two PRs are not duplicates: one changes the PR documentation workflow into a draft-conversion action, while the other adjusts CI branch routing and touches unrelated model code/test plumbing. They do not appear to fix the same underlying problem or implement the same change.", + "confidence": 0.96, "canonical_issue_reason": null, - "canonical_pr_reason": "If a representative PR must be picked, 46845 is the more coherent CI-related change set; 44965 is a separate workflow tweak and not the same change.", + "canonical_pr_reason": "PR #46845 is the better representative because it has a clearer, concrete purpose (CI branch/routing adjustment) and more substantive context than the placeholder-like PR #44965 titled \"try\".", "best_issue_reason": null, - "best_pr_reason": "46845 is the stronger standalone representative because it has a clearer, more substantive purpose; 44965 is unrelated and should not be merged into it.", + "best_pr_reason": "PR #46845 is the strongest standalone PR in this cluster: its intent is more explicit, and the changes are at least internally coherent compared with #44965, which looks exploratory and incomplete.", "soft_edge_verdicts": [ { "left": "pull_request:44965", "right": "pull_request:46845", "accept": false, - "reason": "Different purposes and code paths: 44965 edits the PR-documentation/draft workflow, while 46845 changes CI branch routing and includes unrelated model code/test edits. They are not the same underlying PR." + "reason": "Reject: these PRs target different changes. #44965 rewrites the PR documentation workflow to convert PRs to draft, while #46845 changes CI workflow branch handling and includes an unrelated edit in modeling_utils.py. They do not share the same concrete bug or code path." } ] }, @@ -23283,12 +23364,12 @@ "soft_pairs": [ "pull_request:44979|pull_request:45363" ], - "prepared_review_unit_hash": "98b253d3a2fd1ee8da61f5b5018512d9d290280e77c22c6d42246844dd5cb7eb", + "prepared_review_unit_hash": "ed94334000f7a11579355824b8ab79d82f0942e062ba91a5ac4d0144091426bf", "component_budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 4407, + "serialized_chars": 4408, "estimated_input_tokens": 1102, "estimated_eval_tokens": 2460 }, @@ -23296,7 +23377,7 @@ "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 4407, + "serialized_chars": 4408, "estimated_input_tokens": 1102, "estimated_eval_tokens": 2460 }, @@ -23305,18 +23386,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These are related feature PRs in the module/kernel fusion area, but they do not look like the same concrete change. One is a general Module Fusion API, the other is a specific n-to-1 kernel fusion implementation via KernelConfig.", - "confidence": 0.87, + "summary": "The two pull requests are related to module/kernel fusion, but they are not the same underlying change: one introduces a generic Module Fusion API, while the other wires up n-to-1 kernel fusion through KernelConfig and related integration code. They could coexist, not be merged as duplicates.", + "confidence": 0.91, "canonical_issue_reason": null, - "canonical_pr_reason": "PR #45363 is the more concrete and implementation-complete change: it wires KernelConfig into model loading/integrations and adds an example, whereas #44979 is a broader API scaffold for module fusion.", + "canonical_pr_reason": "PR 44979 is the cleaner canonical representative because it introduces the foundational Module Fusion API itself, which the other PR appears to build on rather than duplicate.", "best_issue_reason": null, - "best_pr_reason": "PR #45363 is the strongest standalone candidate because it targets a specific code path and shipping behavior (n-to-1 kernel fusion), with multiple implementation files and usage wiring, rather than just introducing a generic fusion API layer.", + "best_pr_reason": "PR 44979 is the best cluster representative since it defines the base abstraction; PR 45363 is a narrower follow-on integration/use-case change with example and wiring code.", "soft_edge_verdicts": [ { "left": "pull_request:44979", "right": "pull_request:45363", "accept": false, - "reason": "Related area, but different scope: #44979 introduces a generic Module Fusion API, while #45363 implements a specific KernelConfig-based n-to-1 kernel fusion flow. They are not the same underlying change and would not plausibly be merged as one PR." + "reason": "Different scope and code path: 44979 adds a new generic module-fusion API, while 45363 implements n-to-1 kernel fusion via KernelConfig plus integration/example changes. They are related, but not the same fix/change and would not plausibly be merged into one PR as duplicates." } ] }, @@ -23346,7 +23427,7 @@ "soft_pairs": [ "pull_request:45086|pull_request:45317" ], - "prepared_review_unit_hash": "8e87451fd10477014cec12d99e2666b7390003e0e505de40a8caf05db47f8f01", + "prepared_review_unit_hash": "98fe49e488b0466dc06a0945d1c8819436eb71956388bc98e6474e190b0839f8", "component_budget": { "node_count": 2, "item_count": 2, @@ -23368,18 +23449,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Both PRs fix the same AttributeError in `_patch_mistral_regex` by switching from `tokenizer.backend_tokenizer.pre_tokenizer` to `tokenizer.pre_tokenizer`; they target the same issue and code path, with 45317 adding tests and appearing to be the more complete version.", - "confidence": 0.96, + "summary": "Both pull requests fix the same Mistral regex AttributeError by switching from `backend_tokenizer.pre_tokenizer` to `tokenizer.pre_tokenizer` in the same code path; 45317 is the more complete change because it adds tests.", + "confidence": 0.98, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 45317 is the better canonical PR because it implements the same fix as 45086 and adds test coverage, making it the more complete change.", + "canonical_pr_reason": "PR 45317 is the clearest canonical PR: same core fix as 45086, but with added test coverage and a more explicit title about the failing `fix_mistral_regex=True` path.", "best_issue_reason": null, - "best_pr_reason": "PR 45317 is the best PR in the cluster because it fixes the bug and includes tests; 45086 is effectively a smaller precursor/duplicate of the same patch.", + "best_pr_reason": "PR 45317 is the best representative because it addresses the same bug while also including regression tests, making it the more complete standalone fix.", "soft_edge_verdicts": [ { "left": "pull_request:45086", "right": "pull_request:45317", "accept": true, - "reason": "Same underlying bug and same concrete fix in the same function/file; 45317 is a superset with tests, so these are duplicate PRs for the same change." + "reason": "Same underlying AttributeError in `_patch_mistral_regex`, same file and same line-level fix; 45317 appears to be an expanded version with tests rather than a different change." } ] }, @@ -23409,7 +23490,7 @@ "soft_pairs": [ "pull_request:45115|pull_request:45569" ], - "prepared_review_unit_hash": "94feba84d52122ad8c5d4b664061b63b215b4a440433c1dcbed4b6c807f95f2d", + "prepared_review_unit_hash": "095a59f3d46d1f32a1e8aeaf5eb7339d1b7c38271459b11a07337fdeb045dee5", "component_budget": { "node_count": 2, "item_count": 2, @@ -23431,18 +23512,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Two open PRs touch NemotronH internals, but they are not the same change: one is a refactor to make `nemotron_h` inherit GraniteMoeHybrid-like conversion/mapping behavior, while the other is a broader feature/cleanup PR introducing separate NemotronHDense/NemotronHSparse docs and modules. Related subsystem, different concrete scopes.", - "confidence": 0.86, + "summary": "These PRs are related to Nemotron H, but they address different scopes: #45115 is a conversion-mapping/refactor PR, while #45569 is a broader model split and docs/tests update. They do not look like the same underlying change.", + "confidence": 0.89, "canonical_issue_reason": null, - "canonical_pr_reason": "No issue is present; if a PR must represent the cluster, #45569 is the better umbrella candidate because it covers the broader NemotronH family restructuring and model/doc additions rather than only a narrow conversion-mapping refactor.", + "canonical_pr_reason": "pull_request:45569 is the more canonical PR for the actual model-structure change: it introduces NemotronH dense/sparse model files, updates auto/model registration, adds docs, and includes tests. It is the more complete and user-facing implementation.", "best_issue_reason": null, - "best_pr_reason": "#45569 is the strongest representative of the cluster because it is the more comprehensive NemotronH family change and includes the visible user-facing module/doc split; #45115 is a narrower implementation refactor.", + "best_pr_reason": "pull_request:45569 is the strongest standalone PR because it covers the full architectural split and integration points. pull_request:45115 is narrower and mostly updates checkpoint conversion mappings for renamed weights.", "soft_edge_verdicts": [ { "left": "pull_request:45115", "right": "pull_request:45569", "accept": false, - "reason": "Shared NemotronH files only indicate related model work. #45115 is a conversion/refactor toward GraniteMoeHybrid-compatible naming, while #45569 adds separate dense/sparse variants and documentation. They do not look mergeable into one single fix/change." + "reason": "Reject: they share NemotronH files, but one is about conversion key remapping while the other is about introducing dense/sparse model variants plus docs/tests. Different concrete code-path changes, not mergeable as the same fix." } ] }, @@ -23472,7 +23553,7 @@ "soft_pairs": [ "pull_request:45332|pull_request:45333" ], - "prepared_review_unit_hash": "9e1c8b3704746ed1f8124ee6fed3186c43294499e2af4b0f4a7c826b588d6a27", + "prepared_review_unit_hash": "abcef5461863d14fc1b1e2c7323e994b928b9985e1455696270d622cc97c0d1b", "component_budget": { "node_count": 2, "item_count": 2, @@ -23494,18 +23575,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Two related pull requests for heterogeneous model support, but they are not the same change: one is a broader modeling + config implementation, the other is a narrower per-layer config-only follow-up/slice.", - "confidence": 0.84, + "summary": "These are related but not duplicates: PR 45333 adds per-layer heterogeneous configuration support, while PR 45332 includes that plus broader heterogeneous modeling/runtime changes. They share setup files and some configuration code, but they target different scopes of the feature.", + "confidence": 0.91, "canonical_issue_reason": null, - "canonical_pr_reason": "No clear canonical PR duplicate: PR 45332 is the broader end-to-end heterogeneous model support change, while PR 45333 is a smaller configuration-only implementation with different scope.", + "canonical_pr_reason": "PR 45332 is the broader, more complete heterogeneous-model feature PR; it includes the config work plus modeling/runtime changes across cache, masking, modeling utils, and model support.", "best_issue_reason": null, - "best_pr_reason": "PR 45332 is the best representative of the cluster because it covers the larger, more complete heterogeneous-model feature set and subsumes the config work shown in PR 45333.", + "best_pr_reason": "PR 45332 is the best representative of the cluster because it covers the full end-to-end heterogeneous support rather than only the configuration subset.", "soft_edge_verdicts": [ { "left": "pull_request:45332", "right": "pull_request:45333", "accept": false, - "reason": "Related feature area, but not the same concrete change. PR 45332 adds heterogeneous model support including modeling code paths, while PR 45333 is limited to per-layer configuration utilities; they are too different in scope to merge as duplicates." + "reason": "They overlap on heterogeneous configuration plumbing, but 45332 also changes modeling/runtime paths and is not just a variant of 45333. The PRs are complementary scope, not the same concrete change." } ] }, @@ -23535,12 +23616,12 @@ "soft_pairs": [ "pull_request:45415|pull_request:45425" ], - "prepared_review_unit_hash": "6980168954a912e4e61327bca98bfcb00308b73b1878432b87ebeff065e1671c", + "prepared_review_unit_hash": "452aad0baac796558941499008fa92e596b18bfc5b75b7b73061c85e2d781921", "component_budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 5514, + "serialized_chars": 5513, "estimated_input_tokens": 1379, "estimated_eval_tokens": 3014 }, @@ -23548,7 +23629,7 @@ "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 5514, + "serialized_chars": 5513, "estimated_input_tokens": 1379, "estimated_eval_tokens": 3014 }, @@ -23557,18 +23638,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Both PRs are typing-related, but they address different changes: #45415 is a broad repository-wide type-checking sweep with pyproject ignore tweaks, while #45425 adds specific typing protocols/aliases and related support for modeling_utils/PEFT. They overlap in area, not in the same concrete fix.", - "confidence": 0.93, + "summary": "These are two related typing-focused PRs, but they are not the same underlying change. PR 45415 is a broad repository-wide type-checking pass across many Transformers modules, while PR 45425 adds specific typing helpers/protocols for a narrower modeling_utils/PEFT-related path.", + "confidence": 0.91, "canonical_issue_reason": null, - "canonical_pr_reason": "#45415 is the broader umbrella typing sweep across many transformer modules, so it is the more representative PR if one needs a single anchor for this cluster.", + "canonical_pr_reason": "PR 45415 is the broader umbrella change: it adjusts type-checking config and applies annotations across many files, making it the better canonical representative of the cluster.", "best_issue_reason": null, - "best_pr_reason": "#45415 best represents the cluster because it is the larger, central type-checking cleanup; #45425 is a narrower support PR for specific typing abstractions.", + "best_pr_reason": "PR 45415 is the best overall representative because it is the most comprehensive and central typing cleanup in the set; PR 45425 is narrower and more specialized.", "soft_edge_verdicts": [ { "left": "pull_request:45415", "right": "pull_request:45425", "accept": false, - "reason": "Do not merge: they are related by typing infrastructure, but #45415 is a broad type-checking pass and #45425 is a targeted protocol/alias addition. They do not look like the same underlying bug or change, and they would not plausibly be one combined PR." + "reason": "Reject: both are typing-related, but 45415 is a broad annotation/type-checking sweep while 45425 adds specific protocol/alias definitions for modeling_utils and PEFT. They overlap in a few files, yet they do not appear to fix the same concrete code-path problem or be mergeable as one single PR." } ] }, @@ -23598,7 +23679,7 @@ "soft_pairs": [ "pull_request:45548|pull_request:45662" ], - "prepared_review_unit_hash": "1a8a81b655f25f9893206166ddc48430429edb1e78c53e03b0e972845014670a", + "prepared_review_unit_hash": "e7be306aa6b9347fc0ba5edd7e136b7e1b13f2bb8cd93c2897f39022ec25ef31", "component_budget": { "node_count": 2, "item_count": 2, @@ -23620,18 +23701,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These two PRs are related by expert-parallel loading, but they fix different underlying bugs: one is about EP + DeepSpeed ZeRO-3 loading/serialization, the other about EP + FSDP2 parameters being overwritten by rank-0 broadcast. They should not be merged as duplicates.", - "confidence": 0.96, + "summary": "These are related EP-loading PRs, but they fix different bugs: one preserves distributed config and exposes `has_ep` for Accelerate/ZeRO-3 loading, while the other fixes expert parameters being overwritten in FSDP2/sharded loading. Not duplicates.", + "confidence": 0.97, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 45662 is the better canonical representative because it targets a concrete EP loading overwrite bug with a clear multi-file fix; its title and diff align tightly with the failure mode described in the code changes.", + "canonical_pr_reason": "PR #45662 is the more concrete representative of the cluster: it targets a specific expert-parameter overwrite path during distributed loading, with broader code-path changes across loading and MoE utilities.", "best_issue_reason": null, - "best_pr_reason": "PR 45662 is the strongest standalone candidate: it fixes a specific post-shard overwrite path and has the most direct, code-path-specific implementation. PR 45548 is narrower and addresses a different loading stack.", + "best_pr_reason": "PR #45662 is the strongest standalone PR for this cluster because it addresses a precise runtime failure mode in the loading pipeline and has the clearest implementation impact.", "soft_edge_verdicts": [ { "left": "pull_request:45548", "right": "pull_request:45662", "accept": false, - "reason": "Both involve EP/loading, but the concrete bugs differ: 45548 is about DeepSpeed ZeRO-3 loading via accelerate launch and config serialization, while 45662 is about experts being overwritten by rank-0 broadcast in FSDP2. Different code paths and fixes, so not duplicates." + "reason": "Different underlying bugs and fix strategies: #45548 serializes `distributed_config` and adds `has_ep` to make EP/ZeRO-3 loading work; #45662 wraps loaded params post-shard to prevent expert weights being overwritten under FSDP2. Shared EP context and one filename are not enough to treat them as the same change." } ] }, @@ -23661,12 +23742,12 @@ "soft_pairs": [ "pull_request:45639|pull_request:45645" ], - "prepared_review_unit_hash": "d0c8b17dceb191e5489959d2d3b0c6c17a5d4bb56f8833a006ad4ae51ef3c47f", + "prepared_review_unit_hash": "ee90f8b30f4b0981c528f8c3190c6cef160e39511180076272234823f6162575", "component_budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 4672, + "serialized_chars": 4671, "estimated_input_tokens": 1168, "estimated_eval_tokens": 2592 }, @@ -23674,7 +23755,7 @@ "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 4672, + "serialized_chars": 4671, "estimated_input_tokens": 1168, "estimated_eval_tokens": 2592 }, @@ -23683,18 +23764,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Both PRs address the same xdist/captured_info collision problem around patched testing debug logs; #45645 looks like the fuller follow-up because it also updates CI log collection to read multiple captured_info files.", - "confidence": 0.88, + "summary": "Both PRs address the same xdist-safe captured_info logging problem in transformers testing utilities; one is the narrower code fix, the other is the broader end-to-end version that also updates CI log collection.", + "confidence": 0.92, "canonical_issue_reason": null, - "canonical_pr_reason": "PR #45645 is the better canonical PR: it covers the same xdist-safe output-file change as #45639 and additionally fixes downstream CI artifact collection, making it the more complete end-to-end fix.", + "canonical_pr_reason": "PR 45645 is the most complete representative: it fixes the xdist collision in `testing_utils.py` and also updates CI to read all `captured_info*.txt` artifacts, matching the full user-facing impact of the bug.", "best_issue_reason": null, - "best_pr_reason": "PR #45645 is the strongest standalone representative because it includes the core testing_utils fix plus the workflow update needed to preserve all captured debug logs under xdist.", + "best_pr_reason": "PR 45639 is the cleanest code-path fix and the best narrow representative if you want the smallest change that directly makes patched testing debug logs xdist-safe.", "soft_edge_verdicts": [ { "left": "pull_request:45639", "right": "pull_request:45645", "accept": true, - "reason": "Same underlying bug: xdist collisions for captured_info debug artifacts. The diff on #45645 extends the same core fix in testing_utils and adds the CI-side handling, so these could plausibly be merged into one PR." + "reason": "Same underlying bug and same core implementation area (`testing_utils.py`); the second PR expands the fix to CI artifact handling, so these could plausibly be merged into one PR." } ] }, @@ -23724,12 +23805,12 @@ "soft_pairs": [ "pull_request:45720|pull_request:46259" ], - "prepared_review_unit_hash": "71c3c8b93542773deaa783dd6d5883e39486dc3bfa896c653edfd4b2c017b7d6", + "prepared_review_unit_hash": "e4c22c8aa34b045e0b82e68fc7104be2185bb3e4638de279e8fd6797995bffc0", "component_budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 7234, + "serialized_chars": 7233, "estimated_input_tokens": 1809, "estimated_eval_tokens": 3874 }, @@ -23737,7 +23818,7 @@ "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 7234, + "serialized_chars": 7233, "estimated_input_tokens": 1809, "estimated_eval_tokens": 3874 }, @@ -23746,18 +23827,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Both PRs are linter-rule implementations, but they target different rules and different code paths: one adds `can_return_tuple`/`return_dict` handling for model forwards, the other refactors weight initialization patterns. They overlap in some model files but are not duplicates.", + "summary": "These two PRs are both linter-rule/modeling cleanup changes, but they target different rules and different underlying code paths: PR 45720 adds TRF003 tuple-return handling across several model forwards, while PR 46259 adds TRF018 initialization-rule changes across many model initializers. They are similar in theme but not duplicates.", "confidence": 0.95, "canonical_issue_reason": null, - "canonical_pr_reason": "PR #46259 is the better canonical representative because it is a broader, more substantive rule implementation with a clear model-initialization refactor across many architectures.", + "canonical_pr_reason": "No single canonical PR as a duplicate representative: the changes implement different lint rules (TRF003 vs TRF018) and would not plausibly merge into one patch without conflating unrelated fixes.", "best_issue_reason": null, - "best_pr_reason": "PR #46259 is the strongest overall PR in this cluster: it addresses a concrete modeling initialization rule change and has the wider, more coherent code impact.", + "best_pr_reason": "pull_request:46259 is the broader/more representative linter-rule PR in this cluster, with wider model coverage and a more systematic refactor pattern, though it is not a duplicate of 45720.", "soft_edge_verdicts": [ { "left": "pull_request:45720", "right": "pull_request:46259", "accept": false, - "reason": "Different linter rules and different underlying fixes: TRF003 changes forward return-tuple behavior, while TRF018 changes model weight initialization. Shared files do not imply the same change." + "reason": "Different lint rules and different bug/change classes: TRF003 tuple-return handling in forward methods vs TRF018 weight-initialization behavior. Shared files are incidental to the same modeling subsystem, not the same underlying fix." } ] }, @@ -23856,52 +23937,52 @@ "pull_request:45932|pull_request:46891", "pull_request:46100|pull_request:46891" ], - "prepared_review_unit_hash": "59201dfc572fe3e5cdca81b9d85ec243f6babe019e6715bcdec1b3afd796a2a4", + "prepared_review_unit_hash": "97193e916be9f9fed300720cbded4e9d6a528a29418d79c62fd8b35845b55d3d", "component_budget": { "node_count": 3, "item_count": 3, "soft_pair_count": 3, - "serialized_chars": 7883, - "estimated_input_tokens": 1971, - "estimated_eval_tokens": 4198 + "serialized_chars": 7880, + "estimated_input_tokens": 1970, + "estimated_eval_tokens": 4196 }, "budget": { "node_count": 3, "item_count": 3, "soft_pair_count": 3, - "serialized_chars": 7883, - "estimated_input_tokens": 1971, - "estimated_eval_tokens": 4198 + "serialized_chars": 7880, + "estimated_input_tokens": 1970, + "estimated_eval_tokens": 4196 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These PRs all touch tests/CI, but they are not the same underlying change. Each has a different target: grafana/telemetry CI wiring and model tweaks, a kernel test migration, and PEFT integration CI. No pair looks mergeable as one PR.", - "confidence": 0.92, + "summary": "All three items are open PRs in the same general testing/CI orbit, but they target different code paths and changes: one is a broad temporary model/CI test PR, one is a test migration with new distributed kernel tests, and one is a PEFT CI integration PR. They are not duplicates of the same underlying fix.", + "confidence": 0.91, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 46891 is the clearest, most self-contained change: it targets PEFT integration CI and related test-fetcher plumbing. The other two are broader or unrelated test/CI experiments.", + "canonical_pr_reason": "No single underlying bug/change ties these PRs together. If one representative must be chosen, #46891 is the most self-contained and focused PR, centered on PEFT CI wiring, whereas #45932 and #46100 are broader and touch unrelated test/model areas.", "best_issue_reason": null, - "best_pr_reason": "PR 46891 is the best representative because it has a specific, coherent purpose and smaller scope than the other two PRs, which are broader test-migration or grafana/telemetry-related changes.", + "best_pr_reason": "#46891 is the best cluster representative because it is the narrowest and most coherent change set. The others are also test/CI-related, but they modify different subsystems and solve different problems.", "soft_edge_verdicts": [ { "left": "pull_request:45932", "right": "pull_request:46100", "accept": false, - "reason": "Both are test/CI oriented, but they modify different things: 45932 is a grafana/telemetry workflow plus model config updates, while 46100 is a kernel-test migration with DDP/test files. Different underlying changes." + "reason": "Both are test-oriented, but #45932 is a broad \"testing grafana\" PR touching model code and CI wiring, while #46100 is a test migration adding kernel/distributed test files. They do not appear to fix the same concrete code-path problem and would not plausibly merge as one PR." }, { "left": "pull_request:45932", "right": "pull_request:46891", "accept": false, - "reason": "They only overlap in the workflow file and general CI context. 45932 is about grafana/telemetry and assorted model edits; 46891 is about PEFT integration CI. Not the same bug or change." + "reason": "#45932 is a broad temporary testing/CI PR with unrelated model changes; #46891 is specifically about PEFT integration CI and a small peft integration code path. Shared workflow edits are not enough to make them the same change." }, { "left": "pull_request:46100", "right": "pull_request:46891", "accept": false, - "reason": "Both touch tests, but one is a kernel test migration and the other is PEFT integration CI plumbing. They address different code paths and would not plausibly be merged into one PR." + "reason": "#46100 adds kernel tests and test-migration files, while #46891 changes PEFT integration CI behavior and a PEFT integration helper. They touch different subsystems and address different test/CI concerns." } ] }, @@ -23931,7 +24012,7 @@ "soft_pairs": [ "pull_request:46126|pull_request:46269" ], - "prepared_review_unit_hash": "e5e15b1a95f15756c61343604ea738a747d36a87e78d73a3fd60f340fbb6c46d", + "prepared_review_unit_hash": "93423926f50836b67b67b53d07eb5027ac34f9154d777be0eee97738f07ec042", "component_budget": { "node_count": 2, "item_count": 2, @@ -23953,18 +24034,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Two open PRs overlap in distributed/EP-related files, but they are not the same change: one is a focused EP + Trainer integration, the other is a broad distributed branch base touching many unrelated areas. I do not treat them as duplicates.", - "confidence": 0.94, + "summary": "These are not duplicates: PR #46126 is a focused EP/trainer integration change around router/tensor-parallel code, while PR #46269 is a broad distributed-branch base with many unrelated distributed and CI edits. The overlapping filenames look like branch overlap, not the same underlying fix.", + "confidence": 0.91, "canonical_issue_reason": null, - "canonical_pr_reason": "PR #46126 is the more focused, self-contained change with a clear concrete purpose (EP + Trainer integration). PR #46269 is a broad umbrella branch with many unrelated distributed edits, so it is less suitable as the representative PR for this cluster.", + "canonical_pr_reason": "PR #46126 is the better canonical PR because it has a single, concrete scope centered on EP routing and trainer integration; PR #46269 is a wide base branch touching many subsystems and is not a single actionable change.", "best_issue_reason": null, - "best_pr_reason": "PR #46126 is the best representative because its title and diff describe one coherent feature/change path, making it easier to understand than the sprawling branch-base PR.", + "best_pr_reason": "PR #46126 is the best representative of the cluster because it is the most specific and technically coherent change, whereas #46269 is too broad and heterogeneous to serve as a clean duplicate target.", "soft_edge_verdicts": [ { "left": "pull_request:46126", "right": "pull_request:46269", "accept": false, - "reason": "Shared distributed files alone are not enough: #46126 is a specific EP/Trainer integration, while #46269 is a broad branch-base PR spanning FSDP, MoE, CI, and multiple core files. They do not appear to be the same underlying bug/change and would not plausibly be merged as one PR." + "reason": "Shared distributed files are not enough to treat these as the same change. #46126 targets EP/router and trainer integration; #46269 is a large branch-base PR spanning CI, FSDP, MoE, loading, and other distributed components, so they would not plausibly merge as one PR." } ] }, @@ -24030,208 +24111,208 @@ "pull_request:46504|pull_request:46563", "pull_request:46504|pull_request:46716" ], - "prepared_review_unit_hash": "4e562897e3e4498a2c92612cf76d120d71d391672ff128e9345f85e6405076ba", + "prepared_review_unit_hash": "7ac72a0ea3921b83f4a6e1096cfa6cd1225313f9074e36ae49fcb35c9439dbfc", "component_budget": { "node_count": 10, "item_count": 10, "soft_pair_count": 29, - "serialized_chars": 25929, - "estimated_input_tokens": 6483, - "estimated_eval_tokens": 13222 + "serialized_chars": 25919, + "estimated_input_tokens": 6480, + "estimated_eval_tokens": 13216 }, "budget": { "node_count": 10, "item_count": 10, "soft_pair_count": 29, - "serialized_chars": 25929, - "estimated_input_tokens": 6483, - "estimated_eval_tokens": 13222 + "serialized_chars": 25919, + "estimated_input_tokens": 6480, + "estimated_eval_tokens": 13216 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "All candidates are small, unrelated workflow-hardening chores. The only strong overlap is that several PRs touch .github/workflows/pr-ci-security-gate.yml, but they change different parts of the workflow and do not look like one mergeable fix.", - "confidence": 0.91, + "summary": "Most items are independent workflow-hygiene PRs touching different files and code paths. The only tight cluster is pr-ci-security-gate.yml, but those PRs make different hardening edits rather than the same concrete fix, so none of the soft edges should merge.", + "confidence": 0.94, "canonical_issue_reason": null, - "canonical_pr_reason": "pull_request:46554 is the most central and substantive of the shared pr-ci-security-gate.yml edits, but it is still distinct from the other workflow changes rather than a true duplicate target.", + "canonical_pr_reason": "PR 46554 is the strongest representative because it makes the core security-gate logic change; the other nearby PRs are smaller, complementary edits (permissions, comments, secrets handling) rather than the same fix.", "best_issue_reason": null, - "best_pr_reason": "pull_request:46554 is the best representative anchor for the cluster because it is the most substantial edit among the related workflow-security-gate PRs, even though the cluster does not contain a genuine duplicate PR set.", + "best_pr_reason": "PR 46554 is the best global representative: it changes the substantive execution path in pr-ci-security-gate.yml, while the others are mostly ancillary workflow tweaks or unrelated maintenance changes.", "soft_edge_verdicts": [ { "left": "pull_request:46138", "right": "pull_request:46554", "accept": false, - "reason": "Different workflow files and different fixes: stdin handling in self-comment-ci vs PR security-gate logic." + "reason": "Different workflows and different behavior changes; only a broad CI/workflow similarity, not the same bug or fix." }, { "left": "pull_request:46138", "right": "pull_request:46558", "accept": false, - "reason": "Unrelated workflows; one fixes comment piping, the other changes PR CI gate permissions/approval logic." + "reason": "Unrelated workflow edits: stdin handling in self-comment CI versus security-gate permissions/approval logic." }, { "left": "pull_request:46138", "right": "pull_request:46563", "accept": false, - "reason": "Different workflow and change type; shell input handling is not the same bug as reusable-workflow security comments." + "reason": "Different file and different concern; one is command input handling, the other is a reusable-workflow/security note change." }, { "left": "pull_request:46138", "right": "pull_request:46716", "accept": false, - "reason": "Same file category only in a broad sense; these are different security-related workflow edits with no shared concrete bug." + "reason": "Different workflow and different code path; one changes CI message passing, the other changes PR security-gate secrets handling." }, { "left": "pull_request:46138", "right": "pull_request:46504", "accept": false, - "reason": "Both harden shell input handling, but in different workflows and different code paths; not one mergeable fix." + "reason": "Both are shell-safety workflow tweaks, but they fix different scripts in different workflows and are not the same underlying issue." }, { "left": "pull_request:46432", "right": "pull_request:46503", "accept": false, - "reason": "SSH runner hardening vs benchmark trigger logic; unrelated workflow changes." + "reason": "Different workflow files and different trigger logic; no shared concrete bug." }, { "left": "pull_request:46432", "right": "pull_request:46501", "accept": false, - "reason": "Different workflow files and distinct changes; SSH runner secrets/network handling is not the same issue as anti-slop permissions." + "reason": "Different workflow maintenance changes in different jobs; similar namespace only." }, { "left": "pull_request:46432", "right": "pull_request:46502", "accept": false, - "reason": "No concrete overlap beyond being workflow chores; one is SSH runner hardening, the other changes reviewer assignment checkout behavior." + "reason": "Unrelated workflow edits: SSH runner hardening versus reviewer assignment checkout behavior." }, { "left": "pull_request:46432", "right": "pull_request:46554", "accept": false, - "reason": "Different workflows and problems: SSH runner secret handling vs PR CI security-gate behavior." + "reason": "Both touch CI/workflow infrastructure, but they address different problems: SSH runner hardening versus PR security-gate logic." }, { "left": "pull_request:46432", "right": "pull_request:46558", "accept": false, - "reason": "SSH runner changes are unrelated to PR CI gate permission/approval updates." + "reason": "Different workflow and different fix; security-gate permissions/approval gating is not the SSH runner hardening issue." }, { "left": "pull_request:46501", "right": "pull_request:46503", "accept": false, - "reason": "Both are workflow tweaks, but they address different jobs and different behavior." + "reason": "Both are workflow updates, but one is anti-slop permissions and the other is benchmark dispatch support; not the same change." }, { "left": "pull_request:46501", "right": "pull_request:46502", "accept": false, - "reason": "Different workflows and fixes: anti-slop permissions vs assign-reviewers checkout ref." + "reason": "Different workflows with unrelated behavior changes." }, { "left": "pull_request:46501", "right": "pull_request:46504", "accept": false, - "reason": "No same underlying bug; one adjusts anti-slop permissions, the other changes shell handling in build-ci-docker-images." + "reason": "Different workflow files and different shell/input handling fixes; only superficial similarity." }, { "left": "pull_request:46501", "right": "pull_request:46554", "accept": false, - "reason": "Different workflow files and unrelated fixes." + "reason": "Security-gate logic is unrelated to anti-slop permission scoping." }, { "left": "pull_request:46502", "right": "pull_request:46503", "accept": false, - "reason": "Reviewer assignment checkout ref vs benchmark workflow trigger expansion are unrelated." + "reason": "Different workflow jobs and different fixes; no shared underlying bug." }, { "left": "pull_request:46502", "right": "pull_request:46504", "accept": false, - "reason": "Different workflow concerns; not the same bug or change." + "reason": "Reviewer checkout ref change is unrelated to Docker-image build command quoting." }, { "left": "pull_request:46502", "right": "pull_request:46554", "accept": false, - "reason": "Distinct workflow edits with no shared concrete code-path problem." + "reason": "Different workflow infrastructure changes with no common code-path problem." }, { "left": "pull_request:46503", "right": "pull_request:46504", "accept": false, - "reason": "Benchmark trigger logic and docker-image build message parsing are different workflow behaviors." + "reason": "Benchmark workflow trigger change versus build-image shell quoting; unrelated." }, { "left": "pull_request:46503", "right": "pull_request:46554", "accept": false, - "reason": "Different workflows and unrelated fixes." + "reason": "Different workflows and different functionality; not a duplicate fix." }, { "left": "pull_request:46504", "right": "pull_request:46554", "accept": false, - "reason": "Both are workflow hardening, but they affect different jobs and unrelated execution paths." + "reason": "Both are CI/workflow edits, but they fix different scripts and different jobs; not one mergeable change." }, { "left": "pull_request:46504", "right": "pull_request:46558", "accept": false, - "reason": "Shell quoting in build-ci-docker-images is unrelated to PR CI security-gate permissions/approval logic." + "reason": "Build-script input handling versus security-gate permissions/approval logic are separate concerns." }, { "left": "pull_request:46554", "right": "pull_request:46558", "accept": false, - "reason": "Same workflow file, but different changes: checkout/diff scanning versus permissions and approval gating." + "reason": "Same file, but different changes: 46554 adds checkout/diff-based security checks, while 46558 adjusts permissions and gates approval on success; complementary, not the same fix." }, { "left": "pull_request:46554", "right": "pull_request:46563", "accept": false, - "reason": "Same workflow file, but one adds a checkout/diff check and the other only documents the reusable workflow safety." + "reason": "46563 is just a reusable-workflow/comment clarification, not the same code-path change as 46554's security checks." }, { "left": "pull_request:46554", "right": "pull_request:46716", "accept": false, - "reason": "Both touch the security-gate workflow, but one changes the gate logic while the other removes inherited secrets; not the same fix." + "reason": "Secrets scoping and security-check logic are related thematically but are different concrete changes." }, { "left": "pull_request:46558", "right": "pull_request:46563", "accept": false, - "reason": "Same file but different edits: permission scoping/approval condition vs explanatory comments." + "reason": "Permissions/approval gating is distinct from the reusable-workflow safety comment update." }, { "left": "pull_request:46558", "right": "pull_request:46716", "accept": false, - "reason": "Different concrete changes in the same workflow file; not one duplicate bug." + "reason": "Both are security-gate hardening, but one changes permissions/approval flow and the other changes secrets inheritance; not the same fix." }, { "left": "pull_request:46563", "right": "pull_request:46716", "accept": false, - "reason": "One is documentation/commentary, the other changes secrets handling; clearly separate changes." + "reason": "Comment-only safety note versus actual secrets scoping change; different kinds of edits." }, { "left": "pull_request:46504", "right": "pull_request:46563", "accept": false, - "reason": "Different workflows and different fixes; no shared underlying code-path problem." + "reason": "Unrelated workflow edits; only superficial similarity in being small CI changes." }, { "left": "pull_request:46504", "right": "pull_request:46716", "accept": false, - "reason": "Build-image shell parsing vs PR CI security-gate secrets handling are unrelated." + "reason": "Different workflow and different problem domain; build-image command quoting is unrelated to PR CI secrets scoping." } ] }, @@ -24261,12 +24342,12 @@ "soft_pairs": [ "pull_request:46152|pull_request:46207" ], - "prepared_review_unit_hash": "452ad072d7b10f13c4dcefee8fafc2c4d718baabaac807ef1210fcec8b1e79aa", + "prepared_review_unit_hash": "f95db07950c690999a7689a2013711ae32fbb02d9e4e34cdf1ea69d2a8d89e19", "component_budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 4276, + "serialized_chars": 4275, "estimated_input_tokens": 1069, "estimated_eval_tokens": 2394 }, @@ -24274,7 +24355,7 @@ "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 4276, + "serialized_chars": 4275, "estimated_input_tokens": 1069, "estimated_eval_tokens": 2394 }, @@ -24283,18 +24364,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Both PRs address reading Python source without raw file open(), but they fix different code paths: one in import_utils for resolving package resources, the other in modeling_utils for attention-implementation detection. They are related in theme, not duplicates.", - "confidence": 0.82, + "summary": "Both PRs modernize source inspection to avoid `open()`, but they target different subsystems and distinct code paths, so they are not duplicates.", + "confidence": 0.94, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 46207 is the more focused, self-contained fix: it replaces file-open source inspection in a specific runtime check and includes test coverage, making it a cleaner representative of the source-reading bug class.", + "canonical_pr_reason": "No single canonical PR: 46152 changes `import_utils.py` for traversable resolution, while 46207 changes `modeling_utils.py` for `_can_set_attn_implementation`; they solve different bugs.", "best_issue_reason": null, - "best_pr_reason": "PR 46207 is the best representative because it targets a concrete behavior in one helper, has clearer intent in the title, and appears less infrastructural than PR 46152.", + "best_pr_reason": "46207 is the cleaner representative if one must be chosen because it is a focused, test-backed fix for one concrete code path; 46152 addresses a separate helper in a different module.", "soft_edge_verdicts": [ { "left": "pull_request:46152", "right": "pull_request:46207", "accept": false, - "reason": "Different underlying fixes in different modules: 46152 changes package resource resolution in import_utils, while 46207 changes source inspection for _can_set_attn_implementation in modeling_utils. Similar theme, but not the same code-path problem." + "reason": "Same broad theme (avoid `open()`), but different modules, different functions, and different behaviors; they are not the same underlying change and would not be merged into one PR." } ] }, @@ -24387,12 +24468,12 @@ "soft_pairs": [ "pull_request:46281|pull_request:46890" ], - "prepared_review_unit_hash": "634cb8e05c2824626c842a0f4228e157266cad20026beb1831f5b9e6f2d053fb", + "prepared_review_unit_hash": "f27a94c190d9a11207de7366a1a4e4b0859090bc0a8e2e598d3be1470bb0146b", "component_budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 4271, + "serialized_chars": 4270, "estimated_input_tokens": 1068, "estimated_eval_tokens": 2392 }, @@ -24400,7 +24481,7 @@ "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 4271, + "serialized_chars": 4270, "estimated_input_tokens": 1068, "estimated_eval_tokens": 2392 }, @@ -24409,18 +24490,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Two security-hardening PRs in adjacent Transformers code paths, both about path traversal containment, but they fix different untrusted inputs and are not duplicates.", - "confidence": 0.95, + "summary": "Both PRs harden path handling against traversal, but they fix different inputs in different code paths: one guards sharded checkpoint filenames from the index `weight_map`, the other guards `transformers_weights` in model config resolution. They are related security hardening, not duplicates.", + "confidence": 0.97, "canonical_issue_reason": null, - "canonical_pr_reason": "PR #46281 is the better representative because it is the more concrete and narrowly scoped fix: it hardens sharded-checkpoint loading against traversal through the checkpoint index `weight_map` in a specific loader path.", + "canonical_pr_reason": "46281 is the cleaner representative of the cluster theme: it fixes a concrete checkpoint-shard loading path and the containment check is explicit and well-scoped. Still, it is not a duplicate of 46890.", "best_issue_reason": null, - "best_pr_reason": "PR #46281 is the best overall fit for the cluster because it addresses the clearest, most self-contained bug with an explicit containment check and minimal scope; PR #46890 is related but targets a different config field and code path.", + "best_pr_reason": "46281 is the better standalone PR because the bug is narrowly defined, the affected path is clear, and the fix is self-contained. 46890 addresses a separate config-driven file-resolution path.", "soft_edge_verdicts": [ { "left": "pull_request:46281", "right": "pull_request:46890", "accept": false, - "reason": "Both are path-traversal fixes, but they guard different untrusted sources (`weight_map` vs `transformers_weights`) in different loading paths. They are related hardening work, not the same underlying bug or a mergeable single PR." + "reason": "Different underlying bugs and different code paths: `get_checkpoint_shard_files` uses index `weight_map` entries, while `_get_resolved_checkpoint_files` uses `transformers_weights`. Same vulnerability class, but not the same fix or mergeable into one PR." } ] }, @@ -24450,12 +24531,12 @@ "soft_pairs": [ "pull_request:46310|pull_request:46546" ], - "prepared_review_unit_hash": "61cb7d896af8b593eb7bac617411cd6056997a8e22452ca95f3b95c93995661a", + "prepared_review_unit_hash": "7aaab9e0aeda2c627210bb0d5640ca3a07432b6a9d3922f7580e39a595676187", "component_budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 4851, + "serialized_chars": 4849, "estimated_input_tokens": 1213, "estimated_eval_tokens": 2682 }, @@ -24463,7 +24544,7 @@ "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 4851, + "serialized_chars": 4849, "estimated_input_tokens": 1213, "estimated_eval_tokens": 2682 }, @@ -24472,18 +24553,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Both PRs address the same off-by-one right-window boundary in Moonshine streaming masks, but PR 46546 also extends the fix to Gemma4, so it is broader than PR 46310 and not a clean duplicate.", - "confidence": 0.78, + "summary": "Both PRs fix the same off-by-one bug in Moonshine streaming\u2019s right-side sliding window mask (`-dist <= right_window_size` vs `<`). PR 46546 is the more complete representation because it includes the Moonshine fix plus tests and an additional analogous Gemma4 fix.", + "confidence": 0.92, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 46546 is the broader and more complete fix: it includes the Moonshine streaming boundary correction plus the same boundary adjustment in Gemma4 and added test coverage.", + "canonical_pr_reason": "PR 46546 is the better canonical PR because it contains the Moonshine streaming boundary fix, adds test coverage, and is the more complete patch set.", "best_issue_reason": null, - "best_pr_reason": "PR 46546 best represents the underlying fix because it covers the Moonshine streaming bug and the related Gemma4 instance, making it the most complete candidate in this cluster.", + "best_pr_reason": "PR 46546 best represents the cluster since it covers the same Moonshine streaming bug as 46310 and includes validation plus the related Gemma4 adjustment.", "soft_edge_verdicts": [ { "left": "pull_request:46310", "right": "pull_request:46546", - "accept": false, - "reason": "They share the same Moonshine streaming off-by-one mask fix, but 46546 also changes Gemma4, so the PRs are not the same concrete change and should not be deduplicated." + "accept": true, + "reason": "Same underlying Moonshine streaming right-window boundary bug; the diff is the same off-by-one correction in the same mask code path, and 46546 is essentially a superset of 46310 for that fix." } ] }, @@ -24522,12 +24603,12 @@ "pull_request:46425|pull_request:46850", "pull_request:46850|pull_request:46917" ], - "prepared_review_unit_hash": "b599cab00e4b0092c93792878682a219e4bbf8ba5c415f11233f2e17be32fd76", + "prepared_review_unit_hash": "613d36cd047c28497d745197401c5023aff1e5e68ff677a57055047cf3f5090c", "component_budget": { "node_count": 6, "item_count": 6, "soft_pair_count": 6, - "serialized_chars": 16837, + "serialized_chars": 16838, "estimated_input_tokens": 4210, "estimated_eval_tokens": 8676 }, @@ -24535,7 +24616,7 @@ "node_count": 6, "item_count": 6, "soft_pair_count": 6, - "serialized_chars": 16837, + "serialized_chars": 16838, "estimated_input_tokens": 4210, "estimated_eval_tokens": 8676 }, @@ -24544,48 +24625,48 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "This cluster is heterogeneous: it mixes unrelated feature PRs for continuous batching multimodal support, multiple new model integrations (HunYuan VL, RADIO, Nemotron Omni), a Gemma masking bug fix, and a deprecation cleanup. The soft-similarity links are not true duplicates.", - "confidence": 0.98, + "summary": "These PRs are not duplicates of one another. They span unrelated changes: multimodal continuous batching, HunYuan VL model support, RADIO model support, Nemotron Omni support, Gemma masking fixes, and deprecated API removal. The only similarities are broad feature-addition patterns and shared framework files, which are not enough to merge PRs.", + "confidence": 0.97, "canonical_issue_reason": null, - "canonical_pr_reason": "No single canonical PR is safe here; the items do not describe the same underlying change, so there is no real duplicate anchor. If a weak representative is needed, 46425 is the most central by similarity, but only as a cluster exemplar.", + "canonical_pr_reason": null, "best_issue_reason": null, - "best_pr_reason": "pull_request:46425 is the broadest and most central-looking item in the similarity graph, but it is still only a weak representative; the cluster is too mixed to treat any PR as a true canonical duplicate.", + "best_pr_reason": "PR 46425 is the most self-contained model-support addition and the closest thing to a cluster representative, but it is not a true duplicate canonical target because the set is heterogeneous.", "soft_edge_verdicts": [ { "left": "pull_request:46376", "right": "pull_request:46417", "accept": false, - "reason": "Continuous batching multimodal support vs. a new HunYuan VL model integration/doc PR; different features and code paths." + "reason": "Different features and code paths: continuous batching multimodal support vs adding a new HunYuan VL model package." }, { "left": "pull_request:46376", "right": "pull_request:46425", "accept": false, - "reason": "Continuous batching multimodal support vs. RADIO model support; both touch Transformers internals but are unrelated changes." + "reason": "Both are large feature additions, but one changes continuous batching internals while the other adds a specific vision-language model; no shared bug or mergeable code-path." }, { "left": "pull_request:46425", "right": "pull_request:46509", "accept": false, - "reason": "RADIO model addition and Nemotron Omni support are separate model implementations, not the same bug or change." + "reason": "Distinct model integrations for different architectures (RADIO vs Nemotron Omni); same broad subsystem only, not the same change." }, { "left": "pull_request:46425", - "right": "pull_request:46850", + "right": "pull_request:46917", "accept": false, - "reason": "RADIO model support vs. Gemma masking logic fix; different subsystems and no shared concrete failure." + "reason": "RADIO model support and deprecated API cleanup are unrelated; no common underlying bug or implementation target." }, { "left": "pull_request:46425", - "right": "pull_request:46917", + "right": "pull_request:46850", "accept": false, - "reason": "RADIO model support vs. deprecation/removal cleanup in image processing and training args; not the same underlying issue." + "reason": "Model addition versus a Gemma attention-masking bug fix; completely different concrete behavior and files." }, { "left": "pull_request:46850", "right": "pull_request:46917", "accept": false, - "reason": "Gemma sliding-window masking bug fix vs. deprecated API cleanup; completely different code-path problems." + "reason": "Gemma sliding-window masking fix and deprecated training/image-backend cleanup are unrelated changes." } ] }, @@ -24615,40 +24696,40 @@ "soft_pairs": [ "pull_request:46574|pull_request:46580" ], - "prepared_review_unit_hash": "d4b57953e761df970d44d3d19e82f811b4bb0e91cc1d373e6f494dbc0cea40dd", + "prepared_review_unit_hash": "fd347c25ad1076d0ee8a6814c2a5defa5c91f0eda8a2578129a2047a972aced6", "component_budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 4409, - "estimated_input_tokens": 1103, - "estimated_eval_tokens": 2462 + "serialized_chars": 4408, + "estimated_input_tokens": 1102, + "estimated_eval_tokens": 2460 }, "budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 4409, - "estimated_input_tokens": 1103, - "estimated_eval_tokens": 2462 + "serialized_chars": 4408, + "estimated_input_tokens": 1102, + "estimated_eval_tokens": 2460 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "The two PRs are unrelated tokenizer fixes: one corrects CodeLlama decode behavior, the other fixes DeepSeek checkpoint-to-tokenizer-class resolution.", - "confidence": 0.97, + "summary": "These two PRs fix different tokenizer-related problems in different models: CodeLlama decode-space stripping vs. DeepSeek-Coder/DeepSeek-LLM auto-tokenizer misclassification. They are not duplicates and should not be merged.", + "confidence": 0.98, "canonical_issue_reason": null, - "canonical_pr_reason": "No single canonical PR: 46574 fixes CodeLlama leading-space stripping on decode, while 46580 fixes wrong tokenizer class loading for specific DeepSeek checkpoints.", + "canonical_pr_reason": "No single canonical PR: the changes address separate bugs in different code paths and model families.", "best_issue_reason": null, - "best_pr_reason": "PR 46574 is the better representative of its own bug fix because it is a focused code-path correction with a targeted test, whereas 46580 is a broader auto-tokenizer routing change.", + "best_pr_reason": "Neither PR is a global canonical choice because they solve unrelated issues; each is only best for its own specific bug.", "soft_edge_verdicts": [ { "left": "pull_request:46574", "right": "pull_request:46580", "accept": false, - "reason": "Different underlying bugs and code paths: CodeLlama decode stripping vs DeepSeek checkpoint tokenizer-class selection; not mergeable as one PR." + "reason": "Different failure modes and code paths: CodeLlama decoder setup vs. AutoTokenizer checkpoint-prefix override for DeepSeek models. Not the same underlying change." } ] }, @@ -24678,12 +24759,12 @@ "soft_pairs": [ "pull_request:46603|pull_request:46604" ], - "prepared_review_unit_hash": "06bebfeb2b097f117f9b3f1e85b4e9fba5d304e882be3f4828fdb92b9c6064e3", + "prepared_review_unit_hash": "cc33ca8c203482762107b8b95bdb103ffae57774d83082ea5263be55e87d3f90", "component_budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 5495, + "serialized_chars": 5496, "estimated_input_tokens": 1374, "estimated_eval_tokens": 3004 }, @@ -24691,7 +24772,7 @@ "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 5495, + "serialized_chars": 5496, "estimated_input_tokens": 1374, "estimated_eval_tokens": 3004 }, @@ -24700,18 +24781,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Both PRs are Mistral-related, but they address different changes: one refactors/moves the MistralConverter into a package, while the other adds tekken tokenizer detection, conversion, and save utilities. Shared files are mostly the same integration surface, not the same underlying fix.", - "confidence": 0.96, + "summary": "Two related but distinct Mistral PRs: one moves MistralConverter into a dedicated integrations package, the other adds tekken tokenizer detection/conversion/save utilities. They overlap on files and subsystem but do not appear to be the same fix or mergeable into one PR as-is.", + "confidence": 0.94, "canonical_issue_reason": null, - "canonical_pr_reason": "No single canonical PR: these are separate changes with different goals and code paths, not duplicates of one another.", + "canonical_pr_reason": "PR #46604 is the better representative of the cluster because it introduces the broader tokenizer functionality (detection, conversion, and save utilities) rather than only relocating existing converter code.", "best_issue_reason": null, - "best_pr_reason": "PR 46604 is the more substantive feature/change set and the better representative of the Mistral tokenizer work, though it is not a duplicate of 46603.", + "best_pr_reason": "PR #46604 is the strongest standalone change set: it covers the concrete new Mistral tokenizer behavior, while #46603 is mostly a refactor/package move.", "soft_edge_verdicts": [ { "left": "pull_request:46603", "right": "pull_request:46604", "accept": false, - "reason": "Different underlying changes: 46603 is a package refactor for MistralConverter; 46604 adds new tekken tokenizer functionality and utilities. They share subsystem files but are not mergeable as one duplicate PR." + "reason": "Reject: #46603 is a code move/refactor for MistralConverter, while #46604 adds tekken tokenizer detection, conversion, and save utilities. They share Mistral files but address different underlying changes and would not plausibly be merged as one PR." } ] }, @@ -24733,56 +24814,64 @@ "source": "llm", "cache_hit": false, "model": "gpt-5.4-mini?service_tier=flex", - "cluster_id": "cluster-46741-3", + "cluster_id": "cluster-46741-4", "nodes": [ "pull_request:46741", "pull_request:46773", - "pull_request:46781" + "pull_request:46781", + "pull_request:46937" ], "soft_pairs": [ "pull_request:46741|pull_request:46773", - "pull_request:46741|pull_request:46781" + "pull_request:46741|pull_request:46781", + "pull_request:46773|pull_request:46937" ], - "prepared_review_unit_hash": "14fc807002fc8270b71475adb931e5df05d3de50d191106199a8c759c3e1cb83", + "prepared_review_unit_hash": "540b0506cea828d90cce328a1ba3976f6e7719d81844620af3dbddf39f8cdabb", "component_budget": { - "node_count": 3, - "item_count": 3, - "soft_pair_count": 2, - "serialized_chars": 8452, - "estimated_input_tokens": 2113, - "estimated_eval_tokens": 4482 + "node_count": 4, + "item_count": 4, + "soft_pair_count": 3, + "serialized_chars": 11149, + "estimated_input_tokens": 2788, + "estimated_eval_tokens": 5832 }, "budget": { - "node_count": 3, - "item_count": 3, - "soft_pair_count": 2, - "serialized_chars": 8452, - "estimated_input_tokens": 2113, - "estimated_eval_tokens": 4482 + "node_count": 4, + "item_count": 4, + "soft_pair_count": 3, + "serialized_chars": 11149, + "estimated_input_tokens": 2788, + "estimated_eval_tokens": 5832 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These are three separate pull requests in the cache/attention space, but they fix different bugs: chunked-prefill/speculative decoding state handling, padding-mask application at batch size 1, and QuantizedCache initialization for hybrid models. They should not be clustered as duplicates.", - "confidence": 0.97, + "summary": "Four open PRs cluster around hybrid/linear-attention model behavior, but they address different bugs: chunked-prefill/speculative decoding cache handling, padding-mask application for batch size 1, QuantizedCache construction for hybrid models, and assisted-generation statefulness. No duplicate pair is strong enough to merge.", + "confidence": 0.93, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 46741 is the best cluster representative because it targets a concrete state-handling bug in the Mamba2 chunked-prefill/speculative-decoding path and spans several affected model implementations, making it the broadest and most central code change here.", + "canonical_pr_reason": "PR #46741 is the strongest representative: it fixes a concrete cache/state handling bug in chunked-prefill/speculative decoding and applies it across multiple affected hybrid models with targeted code changes and tests.", "best_issue_reason": null, - "best_pr_reason": "PR 46741 is the strongest standalone fix: it addresses a specific runtime bug in cache/state reuse during chunked prefill/speculative decoding, with direct model-code changes and regression coverage across affected models.", + "best_pr_reason": "PR #46741 is the best standalone change in the cluster because it targets a specific runtime failure mode in a well-defined code path and has the broadest, most coherent fix surface across affected models.", "soft_edge_verdicts": [ { "left": "pull_request:46741", "right": "pull_request:46773", "accept": false, - "reason": "Different bugs and code paths: 46741 changes precomputed state reuse for chunked prefill/speculative decoding, while 46773 fixes padding-mask application being skipped for batch-size-1 inputs. Shared model filenames are incidental." + "reason": "Both touch Bamba and other linear-attention models, but #46741 fixes precomputed-state handling for chunked prefill/speculative decoding, while #46773 fixes padding-mask application when batch size is 1. Different bug, different code path." }, { "left": "pull_request:46741", "right": "pull_request:46781", "accept": false, - "reason": "Different failures: 46741 fixes cache/state handling in model forward paths, while 46781 fixes QuantizedCache layer construction for hybrid SSM+attention models. They are not the same underlying change and would not merge into one PR." + "reason": "#46741 changes model forward/cache-state logic in specific architectures; #46781 changes QuantizedCache layer construction in cache_utils for hybrid models. Related subsystem, but not the same underlying defect." + }, + { + "left": "pull_request:46773", + "right": "pull_request:46937", + "accept": false, + "reason": "#46773 addresses padding-mask skipping; #46937 marks LFM2/LFM2-MoE as stateful to reject assisted generation. These are unrelated behavior fixes." } ] }, @@ -24815,12 +24904,12 @@ "pull_request:46895|pull_request:46916", "pull_request:46898|pull_request:46916" ], - "prepared_review_unit_hash": "85e0ae23309498b13b1637ceab3971a8913338ceabb2b1d5192afab01e5f6c3b", + "prepared_review_unit_hash": "aedf8d5c5e1ff1a317506bb871751cfc19777863aba87605ad51c3cde63b6f03", "component_budget": { "node_count": 3, "item_count": 3, "soft_pair_count": 3, - "serialized_chars": 7715, + "serialized_chars": 7716, "estimated_input_tokens": 1929, "estimated_eval_tokens": 4114 }, @@ -24828,7 +24917,7 @@ "node_count": 3, "item_count": 3, "soft_pair_count": 3, - "serialized_chars": 7715, + "serialized_chars": 7716, "estimated_input_tokens": 1929, "estimated_eval_tokens": 4114 }, @@ -24837,30 +24926,30 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These are related fixes for the same double-shift training-loss pattern, but only the model-specific overlaps are true duplicates. The cross-model PR is the broadest representative, while CohereASR and Florence2 are separate code paths.", - "confidence": 0.91, + "summary": "All three PRs address the same broad training-loss double-shift pattern, but they are model-specific patches and not true duplicates. PR 46916 is the best representative because it spans both CohereASR and Florence2 and subsumes the other two fixes.", + "confidence": 0.83, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 46916 is the most representative canonical PR because it subsumes both concrete fixes and touches both affected model implementations, making it broader than the single-model variants.", + "canonical_pr_reason": "46916 is the broadest and most representative patch: it covers both CohereASR and Florence2 training-loss double-shift fixes, while the others each cover only one model.", "best_issue_reason": null, - "best_pr_reason": "PR 46916 is the strongest overall PR candidate: it addresses the same double-shift loss bug in both CohereASR and Florence2, so it is the most comprehensive fix in the cluster.", + "best_pr_reason": "46916 best captures the cluster theme and includes both code paths, making it the strongest anchor PR.", "soft_edge_verdicts": [ { "left": "pull_request:46895", "right": "pull_request:46898", "accept": false, - "reason": "Both fix the same bug pattern, but in different models/code paths (CohereASR vs Florence2). They are related, not the same underlying change, so they should not be merged as duplicates." + "reason": "Same symptom, but different model code paths (CohereASR vs Florence2); these are parallel model-specific fixes, not one duplicate change." }, { "left": "pull_request:46895", "right": "pull_request:46916", - "accept": true, - "reason": "46916 includes the same CohereASR double-shift loss fix and touches the same CohereASR files, so this is the same underlying change for that code path." + "accept": false, + "reason": "46916 is a broader superset patch that also changes Florence2; it overlaps on CohereASR but is not the same single concrete PR change." }, { "left": "pull_request:46898", "right": "pull_request:46916", - "accept": true, - "reason": "46916 includes the same Florence2 double-shift loss fix and touches the same Florence2 files, so this is the same underlying change for that code path." + "accept": false, + "reason": "46916 includes extra CohereASR work in addition to Florence2, so it is not just a duplicate of the Florence2-only fix." } ] }, @@ -24902,94 +24991,94 @@ "pull_request:46926|pull_request:46928", "pull_request:46927|pull_request:46928" ], - "prepared_review_unit_hash": "1c5440da5337902e1237e329d7d17a696b28377fe86c4f55b5468a58e35c642b", + "prepared_review_unit_hash": "e0a57655ac9c6e71cdce62aa886fdff05dd64e5a1f8f002fa8f66586c9c1180e", "component_budget": { "node_count": 5, "item_count": 5, "soft_pair_count": 10, - "serialized_chars": 11925, - "estimated_input_tokens": 2982, - "estimated_eval_tokens": 6220 + "serialized_chars": 11920, + "estimated_input_tokens": 2980, + "estimated_eval_tokens": 6216 }, "budget": { "node_count": 5, "item_count": 5, "soft_pair_count": 10, - "serialized_chars": 11925, - "estimated_input_tokens": 2982, - "estimated_eval_tokens": 6220 + "serialized_chars": 11920, + "estimated_input_tokens": 2980, + "estimated_eval_tokens": 6216 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These PRs are all superficially similar test/fix patches, but they target different models and different failure modes. None appear to be duplicates of the same underlying change, so this cluster should not be collapsed into a single canonical PR.", - "confidence": 0.98, + "summary": "These PRs are not duplicates despite similar triage labels and soft similarity: each changes a different model or subsystem and fixes a different failure mode. The cluster is a set of unrelated model-specific test/load fixes, not one underlying bug.", + "confidence": 0.94, "canonical_issue_reason": null, - "canonical_pr_reason": "No canonical PR: each PR fixes a different model or code path (EdgeTam kwarg cleanup, DeepSeek/MiniMax offload-folder load issues, GLM46V input dtype cast, BLIP2 tokenizer registration).", + "canonical_pr_reason": null, "best_issue_reason": null, - "best_pr_reason": "None of these is a better duplicate representative than the others; they are separate fixes and should remain distinct.", + "best_pr_reason": "No single PR is a good cluster representative: the changes target different code paths (EdgeTAM forward kwargs, DeepSeek/MiniMax load-time offloading, GLM46V input_id dtype handling, and BLIP-2 tokenizer registration).", "soft_edge_verdicts": [ { "left": "pull_request:46907", "right": "pull_request:46927", "accept": false, - "reason": "Different models and bugs: EdgeTam drops an invalid processor kwarg, while GLM46V casts input_ids to long in forward." + "reason": "Different models and fixes: EdgeTAM strips an invalid processor kwarg before backbone forward, while GLM46V adds input_id casting in its forward method. Not the same bug or mergeable into one PR." }, { "left": "pull_request:46907", "right": "pull_request:46926", "accept": false, - "reason": "Different models and failure modes; one fixes EdgeTam backbone kwargs, the other changes test loading for MiniMax-M3-VL." + "reason": "Different model families and failure modes: EdgeTAM integration-test import/config issue vs MiniMax-M3-VL load_error workaround via offload_folder." }, { "left": "pull_request:46907", "right": "pull_request:46908", "accept": false, - "reason": "EdgeTam runtime kwarg cleanup is unrelated to DeepSeekV3.2 test offload-folder setup." + "reason": "EdgeTAM code-path fix vs DeepSeek-V3.2 test harness load_error fix; unrelated models and unrelated runtime problems." }, { "left": "pull_request:46907", "right": "pull_request:46928", "accept": false, - "reason": "Unrelated changes: EdgeTam model forward path vs BLIP2 tokenizer auto-registration." + "reason": "EdgeTAM forward-kwargs fix is unrelated to BLIP-2 tokenizer auto-registration; different subsystem and different bug." }, { "left": "pull_request:46908", "right": "pull_request:46926", "accept": false, - "reason": "Both add offload_folder in tests, but for different models and separate load_error cases; not the same concrete bug." + "reason": "Both are load_error-related test fixes, but for different models and different loaders/offload behavior. Not the same underlying issue." }, { "left": "pull_request:46908", "right": "pull_request:46928", "accept": false, - "reason": "DeepSeek load_error test setup is unrelated to BLIP2 tokenizer mapping." + "reason": "DeepSeek load-time offloading workaround is unrelated to BLIP-2 tokenizer mapping." }, { "left": "pull_request:46908", "right": "pull_request:46927", "accept": false, - "reason": "Different code paths: DeepSeek test harness offloading vs GLM46V forward dtype normalization." + "reason": "DeepSeek load_error workaround vs GLM46V dtype conversion in forward; different code paths and failure modes." }, { "left": "pull_request:46926", "right": "pull_request:46927", "accept": false, - "reason": "MiniMax test offload-folder fix and GLM46V input dtype fix are different underlying issues." + "reason": "MiniMax-M3-VL offload_folder test fix is unrelated to GLM46V forward input_id casting." }, { "left": "pull_request:46926", "right": "pull_request:46928", "accept": false, - "reason": "Separate model-specific fixes; test offloading is not the same as BLIP2 tokenizer registration." + "reason": "MiniMax load_error workaround vs BLIP-2 tokenizer registry update; no shared underlying bug." }, { "left": "pull_request:46927", "right": "pull_request:46928", "accept": false, - "reason": "GLM46V forward input_ids cast is unrelated to BLIP2 auto tokenizer list update." + "reason": "GLM46V input_id long-casting fix is unrelated to BLIP-2 auto tokenizer support." } ] }, diff --git a/analysis/current/manifest.json b/analysis/current/manifest.json index d6027e725d32f1a77218f796151360de0f6e95d5..8298879a794bc6d7f694f65d6f46be9d012a626c 100644 --- a/analysis/current/manifest.json +++ b/analysis/current/manifest.json @@ -1,8 +1,8 @@ { - "analysis_id": "hybrid-model-20260627t180026z", + "analysis_id": "hybrid-model-20260628t000017z", "archived_artifacts": { - "hybrid": "snapshots/20260627T180026Z/analysis-runs/hybrid-model-20260627t180026z/analysis-report-hybrid.json", - "hybrid_reviews": "snapshots/20260627T180026Z/analysis-runs/hybrid-model-20260627t180026z/analysis-report-hybrid.llm-reviews.json" + "hybrid": "snapshots/20260628T000017Z/analysis-runs/hybrid-model-20260628t000017z/analysis-report-hybrid.json", + "hybrid_reviews": "snapshots/20260628T000017Z/analysis-runs/hybrid-model-20260628t000017z/analysis-report-hybrid.llm-reviews.json" }, "artifacts": { "hybrid": "analysis/current/analysis-report-hybrid.json", @@ -10,9 +10,9 @@ }, "channel": "canonical", "model": null, - "published_at": "2026-06-27T18:10:36Z", + "published_at": "2026-06-28T00:10:09Z", "repo": "huggingface/transformers", "schema_version": 1, - "snapshot_id": "20260627T180026Z", + "snapshot_id": "20260628T000017Z", "variant": "hybrid" } diff --git a/snapshots/20260628T000017Z/analysis-runs/hybrid-model-20260628t000017z/analysis-report-hybrid.json b/snapshots/20260628T000017Z/analysis-runs/hybrid-model-20260628t000017z/analysis-report-hybrid.json new file mode 100644 index 0000000000000000000000000000000000000000..128cbd951af2cb041e4e5068b05dab7f764e8398 --- /dev/null +++ b/snapshots/20260628T000017Z/analysis-runs/hybrid-model-20260628t000017z/analysis-report-hybrid.json @@ -0,0 +1,2698 @@ +{ + "schema_version": "1.0", + "repo": "huggingface/transformers", + "snapshot_id": "20260628T000017Z", + "generated_at": "2026-06-28T00:10:07Z", + "evidence_quality": "full", + "llm_enrichment": true, + "meta_bugs": [ + { + "cluster_id": "cluster-43979-11", + "summary": "Cluster of 1 issues and 10 PRs centered on issue #43979.", + "status": "open", + "confidence": 0.8, + "canonical_issue_number": 43979, + "canonical_pr_number": 44007, + "issue_numbers": [ + 43979 + ], + "pr_numbers": [ + 43996, + 44007, + 44013, + 44044, + 44066, + 44072, + 44085, + 44129, + 44154, + 44722 + ], + "evidence_types": [ + "closing_reference", + "shared_issue_target", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 43996, + "right_pr_number": 44007, + "code_similarity": 0.179, + "size_similarity": 0.576, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.429, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 43996, + "right_pr_number": 44013, + "code_similarity": 0.122, + "size_similarity": 0.318, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.392, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 43996, + "right_pr_number": 44044, + "code_similarity": 0.245, + "size_similarity": 0.864, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.479, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 43996, + "right_pr_number": 44066, + "code_similarity": 0.225, + "size_similarity": 0.818, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.408, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 43996, + "right_pr_number": 44072, + "code_similarity": 0.14, + "size_similarity": 0.303, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.528, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 43996, + "right_pr_number": 44085, + "code_similarity": 0.216, + "size_similarity": 0.783, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.398, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 43996, + "right_pr_number": 44129, + "code_similarity": 0.163, + "size_similarity": 0.643, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.229, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 43996, + "right_pr_number": 44154, + "code_similarity": 0.153, + "size_similarity": 0.535, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.31, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 43996, + "right_pr_number": 44722, + "code_similarity": 0.225, + "size_similarity": 0.848, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.368, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 44007, + "right_pr_number": 44013, + "code_similarity": 0.19, + "size_similarity": 0.553, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.531, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 44007, + "right_pr_number": 44044, + "code_similarity": 0.186, + "size_similarity": 0.667, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.354, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 44007, + "right_pr_number": 44066, + "code_similarity": 0.188, + "size_similarity": 0.704, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.315, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 44007, + "right_pr_number": 44072, + "code_similarity": 0.212, + "size_similarity": 0.526, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.708, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 44007, + "right_pr_number": 44085, + "code_similarity": 0.195, + "size_similarity": 0.735, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.318, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 44007, + "right_pr_number": 44129, + "code_similarity": 0.103, + "size_similarity": 0.37, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.191, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 44007, + "right_pr_number": 44154, + "code_similarity": 0.238, + "size_similarity": 0.93, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.344, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 44007, + "right_pr_number": 44722, + "code_similarity": 0.178, + "size_similarity": 0.679, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.28, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 44013, + "right_pr_number": 44044, + "code_similarity": 0.126, + "size_similarity": 0.368, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.351, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 44013, + "right_pr_number": 44066, + "code_similarity": 0.127, + "size_similarity": 0.389, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.325, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 44013, + "right_pr_number": 44072, + "code_similarity": 0.29, + "size_similarity": 0.952, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.667, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 44013, + "right_pr_number": 44085, + "code_similarity": 0.131, + "size_similarity": 0.406, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.329, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 44013, + "right_pr_number": 44129, + "code_similarity": 0.07, + "size_similarity": 0.205, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.192, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 44013, + "right_pr_number": 44154, + "code_similarity": 0.177, + "size_similarity": 0.594, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.389, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 44013, + "right_pr_number": 44722, + "code_similarity": 0.118, + "size_similarity": 0.375, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.287, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 44044, + "right_pr_number": 44066, + "code_similarity": 0.25, + "size_similarity": 0.947, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.404, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 44044, + "right_pr_number": 44072, + "code_similarity": 0.136, + "size_similarity": 0.351, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.442, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 44044, + "right_pr_number": 44085, + "code_similarity": 0.24, + "size_similarity": 0.906, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.394, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 44044, + "right_pr_number": 44129, + "code_similarity": 0.147, + "size_similarity": 0.555, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.243, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 44044, + "right_pr_number": 44154, + "code_similarity": 0.17, + "size_similarity": 0.62, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.306, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 44044, + "right_pr_number": 44722, + "code_similarity": 0.257, + "size_similarity": 0.982, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.402, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 44066, + "right_pr_number": 44072, + "code_similarity": 0.133, + "size_similarity": 0.37, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.393, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 44066, + "right_pr_number": 44085, + "code_similarity": 0.763, + "size_similarity": 0.957, + "file_overlap": 0.5, + "area_overlap": 0.825, + "patch_similarity": 0.887, + "shared_filenames": [ + "src/transformers/models/gptj/modeling_gptj.py" + ], + "shared_file_areas": [ + { + "filename": "src/transformers/models/gptj/modeling_gptj.py", + "left_ranges": [ + [ + 33, + 41 + ], + [ + 174, + 181 + ], + [ + 250, + 257 + ], + [ + 398, + 419 + ], + [ + 425, + 434 + ], + [ + 459, + 466 + ], + [ + 471, + 488 + ], + [ + 518, + 540 + ], + [ + 554, + 560 + ], + [ + 566, + 575 + ], + [ + 580, + 598 + ], + [ + 601, + 606 + ], + [ + 634, + 640 + ], + [ + 646, + 653 + ], + [ + 658, + 674 + ], + [ + 717, + 722 + ], + [ + 738, + 744 + ], + [ + 749, + 772 + ], + [ + 790, + 795 + ] + ], + "right_ranges": [ + [ + 33, + 40 + ], + [ + 173, + 180 + ], + [ + 249, + 256 + ], + [ + 397, + 405 + ], + [ + 408, + 419 + ], + [ + 425, + 434 + ], + [ + 459, + 465 + ], + [ + 470, + 489 + ], + [ + 517, + 539 + ], + [ + 553, + 559 + ], + [ + 565, + 574 + ], + [ + 579, + 597 + ], + [ + 600, + 611 + ], + [ + 633, + 639 + ], + [ + 645, + 652 + ], + [ + 657, + 673 + ], + [ + 716, + 728 + ], + [ + 737, + 743 + ], + [ + 748, + 771 + ], + [ + 789, + 794 + ] + ] + } + ] + }, + { + "left_pr_number": 44066, + "right_pr_number": 44129, + "code_similarity": 0.145, + "size_similarity": 0.526, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.263, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 44066, + "right_pr_number": 44154, + "code_similarity": 0.174, + "size_similarity": 0.654, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.286, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 44066, + "right_pr_number": 44722, + "code_similarity": 0.907, + "size_similarity": 0.964, + "file_overlap": 1.0, + "area_overlap": 0.808, + "patch_similarity": 0.874, + "shared_filenames": [ + "src/transformers/models/codegen/modeling_codegen.py", + "src/transformers/models/gptj/modeling_gptj.py" + ], + "shared_file_areas": [ + { + "filename": "src/transformers/models/codegen/modeling_codegen.py", + "left_ranges": [ + [ + 245, + 266 + ] + ], + "right_ranges": [ + [ + 228, + 234 + ] + ] + }, + { + "filename": "src/transformers/models/gptj/modeling_gptj.py", + "left_ranges": [ + [ + 33, + 41 + ], + [ + 174, + 181 + ], + [ + 250, + 257 + ], + [ + 398, + 419 + ], + [ + 425, + 434 + ], + [ + 459, + 466 + ], + [ + 471, + 488 + ], + [ + 518, + 540 + ], + [ + 554, + 560 + ], + [ + 566, + 575 + ], + [ + 580, + 598 + ], + [ + 601, + 606 + ], + [ + 634, + 640 + ], + [ + 646, + 653 + ], + [ + 658, + 674 + ], + [ + 717, + 722 + ], + [ + 738, + 744 + ], + [ + 749, + 772 + ], + [ + 790, + 795 + ] + ], + "right_ranges": [ + [ + 33, + 42 + ], + [ + 175, + 182 + ], + [ + 251, + 258 + ], + [ + 399, + 420 + ], + [ + 426, + 435 + ], + [ + 460, + 467 + ], + [ + 472, + 489 + ], + [ + 519, + 540 + ], + [ + 554, + 560 + ], + [ + 566, + 575 + ], + [ + 580, + 598 + ], + [ + 601, + 612 + ], + [ + 634, + 640 + ], + [ + 646, + 653 + ], + [ + 658, + 663 + ], + [ + 666, + 674 + ], + [ + 717, + 722 + ], + [ + 738, + 744 + ], + [ + 749, + 772 + ], + [ + 790, + 795 + ] + ] + } + ] + }, + { + "left_pr_number": 44072, + "right_pr_number": 44085, + "code_similarity": 0.137, + "size_similarity": 0.387, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.398, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 44072, + "right_pr_number": 44129, + "code_similarity": 0.074, + "size_similarity": 0.195, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.231, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 44072, + "right_pr_number": 44154, + "code_similarity": 0.175, + "size_similarity": 0.566, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.414, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 44072, + "right_pr_number": 44722, + "code_similarity": 0.124, + "size_similarity": 0.357, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.347, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 44085, + "right_pr_number": 44129, + "code_similarity": 0.141, + "size_similarity": 0.503, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.272, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 44085, + "right_pr_number": 44154, + "code_similarity": 0.18, + "size_similarity": 0.684, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.289, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 44085, + "right_pr_number": 44722, + "code_similarity": 0.728, + "size_similarity": 0.923, + "file_overlap": 0.5, + "area_overlap": 0.791, + "patch_similarity": 0.78, + "shared_filenames": [ + "src/transformers/models/gptj/modeling_gptj.py" + ], + "shared_file_areas": [ + { + "filename": "src/transformers/models/gptj/modeling_gptj.py", + "left_ranges": [ + [ + 33, + 40 + ], + [ + 173, + 180 + ], + [ + 249, + 256 + ], + [ + 397, + 405 + ], + [ + 408, + 419 + ], + [ + 425, + 434 + ], + [ + 459, + 465 + ], + [ + 470, + 489 + ], + [ + 517, + 539 + ], + [ + 553, + 559 + ], + [ + 565, + 574 + ], + [ + 579, + 597 + ], + [ + 600, + 611 + ], + [ + 633, + 639 + ], + [ + 645, + 652 + ], + [ + 657, + 673 + ], + [ + 716, + 728 + ], + [ + 737, + 743 + ], + [ + 748, + 771 + ], + [ + 789, + 794 + ] + ], + "right_ranges": [ + [ + 33, + 42 + ], + [ + 175, + 182 + ], + [ + 251, + 258 + ], + [ + 399, + 420 + ], + [ + 426, + 435 + ], + [ + 460, + 467 + ], + [ + 472, + 489 + ], + [ + 519, + 540 + ], + [ + 554, + 560 + ], + [ + 566, + 575 + ], + [ + 580, + 598 + ], + [ + 601, + 612 + ], + [ + 634, + 640 + ], + [ + 646, + 653 + ], + [ + 658, + 663 + ], + [ + 666, + 674 + ], + [ + 717, + 722 + ], + [ + 738, + 744 + ], + [ + 749, + 772 + ], + [ + 790, + 795 + ] + ] + } + ] + }, + { + "left_pr_number": 44129, + "right_pr_number": 44154, + "code_similarity": 0.099, + "size_similarity": 0.344, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.199, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 44129, + "right_pr_number": 44722, + "code_similarity": 0.146, + "size_similarity": 0.545, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.247, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 44154, + "right_pr_number": 44722, + "code_similarity": 0.164, + "size_similarity": 0.631, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.255, + "shared_filenames": [], + "shared_file_areas": [] + } + ] + }, + { + "cluster_id": "cluster-41211-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #41211.", + "status": "open", + "confidence": 0.55, + "canonical_issue_number": 41211, + "canonical_pr_number": 44339, + "issue_numbers": [ + 41211 + ], + "pr_numbers": [ + 41356, + 44339 + ], + "evidence_types": [ + "closing_reference" + ], + "pr_comparisons": [ + { + "left_pr_number": 41356, + "right_pr_number": 44339, + "code_similarity": 0.155, + "size_similarity": 0.094, + "file_overlap": 0.273, + "area_overlap": 0.078, + "patch_similarity": 0.183, + "shared_filenames": [ + "docs/source/en/model_doc/deimv2.md", + "src/transformers/models/deimv2/__init__.py", + "src/transformers/models/deimv2/configuration_deimv2.py", + "src/transformers/models/deimv2/modeling_deimv2.py", + "tests/models/deimv2/__init__.py", + "tests/models/deimv2/test_modeling_deimv2.py" + ], + "shared_file_areas": [ + { + "filename": "docs/source/en/model_doc/deimv2.md", + "left_ranges": [ + [ + 1, + 132 + ] + ], + "right_ranges": [ + [ + 1, + 65 + ] + ] + }, + { + "filename": "src/transformers/models/deimv2/__init__.py", + "left_ranges": [ + [ + 1, + 15 + ] + ], + "right_ranges": [ + [ + 1, + 29 + ] + ] + }, + { + "filename": "src/transformers/models/deimv2/configuration_deimv2.py", + "left_ranges": [ + [ + 1, + 74 + ] + ], + "right_ranges": [ + [ + 1, + 266 + ] + ] + }, + { + "filename": "tests/models/deimv2/test_modeling_deimv2.py", + "left_ranges": [ + [ + 1, + 15 + ] + ], + "right_ranges": [ + [ + 1, + 1734 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-46894-5", + "summary": "Cluster of 2 issues and 3 PRs centered on issue #46897.", + "status": "open", + "confidence": 0.6, + "canonical_issue_number": 46897, + "canonical_pr_number": 46895, + "issue_numbers": [ + 46894, + 46897 + ], + "pr_numbers": [ + 46895, + 46898, + 46916 + ], + "evidence_types": [ + "closing_reference", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 46895, + "right_pr_number": 46898, + "code_similarity": 0.117, + "size_similarity": 0.467, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.156, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 46895, + "right_pr_number": 46916, + "code_similarity": 0.335, + "size_similarity": 0.622, + "file_overlap": 0.4, + "area_overlap": 0.194, + "patch_similarity": 0.154, + "shared_filenames": [ + "src/transformers/models/cohere_asr/modeling_cohere_asr.py", + "src/transformers/models/cohere_asr/modular_cohere_asr.py" + ], + "shared_file_areas": [ + { + "filename": "src/transformers/models/cohere_asr/modeling_cohere_asr.py", + "left_ranges": [ + [ + 615, + 627 + ], + [ + 639, + 654 + ] + ], + "right_ranges": [ + [ + 637, + 644 + ] + ] + }, + { + "filename": "src/transformers/models/cohere_asr/modular_cohere_asr.py", + "left_ranges": [ + [ + 478, + 490 + ], + [ + 502, + 517 + ] + ], + "right_ranges": [ + [ + 500, + 507 + ] + ] + } + ] + }, + { + "left_pr_number": 46898, + "right_pr_number": 46916, + "code_similarity": 0.42, + "size_similarity": 0.75, + "file_overlap": 0.5, + "area_overlap": 0.194, + "patch_similarity": 0.35, + "shared_filenames": [ + "src/transformers/models/florence2/modeling_florence2.py", + "src/transformers/models/florence2/modular_florence2.py" + ], + "shared_file_areas": [ + { + "filename": "src/transformers/models/florence2/modeling_florence2.py", + "left_ranges": [ + [ + 872, + 878 + ], + [ + 881, + 887 + ], + [ + 904, + 918 + ] + ], + "right_ranges": [ + [ + 902, + 909 + ] + ] + }, + { + "filename": "src/transformers/models/florence2/modular_florence2.py", + "left_ranges": [ + [ + 1580, + 1586 + ], + [ + 1589, + 1595 + ], + [ + 1612, + 1626 + ] + ], + "right_ranges": [ + [ + 1610, + 1617 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-43240-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #43240.", + "status": "open", + "confidence": 0.75, + "canonical_issue_number": 43240, + "canonical_pr_number": 43251, + "issue_numbers": [ + 43240 + ], + "pr_numbers": [ + 43251, + 43254 + ], + "evidence_types": [ + "closing_reference", + "shared_issue_target" + ], + "pr_comparisons": [ + { + "left_pr_number": 43251, + "right_pr_number": 43254, + "code_similarity": 0.794, + "size_similarity": 0.64, + "file_overlap": 1.0, + "area_overlap": 0.667, + "patch_similarity": 0.882, + "shared_filenames": [ + "src/transformers/loss/loss_utils.py" + ], + "shared_file_areas": [ + { + "filename": "src/transformers/loss/loss_utils.py", + "left_ranges": [ + [ + 30, + 50 + ] + ], + "right_ranges": [ + [ + 30, + 43 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-45647-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #45647.", + "status": "open", + "confidence": 0.75, + "canonical_issue_number": 45647, + "canonical_pr_number": 45738, + "issue_numbers": [ + 45647 + ], + "pr_numbers": [ + 45737, + 45738 + ], + "evidence_types": [ + "closing_reference", + "shared_issue_target" + ], + "pr_comparisons": [ + { + "left_pr_number": 45737, + "right_pr_number": 45738, + "code_similarity": 0.526, + "size_similarity": 0.11, + "file_overlap": 0.5, + "area_overlap": 1.0, + "patch_similarity": 0.024, + "shared_filenames": [ + "src/transformers/models/musicgen_melody/modeling_musicgen_melody.py" + ], + "shared_file_areas": [ + { + "filename": "src/transformers/models/musicgen_melody/modeling_musicgen_melody.py", + "left_ranges": [ + [ + 491, + 497 + ] + ], + "right_ranges": [ + [ + 491, + 497 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-46869-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #46869.", + "status": "open", + "confidence": 0.75, + "canonical_issue_number": 46869, + "canonical_pr_number": 46886, + "issue_numbers": [ + 46869 + ], + "pr_numbers": [ + 46886, + 46922 + ], + "evidence_types": [ + "closing_reference", + "shared_issue_target" + ], + "pr_comparisons": [ + { + "left_pr_number": 46886, + "right_pr_number": 46922, + "code_similarity": 0.711, + "size_similarity": 0.857, + "file_overlap": 1.0, + "area_overlap": 0.5, + "patch_similarity": 0.429, + "shared_filenames": [ + "src/transformers/models/yolos/modeling_yolos.py" + ], + "shared_file_areas": [ + { + "filename": "src/transformers/models/yolos/modeling_yolos.py", + "left_ranges": [ + [ + 446, + 452 + ] + ], + "right_ranges": [ + [ + 434, + 440 + ], + [ + 446, + 452 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-46159-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #46159.", + "status": "open", + "confidence": 0.55, + "canonical_issue_number": 46159, + "canonical_pr_number": 46174, + "issue_numbers": [ + 46159 + ], + "pr_numbers": [ + 46174, + 46780 + ], + "evidence_types": [ + "closing_reference" + ], + "pr_comparisons": [ + { + "left_pr_number": 46174, + "right_pr_number": 46780, + "code_similarity": 0.841, + "size_similarity": 0.87, + "file_overlap": 1.0, + "area_overlap": 0.767, + "patch_similarity": 0.657, + "shared_filenames": [ + "src/transformers/models/aimv2/modeling_aimv2.py", + "src/transformers/models/d_fine/modeling_d_fine.py", + "src/transformers/models/deimv2/modeling_deimv2.py", + "src/transformers/models/pp_doclayout_v2/modeling_pp_doclayout_v2.py", + "src/transformers/models/pp_doclayout_v3/modeling_pp_doclayout_v3.py", + "src/transformers/models/rt_detr/modeling_rt_detr.py", + "src/transformers/models/rt_detr_v2/modeling_rt_detr_v2.py", + "src/transformers/models/vit_mae/modeling_vit_mae.py", + "src/transformers/models/vit_mae/modular_vit_mae.py" + ], + "shared_file_areas": [ + { + "filename": "src/transformers/models/aimv2/modeling_aimv2.py", + "left_ranges": [ + [ + 134, + 141 + ], + [ + 144, + 154 + ], + [ + 157, + 165 + ] + ], + "right_ranges": [ + [ + 134, + 140 + ], + [ + 143, + 153 + ], + [ + 156, + 162 + ] + ] + }, + { + "filename": "src/transformers/models/d_fine/modeling_d_fine.py", + "left_ranges": [ + [ + 631, + 638 + ], + [ + 641, + 651 + ], + [ + 654, + 662 + ] + ], + "right_ranges": [ + [ + 631, + 637 + ], + [ + 640, + 650 + ], + [ + 653, + 659 + ] + ] + }, + { + "filename": "src/transformers/models/deimv2/modeling_deimv2.py", + "left_ranges": [ + [ + 703, + 710 + ], + [ + 713, + 723 + ], + [ + 726, + 734 + ] + ], + "right_ranges": [ + [ + 703, + 709 + ], + [ + 712, + 722 + ], + [ + 725, + 731 + ] + ] + }, + { + "filename": "src/transformers/models/pp_doclayout_v2/modeling_pp_doclayout_v2.py", + "left_ranges": [ + [ + 1551, + 1558 + ], + [ + 1561, + 1571 + ], + [ + 1574, + 1582 + ] + ], + "right_ranges": [ + [ + 1551, + 1557 + ], + [ + 1560, + 1570 + ], + [ + 1573, + 1579 + ] + ] + }, + { + "filename": "src/transformers/models/pp_doclayout_v3/modeling_pp_doclayout_v3.py", + "left_ranges": [ + [ + 804, + 811 + ], + [ + 814, + 824 + ], + [ + 827, + 835 + ] + ], + "right_ranges": [ + [ + 804, + 810 + ], + [ + 813, + 823 + ], + [ + 826, + 832 + ] + ] + }, + { + "filename": "src/transformers/models/rt_detr/modeling_rt_detr.py", + "left_ranges": [ + [ + 847, + 854 + ], + [ + 857, + 867 + ], + [ + 870, + 878 + ] + ], + "right_ranges": [ + [ + 847, + 853 + ], + [ + 856, + 866 + ], + [ + 869, + 875 + ] + ] + }, + { + "filename": "src/transformers/models/rt_detr_v2/modeling_rt_detr_v2.py", + "left_ranges": [ + [ + 976, + 983 + ], + [ + 986, + 996 + ], + [ + 999, + 1007 + ] + ], + "right_ranges": [ + [ + 976, + 982 + ], + [ + 985, + 995 + ], + [ + 998, + 1004 + ] + ] + }, + { + "filename": "src/transformers/models/vit_mae/modeling_vit_mae.py", + "left_ranges": [ + [ + 152, + 159 + ], + [ + 162, + 172 + ], + [ + 175, + 183 + ] + ], + "right_ranges": [ + [ + 152, + 158 + ], + [ + 161, + 171 + ], + [ + 174, + 180 + ] + ] + }, + { + "filename": "src/transformers/models/vit_mae/modular_vit_mae.py", + "left_ranges": [ + [ + 62, + 69 + ], + [ + 72, + 82 + ], + [ + 85, + 93 + ] + ], + "right_ranges": [ + [ + 62, + 68 + ], + [ + 71, + 81 + ], + [ + 84, + 90 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-45561-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #45561.", + "status": "open", + "confidence": 0.8, + "canonical_issue_number": 45561, + "canonical_pr_number": 45645, + "issue_numbers": [ + 45561 + ], + "pr_numbers": [ + 45639, + 45645 + ], + "evidence_types": [ + "closing_reference", + "shared_issue_target", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 45639, + "right_pr_number": 45645, + "code_similarity": 0.721, + "size_similarity": 0.973, + "file_overlap": 0.75, + "area_overlap": 0.563, + "patch_similarity": 0.696, + "shared_filenames": [ + "src/transformers/testing_utils.py", + "tests/utils/test_testing_utils.py", + "utils/notification_service.py" + ], + "shared_file_areas": [ + { + "filename": "src/transformers/testing_utils.py", + "left_ranges": [ + [ + 3611, + 3618 + ], + [ + 3835, + 3869 + ] + ], + "right_ranges": [ + [ + 3525, + 3558 + ], + [ + 3782, + 3788 + ] + ] + }, + { + "filename": "tests/utils/test_testing_utils.py", + "left_ranges": [ + [ + 1, + 144 + ] + ], + "right_ranges": [ + [ + 1, + 114 + ] + ] + }, + { + "filename": "utils/notification_service.py", + "left_ranges": [ + [ + 935, + 967 + ] + ], + "right_ranges": [ + [ + 935, + 967 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-46305-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #46305.", + "status": "open", + "confidence": 0.8, + "canonical_issue_number": 46305, + "canonical_pr_number": 46546, + "issue_numbers": [ + 46305 + ], + "pr_numbers": [ + 46310, + 46546 + ], + "evidence_types": [ + "closing_reference", + "shared_issue_target", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 46310, + "right_pr_number": 46546, + "code_similarity": 0.566, + "size_similarity": 0.444, + "file_overlap": 0.5, + "area_overlap": 0.867, + "patch_similarity": 0.16, + "shared_filenames": [ + "src/transformers/models/moonshine_streaming/modeling_moonshine_streaming.py", + "src/transformers/models/moonshine_streaming/modular_moonshine_streaming.py" + ], + "shared_file_areas": [ + { + "filename": "src/transformers/models/moonshine_streaming/modeling_moonshine_streaming.py", + "left_ranges": [ + [ + 356, + 362 + ] + ], + "right_ranges": [ + [ + 355, + 361 + ] + ] + }, + { + "filename": "src/transformers/models/moonshine_streaming/modular_moonshine_streaming.py", + "left_ranges": [ + [ + 269, + 275 + ] + ], + "right_ranges": [ + [ + 269, + 275 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-45081-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #45081.", + "status": "open", + "confidence": 0.8, + "canonical_issue_number": 45081, + "canonical_pr_number": 45317, + "issue_numbers": [ + 45081 + ], + "pr_numbers": [ + 45086, + 45317 + ], + "evidence_types": [ + "closing_reference", + "shared_issue_target", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 45086, + "right_pr_number": 45317, + "code_similarity": 0.589, + "size_similarity": 0.34, + "file_overlap": 0.5, + "area_overlap": 1.0, + "patch_similarity": 0.136, + "shared_filenames": [ + "src/transformers/tokenization_utils_tokenizers.py" + ], + "shared_file_areas": [ + { + "filename": "src/transformers/tokenization_utils_tokenizers.py", + "left_ranges": [ + [ + 1360, + 1370 + ], + [ + 1374, + 1380 + ] + ], + "right_ranges": [ + [ + 1360, + 1370 + ], + [ + 1374, + 1380 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-43698-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #43698.", + "status": "open", + "confidence": 0.75, + "canonical_issue_number": 43698, + "canonical_pr_number": 43779, + "issue_numbers": [ + 43698 + ], + "pr_numbers": [ + 43779, + 43816 + ], + "evidence_types": [ + "closing_reference", + "shared_issue_target" + ], + "pr_comparisons": [ + { + "left_pr_number": 43779, + "right_pr_number": 43816, + "code_similarity": 0.418, + "size_similarity": 0.538, + "file_overlap": 1.0, + "area_overlap": 0.02, + "patch_similarity": 0.02, + "shared_filenames": [ + "src/transformers/integrations/integration_utils.py" + ], + "shared_file_areas": [ + { + "filename": "src/transformers/integrations/integration_utils.py", + "left_ranges": [ + [ + 2227, + 2241 + ], + [ + 2303, + 2309 + ] + ], + "right_ranges": [ + [ + 2278, + 2291 + ], + [ + 2309, + 2322 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-43612-2", + "summary": "Cluster of 2 related pull requests linked by soft_similarity.", + "status": "open", + "confidence": 0.5, + "canonical_issue_number": null, + "canonical_pr_number": 43612, + "issue_numbers": [], + "pr_numbers": [ + 43612, + 43613 + ], + "evidence_types": [ + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 43612, + "right_pr_number": 43613, + "code_similarity": 0.6, + "size_similarity": 0.822, + "file_overlap": 0.333, + "area_overlap": 0.698, + "patch_similarity": 0.608, + "shared_filenames": [ + "docs/source/en/_toctree.yml", + "docs/source/en/main_classes/pipelines.md", + "docs/source/en/model_doc/auto.md", + "src/transformers/__init__.py", + "src/transformers/models/auto/modeling_auto.py", + "src/transformers/pipelines/__init__.py", + "utils/check_docstrings.py", + "utils/update_metadata.py" + ], + "shared_file_areas": [ + { + "filename": "docs/source/en/_toctree.yml", + "left_ranges": [ + [ + 335, + 342 + ] + ], + "right_ranges": [ + [ + 309, + 316 + ] + ] + }, + { + "filename": "docs/source/en/main_classes/pipelines.md", + "left_ranges": [ + [ + 473, + 484 + ] + ], + "right_ranges": [ + [ + 473, + 484 + ] + ] + }, + { + "filename": "docs/source/en/model_doc/auto.md", + "left_ranges": [ + [ + 113, + 122 + ] + ], + "right_ranges": [ + [ + 113, + 122 + ] + ] + }, + { + "filename": "src/transformers/__init__.py", + "left_ranges": [ + [ + 155, + 161 + ], + [ + 664, + 670 + ] + ], + "right_ranges": [ + [ + 155, + 161 + ], + [ + 664, + 670 + ] + ] + }, + { + "filename": "src/transformers/models/auto/modeling_auto.py", + "left_ranges": [ + [ + 1629, + 1642 + ], + [ + 1807, + 1816 + ], + [ + 1840, + 1849 + ], + [ + 2186, + 2192 + ], + [ + 2233, + 2239 + ] + ], + "right_ranges": [ + [ + 1629, + 1646 + ], + [ + 1811, + 1820 + ], + [ + 1844, + 1853 + ], + [ + 2190, + 2196 + ], + [ + 2237, + 2243 + ] + ] + }, + { + "filename": "src/transformers/pipelines/__init__.py", + "left_ranges": [ + [ + 71, + 77 + ], + [ + 108, + 114 + ], + [ + 300, + 311 + ], + [ + 446, + 453 + ] + ], + "right_ranges": [ + [ + 71, + 77 + ], + [ + 108, + 114 + ], + [ + 300, + 311 + ], + [ + 446, + 453 + ] + ] + }, + { + "filename": "utils/check_docstrings.py", + "left_ranges": [ + [ + 109, + 115 + ] + ], + "right_ranges": [ + [ + 109, + 115 + ] + ] + }, + { + "filename": "utils/update_metadata.py", + "left_ranges": [ + [ + 74, + 84 + ] + ], + "right_ranges": [ + [ + 74, + 84 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-44018-2", + "summary": "Cluster of 2 related pull requests linked by soft_similarity.", + "status": "open", + "confidence": 0.5, + "canonical_issue_number": null, + "canonical_pr_number": 44068, + "issue_numbers": [], + "pr_numbers": [ + 44018, + 44068 + ], + "evidence_types": [ + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 44018, + "right_pr_number": 44068, + "code_similarity": 0.766, + "size_similarity": 0.939, + "file_overlap": 1.0, + "area_overlap": 0.425, + "patch_similarity": 0.866, + "shared_filenames": [ + "src/transformers/models/gpt_neo/modeling_gpt_neo.py", + "tests/models/gpt_neo/test_modeling_gpt_neo.py" + ], + "shared_file_areas": [ + { + "filename": "src/transformers/models/gpt_neo/modeling_gpt_neo.py", + "left_ranges": [ + [ + 26, + 46 + ], + [ + 138, + 143 + ], + [ + 182, + 187 + ], + [ + 283, + 295 + ], + [ + 327, + 341 + ], + [ + 348, + 354 + ], + [ + 360, + 369 + ], + [ + 399, + 406 + ], + [ + 411, + 419 + ], + [ + 428, + 433 + ], + [ + 472, + 492 + ], + [ + 507, + 513 + ], + [ + 519, + 528 + ], + [ + 541, + 559 + ], + [ + 562, + 567 + ], + [ + 595, + 601 + ], + [ + 607, + 614 + ], + [ + 627, + 643 + ], + [ + 685, + 690 + ], + [ + 708, + 714 + ], + [ + 720, + 727 + ], + [ + 740, + 757 + ], + [ + 761, + 766 + ], + [ + 780, + 786 + ], + [ + 791, + 798 + ], + [ + 807, + 822 + ], + [ + 840, + 845 + ] + ], + "right_ranges": [ + [ + 26, + 31 + ], + [ + 34, + 44 + ], + [ + 136, + 141 + ], + [ + 180, + 185 + ], + [ + 281, + 293 + ], + [ + 325, + 339 + ], + [ + 346, + 352 + ], + [ + 358, + 367 + ], + [ + 397, + 404 + ], + [ + 409, + 417 + ], + [ + 426, + 434 + ], + [ + 463, + 483 + ], + [ + 498, + 504 + ], + [ + 510, + 519 + ], + [ + 532, + 550 + ], + [ + 553, + 564 + ], + [ + 586, + 592 + ], + [ + 598, + 605 + ], + [ + 618, + 634 + ], + [ + 676, + 688 + ], + [ + 699, + 705 + ], + [ + 711, + 718 + ], + [ + 731, + 748 + ], + [ + 752, + 762 + ], + [ + 771, + 777 + ], + [ + 782, + 789 + ], + [ + 798, + 813 + ], + [ + 831, + 836 + ] + ] + }, + { + "filename": "tests/models/gpt_neo/test_modeling_gpt_neo.py", + "left_ranges": [ + [ + 458, + 464 + ] + ], + "right_ranges": [ + [ + 458, + 464 + ] + ] + } + ] + } + ] + } + ], + "duplicate_issues": [ + { + "cluster_id": "cluster-46894-5", + "canonical_issue_number": 46897, + "duplicate_issue_numbers": [ + 46894 + ], + "reason": "Issues in cluster-46894-5 are treated as duplicates because they share closing_reference, soft_similarity evidence." + } + ], + "duplicate_prs": [ + { + "cluster_id": "cluster-41211-3", + "canonical_pr_number": 44339, + "duplicate_pr_numbers": [ + 41356 + ], + "target_issue_number": 41211, + "reason": "PRs in cluster-41211-3 are treated as duplicates because they converge on issue #41211 with closing_reference evidence." + }, + { + "cluster_id": "cluster-43240-3", + "canonical_pr_number": 43251, + "duplicate_pr_numbers": [ + 43254 + ], + "target_issue_number": 43240, + "reason": "PRs in cluster-43240-3 are treated as duplicates because they converge on issue #43240 with closing_reference, shared_issue_target evidence." + }, + { + "cluster_id": "cluster-43612-2", + "canonical_pr_number": 43612, + "duplicate_pr_numbers": [ + 43613 + ], + "target_issue_number": null, + "reason": "PRs in cluster-43612-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-43698-3", + "canonical_pr_number": 43779, + "duplicate_pr_numbers": [ + 43816 + ], + "target_issue_number": 43698, + "reason": "PRs in cluster-43698-3 are treated as duplicates because they converge on issue #43698 with closing_reference, shared_issue_target evidence." + }, + { + "cluster_id": "cluster-43979-11", + "canonical_pr_number": 44007, + "duplicate_pr_numbers": [ + 43996, + 44013, + 44044, + 44066, + 44072, + 44085, + 44129, + 44154, + 44722 + ], + "target_issue_number": 43979, + "reason": "PRs in cluster-43979-11 are treated as duplicates because they converge on issue #43979 with closing_reference, shared_issue_target, soft_similarity evidence." + }, + { + "cluster_id": "cluster-44018-2", + "canonical_pr_number": 44068, + "duplicate_pr_numbers": [ + 44018 + ], + "target_issue_number": null, + "reason": "PRs in cluster-44018-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-45081-3", + "canonical_pr_number": 45317, + "duplicate_pr_numbers": [ + 45086 + ], + "target_issue_number": 45081, + "reason": "PRs in cluster-45081-3 are treated as duplicates because they converge on issue #45081 with closing_reference, shared_issue_target, soft_similarity evidence." + }, + { + "cluster_id": "cluster-45561-3", + "canonical_pr_number": 45645, + "duplicate_pr_numbers": [ + 45639 + ], + "target_issue_number": 45561, + "reason": "PRs in cluster-45561-3 are treated as duplicates because they converge on issue #45561 with closing_reference, shared_issue_target, soft_similarity evidence." + }, + { + "cluster_id": "cluster-45647-3", + "canonical_pr_number": 45738, + "duplicate_pr_numbers": [ + 45737 + ], + "target_issue_number": 45647, + "reason": "PRs in cluster-45647-3 are treated as duplicates because they converge on issue #45647 with closing_reference, shared_issue_target evidence." + }, + { + "cluster_id": "cluster-46159-3", + "canonical_pr_number": 46174, + "duplicate_pr_numbers": [ + 46780 + ], + "target_issue_number": 46159, + "reason": "PRs in cluster-46159-3 are treated as duplicates because they converge on issue #46159 with closing_reference evidence." + }, + { + "cluster_id": "cluster-46305-3", + "canonical_pr_number": 46546, + "duplicate_pr_numbers": [ + 46310 + ], + "target_issue_number": 46305, + "reason": "PRs in cluster-46305-3 are treated as duplicates because they converge on issue #46305 with closing_reference, shared_issue_target, soft_similarity evidence." + }, + { + "cluster_id": "cluster-46869-3", + "canonical_pr_number": 46886, + "duplicate_pr_numbers": [ + 46922 + ], + "target_issue_number": 46869, + "reason": "PRs in cluster-46869-3 are treated as duplicates because they converge on issue #46869 with closing_reference, shared_issue_target evidence." + }, + { + "cluster_id": "cluster-46894-5", + "canonical_pr_number": 46895, + "duplicate_pr_numbers": [ + 46898, + 46916 + ], + "target_issue_number": 46897, + "reason": "PRs in cluster-46894-5 are treated as duplicates because they converge on issue #46897 with closing_reference, soft_similarity evidence." + } + ], + "best_issue": { + "cluster_id": "cluster-46894-5", + "issue_number": 46897, + "reason": "Issue #46897 is the strongest global issue candidate because it is open, belongs to a cluster with 5 artifacts, and carries 0 discussion comments plus 9 inbound references.", + "score": 159.0 + }, + "best_pr": { + "cluster_id": "cluster-43979-11", + "pr_number": 44007, + "reason": "PR #44007 is the strongest global PR candidate because it is open, belongs to a cluster with 11 artifacts, links to 1 issues, and carries 2 review/discussion events.", + "score": 238.886 + } +} diff --git a/snapshots/20260628T000017Z/analysis-runs/hybrid-model-20260628t000017z/analysis-report-hybrid.llm-reviews.json b/snapshots/20260628T000017Z/analysis-runs/hybrid-model-20260628t000017z/analysis-report-hybrid.llm-reviews.json new file mode 100644 index 0000000000000000000000000000000000000000..08ddb89f2d0642bb166b42a6709f56bd7eaca65c --- /dev/null +++ b/snapshots/20260628T000017Z/analysis-runs/hybrid-model-20260628t000017z/analysis-report-hybrid.llm-reviews.json @@ -0,0 +1,25093 @@ +{ + "schema_version": "1.0", + "repo": "huggingface/transformers", + "snapshot_id": "20260628T000017Z", + "generated_at": "2026-06-28T00:10:07Z", + "model": "gpt-5.4-mini?service_tier=flex", + "reviews": [ + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 1, + "review_unit_count": 61, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-502", + "nodes": [ + "issue:24643", + "issue:30277", + "issue:34634", + "issue:34928", + "issue:35545", + "issue:36010", + "issue:36331", + "issue:39290", + "issue:41720", + "issue:41762", + "issue:42915", + "issue:43039", + "issue:43425", + "issue:43541", + "issue:43596", + "issue:43638", + "issue:43716", + "issue:43827", + "issue:43828", + "issue:43975", + "issue:44292", + "issue:44322", + "issue:44457", + "issue:44509", + "issue:44512", + "issue:44805", + "issue:44841", + "issue:44918", + "issue:45081", + "issue:45161", + "issue:45237", + "issue:45464", + "issue:45507", + "issue:45685", + "issue:45800", + "issue:45823", + "issue:45859", + "issue:45874", + "issue:45901", + "issue:45902", + "issue:46015", + "issue:46165", + "issue:46294", + "issue:46468", + "issue:46515", + "issue:46661", + "issue:46869" + ], + "soft_pairs": [ + "issue:44322|issue:45464", + "issue:45237|issue:45685", + "issue:43716|issue:45237", + "issue:43828|issue:45237", + "issue:45237|issue:46468", + "issue:45237|issue:45507", + "issue:30277|issue:35545", + "issue:35545|issue:42915", + "issue:35545|issue:41720", + "issue:45901|issue:46015", + "issue:44509|issue:44512", + "issue:35545|issue:36010", + "issue:44918|issue:46165", + "issue:46165|issue:46869", + "issue:44805|issue:45823", + "issue:43638|issue:44805", + "issue:43638|issue:46165", + "issue:43596|issue:44805", + "issue:44805|issue:46294", + "issue:44805|issue:45161", + "issue:46165|issue:46468", + "issue:43596|issue:46165", + "issue:34634|issue:35545", + "issue:30277|issue:43638", + "issue:43596|issue:43638", + "issue:43039|issue:45161", + "issue:43975|issue:44457", + "issue:44292|issue:44457", + "issue:44841|issue:46661", + "issue:43716|issue:46869", + "issue:24643|issue:30277", + "issue:34928|issue:44805", + "issue:45859|issue:45902", + "issue:45685|issue:46869", + "issue:43638|issue:45161", + "issue:34928|issue:36331", + "issue:43596|issue:45161", + "issue:43716|issue:45507", + "issue:44292|issue:45237", + "issue:45874|issue:46294", + "issue:39290|issue:45081", + "issue:45800|issue:46515", + "issue:45823|issue:45874", + "issue:43425|issue:44292", + "issue:34928|issue:41720", + "issue:43828|issue:45161", + "issue:43828|issue:44292", + "issue:43541|issue:43716", + "issue:43716|issue:46468", + "issue:44292|issue:44918", + "issue:45507|issue:45685", + "issue:43716|issue:44292", + "issue:43716|issue:45685", + "issue:44918|issue:45161", + "issue:43039|issue:44918", + "issue:41762|issue:43638", + "issue:43827|issue:44512", + "issue:43828|issue:45685", + "issue:43039|issue:43828", + "issue:30277|issue:36331", + "issue:34928|issue:36010", + "issue:45237|issue:46015", + "issue:43039|issue:44805" + ], + "prepared_review_unit_hash": "517c12978990242cc8684f5400b8837991f5f08e1019f4aebf3d2c16da4cefe9", + "component_budget": { + "node_count": 502, + "item_count": 502, + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 63, + "serialized_chars": 46010, + "estimated_input_tokens": 11503, + "estimated_eval_tokens": 23262 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This is a heterogeneous issue cluster, not one coherent duplicate set. Only a few soft edges look like true duplicates: the Bert/ZeRO-3 index error pair, the Qwen3.5 MoE rotary-embedding/offload pair, and the torch 2.4.1 compatibility pair.", + "confidence": 0.83, + "canonical_issue_reason": "No single canonical issue fits the whole cluster; 43596 is the best anchor because it is the broadest version of the clearest duplicate family (paired with 43638 on the same ZeRO-3/BERT IndexError).", + "canonical_pr_reason": null, + "best_issue_reason": "43596 is the strongest global representative among the duplicate-like items: it states the core ZeRO-3/BERT index-out-of-bounds bug in the most general way.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44322", + "right": "issue:45464", + "accept": false, + "reason": "Different failures: config attribute mismatch in Qwen3.5 vs streaming chat/completions API error." + }, + { + "left": "issue:45237", + "right": "issue:45685", + "accept": false, + "reason": "Different model/backend problems: AMD GPU failure vs an MPS histogram/int dtype issue." + }, + { + "left": "issue:43716", + "right": "issue:45237", + "accept": false, + "reason": "Different code paths and symptoms: image preprocessor dtype mismatch vs GPT-OSS AMD GPU failure." + }, + { + "left": "issue:43828", + "right": "issue:45237", + "accept": false, + "reason": "Autocast dtype mismatch in Phi-tiny-MoE is unrelated to GPT-OSS-on-AMD failure." + }, + { + "left": "issue:45237", + "right": "issue:46468", + "accept": false, + "reason": "Unrelated model/example failures; not the same underlying bug." + }, + { + "left": "issue:45237", + "right": "issue:45507", + "accept": false, + "reason": "Different components: GPT-OSS GPU issue vs GraniteMoEHybrid invalid method call." + }, + { + "left": "issue:30277", + "right": "issue:35545", + "accept": false, + "reason": "DeepSpeed collective mismatch and ModernBERT ONNX export are unrelated." + }, + { + "left": "issue:35545", + "right": "issue:42915", + "accept": false, + "reason": "ONNX export failure vs Qwen3Moe FP8 failure; different bug families." + }, + { + "left": "issue:35545", + "right": "issue:41720", + "accept": false, + "reason": "Export-time error and Qwen3 auto device-map cuda assert are different problems." + }, + { + "left": "issue:45901", + "right": "issue:46015", + "accept": false, + "reason": "Table-QA crash and outdated RAG docs are unrelated." + }, + { + "left": "issue:44509", + "right": "issue:44512", + "accept": false, + "reason": "Both are docs cleanups, but they refer to different removed docs features and are not the same concrete issue." + }, + { + "left": "issue:35545", + "right": "issue:36010", + "accept": false, + "reason": "ONNX export and GenerationMixin import error are different code paths." + }, + { + "left": "issue:44918", + "right": "issue:46165", + "accept": false, + "reason": "TRL SFT embedding unpacking and remote-code resume training are unrelated." + }, + { + "left": "issue:46165", + "right": "issue:46869", + "accept": false, + "reason": "Remote-code resume training and YOLO multi-GPU device-map failure are unrelated." + }, + { + "left": "issue:44805", + "right": "issue:45823", + "accept": false, + "reason": "Both mention mismatches, but the models and failure modes differ." + }, + { + "left": "issue:43638", + "right": "issue:44805", + "accept": false, + "reason": "ZeRO-3/BERT index error vs mask/tensor shape mismatch are different bugs." + }, + { + "left": "issue:43596", + "right": "issue:44805", + "accept": false, + "reason": "Same broad symptom class, but different model and error path." + }, + { + "left": "issue:44805", + "right": "issue:46294", + "accept": false, + "reason": "Mask shape mismatch and video token/template mismatch are unrelated." + }, + { + "left": "issue:44805", + "right": "issue:45161", + "accept": false, + "reason": "Different issues: tensor mask mismatch vs GPT-OSS tensor-parallel failure." + }, + { + "left": "issue:46165", + "right": "issue:46468", + "accept": false, + "reason": "Unrelated training resume and Nemotron example failures." + }, + { + "left": "issue:43596", + "right": "issue:46165", + "accept": false, + "reason": "ZeRO-3/BERT init bug vs resume-training remote-code issue." + }, + { + "left": "issue:34634", + "right": "issue:35545", + "accept": false, + "reason": "BarkProcessor voice_preset and ModernBERT export are unrelated." + }, + { + "left": "issue:30277", + "right": "issue:43638", + "accept": false, + "reason": "Both involve DeepSpeed, but one is a collective mismatch and the other is a ZeRO-3 empty-index error." + }, + { + "left": "issue:43596", + "right": "issue:43638", + "accept": true, + "reason": "Same underlying ZeRO-3/BERT IndexError family; 43638 is the narrower non-pretrained Bert variant of 43596." + }, + { + "left": "issue:43039", + "right": "issue:45161", + "accept": false, + "reason": "Liger Kernel cross-entropy and GPT-OSS tensor-parallel failure are unrelated." + }, + { + "left": "issue:43975", + "right": "issue:44457", + "accept": false, + "reason": "Incorrect detokenization and LoRA save/reload mismatch are different bugs." + }, + { + "left": "issue:44292", + "right": "issue:44457", + "accept": false, + "reason": "Qwen NVFP4 runtime error and LoRA save/reload mismatch are unrelated." + }, + { + "left": "issue:44841", + "right": "issue:46661", + "accept": false, + "reason": "Processor failure for Voxtral and remote-code config_class import error are different problems." + }, + { + "left": "issue:43716", + "right": "issue:46869", + "accept": false, + "reason": "Different model families and failure modes." + }, + { + "left": "issue:24643", + "right": "issue:30277", + "accept": false, + "reason": "Both are DeepSpeed-related, but the concrete errors and code paths differ." + }, + { + "left": "issue:34928", + "right": "issue:44805", + "accept": false, + "reason": "Activation-checkpointing/FSDP recompute mismatch is not the same as the mask-shape IndexError." + }, + { + "left": "issue:45859", + "right": "issue:45902", + "accept": true, + "reason": "Same Qwen3.5 MoE rotary-embedding/materialization bug; CPU offload failure and uninitialized inv_freq point to the same root cause." + }, + { + "left": "issue:45685", + "right": "issue:46869", + "accept": false, + "reason": "MPS histogram/int issue and YOLO multi-GPU device-map failure are unrelated." + }, + { + "left": "issue:43638", + "right": "issue:45161", + "accept": false, + "reason": "ZeRO-3 Bert init IndexError and GPT-OSS tensor-parallel issue are unrelated." + }, + { + "left": "issue:34928", + "right": "issue:36331", + "accept": false, + "reason": "FSDP activation-checkpointing mismatch and CustomTrainer compute_loss signature change are unrelated." + }, + { + "left": "issue:43596", + "right": "issue:45161", + "accept": false, + "reason": "ZeRO-3/BERT empty-index bug and GPT-OSS tensor-parallel failure are unrelated." + }, + { + "left": "issue:43716", + "right": "issue:45507", + "accept": false, + "reason": "Mistral dtype mismatch and GraniteMoEHybrid invalid method call are different bugs." + }, + { + "left": "issue:44292", + "right": "issue:45237", + "accept": false, + "reason": "Qwen NVFP4 runtime error and GPT-OSS AMD GPU failure are unrelated." + }, + { + "left": "issue:45874", + "right": "issue:46294", + "accept": false, + "reason": "Gemma4 reverse-embedding lookup and Qwen3.5 video token mismatch are unrelated." + }, + { + "left": "issue:39290", + "right": "issue:45081", + "accept": false, + "reason": "Gemma3 sliding-window config error and Mistral regex patch crash are different issues." + }, + { + "left": "issue:45800", + "right": "issue:46515", + "accept": true, + "reason": "These are the same torch 2.4.1 incompatibility issue; 46515 is just the shorter duplicate report." + }, + { + "left": "issue:45823", + "right": "issue:45874", + "accept": false, + "reason": "Same model family, but device mismatch and expensive reverse embedding lookup are different bugs." + }, + { + "left": "issue:43425", + "right": "issue:44292", + "accept": false, + "reason": "Torch-version incompatibility and Qwen NVFP4 runtime error are unrelated." + }, + { + "left": "issue:34928", + "right": "issue:41720", + "accept": false, + "reason": "FSDP recompute mismatch and Qwen3 auto device-map cuda assert are different failures." + }, + { + "left": "issue:43828", + "right": "issue:45161", + "accept": false, + "reason": "Phi autocast dtype mismatch and GPT-OSS tensor-parallel failure are unrelated." + }, + { + "left": "issue:43828", + "right": "issue:44292", + "accept": false, + "reason": "Both are runtime errors, but the models and mechanisms differ." + }, + { + "left": "issue:43541", + "right": "issue:43716", + "accept": false, + "reason": "Mixtral grouped_mm tracing error and Mistral dtype mismatch are unrelated." + }, + { + "left": "issue:43716", + "right": "issue:46468", + "accept": false, + "reason": "Mistral dtype mismatch and Nemotron example failure are unrelated." + }, + { + "left": "issue:44292", + "right": "issue:44918", + "accept": false, + "reason": "Qwen NVFP4 runtime error and TRL SFT embedding unpacking are unrelated." + }, + { + "left": "issue:45507", + "right": "issue:45685", + "accept": false, + "reason": "GraniteMoEHybrid invalid method and MPS histogram/int error are unrelated." + }, + { + "left": "issue:43716", + "right": "issue:44292", + "accept": false, + "reason": "Different models and failure modes." + }, + { + "left": "issue:43716", + "right": "issue:45685", + "accept": false, + "reason": "Mistral dtype mismatch and MPS histogram/int issue are unrelated." + }, + { + "left": "issue:44918", + "right": "issue:45161", + "accept": false, + "reason": "TRL SFT embedding unpacking and GPT-OSS tensor-parallel failure are unrelated." + }, + { + "left": "issue:43039", + "right": "issue:44918", + "accept": false, + "reason": "Liger Kernel cross-entropy issue and Qwen3.5/TRl embedding unpacking are unrelated." + }, + { + "left": "issue:41762", + "right": "issue:43638", + "accept": false, + "reason": "Both are ZeRO-3 IndexErrors, but the models and exact code paths differ enough to avoid merging them." + }, + { + "left": "issue:43827", + "right": "issue:44512", + "accept": false, + "reason": "Both are docs regressions, but they concern different removed features and pages." + }, + { + "left": "issue:43828", + "right": "issue:45685", + "accept": false, + "reason": "Phi autocast dtype mismatch and MPS histogram/int error are unrelated." + }, + { + "left": "issue:43039", + "right": "issue:43828", + "accept": false, + "reason": "Liger Kernel cross-entropy and Phi autocast dtype mismatch are unrelated." + }, + { + "left": "issue:30277", + "right": "issue:36331", + "accept": false, + "reason": "DeepSpeed collective mismatch and compute_loss signature change are unrelated." + }, + { + "left": "issue:34928", + "right": "issue:36010", + "accept": false, + "reason": "FSDP recompute mismatch and GenerationMixin import error are unrelated." + }, + { + "left": "issue:45237", + "right": "issue:46015", + "accept": false, + "reason": "GPT-OSS AMD GPU failure and RAG docs cleanup are unrelated." + }, + { + "left": "issue:43039", + "right": "issue:44805", + "accept": false, + "reason": "Liger Kernel cross-entropy and mask-shape IndexError are unrelated." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays conservative about the soft edges. The three accepted duplicate-like pairs are well supported by the titles, and the rejected pairs are generally justified as different bug families or code paths." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 2, + "review_unit_count": 61, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-502", + "nodes": [ + "issue:36010", + "issue:39290", + "issue:41093", + "issue:41720", + "issue:41762", + "issue:43039", + "issue:43296", + "issue:43404", + "issue:43425", + "issue:43531", + "issue:43541", + "issue:43572", + "issue:43596", + "issue:43638", + "issue:43792", + "issue:43828", + "issue:43854", + "issue:43866", + "issue:43872", + "issue:43901", + "issue:43975", + "issue:44292", + "issue:44512", + "issue:44560", + "issue:44589", + "issue:44841", + "issue:44863", + "issue:45020", + "issue:45070", + "issue:45084", + "issue:45237", + "issue:45362", + "issue:45399", + "issue:45464", + "issue:45507", + "issue:45685", + "issue:45800", + "issue:45812", + "issue:45823", + "issue:45901", + "issue:45920", + "issue:46015", + "issue:46294", + "issue:46304", + "issue:46402", + "issue:46661", + "issue:46772", + "issue:46869" + ], + "soft_pairs": [ + "issue:43866|issue:46402", + "issue:45800|issue:46304", + "issue:43296|issue:46304", + "issue:41762|issue:43596", + "issue:39290|issue:41720", + "issue:44589|issue:46304", + "issue:43901|issue:44512", + "issue:45237|issue:45901", + "issue:43975|issue:46015", + "issue:43039|issue:44560", + "issue:45823|issue:46294", + "issue:44841|issue:45823", + "issue:43866|issue:46772", + "issue:44863|issue:46015", + "issue:45020|issue:46304", + "issue:45812|issue:45920", + "issue:43866|issue:44863", + "issue:43872|issue:46304", + "issue:44292|issue:46015", + "issue:39290|issue:43531", + "issue:41093|issue:43638", + "issue:43792|issue:45901", + "issue:43854|issue:43866", + "issue:43425|issue:45070", + "issue:43975|issue:45901", + "issue:43792|issue:45070", + "issue:43596|issue:45800", + "issue:45362|issue:45464", + "issue:43541|issue:46869", + "issue:44292|issue:45901", + "issue:39290|issue:45823", + "issue:45464|issue:45685", + "issue:36010|issue:39290", + "issue:43541|issue:45685", + "issue:43541|issue:43828", + "issue:44841|issue:45084", + "issue:45507|issue:46869", + "issue:43404|issue:43866", + "issue:41093|issue:41762", + "issue:45399|issue:46661", + "issue:43572|issue:45070", + "issue:45800|issue:45823" + ], + "prepared_review_unit_hash": "9efb9a731d66e7097e6ff5b62f4e59e1a73167e88f95a8f1bdd5b3190de84c38", + "component_budget": { + "node_count": 502, + "item_count": 502, + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 42, + "serialized_chars": 40044, + "estimated_input_tokens": 10011, + "estimated_eval_tokens": 20278 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly heterogeneous: one real duplicate-like tokenizer regression pair stands out, and one ZeRO-3 loading pair is plausibly the same bug. The rest look like unrelated model-specific failures or docs issues.", + "confidence": 0.74, + "canonical_issue_reason": "Issue 45920 is the best canonical anchor: it describes the broad v5 AutoTokenizer token-ID regression across multiple models, while 45812 is the narrower Granite-specific report of the same underlying bug.", + "canonical_pr_reason": null, + "best_issue_reason": "45920 is the strongest representative because it captures the widest symptom set for the tokenizer regression and is the clearest issue to attach follow-up fixes to.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43866", + "right": "issue:46402", + "accept": false, + "reason": "Different failures: corrupted checkpoint vs missing loaded weights in a different model." + }, + { + "left": "issue:45800", + "right": "issue:46304", + "accept": false, + "reason": "Both mention torch/transformers compatibility, but they describe different breakages and code paths." + }, + { + "left": "issue:43296", + "right": "issue:46304", + "accept": false, + "reason": "Different runtime stacks and symptoms; no clear shared underlying bug." + }, + { + "left": "issue:41762", + "right": "issue:43596", + "accept": true, + "reason": "Both report the same ZeRO-3 loading path hitting an index-out-of-bounds size-0 error during model initialization." + }, + { + "left": "issue:39290", + "right": "issue:41720", + "accept": false, + "reason": "Unrelated failures: a Gemma3/vLLM config attribute error versus a Qwen3 device-mapping CUDA assert." + }, + { + "left": "issue:44589", + "right": "issue:46304", + "accept": false, + "reason": "Storage-backend/type lookup failure is unrelated to the distributed build regression in 46304." + }, + { + "left": "issue:43901", + "right": "issue:44512", + "accept": false, + "reason": "Both are docs-related, but they target different documentation problems and should not be merged." + }, + { + "left": "issue:45237", + "right": "issue:45901", + "accept": false, + "reason": "Completely different subsystems: AMD GPU support versus table-question-answering pipeline crash." + }, + { + "left": "issue:43975", + "right": "issue:46015", + "accept": false, + "reason": "Tokenizer detokenization bug and outdated RAG docs are unrelated." + }, + { + "left": "issue:43039", + "right": "issue:44560", + "accept": false, + "reason": "Liger kernel cross-entropy routing is unrelated to the video StopIteration failure." + }, + { + "left": "issue:45823", + "right": "issue:46294", + "accept": false, + "reason": "Both are multimodal model issues, but one is device placement and the other is chat-template video token mismatch." + }, + { + "left": "issue:44841", + "right": "issue:45823", + "accept": false, + "reason": "Different models and different failure modes; no shared code-path bug is evident." + }, + { + "left": "issue:43866", + "right": "issue:46772", + "accept": false, + "reason": "Checkpoint corruption is not the same as NVFP4/ModelOpt compatibility." + }, + { + "left": "issue:44863", + "right": "issue:46015", + "accept": false, + "reason": "Model-loading failure and docs cleanup are unrelated." + }, + { + "left": "issue:45020", + "right": "issue:46304", + "accept": false, + "reason": "Remote-code breakage and USE_DISTRIBUTED=0 support regression are distinct issues." + }, + { + "left": "issue:45812", + "right": "issue:45920", + "accept": true, + "reason": "Same underlying v5 AutoTokenizer regression: both report wrong token IDs across models, with 45920 being the broader report." + }, + { + "left": "issue:43866", + "right": "issue:44863", + "accept": false, + "reason": "Corrupted checkpoint versus missing checkpoint-loading support; not the same bug." + }, + { + "left": "issue:43872", + "right": "issue:46304", + "accept": false, + "reason": "bitsandbytes Int8Params API incompatibility is unrelated to the PyTorch distributed regression." + }, + { + "left": "issue:44292", + "right": "issue:46015", + "accept": false, + "reason": "NVFP4 runtime error and RAG docs are unrelated." + }, + { + "left": "issue:39290", + "right": "issue:43531", + "accept": false, + "reason": "Both mention sliding-window behavior, but the affected models and concrete failures differ." + }, + { + "left": "issue:41093", + "right": "issue:43638", + "accept": false, + "reason": "Both are IndexError reports, but one is a mask/tensor shape mismatch and the other is a ZeRO-3 model-init failure." + }, + { + "left": "issue:43792", + "right": "issue:45901", + "accept": false, + "reason": "Whisper load failure and table-question-answering crash are unrelated." + }, + { + "left": "issue:43854", + "right": "issue:43866", + "accept": false, + "reason": "Different model families and different failure modes." + }, + { + "left": "issue:43425", + "right": "issue:45070", + "accept": false, + "reason": "Torch version compatibility and pydantic PretrainedConfig regression are separate problems." + }, + { + "left": "issue:43975", + "right": "issue:45901", + "accept": false, + "reason": "Tokenizer detokenization regression is unrelated to the table QA crash." + }, + { + "left": "issue:43792", + "right": "issue:45070", + "accept": false, + "reason": "Whisper runtime failure does not match the pydantic model regression." + }, + { + "left": "issue:43596", + "right": "issue:45800", + "accept": false, + "reason": "ZeRO-3 loading IndexError and torch/transformers incompatibility are not the same bug." + }, + { + "left": "issue:45362", + "right": "issue:45464", + "accept": false, + "reason": "Both involve Qwen3.5, but one is chat crashes and the other is streaming chat/completions API failure." + }, + { + "left": "issue:43541", + "right": "issue:46869", + "accept": false, + "reason": "Grouped_mm tracing failure in Mixtral is unrelated to YOLO multi-GPU device-mapping." + }, + { + "left": "issue:44292", + "right": "issue:45901", + "accept": false, + "reason": "NVFP4 model runtime issue and table QA crash are unrelated." + }, + { + "left": "issue:39290", + "right": "issue:45823", + "accept": false, + "reason": "Gemma3/vLLM config error and Gemma4 device-mismatch are different bugs." + }, + { + "left": "issue:45464", + "right": "issue:45685", + "accept": false, + "reason": "Streaming chat API failure is unrelated to the MPS histogram Int error." + }, + { + "left": "issue:36010", + "right": "issue:39290", + "accept": false, + "reason": "ImportError for GenerationMixin is unrelated to the Gemma3/vLLM config issue." + }, + { + "left": "issue:43541", + "right": "issue:45685", + "accept": false, + "reason": "Different model families and different backend failures." + }, + { + "left": "issue:43541", + "right": "issue:43828", + "accept": false, + "reason": "Torch dynamo grouped_mm tracing failure and autocast dtype mismatch are distinct issues." + }, + { + "left": "issue:44841", + "right": "issue:45084", + "accept": false, + "reason": "Processor failure and non-template-node compile error are unrelated." + }, + { + "left": "issue:45507", + "right": "issue:46869", + "accept": false, + "reason": "Invalid method call in GraniteMoEHybrid is not the same as YOLO multi-GPU device-map failure." + }, + { + "left": "issue:43404", + "right": "issue:43866", + "accept": false, + "reason": "Weight-tying bug in Mistral3 is unrelated to checkpoint corruption." + }, + { + "left": "issue:41093", + "right": "issue:41762", + "accept": false, + "reason": "Both are IndexErrors, but the observed failure surfaces and contexts do not match." + }, + { + "left": "issue:45399", + "right": "issue:46661", + "accept": false, + "reason": "Flash-attn fallback logic and remote-code model_class/config_class handling are unrelated." + }, + { + "left": "issue:43572", + "right": "issue:45070", + "accept": false, + "reason": "Missing StableLmConfig field and pydantic PretrainedConfig regression are different configuration bugs." + }, + { + "left": "issue:45800", + "right": "issue:45823", + "accept": false, + "reason": "Torch version compatibility and Gemma4 device placement are unrelated." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays conservative overall. The canonicalization around 45812/45920 is well supported by the issue titles, and the ZeRO-3 pair is a plausible duplicate-style match given the identical index-out-of-bounds/sized-0 failure pattern during model loading. No soft-edge verdicts appear overstated based on the available evidence." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 3, + "review_unit_count": 61, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-502", + "nodes": [ + "issue:39186", + "issue:40990", + "issue:41093", + "issue:42915", + "issue:43054", + "issue:43296", + "issue:43366", + "issue:43404", + "issue:43531", + "issue:43541", + "issue:43596", + "issue:43645", + "issue:43716", + "issue:43828", + "issue:43873", + "issue:43994", + "issue:44291", + "issue:44360", + "issue:44387", + "issue:44479", + "issue:44492", + "issue:44512", + "issue:44530", + "issue:44805", + "issue:44863", + "issue:44918", + "issue:44960", + "issue:45005", + "issue:45020", + "issue:45070", + "issue:45161", + "issue:45183", + "issue:45399", + "issue:45464", + "issue:45507", + "issue:45632", + "issue:45698", + "issue:45759", + "issue:45800", + "issue:45902", + "issue:46294", + "issue:46463", + "issue:46468", + "issue:46489", + "issue:46491", + "issue:46640", + "issue:46661", + "issue:46869" + ], + "soft_pairs": [ + "issue:43296|issue:43366", + "issue:44387|issue:45005", + "issue:44805|issue:44918", + "issue:43541|issue:45161", + "issue:43828|issue:45464", + "issue:43366|issue:45464", + "issue:44291|issue:45070", + "issue:43531|issue:45070", + "issue:41093|issue:43596", + "issue:45020|issue:45800", + "issue:43716|issue:46294", + "issue:43054|issue:43994", + "issue:44479|issue:46294", + "issue:46468|issue:46869", + "issue:45759|issue:45902", + "issue:46640|issue:46661", + "issue:43716|issue:44918", + "issue:39186|issue:44918", + "issue:40990|issue:44960", + "issue:43645|issue:45399", + "issue:45632|issue:45698", + "issue:43404|issue:45005", + "issue:42915|issue:45399", + "issue:43873|issue:45005", + "issue:44530|issue:45005", + "issue:44360|issue:44512", + "issue:44492|issue:44512", + "issue:44863|issue:45005", + "issue:45183|issue:46463", + "issue:46489|issue:46491", + "issue:44918|issue:45507" + ], + "prepared_review_unit_hash": "311ee8df90ca91f2472bfe4d4dd60b368ea9f3628129752a7841e4f2b4bd6434", + "component_budget": { + "node_count": 502, + "item_count": 502, + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 31, + "serialized_chars": 36846, + "estimated_input_tokens": 9212, + "estimated_eval_tokens": 18680 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is heterogeneous: it spans unrelated model-loading, tokenizer, quantization, multimodal, and runtime errors. I do not see a single duplicate issue to collapse the set, and none of the soft pairs look like the same concrete bug or change.", + "confidence": 0.95, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": "No issue is a strong canonical representative for the whole cluster because the topics diverge substantially. If forced, issue 45698 is one of the clearest self-contained bug reports, but it is not a true cluster anchor.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43296", + "right": "issue:43366", + "accept": false, + "reason": "Different failures: PaddleOCR-VL load error vs GGUF gpt-oss architecture support. Same broad area (model support), but not the same bug." + }, + { + "left": "issue:44387", + "right": "issue:45005", + "accept": false, + "reason": "Both mention v5 and weight/quantization-related regressions, but one is int4 CUDA reserved memory OOM and the other is tied weights on translation models. Not the same underlying defect." + }, + { + "left": "issue:44805", + "right": "issue:44918", + "accept": false, + "reason": "Both involve shape/packing-style errors, but one is a mask-vs-tensor indexing mismatch and the other is Qwen3.5 input embedding unpacking with TRL SFT. Different code paths and symptoms." + }, + { + "left": "issue:43541", + "right": "issue:45161", + "accept": false, + "reason": "Both are MoE/runtime issues, but one is grouped_mm tracing failure for Mixtral float32 and the other is TP not working for GPT-OSS MoE. Not the same bug." + }, + { + "left": "issue:43828", + "right": "issue:45464", + "accept": false, + "reason": "Different modalities and failures: autocast dtype mismatch in Phi-tiny-MoE vs streaming chat/completions failure on Qwen3.5-0.8B." + }, + { + "left": "issue:43366", + "right": "issue:45464", + "accept": false, + "reason": "GGUF gpt-oss architecture support vs Qwen3.5 streaming inference. Shared inference context only, not the same issue." + }, + { + "left": "issue:44291", + "right": "issue:45070", + "accept": false, + "reason": "Both are v5 compatibility regressions, but one is init_empty_weights/_is_hf_initialized TypeError and the other is PretrainedConfig field breakage in pydantic models." + }, + { + "left": "issue:43531", + "right": "issue:45070", + "accept": false, + "reason": "Sliding-window behavior in Qwen3-MoE is unrelated to the pydantic PretrainedConfig regression." + }, + { + "left": "issue:41093", + "right": "issue:43596", + "accept": false, + "reason": "Both are IndexError reports, but the first is a mask-length mismatch and the second is zero-sized tensor indexing during deepspeed zero3 init. Different causes." + }, + { + "left": "issue:45020", + "right": "issue:45800", + "accept": false, + "reason": "Both are broad compatibility complaints, but one is remote_code breakage and the other is torch 2.4.1 vs transformers 5.8.0 incompatibility. Too generic to merge." + }, + { + "left": "issue:43716", + "right": "issue:46294", + "accept": false, + "reason": "Both touch Qwen3.5/multimodal input handling, but one is image-preprocessor dtype mismatch and the other is video feature/token count mismatch." + }, + { + "left": "issue:43054", + "right": "issue:43994", + "accept": false, + "reason": "Both are SigLIP/SigLIP2 quality regressions, but one is text embedding quality and the other is nonsensical AutoModel/pipeline output. Related family, different bug." + }, + { + "left": "issue:44479", + "right": "issue:46294", + "accept": false, + "reason": "Both involve Qwen3.5 multimodal video, but one is a v5.3.0 video-input regression across several models and the other is a chat template feature/token mismatch. Not the same concrete defect." + }, + { + "left": "issue:46468", + "right": "issue:46869", + "accept": false, + "reason": "Different model stacks and failures: broken Nemotron example vs YOLO multi-GPU device_map auto." + }, + { + "left": "issue:45759", + "right": "issue:45902", + "accept": false, + "reason": "Both concern Qwen 3.5/3.6 internals, but one is from_config text_config unwrapping and the other is uninitialized rotary embedding memory after to_empty materialization." + }, + { + "left": "issue:46640", + "right": "issue:46661", + "accept": false, + "reason": "Flash Attention 2 issue vs remote-code model_class.config_class AttributeError. No shared code-path." + }, + { + "left": "issue:43716", + "right": "issue:44918", + "accept": false, + "reason": "Different multimodal/training failures: image preprocessor dtype mismatch vs TRL SFT unpacking of Qwen3.5 embeddings." + }, + { + "left": "issue:39186", + "right": "issue:44918", + "accept": false, + "reason": "FSDP weight-dimension RuntimeError is unrelated to Qwen3.5 embedding unpacking." + }, + { + "left": "issue:40990", + "right": "issue:44960", + "accept": false, + "reason": "High perplexity on gpt-oss-20b and a generic GLM5 issue are unrelated and at very different abstraction levels." + }, + { + "left": "issue:43645", + "right": "issue:45399", + "accept": false, + "reason": "Jupyter custom-model initialization breakage is unrelated to flash-attn2 fallback gating." + }, + { + "left": "issue:45632", + "right": "issue:45698", + "accept": false, + "reason": "Both involve remote-code/custom-module loading, but one is cache-path collision for local models sharing a leaf name and the other is loading the wrong custom module after save_pretrained. Similar area, not the same bug." + }, + { + "left": "issue:43404", + "right": "issue:45005", + "accept": false, + "reason": "Mistral3 tied lm_head weight bug vs translation-model tied weights in v5. Different model families and failure modes." + }, + { + "left": "issue:42915", + "right": "issue:45399", + "accept": false, + "reason": "Qwen3-MoE fine-grained FP8 config failure is unrelated to flash-attn2 fallback blocking." + }, + { + "left": "issue:43873", + "right": "issue:45005", + "accept": false, + "reason": "Quantization offloading behavior vs tied-weights regression. Both are v5-era infra issues, but not the same defect." + }, + { + "left": "issue:44530", + "right": "issue:45005", + "accept": false, + "reason": "PagedAttentionCache linear_attention crash on Qwen3.5 is unrelated to tied weights on translation models." + }, + { + "left": "issue:44360", + "right": "issue:44512", + "accept": false, + "reason": "DSA indexer ReLU discussion is unrelated to docs mentioning a removed run command." + }, + { + "left": "issue:44492", + "right": "issue:44512", + "accept": false, + "reason": "Cache-strategy typo and removed docs command are both documentation-ish, but not the same issue." + }, + { + "left": "issue:44863", + "right": "issue:45005", + "accept": false, + "reason": "NemotronH checkpoint-loading failure is unrelated to translation-model tied weights." + }, + { + "left": "issue:45183", + "right": "issue:46463", + "accept": false, + "reason": "Both are validation/error-message issues, but one is XOR validation for input_ids/inputs_embeds across models and the other is decoder input validation messaging for T5. Similar pattern, different target and likely different fix." + }, + { + "left": "issue:46489", + "right": "issue:46491", + "accept": false, + "reason": "Both are tokenizer regressions, but they affect different model families and different exact round-trip behaviors. Too broad to treat as one duplicate." + }, + { + "left": "issue:44918", + "right": "issue:45507", + "accept": false, + "reason": "Qwen3.5 embedding unpacking with TRL SFT is unrelated to GraniteMoEHybrid calling an invalid method." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 4, + "review_unit_count": 61, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-502", + "nodes": [ + "issue:24643", + "issue:28530", + "issue:36331", + "issue:38175", + "issue:39692", + "issue:41553", + "issue:41762", + "issue:42915", + "issue:42947", + "issue:43054", + "issue:43493", + "issue:43643", + "issue:43644", + "issue:43782", + "issue:43824", + "issue:43828", + "issue:43856", + "issue:43872", + "issue:43881", + "issue:43883", + "issue:43950", + "issue:43975", + "issue:44292", + "issue:44387", + "issue:44403", + "issue:44451", + "issue:44488", + "issue:44534", + "issue:44589", + "issue:44910", + "issue:44928", + "issue:44960", + "issue:44991", + "issue:45020", + "issue:45399", + "issue:45520", + "issue:45663", + "issue:45685", + "issue:45701", + "issue:45823", + "issue:45859", + "issue:45902", + "issue:45920", + "issue:45987", + "issue:46489", + "issue:46858", + "issue:46899", + "issue:46932" + ], + "soft_pairs": [ + "issue:43975|issue:44292", + "issue:43950|issue:44534", + "issue:39692|issue:43054", + "issue:43644|issue:43950", + "issue:36331|issue:41762", + "issue:43872|issue:44589", + "issue:46899|issue:46932", + "issue:43828|issue:44928", + "issue:45920|issue:46489", + "issue:45987|issue:46858", + "issue:43054|issue:45701", + "issue:38175|issue:43054", + "issue:43950|issue:45902", + "issue:41553|issue:43883", + "issue:43054|issue:43493", + "issue:44910|issue:45859", + "issue:42947|issue:44387", + "issue:45663|issue:45902", + "issue:41553|issue:43881", + "issue:44403|issue:44451", + "issue:43975|issue:45520", + "issue:44403|issue:44488", + "issue:24643|issue:28530", + "issue:42947|issue:43856", + "issue:41762|issue:42915", + "issue:43643|issue:45020", + "issue:44991|issue:45399", + "issue:43824|issue:43872", + "issue:43782|issue:44960", + "issue:43881|issue:43883", + "issue:45685|issue:45823" + ], + "prepared_review_unit_hash": "7b75f5370cfc9d858b6aeeb348c9e27af4c72107f76aaf9e1736e51305a7ce83", + "component_budget": { + "node_count": 502, + "item_count": 502, + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 31, + "serialized_chars": 37127, + "estimated_input_tokens": 9282, + "estimated_eval_tokens": 18820 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Mostly a heterogeneous batch with a few clear duplicate pairs: non-persistent buffer corruption in v5, SigLIP2 quantized embedder dtype casting, and cache length return-type regressions.", + "confidence": 0.78, + "canonical_issue_reason": "43950 is the best anchor for the non-persistent-buffer regression cluster: it states the concrete v5 corruption bug clearly, and the nearby reports 43644 and 44534 are the same issue phrased more tersely.", + "canonical_pr_reason": null, + "best_issue_reason": "43950 is the most canonical issue overall because it is the most specific, technical description of a real regression and is supported by multiple near-identical reports.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43975", + "right": "issue:44292", + "accept": false, + "reason": "Different models and failure modes: tokenizer detokenization vs loading a Qwen NVFP4 model." + }, + { + "left": "issue:43950", + "right": "issue:44534", + "accept": true, + "reason": "Same v5 regression corrupting non-persistent buffers; 44534 is just a shorter restatement." + }, + { + "left": "issue:39692", + "right": "issue:43054", + "accept": false, + "reason": "Docs-example model/processor mismatch and quantization errors are not the same as a SigLIP2 text-embedding quality regression." + }, + { + "left": "issue:43644", + "right": "issue:43950", + "accept": true, + "reason": "Same non-persistent-buffer junk/corruption bug in Transformers 5.x." + }, + { + "left": "issue:36331", + "right": "issue:41762", + "accept": false, + "reason": "Unrelated training API signature issue vs DeepSpeed ZeRO-3 loading crash." + }, + { + "left": "issue:43872", + "right": "issue:44589", + "accept": false, + "reason": "Both are quantization/loading problems, but the error types and code paths are different." + }, + { + "left": "issue:46899", + "right": "issue:46932", + "accept": true, + "reason": "Same Gemma4 quantized vision/audio embedder dtype-casting bug; 46932 explicitly calls it out as the same issue." + }, + { + "left": "issue:43828", + "right": "issue:44928", + "accept": false, + "reason": "Different models and different root causes: autocast dtype mismatch vs 3D position_ids causing flash-attn/SDPA issues." + }, + { + "left": "issue:45920", + "right": "issue:46489", + "accept": false, + "reason": "Both concern tokenizer correctness, but 46489 is a model-specific follow-up and not clearly the same bug as the broader multi-model regression." + }, + { + "left": "issue:45987", + "right": "issue:46858", + "accept": true, + "reason": "Same cache-seq-length contract bug: tensor return value instead of int, with 46858 being a backend-specific manifestation." + }, + { + "left": "issue:43054", + "right": "issue:45701", + "accept": false, + "reason": "SigLIP2 text-embedding quality issue is unrelated to a generic tokenizer-change report." + }, + { + "left": "issue:38175", + "right": "issue:43054", + "accept": false, + "reason": "Both are SigLIP2-related, but zero probabilities and degraded text embeddings are not clearly the same underlying bug." + }, + { + "left": "issue:43950", + "right": "issue:45902", + "accept": false, + "reason": "Shared theme of non-persistent buffers, but 45902 is a specific materialization/uninitialized-memory bug rather than the same from_pretrained corruption regression." + }, + { + "left": "issue:41553", + "right": "issue:43883", + "accept": false, + "reason": "Bad AutoTokenizer error message for Voxtral is unrelated to Molmo tied-weights loading failure." + }, + { + "left": "issue:43054", + "right": "issue:43493", + "accept": true, + "reason": "Both point to the same SigLIP2 implementation discrepancy; the JAX-vs-HF mismatch plausibly explains the degraded text embeddings." + }, + { + "left": "issue:44910", + "right": "issue:45859", + "accept": false, + "reason": "Different Qwen3.5 MoE issues: flash-attn packed-sequence misinterpretation vs CPU-offload incompatibility in rotary embedding." + }, + { + "left": "issue:42947", + "right": "issue:44387", + "accept": false, + "reason": "Both are memory-related, but gradient checkpointing/LoRA inefficiency is not the same bug as int4 reserved-memory OOM." + }, + { + "left": "issue:45663", + "right": "issue:45902", + "accept": false, + "reason": "Different Gemma4/FSDP and buffer-materialization problems with no clear shared code-path." + }, + { + "left": "issue:41553", + "right": "issue:43881", + "accept": false, + "reason": "Different models and failures: Voxtral tokenizer loading vs glm-4v-9b loading." + }, + { + "left": "issue:44403", + "right": "issue:44451", + "accept": false, + "reason": "Generic loading noise is not the same underlying problem as a specific model load failure." + }, + { + "left": "issue:43975", + "right": "issue:45520", + "accept": false, + "reason": "Tokenizer detokenization bug and flash-attn import KeyError are unrelated." + }, + { + "left": "issue:44403", + "right": "issue:44488", + "accept": false, + "reason": "Again, generic loader noise vs a specific Sleng-BERT load failure." + }, + { + "left": "issue:24643", + "right": "issue:28530", + "accept": false, + "reason": "Completely different training errors: DeepSpeed weight shape vs early-stopping metric selection." + }, + { + "left": "issue:42947", + "right": "issue:43856", + "accept": false, + "reason": "Memory usage symptoms are too broad; LoRA checkpointing inefficiency is not the same as Qwen3 MoE training memory behavior." + }, + { + "left": "issue:41762", + "right": "issue:42915", + "accept": false, + "reason": "Different model families and different failure mechanisms." + }, + { + "left": "issue:43643", + "right": "issue:45020", + "accept": false, + "reason": "43643 is a specific missing-fields bug in AutoConfig with trust_remote_code; 45020 is a broad remote_code regression report, not clearly the same underlying issue." + }, + { + "left": "issue:44991", + "right": "issue:45399", + "accept": false, + "reason": "Tokenizer loading failure and flash-attn fallback logic are unrelated." + }, + { + "left": "issue:43824", + "right": "issue:43872", + "accept": false, + "reason": "Importing a missing VL class is unrelated to bitsandbytes Int8Params constructor incompatibility." + }, + { + "left": "issue:43782", + "right": "issue:44960", + "accept": false, + "reason": "Qwen3VL weight_only loading failure and a GLM5 issue are not the same bug." + }, + { + "left": "issue:43881", + "right": "issue:43883", + "accept": false, + "reason": "Different model loading failures with different missing components." + }, + { + "left": "issue:45685", + "right": "issue:45823", + "accept": false, + "reason": "MPS histogram-int support error and Gemma4 device mismatch are distinct backend/model issues." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and the accepted soft-edge calls are reasonably conservative. The main duplicate clusters called out are supported by the titles: non-persistent-buffer corruption (43950/43644/44534), SigLIP2-related discrepancy/casting issues (43054/43493 and 46899/46932), and cache-length return-type regressions (45987/46858). Rejections are also appropriately cautious." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 5, + "review_unit_count": 61, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-502", + "nodes": [ + "issue:24643", + "issue:30064", + "issue:36331", + "issue:41093", + "issue:41553", + "issue:42491", + "issue:42947", + "issue:43257", + "issue:43404", + "issue:43792", + "issue:43827", + "issue:43828", + "issue:43854", + "issue:43872", + "issue:43940", + "issue:43950", + "issue:44220", + "issue:44291", + "issue:44368", + "issue:44479", + "issue:44488", + "issue:44509", + "issue:44530", + "issue:44560", + "issue:44661", + "issue:44805", + "issue:44863", + "issue:44928", + "issue:44991", + "issue:45020", + "issue:45081", + "issue:45305", + "issue:45313", + "issue:45356", + "issue:45399", + "issue:45488", + "issue:45588", + "issue:45715", + "issue:45759", + "issue:45823", + "issue:45923", + "issue:46315", + "issue:46402", + "issue:46489", + "issue:46661", + "issue:46869", + "issue:46894", + "issue:46897" + ], + "soft_pairs": [ + "issue:45081|issue:45399", + "issue:45020|issue:45823", + "issue:44661|issue:45020", + "issue:42491|issue:44368", + "issue:43828|issue:45823", + "issue:45020|issue:45356", + "issue:45715|issue:46315", + "issue:45488|issue:46489", + "issue:43827|issue:44509", + "issue:43872|issue:46869", + "issue:46894|issue:46897", + "issue:45759|issue:46661", + "issue:43404|issue:43950", + "issue:43872|issue:44291", + "issue:45020|issue:46661", + "issue:43792|issue:44220", + "issue:43257|issue:45305", + "issue:41553|issue:44991", + "issue:36331|issue:41093", + "issue:42947|issue:45923", + "issue:43854|issue:44863", + "issue:43940|issue:45313", + "issue:41553|issue:44488", + "issue:44928|issue:45588", + "issue:41093|issue:44805", + "issue:44479|issue:44560", + "issue:24643|issue:30064", + "issue:44530|issue:44863", + "issue:44863|issue:46402" + ], + "prepared_review_unit_hash": "e7d52ec639e9bf6032ed3a883fb19eef029eac328fc1110dcf078e9142c60bfa", + "component_budget": { + "node_count": 502, + "item_count": 502, + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 29, + "serialized_chars": 37086, + "estimated_input_tokens": 9272, + "estimated_eval_tokens": 18800 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly heterogeneous, with a few clear duplicate subgroups: docs for removed pipeline tasks, the `_is_hf_initialized` / `init_empty_weights` loading regression, a mask-shape IndexError, and a pair of nearly identical training-loss double-shift reports.", + "confidence": 0.84, + "canonical_issue_reason": "No single cluster-wide canonical issue fits the whole set; if choosing the clearest duplicate subgroup, issue 43827 is the best root-cause report for the v5 docs/pipeline-task removal regression.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 43827 is the strongest issue candidate because it states the underlying docs regression precisely and matches a near-identical follow-up report (44509).", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:45081", + "right": "issue:45399", + "accept": false, + "reason": "Different failures: tokenizer regex crash vs flash-attn2 fallback gating; same broad area, not the same bug." + }, + { + "left": "issue:45020", + "right": "issue:45823", + "accept": false, + "reason": "`remote_code` compatibility regression is unrelated to a Gemma4 device-map/device placement mismatch." + }, + { + "left": "issue:44661", + "right": "issue:45020", + "accept": false, + "reason": "`TOKENIZER_MAPPING_NAMES` handling is a specific tokenizer registration bug, not the same as the broader remote-code regression." + }, + { + "left": "issue:42491", + "right": "issue:44368", + "accept": false, + "reason": "Qwen3 MoE hf4\u2192hf5 LoRA incompatibility and a tie_word_embeddings warning are different issues." + }, + { + "left": "issue:43828", + "right": "issue:45823", + "accept": false, + "reason": "Autocast dtype mismatch on Phi-tiny-MoE is a different code-path from the Gemma4 device-map issue." + }, + { + "left": "issue:45020", + "right": "issue:45356", + "accept": false, + "reason": "A broad remote-code regression is not the same as the Kimi-K2.5 tokenizer codec regression." + }, + { + "left": "issue:45715", + "right": "issue:46315", + "accept": false, + "reason": "Both are tokenizer performance regressions, but they hit different methods and symptoms (`convert_ids_to_tokens` loop rebuild vs `convert_tokens_to_ids` slowdown)." + }, + { + "left": "issue:45488", + "right": "issue:46489", + "accept": false, + "reason": "Both are DeepSeek tokenizer regressions, but the titles point to different affected families and one is framed as a gap in a prior fix; not clearly the same concrete bug." + }, + { + "left": "issue:43827", + "right": "issue:44509", + "accept": true, + "reason": "Both report the same v5 docs regression: obsolete pipeline task references after pipeline removals." + }, + { + "left": "issue:43872", + "right": "issue:46869", + "accept": false, + "reason": "Bitsandbytes `_is_hf_initialized` incompatibility is unrelated to a YOLO multi-GPU `device_map='auto'` failure." + }, + { + "left": "issue:46894", + "right": "issue:46897", + "accept": true, + "reason": "Near-identical training-loss double-shift reports with the same failure pattern; only the example model differs." + }, + { + "left": "issue:45759", + "right": "issue:46661", + "accept": false, + "reason": "`text_config` unwrapping for composite Qwen configs is not the same as the remote-code `config_class` AttributeError." + }, + { + "left": "issue:43404", + "right": "issue:43950", + "accept": false, + "reason": "`lm_head` tying in Mistral3 and non-persistent buffer corruption are different model-loading bugs." + }, + { + "left": "issue:43872", + "right": "issue:44291", + "accept": true, + "reason": "These describe the same `_is_hf_initialized` / `init_empty_weights` incompatibility; 43872 is the concrete bitsandbytes symptom and 44291 names the root cause." + }, + { + "left": "issue:45020", + "right": "issue:46661", + "accept": false, + "reason": "The remote-code umbrella issue is broader; 46661 is one specific `config_class` failure mode, not enough to merge as a duplicate." + }, + { + "left": "issue:43792", + "right": "issue:44220", + "accept": false, + "reason": "A model-level Whisper failure and a low-level fbank feature-extraction issue are related but not clearly the same bug." + }, + { + "left": "issue:43257", + "right": "issue:45305", + "accept": false, + "reason": "DeepSpeed ZeRO-3 weight loading and DeepSpeed gradient averaging are different problems." + }, + { + "left": "issue:41553", + "right": "issue:44991", + "accept": false, + "reason": "Bad error messaging for Voxtral tokenizer loading and a tokenizer load failure for est-roberta are different regressions." + }, + { + "left": "issue:36331", + "right": "issue:41093", + "accept": false, + "reason": "CustomTrainer API breakage and a mask/tensor shape mismatch are unrelated." + }, + { + "left": "issue:42947", + "right": "issue:45923", + "accept": false, + "reason": "Gradient checkpointing ineffective with PEFT LoRA is not the same as a missing `supports_gradient_checkpointing` flag on a trust_remote_code variant." + }, + { + "left": "issue:43854", + "right": "issue:44863", + "accept": false, + "reason": "Different model-loading failures for different model families; no shared concrete bug is evident." + }, + { + "left": "issue:43940", + "right": "issue:45313", + "accept": false, + "reason": "Both mention ZeRO-3 weight loading, but one is Qwen3-Next all-params-missing and the other is Qwen3.5 language_model weights missing; too different to collapse." + }, + { + "left": "issue:41553", + "right": "issue:44488", + "accept": false, + "reason": "Voxtral tokenizer error messaging and sleng-bert load failure are unrelated." + }, + { + "left": "issue:44928", + "right": "issue:45588", + "accept": false, + "reason": "RLHF gradient explosion from 3D position_ids / SDPA fallback is a different bug from the flash-attention `s_aux=None` crash." + }, + { + "left": "issue:41093", + "right": "issue:44805", + "accept": true, + "reason": "Same mask-shape mismatch IndexError; one report gives concrete dimensions, the other is the same error class more generally." + }, + { + "left": "issue:44479", + "right": "issue:44560", + "accept": false, + "reason": "Both are video-related Qwen regressions, but the reported failures differ (`video input` regression vs `StopIteration` in embedding)." + }, + { + "left": "issue:24643", + "right": "issue:30064", + "accept": false, + "reason": "DeepSpeed training `weight must be 2-D` and void segmentation-map processing are unrelated." + }, + { + "left": "issue:44530", + "right": "issue:44863", + "accept": false, + "reason": "PagedAttentionCache `linear_attention` crash and NemotronH checkpoint loading are different failures." + }, + { + "left": "issue:43872", + "right": "issue:44291", + "accept": true, + "reason": "These describe the same `_is_hf_initialized` / `init_empty_weights` incompatibility; 43872 is the concrete bitsandbytes symptom and 44291 names the root cause." + }, + { + "left": "issue:44863", + "right": "issue:46402", + "accept": false, + "reason": "NemotronH checkpoint loading and Sam3Model text-encoder weights not loading are separate model-specific regressions." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays conservative: it treats the cluster as mostly heterogeneous and only marks the clearest duplicate subgroups as matches. The accepted soft pairs are well-supported by the titles (`43827`/`44509`, `43872`/`44291`, `41093`/`44805`, and the two double-shift reports). No obvious overstatement of evidence." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 6, + "review_unit_count": 61, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-502", + "nodes": [ + "issue:36010", + "issue:38175", + "issue:39692", + "issue:42491", + "issue:42617", + "issue:42915", + "issue:43317", + "issue:43366", + "issue:43531", + "issue:43550", + "issue:43596", + "issue:43632", + "issue:43638", + "issue:43644", + "issue:43645", + "issue:43646", + "issue:43716", + "issue:43720", + "issue:43824", + "issue:43828", + "issue:43950", + "issue:43957", + "issue:43994", + "issue:44155", + "issue:44534", + "issue:44560", + "issue:44589", + "issue:44661", + "issue:44683", + "issue:44756", + "issue:44805", + "issue:44877", + "issue:44910", + "issue:44918", + "issue:45030", + "issue:45325", + "issue:45584", + "issue:45663", + "issue:45672", + "issue:45685", + "issue:45812", + "issue:45823", + "issue:45834", + "issue:45902", + "issue:46489", + "issue:46515", + "issue:46858", + "issue:46899" + ], + "soft_pairs": [ + "issue:43366|issue:43828", + "issue:43596|issue:46515", + "issue:45325|issue:45902", + "issue:43366|issue:43531", + "issue:43638|issue:46515", + "issue:45663|issue:45823", + "issue:44155|issue:45584", + "issue:43366|issue:44589", + "issue:43824|issue:43957", + "issue:43950|issue:45834", + "issue:42915|issue:43957", + "issue:43716|issue:44560", + "issue:44910|issue:46858", + "issue:43645|issue:43646", + "issue:44560|issue:45685", + "issue:43550|issue:44910", + "issue:44683|issue:46515", + "issue:42617|issue:43366", + "issue:44560|issue:44918", + "issue:36010|issue:42915", + "issue:45812|issue:46489", + "issue:45823|issue:46515", + "issue:44661|issue:45823", + "issue:44560|issue:44805", + "issue:38175|issue:43994", + "issue:45672|issue:46899", + "issue:44877|issue:45030", + "issue:43632|issue:45834", + "issue:43638|issue:44661", + "issue:43596|issue:44661", + "issue:42491|issue:43720", + "issue:44661|issue:44805", + "issue:43317|issue:44756", + "issue:38175|issue:39692", + "issue:43644|issue:44534" + ], + "prepared_review_unit_hash": "819b1752c1928ac84ddb0cd026f23f50f452b4c62f9712e02927cd3249d1f6a7", + "component_budget": { + "node_count": 502, + "item_count": 502, + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 35, + "serialized_chars": 38345, + "estimated_input_tokens": 9587, + "estimated_eval_tokens": 19430 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly heterogeneous: most pairs are related by broad subsystem or model family only, not the same bug. I only consider four pairs true duplicates: custom model initialization, SigLIP2 bad outputs, non-persistent buffer corruption, and Gemma4 4-bit quantization/dtype casting.", + "confidence": 0.88, + "canonical_issue_reason": "issue:43646 is the clearest general representative of a true duplicate subthread: it states the broad custom-model initialization regression without the notebook-specific framing of issue:43645.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:43646 is the best issue to anchor the cluster because it is the most general and readable among the actual duplicate candidates; the rest of the cluster is too mixed to serve as one canonical bug.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43366", + "right": "issue:43828", + "accept": false, + "reason": "Different problems: GGUF gpt-oss support vs Phi autocast dtype mismatch." + }, + { + "left": "issue:43596", + "right": "issue:46515", + "accept": false, + "reason": "Both are generic failure reports, but the concrete bugs differ: deepspeed zero3/Bert indexing vs torch 2.4.1 incompatibility." + }, + { + "left": "issue:45325", + "right": "issue:45902", + "accept": false, + "reason": "Different models and code paths: Qwen2.5-VL rope scaling vs Qwen3_5 MoE rotary embedding materialization." + }, + { + "left": "issue:43366", + "right": "issue:43531", + "accept": false, + "reason": "Unrelated issues despite both mentioning Qwen3-MoE." + }, + { + "left": "issue:43638", + "right": "issue:46515", + "accept": false, + "reason": "Separate bugs: zero3/non-pretrained Bert indexing vs a torch-version incompatibility report." + }, + { + "left": "issue:45663", + "right": "issue:45823", + "accept": false, + "reason": "Both are Gemma4-related, but one is an FSDP2 KeyError in text attention and the other is a device_map auto placement mismatch." + }, + { + "left": "issue:44155", + "right": "issue:45584", + "accept": false, + "reason": "Different failure modes and models: AudioFlamingo3 batch leakage vs Whisper empty-transcription generation." + }, + { + "left": "issue:43366", + "right": "issue:44589", + "accept": false, + "reason": "GGUF gpt-oss support is unrelated to Float8 storage loading errors." + }, + { + "left": "issue:43824", + "right": "issue:43957", + "accept": false, + "reason": "Import error for Qwen2.5-VL is not the same as meta-device loading failures in other models." + }, + { + "left": "issue:43950", + "right": "issue:45834", + "accept": false, + "reason": "Non-persistent buffer corruption and Kosmos2.5 OCR index errors are unrelated bugs." + }, + { + "left": "issue:42915", + "right": "issue:43957", + "accept": false, + "reason": "Qwen3Moe FP8 config failure is distinct from meta-device loading regressions." + }, + { + "left": "issue:43716", + "right": "issue:44560", + "accept": false, + "reason": "Mistral-3 image-preprocessor dtype mismatch is not the same as Qwen3-vl video StopIteration." + }, + { + "left": "issue:44910", + "right": "issue:46858", + "accept": false, + "reason": "Both involve caching/position arithmetic, but one is a flash-attention illegal access on Qwen3.5 and the other is a static-cache torch.compile backend issue." + }, + { + "left": "issue:43645", + "right": "issue:43646", + "accept": true, + "reason": "Same underlying custom-model initialization regression; 43645 is the notebook-specific manifestation of 43646." + }, + { + "left": "issue:44560", + "right": "issue:45685", + "accept": false, + "reason": "Different subsystems and error types: Qwen3-vl video inference vs MPS histogram on int tensors." + }, + { + "left": "issue:43550", + "right": "issue:44910", + "accept": false, + "reason": "Both mention attention/compile failures, but the concrete code paths differ: SDPA torch.compile on Bamba vs Qwen3.5 3D position_ids misread by flash attention." + }, + { + "left": "issue:44683", + "right": "issue:46515", + "accept": false, + "reason": "torch>=2.9 flex_attention failure is unrelated to a torch 2.4.1 incompatibility report." + }, + { + "left": "issue:42617", + "right": "issue:43366", + "accept": false, + "reason": "Generic runtime failure for 3d_parallel.py is unrelated to gpt-oss GGUF support." + }, + { + "left": "issue:44560", + "right": "issue:44918", + "accept": false, + "reason": "Both are Qwen3.x issues, but one is a video StopIteration bug and the other is SFT input-embedding unpacking." + }, + { + "left": "issue:36010", + "right": "issue:42915", + "accept": false, + "reason": "ImportError for GenerationMixin is unrelated to Qwen3Moe FP8 failures." + }, + { + "left": "issue:45812", + "right": "issue:46489", + "accept": false, + "reason": "Different tokenizer regressions affecting different model families; not the same bug." + }, + { + "left": "issue:45823", + "right": "issue:46515", + "accept": false, + "reason": "Gemma4 device placement mismatch is separate from the torch 2.4.1 compatibility issue." + }, + { + "left": "issue:44661", + "right": "issue:45823", + "accept": false, + "reason": "Tokenizer-mapping add-new-model bug is unrelated to Gemma4 PLE device mismatch." + }, + { + "left": "issue:44560", + "right": "issue:44805", + "accept": false, + "reason": "Different symptoms and likely causes: video StopIteration vs mask/tensor shape mismatch." + }, + { + "left": "issue:38175", + "right": "issue:43994", + "accept": true, + "reason": "Same SigLIP2 bad-output regression; both describe nonsensical/zero-probability behavior for google/siglip2-base-patch16-224." + }, + { + "left": "issue:45672", + "right": "issue:46899", + "accept": true, + "reason": "Same Gemma4 4-bit quantization/dtype-cast bug in multimodal embedders; the second issue is a broader restatement of the first." + }, + { + "left": "issue:44877", + "right": "issue:45030", + "accept": false, + "reason": "Both mention config validation, but they concern different models and failure conditions." + }, + { + "left": "issue:43632", + "right": "issue:45834", + "accept": false, + "reason": "_is_hf_initialized regression is unrelated to Kosmos2.5 long-OCR index errors." + }, + { + "left": "issue:43638", + "right": "issue:44661", + "accept": false, + "reason": "Zero3/Bert indexing failure is unrelated to tokenizer-mapping name handling." + }, + { + "left": "issue:43596", + "right": "issue:44805", + "accept": false, + "reason": "Both are IndexError reports, but the shapes, APIs, and model contexts differ." + }, + { + "left": "issue:42491", + "right": "issue:43720", + "accept": false, + "reason": "LoRA/qwen3_moe hf4.x versus hf5.x incompatibility is a different problem from BitNet packed-weight loading." + }, + { + "left": "issue:44661", + "right": "issue:44805", + "accept": false, + "reason": "Tokenizer mapping bug and mask/tensor shape mismatch are unrelated." + }, + { + "left": "issue:43317", + "right": "issue:44756", + "accept": false, + "reason": "device_map offload failure and mmap OOM on Strix Halo are different memory/loading bugs." + }, + { + "left": "issue:38175", + "right": "issue:39692", + "accept": false, + "reason": "Both involve SigLIP2, but one is a model output regression while the other is a documentation example with multiple issues." + }, + { + "left": "issue:43644", + "right": "issue:44534", + "accept": true, + "reason": "Same non-persistent-buffer corruption regression in Transformers 5.x; titles differ only in phrasing." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The cluster summary is grounded in the packet, and the soft-edge decisions stay conservative overall. The four accepted duplicate pairs are reasonably supported by the issue titles, while the rest are appropriately rejected as related but distinct." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 7, + "review_unit_count": 61, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-502", + "nodes": [ + "issue:39692", + "issue:41720", + "issue:42491", + "issue:42915", + "issue:43257", + "issue:43425", + "issue:43644", + "issue:43645", + "issue:43708", + "issue:43854", + "issue:43856", + "issue:43873", + "issue:43931", + "issue:43957", + "issue:43994", + "issue:44075", + "issue:44155", + "issue:44164", + "issue:44291", + "issue:44387", + "issue:44402", + "issue:44451", + "issue:44488", + "issue:44756", + "issue:44863", + "issue:44910", + "issue:44960", + "issue:44977", + "issue:44991", + "issue:45276", + "issue:45292", + "issue:45313", + "issue:45335", + "issue:45357", + "issue:45362", + "issue:45406", + "issue:45488", + "issue:45507", + "issue:45588", + "issue:45812", + "issue:46050", + "issue:46459", + "issue:46468", + "issue:46512", + "issue:46846", + "issue:46897", + "issue:46901" + ], + "soft_pairs": [ + "issue:42491|issue:45276", + "issue:41720|issue:42915", + "issue:42915|issue:44756", + "issue:44451|issue:44488", + "issue:44960|issue:44977", + "issue:44863|issue:46846", + "issue:46459|issue:46512", + "issue:45313|issue:45357", + "issue:44488|issue:44991", + "issue:45507|issue:46468", + "issue:43644|issue:43645", + "issue:42491|issue:45292", + "issue:39692|issue:43994", + "issue:43257|issue:43854", + "issue:43856|issue:43873", + "issue:43708|issue:44075", + "issue:43856|issue:44387", + "issue:43854|issue:43931", + "issue:42491|issue:45335", + "issue:44960|issue:46050", + "issue:46897|issue:46901", + "issue:45488|issue:45812", + "issue:45362|issue:45406", + "issue:45276|issue:45335", + "issue:44164|issue:44291", + "issue:44910|issue:45588", + "issue:44402|issue:44488", + "issue:43425|issue:43957", + "issue:39692|issue:44960", + "issue:41720|issue:44155" + ], + "prepared_review_unit_hash": "4caa714ada8b33ec632f260e419d6d9e9af889c4ec71010529d5ff7604aade15", + "component_budget": { + "node_count": 502, + "item_count": 502, + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 30, + "serialized_chars": 36102, + "estimated_input_tokens": 9026, + "estimated_eval_tokens": 18308 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is mostly heterogeneous Transformers 5.x regressions spanning tokenizer loading, quantization/offloading, model-specific loading failures, and training bugs. Only the two BERT tokenizer-loading reports look like true duplicates; the rest are too model- or code-path-specific to merge.", + "confidence": 0.84, + "canonical_issue_reason": "No single issue cleanly represents the whole cluster; issue 43873 is the broadest still-open anchor, but it does not subsume the other reports.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 43873 is the best representative because it is open, broadly scoped, and has the most ongoing discussion/inbound references among the open items.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:42491", + "right": "issue:45276", + "accept": false, + "reason": "Different bugs: Qwen3 MoE LoRA incompatibility vs Gemma4 token-embedding resize behavior." + }, + { + "left": "issue:41720", + "right": "issue:42915", + "accept": false, + "reason": "Both are Qwen3-related, but one is an A800 auto-device-map crash and the other is a FineGrainedFP8Config failure." + }, + { + "left": "issue:42915", + "right": "issue:44756", + "accept": false, + "reason": "Qwen3 FP8 config failure and Strix Halo mmap OOM are unrelated code paths." + }, + { + "left": "issue:44451", + "right": "issue:44488", + "accept": true, + "reason": "Both report the same generic model-loading failure for different BERT repos, with no sign of distinct root causes." + }, + { + "left": "issue:44960", + "right": "issue:44977", + "accept": false, + "reason": "Different models and symptoms; too little evidence of a shared concrete bug." + }, + { + "left": "issue:44863", + "right": "issue:46846", + "accept": false, + "reason": "NemotronH checkpoint loading and Qwen3.5 TP-plan missing entries are separate issues." + }, + { + "left": "issue:46459", + "right": "issue:46512", + "accept": false, + "reason": "Both involve dtype handling, but one is loading composite checkpoints and the other is from_config config-state drift." + }, + { + "left": "issue:45313", + "right": "issue:45357", + "accept": false, + "reason": "DeepSpeed ZeRO-3 weight loading and save_pretrained visual-key regression are different failures." + }, + { + "left": "issue:44488", + "right": "issue:44991", + "accept": true, + "reason": "Both are the same class of tokenizer-loading regression affecting specific BERT-like repos in transformers>=5.0." + }, + { + "left": "issue:45507", + "right": "issue:46468", + "accept": false, + "reason": "GraniteMoEHybrid invalid method call and broken Nemotron example do not share the same concrete bug." + }, + { + "left": "issue:43644", + "right": "issue:43645", + "accept": false, + "reason": "Non-persistent buffer junk and Jupyter custom-model initialization are unrelated regressions." + }, + { + "left": "issue:42491", + "right": "issue:45292", + "accept": false, + "reason": "Unrelated: Qwen3 MoE LoRA reuse vs a generic resize_token_embeddings/output_embeddings bug." + }, + { + "left": "issue:39692", + "right": "issue:43994", + "accept": false, + "reason": "Different SigLIP2 example error vs SigLIP2 AutoModel/pipeline wrong results; not the same bug." + }, + { + "left": "issue:43257", + "right": "issue:43854", + "accept": false, + "reason": "Qwen3 MoE offloading/accelerate-deepspeed issue vs GLM-4.7-Flash unit-test load failure." + }, + { + "left": "issue:43856", + "right": "issue:43873", + "accept": false, + "reason": "Both concern Qwen3/quantization, but one is training memory usage and the other is offloading behavior." + }, + { + "left": "issue:43708", + "right": "issue:44075", + "accept": false, + "reason": "Trainer checkpoint resume math and SGD argument handling are unrelated." + }, + { + "left": "issue:43856", + "right": "issue:44387", + "accept": false, + "reason": "Qwen3 MoE memory usage and int4 reserved-memory/OOM regression are different issues." + }, + { + "left": "issue:43854", + "right": "issue:43931", + "accept": false, + "reason": "GLM-4.7-Flash test loading and Qwen3-VL weight-shape mismatch are separate model-loading problems." + }, + { + "left": "issue:42491", + "right": "issue:45335", + "accept": false, + "reason": "Different architectures and failures; not the same underlying bug." + }, + { + "left": "issue:44960", + "right": "issue:46050", + "accept": false, + "reason": "GLM5 loading/inference and sinq save_pretrained/from_pretrained corruption are unrelated." + }, + { + "left": "issue:46897", + "right": "issue:46901", + "accept": false, + "reason": "Same symptom family, but these are separate model implementations and not clearly the same code-path bug." + }, + { + "left": "issue:45488", + "right": "issue:45812", + "accept": false, + "reason": "Different tokenizer regressions affecting different model families; too broad to merge." + }, + { + "left": "issue:45362", + "right": "issue:45406", + "accept": false, + "reason": "Qwen3.5 chat crash and Gemma4Processor missing _tokenizer are separate issues." + }, + { + "left": "issue:45276", + "right": "issue:45335", + "accept": false, + "reason": "Both involve resize_token_embeddings, but they are distinct model-specific embedding propagation bugs." + }, + { + "left": "issue:44164", + "right": "issue:44291", + "accept": false, + "reason": "save/from_pretrained extra_state handling and init_empty_weights argument mismatch are different loader bugs." + }, + { + "left": "issue:44910", + "right": "issue:45588", + "accept": false, + "reason": "Different flash-attention failures with different triggers and models." + }, + { + "left": "issue:44402", + "right": "issue:44488", + "accept": false, + "reason": "Tokenizer vocab-size mismatch is a different bug from the BERT loading failure." + }, + { + "left": "issue:43425", + "right": "issue:43957", + "accept": false, + "reason": "Torch version incompatibility and meta-device loading failure are not the same defect." + }, + { + "left": "issue:39692", + "right": "issue:44960", + "accept": false, + "reason": "SigLIP2 documentation errors and GLM5 issue are unrelated." + }, + { + "left": "issue:41720", + "right": "issue:44155", + "accept": false, + "reason": "Qwen3 device-map assert and AudioFlamingo3 batched inference leak are unrelated." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet, and the pairwise judgments are generally conservative. The open-item anchor choice of issue 43873 is supported by its higher discussion/inbound-reference activity, and the accepted BERT-loading pair is plausibly the same regression class. No mergeability overreach is evident." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 8, + "review_unit_count": 61, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-502", + "nodes": [ + "issue:36010", + "issue:39404", + "issue:42796", + "issue:42947", + "issue:43335", + "issue:43381", + "issue:43404", + "issue:43425", + "issue:43454", + "issue:43645", + "issue:43646", + "issue:43708", + "issue:43828", + "issue:43854", + "issue:43873", + "issue:43957", + "issue:44079", + "issue:44155", + "issue:44164", + "issue:44291", + "issue:44292", + "issue:44387", + "issue:44402", + "issue:44451", + "issue:44479", + "issue:44589", + "issue:44756", + "issue:44863", + "issue:44877", + "issue:44912", + "issue:44928", + "issue:45216", + "issue:45276", + "issue:45292", + "issue:45335", + "issue:45588", + "issue:45663", + "issue:45684", + "issue:45698", + "issue:45704", + "issue:45759", + "issue:45823", + "issue:45859", + "issue:46326", + "issue:46468", + "issue:46829", + "issue:46894", + "issue:46901" + ], + "soft_pairs": [ + "issue:45663|issue:45759", + "issue:43404|issue:43454", + "issue:44079|issue:46829", + "issue:44291|issue:44756", + "issue:42947|issue:45704", + "issue:43873|issue:44387", + "issue:44292|issue:44912", + "issue:36010|issue:39404", + "issue:45663|issue:45859", + "issue:44387|issue:44756", + "issue:42947|issue:43381", + "issue:44155|issue:45823", + "issue:43957|issue:44756", + "issue:43335|issue:45663", + "issue:45276|issue:45292", + "issue:44863|issue:46468", + "issue:44479|issue:44877", + "issue:43381|issue:44928", + "issue:44164|issue:44756", + "issue:45684|issue:45698", + "issue:45588|issue:45663", + "issue:42796|issue:46326", + "issue:43854|issue:45216", + "issue:45292|issue:45335", + "issue:46894|issue:46901", + "issue:43828|issue:43957", + "issue:44402|issue:44451", + "issue:43646|issue:43708", + "issue:43645|issue:44164", + "issue:44589|issue:44756", + "issue:43425|issue:43828" + ], + "prepared_review_unit_hash": "c0dbae09dc1ca99f53e624f04dea6b358b5b70fad7ad89aeca6ab544714ff3b1", + "component_budget": { + "node_count": 502, + "item_count": 502, + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 31, + "serialized_chars": 37128, + "estimated_input_tokens": 9282, + "estimated_eval_tokens": 18820 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a loose mix of unrelated Transformers issues. The only clear duplicate family is the resize_token_embeddings propagation bug across related Gemma/t5gemma reports; most other pairs share only a broad subsystem or symptom class and should stay separate.", + "confidence": 0.91, + "canonical_issue_reason": "Issue 45292 is the broadest, most neutral statement of the shared resize_token_embeddings/output_embeddings bug, while 45276 and 45335 read as model-specific variants of the same underlying defect.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 45292 is the best canonical issue because it describes the core bug without overfitting to one model family and cleanly subsumes the two related resize_token_embeddings reports.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:45663", + "right": "issue:45759", + "accept": false, + "reason": "Both involve model/config handling, but one is an FSDP2 KeyError in Gemma4 attention and the other is a Qwen multimodal config unwrap issue; different code paths and fixes." + }, + { + "left": "issue:43404", + "right": "issue:43454", + "accept": false, + "reason": "Both mention lm_head tying, but they affect different model families (Mistral3 vs AyaVision) and likely different tying logic." + }, + { + "left": "issue:44079", + "right": "issue:46829", + "accept": false, + "reason": "These are different bugs: a generic ModelOutput key assignment issue versus a VideoPrism classifier output field mix-up." + }, + { + "left": "issue:44291", + "right": "issue:44756", + "accept": false, + "reason": "init_empty_weights/_is_hf_initialized and Strix Halo mmap/OOM are unrelated failure modes." + }, + { + "left": "issue:42947", + "right": "issue:45704", + "accept": false, + "reason": "Gradient checkpointing with PEFT LoRA and a T5 apex RMSNorm memory leak are separate bugs in different areas." + }, + { + "left": "issue:43873", + "right": "issue:44387", + "accept": false, + "reason": "Both concern quantization, but one is offloading behavior and the other is increased reserved memory/OOM under int4; not the same defect." + }, + { + "left": "issue:44292", + "right": "issue:44912", + "accept": false, + "reason": "Different quantization formats and symptoms: NVFP4 load error versus MXFP4 fallback to bf16." + }, + { + "left": "issue:36010", + "right": "issue:39404", + "accept": false, + "reason": "Importing GenerationMixin and Whisper return_language pipeline regression are unrelated issues." + }, + { + "left": "issue:45663", + "right": "issue:45859", + "accept": false, + "reason": "Both are model-specific runtime problems, but one is Gemma4 FSDP2 state handling and the other is Qwen3.5-MoE CPU offload compatibility." + }, + { + "left": "issue:44387", + "right": "issue:44756", + "accept": false, + "reason": "Int4 quantization OOM and disabling mmap on Strix Halo are different memory-path issues." + }, + { + "left": "issue:42947", + "right": "issue:43381", + "accept": false, + "reason": "One is LoRA/gradient-checkpointing ineffectiveness; the other is eval-mode incompatibility. Similar theme, not same bug." + }, + { + "left": "issue:44155", + "right": "issue:45823", + "accept": false, + "reason": "AudioFlamingo3 batched inference leakage and Gemma4 PLE device mismatch are unrelated model-specific bugs." + }, + { + "left": "issue:43957", + "right": "issue:44756", + "accept": false, + "reason": "Meta-device loading failures and Strix Halo mmap OOM are distinct loading/runtime problems." + }, + { + "left": "issue:43335", + "right": "issue:45663", + "accept": false, + "reason": "SwitchTransformers sparse-layer construction bug and Gemma4 FSDP2 attention state bug are unrelated." + }, + { + "left": "issue:45276", + "right": "issue:45292", + "accept": true, + "reason": "Same underlying bug: resize_token_embeddings does not propagate updated embeddings correctly; 45276 is the Gemma4-specific manifestation and 45292 states the general failure." + }, + { + "left": "issue:44863", + "right": "issue:46468", + "accept": false, + "reason": "Both are Nemotron-related, but one is checkpoint loading and the other is a broken official example; insufficient evidence of the same code-path." + }, + { + "left": "issue:44479", + "right": "issue:44877", + "accept": false, + "reason": "Video-input regression for Qwen VL variants and strict Granite Speech config loading are separate issues." + }, + { + "left": "issue:43381", + "right": "issue:44928", + "accept": false, + "reason": "Eval-mode gradient checkpointing and RLHF NaN explosion from 3D position_ids/SDPA fallback are different bugs." + }, + { + "left": "issue:44164", + "right": "issue:44756", + "accept": false, + "reason": "save/from_pretrained extra_state handling and Strix Halo mmap/OOM are unrelated." + }, + { + "left": "issue:45684", + "right": "issue:45698", + "accept": false, + "reason": "Both involve custom-model save/load flows, but one is permissions propagation and the other is wrong module reloading; different defects." + }, + { + "left": "issue:45588", + "right": "issue:45663", + "accept": false, + "reason": "flash_attention.py s_aux=None crash and Gemma4 FSDP2 shared_kv_states KeyError are unrelated." + }, + { + "left": "issue:42796", + "right": "issue:46326", + "accept": false, + "reason": "Tokenizer BPE-marker handling and continue_final_message behavior are different tokenizer bugs." + }, + { + "left": "issue:43854", + "right": "issue:45216", + "accept": false, + "reason": "GLM-4.7-Flash unit-test loading failure and Qwen3.5 save_pretrained regression are unrelated." + }, + { + "left": "issue:45292", + "right": "issue:45335", + "accept": true, + "reason": "Same resize_token_embeddings propagation bug: both report that resized embeddings do not update the relevant output/decoder embedding weights." + }, + { + "left": "issue:46894", + "right": "issue:46901", + "accept": false, + "reason": "Both describe a training-loss double-shift pattern, but they are on different models and lack evidence of a single shared code-path." + }, + { + "left": "issue:43828", + "right": "issue:43957", + "accept": false, + "reason": "Autocast dtype mismatch and meta-device loading failures are distinct runtime issues." + }, + { + "left": "issue:44402", + "right": "issue:44451", + "accept": false, + "reason": "Tokenizer vocab mismatch for specific tokens and a model load failure are not the same bug." + }, + { + "left": "issue:43646", + "right": "issue:43708", + "accept": false, + "reason": "Custom model initialization breakage in Transformers 5.0 and Trainer resume/max_steps miscalculation are unrelated." + }, + { + "left": "issue:43645", + "right": "issue:44164", + "accept": false, + "reason": "Notebook custom-model initialization and extra_state save/load handling are separate issues." + }, + { + "left": "issue:44589", + "right": "issue:44756", + "accept": false, + "reason": "Float8 storage lookup failure and Strix Halo mmap/OOM are different problems." + }, + { + "left": "issue:43425", + "right": "issue:43828", + "accept": false, + "reason": "Torch 2.10 compatibility and autocast dtype mismatch are not the same underlying bug." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays mostly conservative. The resize_token_embeddings family is the only clearly supported duplicate cluster, and the other pair rejections avoid overclaiming shared root causes. No obvious unsupported mergeability claims stand out." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 9, + "review_unit_count": 61, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-502", + "nodes": [ + "issue:36032", + "issue:39404", + "issue:43299", + "issue:43425", + "issue:43452", + "issue:43596", + "issue:43611", + "issue:43644", + "issue:43645", + "issue:43716", + "issue:43856", + "issue:43901", + "issue:43906", + "issue:43931", + "issue:43994", + "issue:44079", + "issue:44164", + "issue:44265", + "issue:44479", + "issue:44488", + "issue:44509", + "issue:44530", + "issue:44589", + "issue:44671", + "issue:44683", + "issue:44756", + "issue:44805", + "issue:44898", + "issue:44912", + "issue:44987", + "issue:44991", + "issue:45081", + "issue:45084", + "issue:45479", + "issue:45701", + "issue:45759", + "issue:45812", + "issue:45823", + "issue:46050", + "issue:46315", + "issue:46326", + "issue:46491", + "issue:46665", + "issue:46821", + "issue:46869", + "issue:46894", + "issue:46897", + "issue:46899" + ], + "soft_pairs": [ + "issue:45812|issue:46315", + "issue:46315|issue:46491", + "issue:39404|issue:45701", + "issue:44530|issue:45084", + "issue:43611|issue:43645", + "issue:39404|issue:46665", + "issue:44683|issue:45823", + "issue:43425|issue:46869", + "issue:43425|issue:43716", + "issue:43425|issue:45823", + "issue:44589|issue:45084", + "issue:44479|issue:44805", + "issue:43906|issue:44530", + "issue:43901|issue:44509", + "issue:44671|issue:44991", + "issue:43994|issue:44079", + "issue:36032|issue:43452", + "issue:44265|issue:44898", + "issue:39404|issue:44912", + "issue:46050|issue:46326", + "issue:44164|issue:45759", + "issue:43906|issue:46326", + "issue:44991|issue:45081", + "issue:43299|issue:43931", + "issue:45823|issue:46899", + "issue:45479|issue:46894", + "issue:43856|issue:44756", + "issue:43596|issue:44683", + "issue:45479|issue:46897", + "issue:44164|issue:46821", + "issue:44488|issue:44987", + "issue:44488|issue:45081", + "issue:43611|issue:43644" + ], + "prepared_review_unit_hash": "256aaa98dbec636a3c3becea3ad982a3520891ffab08f73535139c5b3869d20f", + "component_budget": { + "node_count": 502, + "item_count": 502, + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 33, + "serialized_chars": 37736, + "estimated_input_tokens": 9434, + "estimated_eval_tokens": 19124 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The items are mostly distinct Transformers v5 regressions across different model families and subsystems; none of the soft pairs look like true duplicates.", + "confidence": 0.93, + "canonical_issue_reason": "issue:45812 is the strongest representative anchor: it is a concrete, high-impact v4\u2192v5 tokenizer regression with multiple inbound references and a clear failure mode.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:45812 is the best standalone issue to surface because it is specific, well-scoped, and broadly relevant compared with the more specialized reports.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:45812", + "right": "issue:46315", + "accept": false, + "reason": "Different bug class: wrong token IDs for Granite vs a tokenizer performance regression in convert_tokens_to_ids." + }, + { + "left": "issue:46315", + "right": "issue:46491", + "accept": false, + "reason": "Performance regression in added-token lookup vs a CodeLlama encode/decode spacing bug." + }, + { + "left": "issue:39404", + "right": "issue:45701", + "accept": false, + "reason": "Whisper pipeline return_language failure is unrelated to a general tokenization regression report." + }, + { + "left": "issue:44530", + "right": "issue:45084", + "accept": false, + "reason": "Different code paths: PagedAttentionCache group-type crash vs template compilation error." + }, + { + "left": "issue:43611", + "right": "issue:43645", + "accept": false, + "reason": "Base_model_prefix loading issue and Jupyter custom-model init issue are separate regressions." + }, + { + "left": "issue:39404", + "right": "issue:46665", + "accept": false, + "reason": "Pipeline return_language breakage is not the same as from_pretrained behavior inside Python classes." + }, + { + "left": "issue:44683", + "right": "issue:45823", + "accept": false, + "reason": "Torch version incompatibility in flex_attention is unrelated to Gemma4 device-map forwarding." + }, + { + "left": "issue:43425", + "right": "issue:46869", + "accept": false, + "reason": "Torch 2.10 incompatibility is not the same failure as YOLO multi-GPU device-map handling." + }, + { + "left": "issue:43425", + "right": "issue:43716", + "accept": false, + "reason": "Generic torch incompatibility does not match the specific dtype mismatch in Mistral-3 preprocessing." + }, + { + "left": "issue:43425", + "right": "issue:45823", + "accept": false, + "reason": "Torch compatibility and Gemma4 device mismatch are different problems." + }, + { + "left": "issue:44589", + "right": "issue:45084", + "accept": false, + "reason": "Float8 storage lookup failure is unrelated to non-template-node compilation." + }, + { + "left": "issue:44479", + "right": "issue:44805", + "accept": false, + "reason": "Qwen video-input regression and mask-shape IndexError are different model/input failures." + }, + { + "left": "issue:43906", + "right": "issue:44530", + "accept": false, + "reason": "An isolated reproduction of another issue is not the same as the linear_attention cache crash." + }, + { + "left": "issue:43901", + "right": "issue:44509", + "accept": false, + "reason": "Both are docs-related, but they concern different pipeline/deprecation topics." + }, + { + "left": "issue:44671", + "right": "issue:44991", + "accept": false, + "reason": "CamemBERT masked-LM predictions and est-roberta tokenizer loading are unrelated." + }, + { + "left": "issue:43994", + "right": "issue:44079", + "accept": false, + "reason": "Wrong SigLIP2 outputs are not the same as ModelOutput key assignment behavior." + }, + { + "left": "issue:36032", + "right": "issue:43452", + "accept": false, + "reason": "T5 tokenizer method conflict and gguf_file loading breakage are different tokenizer/model-loading failures." + }, + { + "left": "issue:44265", + "right": "issue:44898", + "accept": false, + "reason": "torch.export with torch_compilable_check is unrelated to Perceiver interpolation at non-default resolution." + }, + { + "left": "issue:39404", + "right": "issue:44912", + "accept": false, + "reason": "Whisper pipeline regression and MXFP4 quantization fallback are unrelated." + }, + { + "left": "issue:46050", + "right": "issue:46326", + "accept": false, + "reason": "Quantized model save/load failure is not the same as mistral_common continue_final_message behavior." + }, + { + "left": "issue:44164", + "right": "issue:45759", + "accept": false, + "reason": "extra_state save/load handling and composite config unwrapping are different APIs and bugs." + }, + { + "left": "issue:43906", + "right": "issue:46326", + "accept": false, + "reason": "A reproduction issue and a tokenizer chat-template bug are not duplicates." + }, + { + "left": "issue:44991", + "right": "issue:45081", + "accept": false, + "reason": "Different tokenizer failures: est-roberta loading vs Mistral regex patching on a backend tokenizer object." + }, + { + "left": "issue:43299", + "right": "issue:43931", + "accept": false, + "reason": "Qwen3VL model-loading breakage and Qwen3-VL weight-shape mismatch are related by family only, not the same concrete bug." + }, + { + "left": "issue:45823", + "right": "issue:46899", + "accept": false, + "reason": "Gemma4 device-map mismatch and Gemma4 4-bit multimodal dtype-cast issues are separate failure modes." + }, + { + "left": "issue:45479", + "right": "issue:46894", + "accept": false, + "reason": "Zero-loss classification bug is unrelated to the CohereASR training-loss double-shift issue." + }, + { + "left": "issue:43856", + "right": "issue:44756", + "accept": false, + "reason": "Qwen3 MoE training memory inefficiency and Strix Halo mmap OOM are different memory problems." + }, + { + "left": "issue:43596", + "right": "issue:44683", + "accept": false, + "reason": "Zero3 init IndexError in BertModel is unrelated to compiled flex_attention on torch>=2.9." + }, + { + "left": "issue:45479", + "right": "issue:46897", + "accept": false, + "reason": "Sequence-classification loss bug does not match the Florence2 double-shift training-loss bug." + }, + { + "left": "issue:44164", + "right": "issue:46821", + "accept": false, + "reason": "extra_state persistence is unrelated to the TimesFM window_size attribute error." + }, + { + "left": "issue:44488", + "right": "issue:44987", + "accept": false, + "reason": "Both are load failures, but they affect different models and likely different root causes." + }, + { + "left": "issue:44488", + "right": "issue:45081", + "accept": false, + "reason": "Generic model-loading failure is not the same as the Mistral regex patch crash." + }, + { + "left": "issue:43611", + "right": "issue:43644", + "accept": false, + "reason": "Loading models with base_model_prefix and non-persistent buffer junk are distinct regressions." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 10, + "review_unit_count": 61, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-502", + "nodes": [ + "issue:32090", + "issue:34567", + "issue:36032", + "issue:39404", + "issue:40444", + "issue:40990", + "issue:42175", + "issue:42796", + "issue:42915", + "issue:43262", + "issue:43317", + "issue:43377", + "issue:43404", + "issue:43421", + "issue:43452", + "issue:43482", + "issue:43541", + "issue:43582", + "issue:43632", + "issue:43643", + "issue:43644", + "issue:43645", + "issue:43646", + "issue:43716", + "issue:43746", + "issue:43883", + "issue:44206", + "issue:44220", + "issue:44451", + "issue:44462", + "issue:44488", + "issue:44530", + "issue:44661", + "issue:44683", + "issue:44737", + "issue:44811", + "issue:44987", + "issue:44991", + "issue:45084", + "issue:45507", + "issue:45584", + "issue:45859", + "issue:46178", + "issue:46285", + "issue:46326", + "issue:46402", + "issue:46489", + "issue:46869" + ], + "soft_pairs": [ + "issue:44462|issue:46489", + "issue:42796|issue:43404", + "issue:46285|issue:46402", + "issue:36032|issue:42915", + "issue:43716|issue:45084", + "issue:32090|issue:39404", + "issue:44661|issue:44683", + "issue:44737|issue:45859", + "issue:43643|issue:44530", + "issue:44451|issue:44987", + "issue:40990|issue:43421", + "issue:42915|issue:44530", + "issue:43541|issue:45084", + "issue:43644|issue:43646", + "issue:43582|issue:43632", + "issue:42175|issue:46178", + "issue:34567|issue:40444", + "issue:43883|issue:44488", + "issue:40444|issue:46326", + "issue:44206|issue:44220", + "issue:44811|issue:45584", + "issue:36032|issue:43645", + "issue:43452|issue:43482", + "issue:44987|issue:44991", + "issue:43317|issue:43482", + "issue:45084|issue:45507", + "issue:43262|issue:43377", + "issue:43746|issue:46326", + "issue:43883|issue:44991", + "issue:45084|issue:46869" + ], + "prepared_review_unit_hash": "873c579097a1926b6d318b965c440e6f64dd1f3fd9364db68971111709f6b82a", + "component_budget": { + "node_count": 502, + "item_count": 502, + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 30, + "serialized_chars": 36689, + "estimated_input_tokens": 9173, + "estimated_eval_tokens": 18602 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Mostly unrelated issue set with a few likely duplicate pairs around v5 init regressions and audio feature extraction. I would keep this cluster centered on the generic Transformers v5 custom-model initialization regression.", + "confidence": 0.61, + "canonical_issue_reason": "issue:43646 is the broadest, most representative regression here: it describes the general Transformers 5.0.0 custom-model initialization breakage that likely subsumes the more specific buffer-init symptom in issue:43644.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:43646 is the best global representative because it is the clearest high-level bug report in the set and maps to a concrete, broadly relevant v5 regression rather than a model-specific symptom.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44462", + "right": "issue:46489", + "accept": false, + "reason": "Both are tokenizer-loading regressions, but one is about ignoring tokenizer.json and the other is a DeepSeek-Coder-specific output mismatch; not the same concrete bug." + }, + { + "left": "issue:42796", + "right": "issue:43404", + "accept": false, + "reason": "Different subsystems and failures: tokenizer BPE marker handling vs a tied lm_head weight issue in Mistral3." + }, + { + "left": "issue:46285", + "right": "issue:46402", + "accept": false, + "reason": "Both mention model structure/loading, but CLIP structure inconsistency and Sam3 text-encoder weights missing are distinct regressions." + }, + { + "left": "issue:36032", + "right": "issue:42915", + "accept": false, + "reason": "Unrelated: T5 tokenizer special-token API conflict vs Qwen3Moe FP8 config failure." + }, + { + "left": "issue:43716", + "right": "issue:45084", + "accept": false, + "reason": "Different code paths: image preprocessor/model dtype mismatch vs a template compilation error." + }, + { + "left": "issue:32090", + "right": "issue:39404", + "accept": false, + "reason": "Trainer GPU broadcast TypeError and Whisper return_language pipeline behavior are unrelated." + }, + { + "left": "issue:44661", + "right": "issue:44683", + "accept": false, + "reason": "Different areas: tokenizer-mapping model addition vs compiled flex_attention on torch>=2.9." + }, + { + "left": "issue:44737", + "right": "issue:45859", + "accept": false, + "reason": "Both involve device placement, but XLNet CPU arange and Qwen3_5Moe CPU offload are separate model-specific bugs." + }, + { + "left": "issue:43643", + "right": "issue:44530", + "accept": false, + "reason": "Trust-remote-code config field loss is unrelated to Qwen3.5 PagedAttentionCache linear_attention crashes." + }, + { + "left": "issue:44451", + "right": "issue:44987", + "accept": false, + "reason": "Both are model-loading complaints, but they target different models and likely different loader regressions." + }, + { + "left": "issue:40990", + "right": "issue:43421", + "accept": false, + "reason": "Perplexity regression on GPT-OSS is unrelated to runtime post-processor updates in TokenizersBackend." + }, + { + "left": "issue:42915", + "right": "issue:44530", + "accept": false, + "reason": "Both involve Qwen models, but FineGrainedFP8Config failure and linear_attention group-type crash are distinct problems." + }, + { + "left": "issue:43541", + "right": "issue:45084", + "accept": false, + "reason": "Mixtral torch-dynamo grouped_mm tracing error is unrelated to the non-template-node compilation error." + }, + { + "left": "issue:43644", + "right": "issue:43646", + "accept": true, + "reason": "Both describe the same v5 custom-model initialization regression; 43644 is a concrete symptom of the broader initialization breakage in 43646." + }, + { + "left": "issue:43582", + "right": "issue:43632", + "accept": false, + "reason": "Apple Silicon allocator warmup TypeError and the _is_hf_initialized flag regression are unrelated." + }, + { + "left": "issue:42175", + "right": "issue:46178", + "accept": false, + "reason": "Both are backend/package-related, but missing tensorflow in torch extras and missing tensorflow_text in BACKENDS_MAPPING are different issues." + }, + { + "left": "issue:34567", + "right": "issue:40444", + "accept": false, + "reason": "TrainerState token-count tracking and Qwen2.5-VL multi-image iterable dataset failures are unrelated." + }, + { + "left": "issue:43883", + "right": "issue:44488", + "accept": false, + "reason": "Molmo tied-weights metadata and loading cjvt/sleng-bert are unrelated loading issues." + }, + { + "left": "issue:40444", + "right": "issue:46326", + "accept": false, + "reason": "Different domains and failures: multimodal fine-tuning with multiple images vs mistral_common continue_final_message behavior." + }, + { + "left": "issue:44206", + "right": "issue:44220", + "accept": true, + "reason": "Both point to the same audio feature-extraction path; 44206 gives the concrete unsupported-center crash that fits the generic _torch_extract_fbank_features() report." + }, + { + "left": "issue:44811", + "right": "issue:45584", + "accept": false, + "reason": "Whisper batch_decode skip_special_tokens handling and empty-transcription generation after align_special_tokens are different bugs." + }, + { + "left": "issue:36032", + "right": "issue:43645", + "accept": false, + "reason": "Tokenizer special-token method conflict is unrelated to Jupyter custom-model initialization breakage." + }, + { + "left": "issue:43452", + "right": "issue:43482", + "accept": false, + "reason": "Both mention GGUF loading, but one is a generic gguf_file/from_pretrained breakage and the other is a specific Qwen2.5-GGUF load failure; not clearly the same concrete bug." + }, + { + "left": "issue:43317", + "right": "issue:43482", + "accept": false, + "reason": "Dequantized model offload failure and Qwen2.5-GGUF loading failure are different loader problems." + }, + { + "left": "issue:45084", + "right": "issue:45507", + "accept": false, + "reason": "Template compilation TypeError is unrelated to GraniteMoEHybrid calling an invalid method." + }, + { + "left": "issue:43262", + "right": "issue:43377", + "accept": false, + "reason": "Audio chat-template sampling-rate defaulting and MIMI batched-vs-single mismatch are distinct issues." + }, + { + "left": "issue:43746", + "right": "issue:46326", + "accept": false, + "reason": "PEFT checkpoint loading for GraniteSpeech and mistral_common continue_final_message are unrelated." + }, + { + "left": "issue:43883", + "right": "issue:44991", + "accept": false, + "reason": "Missing all_tied_weights_keys for Molmo and tokenizer loading failure for EMBEDDIA/est-roberta are unrelated." + }, + { + "left": "issue:45084", + "right": "issue:46869", + "accept": false, + "reason": "Non-template-node compilation failure and YOLO multi-GPU device-map loading failure are unrelated." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is conservative and grounded in the packet. The choice of issue:43646 as the representative v5 custom-model initialization regression is reasonable, and the two accepted soft pairs (43644/43646 and 44206/44220) are plausibly same-bug clusters without overclaiming. The rejected pairs are mostly framed narrowly enough to avoid overstating duplicates." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 11, + "review_unit_count": 61, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-502", + "nodes": [ + "issue:28530", + "issue:34567", + "issue:36032", + "issue:41720", + "issue:42915", + "issue:43089", + "issue:43296", + "issue:43322", + "issue:43377", + "issue:43404", + "issue:43425", + "issue:43531", + "issue:43540", + "issue:43611", + "issue:43632", + "issue:43645", + "issue:43646", + "issue:43742", + "issue:43792", + "issue:43819", + "issue:43994", + "issue:44008", + "issue:44062", + "issue:44220", + "issue:44265", + "issue:44291", + "issue:44403", + "issue:44451", + "issue:44464", + "issue:44466", + "issue:44479", + "issue:44534", + "issue:44589", + "issue:44610", + "issue:44683", + "issue:44849", + "issue:44877", + "issue:44987", + "issue:44991", + "issue:45081", + "issue:45216", + "issue:45325", + "issue:45357", + "issue:45987", + "issue:46315", + "issue:46821", + "issue:46869", + "issue:46899" + ], + "soft_pairs": [ + "issue:43646|issue:44534", + "issue:43792|issue:46869", + "issue:44403|issue:45081", + "issue:43645|issue:44534", + "issue:44849|issue:45325", + "issue:43377|issue:43819", + "issue:43404|issue:44466", + "issue:44464|issue:45987", + "issue:44987|issue:45081", + "issue:43994|issue:44610", + "issue:44451|issue:45081", + "issue:42915|issue:44291", + "issue:28530|issue:34567", + "issue:44008|issue:46899", + "issue:36032|issue:43742", + "issue:44062|issue:44987", + "issue:44265|issue:44610", + "issue:42915|issue:44589", + "issue:43089|issue:45987", + "issue:43322|issue:43540", + "issue:44479|issue:45325", + "issue:43632|issue:46821", + "issue:36032|issue:41720", + "issue:45216|issue:45357", + "issue:43531|issue:44877", + "issue:44991|issue:46315", + "issue:43425|issue:44403", + "issue:43296|issue:43531", + "issue:44220|issue:44683", + "issue:43611|issue:44534" + ], + "prepared_review_unit_hash": "2cdb6e76c9eaff568a31695c4add9b39fadce31028f7c36d1da54b9377875f3f", + "component_budget": { + "node_count": 502, + "item_count": 502, + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 30, + "serialized_chars": 36746, + "estimated_input_tokens": 9187, + "estimated_eval_tokens": 18630 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The set is mostly a heterogeneous mix of unrelated Transformers bugs. I only accept one soft-duplicate pair: the two Qwen3.5 save_pretrained regression reports. Everything else is too model-, path-, or symptom-specific to safely merge.", + "confidence": 0.71, + "canonical_issue_reason": "Issue 45216 is the clearest canonical issue because it states the broad Qwen3.5 save_pretrained regression, and 45357 reads like a narrower follow-up on the same serialization problem.", + "canonical_pr_reason": null, + "best_issue_reason": "45216 is the best global issue anchor here: it is the broadest actionable duplicate target among the only clearly overlapping pair.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43646", + "right": "issue:44534", + "accept": false, + "reason": "Both are v5 regressions, but one is custom model initialization and the other is non-persistent buffer corruption; different bugs." + }, + { + "left": "issue:43792", + "right": "issue:46869", + "accept": false, + "reason": "Whisper loading failure and YOLO multi-GPU device-map assertion are unrelated models and code paths." + }, + { + "left": "issue:44403", + "right": "issue:45081", + "accept": false, + "reason": "Generic loading noise and a Mistral tokenizer regex patch crash are not the same underlying issue." + }, + { + "left": "issue:43645", + "right": "issue:44534", + "accept": false, + "reason": "Notebook custom-model initialization failure is distinct from junk-filled non-persistent buffers." + }, + { + "left": "issue:44849", + "right": "issue:45325", + "accept": false, + "reason": "Qwen3.5 output_hidden_states bug and Qwen2.5-VL rope scaling bug are different model families and defects." + }, + { + "left": "issue:43377", + "right": "issue:43819", + "accept": false, + "reason": "MIMI batching/padding-mask mismatch and DAC from_latents/STE mismatch are unrelated audio-model bugs." + }, + { + "left": "issue:43404", + "right": "issue:44466", + "accept": false, + "reason": "Both involve lm_head weight behavior, but one is a Mistral3 tying bug and the other is device-dependent serialization inconsistency." + }, + { + "left": "issue:44464", + "right": "issue:45987", + "accept": false, + "reason": "Chunked generation inconsistency with compiled forward and StaticCache.get_seq_length returning a tensor are separate generation/caching issues." + }, + { + "left": "issue:44987", + "right": "issue:45081", + "accept": false, + "reason": "A model-loading failure and a tokenizer regex patch crash do not look like the same bug." + }, + { + "left": "issue:43994", + "right": "issue:44610", + "accept": false, + "reason": "SigLIP2 nonsensical outputs and OmDet-Turbo processor size mismatch are different model-specific problems." + }, + { + "left": "issue:44451", + "right": "issue:45081", + "accept": false, + "reason": "Loading ScandiBERT and Mistral tokenizer regex patching are unrelated failures." + }, + { + "left": "issue:42915", + "right": "issue:44291", + "accept": false, + "reason": "Qwen3Moe FP8 config failure and init_empty_weights/_is_hf_initialized argument breakage are different paths." + }, + { + "left": "issue:28530", + "right": "issue:34567", + "accept": false, + "reason": "Early stopping metric lookup and num_input_tokens_seen not updating are unrelated trainer bugs." + }, + { + "left": "issue:44008", + "right": "issue:46899", + "accept": false, + "reason": "Gemma3n variable collision and Gemma4 quantization dtype-cast issues are different multimodal bugs." + }, + { + "left": "issue:36032", + "right": "issue:43742", + "accept": false, + "reason": "T5 tokenizer special-token conflict and MobileLLM key error are not the same defect." + }, + { + "left": "issue:44062", + "right": "issue:44987", + "accept": false, + "reason": "AddedToken keyword conflict and a model-loading failure for physical-intelligence/fast are unrelated." + }, + { + "left": "issue:44265", + "right": "issue:44610", + "accept": false, + "reason": "torch.export with torch_compilable_check and OmDet-Turbo image-size mismatch are different issues." + }, + { + "left": "issue:42915", + "right": "issue:44589", + "accept": false, + "reason": "Qwen3Moe FP8 config failure and Float8 storage lookup failure are distinct runtime problems." + }, + { + "left": "issue:43089", + "right": "issue:45987", + "accept": false, + "reason": "Generation overhead/performance regression is not the same as a return-type contract bug in StaticCache." + }, + { + "left": "issue:43322", + "right": "issue:43540", + "accept": false, + "reason": "Llava Next loading segfault and Qwen3OmniMoe video-input ValueError are unrelated multimodal failures." + }, + { + "left": "issue:44479", + "right": "issue:45325", + "accept": false, + "reason": "A broad video-input regression across Qwen models is not the same as the still-image temporal position_ids bug." + }, + { + "left": "issue:43632", + "right": "issue:46821", + "accept": false, + "reason": "_is_hf_initialized regression and TimesFM window_size AttributeError are unrelated." + }, + { + "left": "issue:36032", + "right": "issue:41720", + "accept": false, + "reason": "Tokenizer constructor conflict and Qwen3 auto device mapping CUDA assert are different bugs." + }, + { + "left": "issue:45216", + "right": "issue:45357", + "accept": true, + "reason": "Both report the same Qwen3.5 save_pretrained serialization regression; 45357 is a narrower manifestation about incorrect visual encoder keys." + }, + { + "left": "issue:43531", + "right": "issue:44877", + "accept": false, + "reason": "Qwen3-MoE sliding_window behavior and granite_speech config loading are unrelated config/model issues." + }, + { + "left": "issue:44991", + "right": "issue:46315", + "accept": false, + "reason": "Tokenizer loading failure and convert_tokens_to_ids slowdown are different symptoms and fixes." + }, + { + "left": "issue:43425", + "right": "issue:44403", + "accept": false, + "reason": "Torch 2.10 incompatibility and loading noise are unrelated concerns." + }, + { + "left": "issue:43296", + "right": "issue:43531", + "accept": false, + "reason": "PaddleOCR-VL load failure and Qwen3-MoE sliding_window issue are distinct problems." + }, + { + "left": "issue:44220", + "right": "issue:44683", + "accept": false, + "reason": "fbank feature extraction and compiled flex_attention failures are unrelated subsystems." + }, + { + "left": "issue:43611", + "right": "issue:44534", + "accept": false, + "reason": "base_model_prefix loading breakage and non-persistent buffer junk are different v5 regressions." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is well grounded in the packet and stays conservative. The lone accepted pair, issue:45216 vs issue:45357, is a plausible same-bug family around Qwen3.5 `save_pretrained` serialization, and the remaining soft-edge rejections do not overstate duplication based on the titles/body summaries provided." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 12, + "review_unit_count": 61, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-502", + "nodes": [ + "issue:33666", + "issue:35707", + "issue:40444", + "issue:40990", + "issue:41553", + "issue:41720", + "issue:42175", + "issue:42796", + "issue:43296", + "issue:43317", + "issue:43377", + "issue:43381", + "issue:43421", + "issue:43482", + "issue:43525", + "issue:43531", + "issue:43582", + "issue:43611", + "issue:43632", + "issue:43646", + "issue:43716", + "issue:43746", + "issue:43792", + "issue:43856", + "issue:44079", + "issue:44291", + "issue:44403", + "issue:44462", + "issue:44509", + "issue:44534", + "issue:44568", + "issue:44589", + "issue:44756", + "issue:44811", + "issue:44991", + "issue:45081", + "issue:45325", + "issue:45491", + "issue:45538", + "issue:45542", + "issue:45663", + "issue:45812", + "issue:45923", + "issue:46491", + "issue:46710", + "issue:46821", + "issue:46899", + "issue:46909" + ], + "soft_pairs": [ + "issue:40444|issue:45325", + "issue:43716|issue:46899", + "issue:45812|issue:46491", + "issue:44403|issue:44991", + "issue:43381|issue:45923", + "issue:43632|issue:44756", + "issue:42796|issue:43746", + "issue:43792|issue:44589", + "issue:33666|issue:40444", + "issue:40990|issue:44811", + "issue:44291|issue:44589", + "issue:42796|issue:46710", + "issue:43482|issue:43525", + "issue:35707|issue:40444", + "issue:44079|issue:45663", + "issue:41553|issue:43582", + "issue:43482|issue:45538", + "issue:43421|issue:44568", + "issue:43531|issue:44291", + "issue:43632|issue:44534", + "issue:42175|issue:45542", + "issue:41720|issue:45081", + "issue:44462|issue:45812", + "issue:43317|issue:43856", + "issue:44079|issue:44534", + "issue:44509|issue:46909", + "issue:43296|issue:46821", + "issue:43377|issue:45491", + "issue:43611|issue:43646" + ], + "prepared_review_unit_hash": "712dd4a58f61c34d18688e023090735e7db41f5a997507b490a21b3286f4a08b", + "component_budget": { + "node_count": 502, + "item_count": 502, + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 29, + "serialized_chars": 36441, + "estimated_input_tokens": 9111, + "estimated_eval_tokens": 18478 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Most items are unrelated issue reports; the only plausible duplicate pair is the Transformers v5 model-initialization regression around `base_model_prefix` / custom model init. No PRs are present.", + "confidence": 0.67, + "canonical_issue_reason": "issue:43646 is the broadest and most representative report of the v5 model-initialization regression, and it plausibly subsumes the more specific loading failure in issue:43611.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:43646 best serves as the cluster anchor because it describes the underlying API breakage at the right abstraction level and is likely to cover the closely related `base_model_prefix` loading symptom.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:40444", + "right": "issue:45325", + "accept": false, + "reason": "Both are Qwen2.5-VL, but one is a dataloader/multimodal-training failure and the other is a RoPE position-id bug; different code paths." + }, + { + "left": "issue:43716", + "right": "issue:46899", + "accept": false, + "reason": "Both involve multimodal dtype issues, but they hit different models and different mechanisms (preprocessor/model dtype mismatch vs unsafe cast in 4-bit embedders)." + }, + { + "left": "issue:45812", + "right": "issue:46491", + "accept": false, + "reason": "Both are tokenizer regressions, but Granite wrong IDs and CodeLlama leading-space round-trip are distinct bugs in different tokenizers." + }, + { + "left": "issue:44403", + "right": "issue:44991", + "accept": false, + "reason": "Unnecessary loading noise and a tokenizer-load failure for a specific model are not the same bug." + }, + { + "left": "issue:43381", + "right": "issue:45923", + "accept": false, + "reason": "Gradient checkpointing eval-mode behavior and a missing capability flag in a trust_remote_code model are related but not the same defect." + }, + { + "left": "issue:43632", + "right": "issue:44756", + "accept": false, + "reason": "A v5 initialization flag regression and an AppleSilicon mmap OOM workaround are unrelated." + }, + { + "left": "issue:42796", + "right": "issue:43746", + "accept": false, + "reason": "Tokenizer BPE marker handling and GraniteSpeech PEFT checkpoint loading are different issues." + }, + { + "left": "issue:43792", + "right": "issue:44589", + "accept": false, + "reason": "Whisper runtime failure and missing Float8 storage registration are unrelated." + }, + { + "left": "issue:33666", + "right": "issue:40444", + "accept": false, + "reason": "Both are Qwen VL training issues, but one is multi-GPU training and the other is multi-image iterable dataset failure; not the same bug." + }, + { + "left": "issue:40990", + "right": "issue:44811", + "accept": false, + "reason": "Perplexity on gpt-oss-20b and Whisper batch_decode skipping specials are unrelated." + }, + { + "left": "issue:44291", + "right": "issue:44589", + "accept": false, + "reason": "Both touch v5 loading/runtime regressions, but one is an init_empty_weights argument break and the other is a Float8 storage lookup failure." + }, + { + "left": "issue:42796", + "right": "issue:46710", + "accept": false, + "reason": "Tokenizer marker handling and DeepSeek accuracy regression are not the same underlying issue." + }, + { + "left": "issue:43482", + "right": "issue:43525", + "accept": false, + "reason": "Qwen2.5-GGUF loading failure and missing `pad_token_id` on Llama4Config are separate model-specific regressions." + }, + { + "left": "issue:35707", + "right": "issue:40444", + "accept": false, + "reason": "Progressive generation with `inputs_embeds`/`past_key_values` is unrelated to Qwen2.5-VL multi-image finetuning." + }, + { + "left": "issue:44079", + "right": "issue:45663", + "accept": false, + "reason": "ModelOutput key assignment and Gemma4 FSDP2 shared_kv state rebuilding are unrelated." + }, + { + "left": "issue:41553", + "right": "issue:43582", + "accept": false, + "reason": "A bad tokenizer-loading error message and an AppleSilicon allocator TypeError are unrelated." + }, + { + "left": "issue:43482", + "right": "issue:45538", + "accept": false, + "reason": "GGUF loading failure and CLIPTokenizer max-length default are different tokenizer/model-loading issues." + }, + { + "left": "issue:43421", + "right": "issue:44568", + "accept": false, + "reason": "Runtime special-token post-processing and mdeberta BOS/EOS insertion are distinct tokenizer behaviors." + }, + { + "left": "issue:43531", + "right": "issue:44291", + "accept": false, + "reason": "Qwen3-MoE sliding window behavior and an init_empty_weights argument regression do not match." + }, + { + "left": "issue:43632", + "right": "issue:44534", + "accept": false, + "reason": "The `_is_hf_initialized` regression and non-persistent buffer corruption are different v5 bugs." + }, + { + "left": "issue:42175", + "right": "issue:45542", + "accept": false, + "reason": "Both involve TensorFlow/backend packaging, but they describe different install paths and symptoms; not clearly the same bug." + }, + { + "left": "issue:41720", + "right": "issue:45081", + "accept": false, + "reason": "Qwen3 auto device-mapping asserts and a Mistral regex-patching AttributeError are unrelated." + }, + { + "left": "issue:44462", + "right": "issue:45812", + "accept": false, + "reason": "Ignoring repository tokenizer.json and wrong Granite token IDs are different AutoTokenizer regressions." + }, + { + "left": "issue:43317", + "right": "issue:43856", + "accept": false, + "reason": "GPU+CPU offload loading failure and Qwen3 MoE training memory inefficiency are not the same issue." + }, + { + "left": "issue:44079", + "right": "issue:44534", + "accept": false, + "reason": "ModelOutput key handling and junk-filled non-persistent buffers are unrelated." + }, + { + "left": "issue:44509", + "right": "issue:46909", + "accept": false, + "reason": "Removed v5 pipeline docs and broken DIA documentation examples are different documentation issues." + }, + { + "left": "issue:43296", + "right": "issue:46821", + "accept": false, + "reason": "PaddleOCR-VL load failure and TimesFM window_size AttributeError are unrelated model-specific bugs." + }, + { + "left": "issue:43377", + "right": "issue:45491", + "accept": false, + "reason": "MIMI batching/padding-mask mismatch and Gemma3 mixed-length NaN embeddings are both batching-related, but not the same underlying bug." + }, + { + "left": "issue:43611", + "right": "issue:43646", + "accept": true, + "reason": "Both report the same Transformers v5 model-loading/initialization regression; `base_model_prefix` failures are a concrete symptom of the broader custom-init breakage." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet, and the soft-edge verdicts are generally conservative. The only accepted pair, issue:43611 with issue:43646, is a reasonable duplicate call at the v5 custom-init/model-loading level. No PRs are present, so there is no mergeability concern." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 13, + "review_unit_count": 61, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-502", + "nodes": [ + "issue:34634", + "issue:35707", + "issue:39692", + "issue:40444", + "issue:40990", + "issue:41720", + "issue:42175", + "issue:42796", + "issue:42915", + "issue:43262", + "issue:43296", + "issue:43377", + "issue:43454", + "issue:43482", + "issue:43493", + "issue:43550", + "issue:43606", + "issue:43644", + "issue:43701", + "issue:43746", + "issue:43818", + "issue:43819", + "issue:43828", + "issue:43873", + "issue:43881", + "issue:43994", + "issue:44079", + "issue:44186", + "issue:44265", + "issue:44291", + "issue:44466", + "issue:44479", + "issue:44488", + "issue:44610", + "issue:44829", + "issue:44877", + "issue:45072", + "issue:45325", + "issue:45357", + "issue:45381", + "issue:45479", + "issue:45584", + "issue:45701", + "issue:45823", + "issue:45834", + "issue:46402", + "issue:46821" + ], + "soft_pairs": [ + "issue:43818|issue:46402", + "issue:44829|issue:45479", + "issue:42175|issue:43296", + "issue:43644|issue:44079", + "issue:43818|issue:45357", + "issue:44265|issue:45072", + "issue:43493|issue:43994", + "issue:40444|issue:45381", + "issue:45325|issue:45381", + "issue:45325|issue:45834", + "issue:43454|issue:44466", + "issue:42915|issue:43482", + "issue:44291|issue:44877", + "issue:43701|issue:45325", + "issue:43873|issue:45701", + "issue:42796|issue:45381", + "issue:43606|issue:44265", + "issue:40990|issue:45584", + "issue:43262|issue:43746", + "issue:44479|issue:45823", + "issue:43881|issue:44488", + "issue:43828|issue:46821", + "issue:35707|issue:41720", + "issue:34634|issue:35707", + "issue:39692|issue:40444", + "issue:43454|issue:43550", + "issue:43606|issue:43819", + "issue:44186|issue:44265", + "issue:41720|issue:45823", + "issue:43377|issue:44610" + ], + "prepared_review_unit_hash": "e81492b087f7236954c284a39eaa5d15640127889642451f5b437c2f324e5173", + "component_budget": { + "node_count": 502, + "item_count": 502, + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 30, + "serialized_chars": 36280, + "estimated_input_tokens": 9070, + "estimated_eval_tokens": 18396 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Most soft pairs are broad-similarity false positives across unrelated models/subsystems. The only clear duplicate-like pair is the SigLIP2 implementation mismatch vs. nonsensical AutoModel/pipeline outputs.", + "confidence": 0.62, + "canonical_issue_reason": "issue:43493 is the clearest, most specific statement of the underlying SigLIP2 bug and best matches the duplicate candidate that reports bad downstream outputs.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:43493 is the best representative issue in this set because it precisely names the root discrepancy and can subsume the more user-facing symptom report.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43818", + "right": "issue:46402", + "accept": false, + "reason": "Different models and failures: Video-LLaVA missing temporal attention/weight sharing vs Sam3 text encoder weights not loaded." + }, + { + "left": "issue:44829", + "right": "issue:45479", + "accept": false, + "reason": "Both affect training, but the causes differ: flash_attention_3 behavior vs single-label classification with num_labels=1." + }, + { + "left": "issue:42175", + "right": "issue:43296", + "accept": false, + "reason": "Unrelated areas: backend packaging for TensorFlow vs loading PaddleOCR-VL in vLLM." + }, + { + "left": "issue:43644", + "right": "issue:44079", + "accept": false, + "reason": "One is a v5 buffer initialization issue; the other is ModelOutput key assignment behavior. Not the same bug." + }, + { + "left": "issue:43818", + "right": "issue:45357", + "accept": false, + "reason": "Both mention visual models, but one is Video-LLaVA architecture problems and the other is Qwen3.5 save_pretrained key serialization." + }, + { + "left": "issue:44265", + "right": "issue:45072", + "accept": false, + "reason": "Different failure modes: torch.export with torch_compilable_check vs dtype mismatches in bfloat16 inference." + }, + { + "left": "issue:43493", + "right": "issue:43994", + "accept": true, + "reason": "Same underlying SigLIP2 implementation problem: HF/JAX discrepancy aligns with nonsensical outputs in AutoModel/pipeline." + }, + { + "left": "issue:40444", + "right": "issue:45381", + "accept": false, + "reason": "Different Qwen2.5-VL issues: finetuning with multi-image iterable datasets vs video vision_position_ids regression." + }, + { + "left": "issue:45325", + "right": "issue:45381", + "accept": false, + "reason": "Both are position-id bugs in Qwen2.5-VL, but one is still-image temporal scaling and the other is video vision_position_ids; not clearly the same concrete bug." + }, + { + "left": "issue:45325", + "right": "issue:45834", + "accept": false, + "reason": "Different models and symptoms: Qwen2.5-VL rope indexing vs Kosmos2.5 OCR long-input index error." + }, + { + "left": "issue:43454", + "right": "issue:44466", + "accept": false, + "reason": "Related to tied lm_head weights, but one is a model config failing to tie weights and the other is inconsistent serialization behavior." + }, + { + "left": "issue:42915", + "right": "issue:43482", + "accept": false, + "reason": "Different models and problems: FineGrainedFP8Config failure vs GGUF loading on transformers v5." + }, + { + "left": "issue:44291", + "right": "issue:44877", + "accept": false, + "reason": "Different loading failures: init_empty_weights argument mismatch vs strict config blocking granite_speech." + }, + { + "left": "issue:43701", + "right": "issue:45325", + "accept": false, + "reason": "No meaningful overlap: checkpoint key mismatch vs Qwen2.5-VL rope-index scaling." + }, + { + "left": "issue:43873", + "right": "issue:45701", + "accept": false, + "reason": "Quantization/offloading behavior is unrelated to tokenization changes." + }, + { + "left": "issue:42796", + "right": "issue:45381", + "accept": false, + "reason": "Tokenizer BPE marker handling is unrelated to Qwen2.5-VL video position IDs." + }, + { + "left": "issue:43606", + "right": "issue:44265", + "accept": false, + "reason": "Bark CPU offload device mismatch and torch.export/torch_compilable_check are separate code paths." + }, + { + "left": "issue:40990", + "right": "issue:45584", + "accept": false, + "reason": "Different tasks and bugs: perplexity on GPT-OSS vs Whisper empty-transcription failure." + }, + { + "left": "issue:43262", + "right": "issue:43746", + "accept": false, + "reason": "Audio processor sampling-rate defaults and GraniteSpeech checkpoint loading are unrelated." + }, + { + "left": "issue:44479", + "right": "issue:45823", + "accept": false, + "reason": "Different regressions: Qwen video-input behavior vs Gemma4 PLE device mismatch during forward." + }, + { + "left": "issue:43881", + "right": "issue:44488", + "accept": false, + "reason": "Different model loading failures for unrelated models." + }, + { + "left": "issue:43828", + "right": "issue:46821", + "accept": false, + "reason": "Autocast dtype mismatch is unrelated to a TimesFM window_size AttributeError." + }, + { + "left": "issue:35707", + "right": "issue:41720", + "accept": false, + "reason": "Progressive generation with inputs_embeds/past_key_values is unrelated to Qwen3 auto device mapping cuda assert." + }, + { + "left": "issue:34634", + "right": "issue:35707", + "accept": false, + "reason": "Bark voice_preset is unrelated to generation with inputs_embeds and past_key_values." + }, + { + "left": "issue:39692", + "right": "issue:40444", + "accept": false, + "reason": "SigLIP2 doc-example errors are unrelated to Qwen2.5-VL multi-image finetuning failures." + }, + { + "left": "issue:43454", + "right": "issue:43550", + "accept": false, + "reason": "Tied-weight serialization/loading issue is unrelated to Bamba torch.compile SDPA failure." + }, + { + "left": "issue:43606", + "right": "issue:43819", + "accept": false, + "reason": "Device offload mismatch and DAC forward/from_latents STE mismatch are different bugs." + }, + { + "left": "issue:44186", + "right": "issue:44265", + "accept": false, + "reason": "Tokenizer NER/padding crashes are unrelated to torch.export and torch_compilable_check." + }, + { + "left": "issue:41720", + "right": "issue:45823", + "accept": false, + "reason": "Both are device-map related, but they affect different models and different errors." + }, + { + "left": "issue:43377", + "right": "issue:44610", + "accept": false, + "reason": "MIMI padding-mask batching bug and OmDet-Turbo processor/model size mismatch are unrelated." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The analysis is grounded in the packet and stays conservative on the soft-edge pairs. The SigLIP2 pair (43493/43994) is the only clearly duplicate-like match from the available titles, and the rest of the rejections are appropriately cautious rather than overstated. The representative/canonical choice is also well supported by the issue titles." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 14, + "review_unit_count": 61, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-502", + "nodes": [ + "issue:26413", + "issue:28530", + "issue:36331", + "issue:40990", + "issue:41720", + "issue:42617", + "issue:42796", + "issue:42915", + "issue:42947", + "issue:43065", + "issue:43257", + "issue:43317", + "issue:43322", + "issue:43377", + "issue:43450", + "issue:43454", + "issue:43526", + "issue:43550", + "issue:43572", + "issue:43720", + "issue:43749", + "issue:43873", + "issue:43881", + "issue:43883", + "issue:44186", + "issue:44387", + "issue:44466", + "issue:44534", + "issue:44568", + "issue:44756", + "issue:44877", + "issue:44898", + "issue:44977", + "issue:44991", + "issue:45072", + "issue:45357", + "issue:45701", + "issue:45704", + "issue:45759", + "issue:45823", + "issue:45920", + "issue:45923", + "issue:46050", + "issue:46209", + "issue:46305", + "issue:46315", + "issue:46402", + "issue:46821" + ], + "soft_pairs": [ + "issue:46305|issue:46315", + "issue:43257|issue:45923", + "issue:44568|issue:45920", + "issue:42796|issue:46209", + "issue:43454|issue:45072", + "issue:45823|issue:46821", + "issue:43377|issue:44898", + "issue:43454|issue:43720", + "issue:40990|issue:42947", + "issue:42617|issue:42915", + "issue:43317|issue:43873", + "issue:43749|issue:44534", + "issue:43572|issue:45759", + "issue:43526|issue:43883", + "issue:43322|issue:46050", + "issue:41720|issue:43322", + "issue:43881|issue:44991", + "issue:36331|issue:42617", + "issue:44387|issue:45704", + "issue:41720|issue:44977", + "issue:36331|issue:42915", + "issue:44466|issue:45357", + "issue:43450|issue:45701", + "issue:44877|issue:44991", + "issue:43065|issue:46402", + "issue:40990|issue:45701", + "issue:44186|issue:45072", + "issue:41720|issue:42617", + "issue:26413|issue:28530", + "issue:43550|issue:43720", + "issue:43873|issue:44756" + ], + "prepared_review_unit_hash": "e28134e280f1359b6308cd258cfc54909d44a529eba2fd1943b33da8f73a24d7", + "component_budget": { + "node_count": 502, + "item_count": 502, + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 31, + "serialized_chars": 36967, + "estimated_input_tokens": 9242, + "estimated_eval_tokens": 18740 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a loose collection of Transformer v5 regressions across tokenization, loading/offloading, config handling, and multimodal model bugs. The soft edges look thematically similar in a few places, but they are not the same underlying defects, so I would not merge any of them as duplicates.", + "confidence": 0.79, + "canonical_issue_reason": "issue:45920 is the broadest, most representative issue here: it captures a v5 tokenizer regression affecting multiple model families, which is the closest shared theme among the tokenizer-related items in the cluster.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:45920 is the best anchor for this cluster because it is concrete, well-scoped, and broad enough to represent the recurring tokenizer regression theme better than the more model-specific or narrow reports.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:46305", + "right": "issue:46315", + "accept": false, + "reason": "Different subsystems: a streaming mask off-by-one bug vs a tokenizer performance regression." + }, + { + "left": "issue:43257", + "right": "issue:45923", + "accept": false, + "reason": "Different failures: Qwen3 MOE weight conversion/loading vs a missing gradient-checkpointing flag." + }, + { + "left": "issue:44568", + "right": "issue:45920", + "accept": false, + "reason": "Both are tokenizer regressions, but they affect different model families and different concrete behaviors." + }, + { + "left": "issue:42796", + "right": "issue:46209", + "accept": false, + "reason": "Tokenizer BPE-marker handling is unrelated to FP8 inference failure on B200." + }, + { + "left": "issue:43454", + "right": "issue:45072", + "accept": false, + "reason": "lm_head tying bug vs dtype mismatch in inference; not the same code-path." + }, + { + "left": "issue:45823", + "right": "issue:46821", + "accept": false, + "reason": "Device-map forward mismatch and a window_size AttributeError are unrelated." + }, + { + "left": "issue:43377", + "right": "issue:44898", + "accept": false, + "reason": "Both involve model input handling, but one is padding-mask batching in MIMI and the other is Perceiver positional interpolation." + }, + { + "left": "issue:43454", + "right": "issue:43720", + "accept": false, + "reason": "Different defects: tied-weight serialization vs packed-weight unpacking during accelerate loading." + }, + { + "left": "issue:40990", + "right": "issue:42947", + "accept": false, + "reason": "Perplexity quality issue vs gradient checkpointing behavior; no shared bug." + }, + { + "left": "issue:42617", + "right": "issue:42915", + "accept": false, + "reason": "A 3d_parallel.py runtime failure is unrelated to Qwen3Moe FineGrainedFP8 loading." + }, + { + "left": "issue:43317", + "right": "issue:43873", + "accept": false, + "reason": "Both mention offloading/quantization, but the concrete failures and code paths differ too much to treat as the same bug." + }, + { + "left": "issue:43749", + "right": "issue:44534", + "accept": false, + "reason": "FSDP CPU RAM efficient loading vs non-persistent buffer initialization are different issues." + }, + { + "left": "issue:43572", + "right": "issue:45759", + "accept": false, + "reason": "Missing config field vs incorrect unwrapping of text_config in composite multimodal configs." + }, + { + "left": "issue:43526", + "right": "issue:43883", + "accept": false, + "reason": "Image processor label reduction bug vs missing all_tied_weights_keys attribute; unrelated." + }, + { + "left": "issue:43322", + "right": "issue:46050", + "accept": false, + "reason": "A Llava Next segmentation fault is not the same as a sinq save/load quantization regression." + }, + { + "left": "issue:41720", + "right": "issue:43322", + "accept": false, + "reason": "Different model families and different failure modes: cudaErrorAssert vs segmentation fault." + }, + { + "left": "issue:43881", + "right": "issue:44991", + "accept": false, + "reason": "Model loading failure and tokenizer loading failure are unrelated." + }, + { + "left": "issue:36331", + "right": "issue:42617", + "accept": false, + "reason": "Trainer API signature breakage is unrelated to the 3d_parallel.py execution failure." + }, + { + "left": "issue:44387", + "right": "issue:45704", + "accept": false, + "reason": "Quantization memory regression vs apex RMSNorm memory leak are different problems." + }, + { + "left": "issue:41720", + "right": "issue:44977", + "accept": false, + "reason": "Qwen3 auto device-map assertion and Qwen3.5 flash-attention generation issues are not the same concrete bug." + }, + { + "left": "issue:44466", + "right": "issue:45357", + "accept": false, + "reason": "Both concern serialization, but one is lm_head.weight and the other is visual encoder keys; different model parts and fixes." + }, + { + "left": "issue:43450", + "right": "issue:45701", + "accept": false, + "reason": "Video batching shape bug vs tokenizer behavior change; unrelated." + }, + { + "left": "issue:44877", + "right": "issue:44991", + "accept": false, + "reason": "Strict config loading vs tokenizer loading regression are different failure modes." + }, + { + "left": "issue:44186", + "right": "issue:45072", + "accept": false, + "reason": "Tokenizer crashes with NER/padding are unrelated to dtype mismatches in bfloat16 inference." + }, + { + "left": "issue:26413", + "right": "issue:28530", + "accept": false, + "reason": "Checkpoint resume iterator exhaustion and early-stopping metric lookup are different training-loop bugs." + }, + { + "left": "issue:43550", + "right": "issue:43720", + "accept": false, + "reason": "torch.compile/SDPA failure vs accelerate loading of packed weights are unrelated." + }, + { + "left": "issue:43873", + "right": "issue:44756", + "accept": false, + "reason": "Offloading/quantization problems are not the same as disabling mmap to avoid OOM on Strix Halo." + }, + { + "left": "issue:40990", + "right": "issue:45701", + "accept": false, + "reason": "Perplexity regression vs tokenization changes; no shared underlying defect." + }, + { + "left": "issue:43065", + "right": "issue:46402", + "accept": false, + "reason": "A dummy Conv2d in Sam3PixelDecoder is unrelated to missing text encoder weights in Sam3Model." + }, + { + "left": "issue:41720", + "right": "issue:42617", + "accept": false, + "reason": "Qwen3 auto device-map loading failure and 3d_parallel.py runtime failure are unrelated." + }, + { + "left": "issue:44898", + "right": "issue:45072", + "accept": false, + "reason": "Perceiver interpolation failure and dtype mismatch are different issues." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 15, + "review_unit_count": 61, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-502", + "nodes": [ + "issue:28530", + "issue:30064", + "issue:33666", + "issue:36010", + "issue:36331", + "issue:41720", + "issue:42617", + "issue:42915", + "issue:43295", + "issue:43316", + "issue:43317", + "issue:43322", + "issue:43404", + "issue:43425", + "issue:43526", + "issue:43531", + "issue:43550", + "issue:43572", + "issue:43577", + "issue:43606", + "issue:43646", + "issue:43708", + "issue:43881", + "issue:44008", + "issue:44186", + "issue:44361", + "issue:44423", + "issue:44464", + "issue:44610", + "issue:44671", + "issue:44734", + "issue:44743", + "issue:44898", + "issue:44913", + "issue:45161", + "issue:45357", + "issue:45406", + "issue:45632", + "issue:45684", + "issue:45685", + "issue:45701", + "issue:45823", + "issue:46291", + "issue:46612", + "issue:46665", + "issue:46821", + "issue:46858", + "issue:46869" + ], + "soft_pairs": [ + "issue:46612|issue:46858", + "issue:43577|issue:43708", + "issue:43295|issue:45701", + "issue:45685|issue:46821", + "issue:43550|issue:43606", + "issue:44186|issue:44610", + "issue:43646|issue:46665", + "issue:43404|issue:45357", + "issue:42915|issue:43317", + "issue:44610|issue:44898", + "issue:44734|issue:46858", + "issue:36010|issue:42617", + "issue:44423|issue:45406", + "issue:46821|issue:46869", + "issue:44008|issue:45823", + "issue:45632|issue:45684", + "issue:44464|issue:46858", + "issue:41720|issue:45161", + "issue:28530|issue:33666", + "issue:44671|issue:44743", + "issue:43531|issue:43572", + "issue:42915|issue:46291", + "issue:36331|issue:41720", + "issue:44186|issue:44361", + "issue:43526|issue:43881", + "issue:43316|issue:44913", + "issue:43425|issue:46821", + "issue:42915|issue:43322", + "issue:28530|issue:30064" + ], + "prepared_review_unit_hash": "f46c4347ef9930f033409d77ebca8dd6769378c16dc849873dd88451b6796729", + "component_budget": { + "node_count": 502, + "item_count": 502, + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 29, + "serialized_chars": 36438, + "estimated_input_tokens": 9110, + "estimated_eval_tokens": 18476 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "All soft pairs are cross-subsystem or cross-model reports with different failure modes, so none look like the same underlying bug/change. I would reject every soft edge. The most suitable standalone issue anchor is the cache/torch.compile backend bug in 46858 because it is the most concrete and actionable report.", + "confidence": 0.97, + "canonical_issue_reason": "issue:46858 is the most concrete, narrowly scoped bug report: it names the exact failing method, the tensor/device mismatch, and the backend condition (static cache + torch.compile on non-CUDA).", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 46858 is the strongest representative report overall because it has a precise failure signature and clear reproduction context, making it a good triage anchor.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:46612", + "right": "issue:46858", + "accept": false, + "reason": "Both involve generation/cache logic, but one is cache reordering semantics across model families and the other is a device-tensor arithmetic bug in prefill; different code paths." + }, + { + "left": "issue:43577", + "right": "issue:43708", + "accept": false, + "reason": "One is a model dtype loading bug in Blip2; the other is Trainer checkpoint resumption max_steps math. Unrelated failure modes." + }, + { + "left": "issue:43295", + "right": "issue:45701", + "accept": false, + "reason": "Different problems: processor/tokenizer regression for multimodal custom code vs a tokenization-version behavior change." + }, + { + "left": "issue:45685", + "right": "issue:46821", + "accept": false, + "reason": "MPS histogram implementation error for MoE internals vs a TimesFM window_size attribute error. No shared underlying bug." + }, + { + "left": "issue:43550", + "right": "issue:43606", + "accept": false, + "reason": "Both are runtime failures, but one is torch.compile/SDPA on Bamba and the other is CPU-offload device mismatch on bark-small." + }, + { + "left": "issue:44186", + "right": "issue:44610", + "accept": false, + "reason": "Tokenizer crash on NER/batching is separate from an OmDet image-size mismatch in its processor." + }, + { + "left": "issue:43646", + "right": "issue:46665", + "accept": false, + "reason": "Transformers 5.0 custom model initialization breakage is not the same as from_pretrained behaving inconsistently inside Python classes." + }, + { + "left": "issue:43404", + "right": "issue:45357", + "accept": false, + "reason": "Weight tying bug in Mistral3 is unrelated to Qwen3.5 save_pretrained writing incorrect visual encoder keys." + }, + { + "left": "issue:42915", + "right": "issue:43317", + "accept": false, + "reason": "Different loading/quantization failures on different models; one is FineGrainedFP8Config, the other is dequantized gpu+cpu offload." + }, + { + "left": "issue:44610", + "right": "issue:44898", + "accept": false, + "reason": "Both touch image preprocessing, but one is a fixed wrong output resolution and the other is a non-default-resolution interpolation failure." + }, + { + "left": "issue:44734", + "right": "issue:46858", + "accept": false, + "reason": "Both mention cache/indexing, but one crashes in serve KV-cache continuation and the other is a prefill slice arithmetic issue on non-CUDA backends." + }, + { + "left": "issue:36010", + "right": "issue:42617", + "accept": false, + "reason": "ImportError for GenerationMixin is unrelated to failure running 3d_parallel.py." + }, + { + "left": "issue:44423", + "right": "issue:45406", + "accept": false, + "reason": "Both are serve-related multimodal crashes, but the concrete bugs differ: tensor conversion on one side and missing _tokenizer on the other." + }, + { + "left": "issue:46821", + "right": "issue:46869", + "accept": false, + "reason": "TimesFM window_size attribute error and YOLO multi-GPU device-map failure are different models and different bug classes." + }, + { + "left": "issue:44008", + "right": "issue:45823", + "accept": false, + "reason": "Variable-name collision causing audio_mel_mask AttributeError is unrelated to Gemma4 PLE device mismatch under auto device mapping." + }, + { + "left": "issue:45632", + "right": "issue:45684", + "accept": false, + "reason": "Cache-path collision for trust_remote_code local models is unrelated to permission propagation in save_pretrained/register_for_auto_class." + }, + { + "left": "issue:44464", + "right": "issue:46858", + "accept": false, + "reason": "Both involve compiled-generation/caching, but one is inconsistent chunked generation outputs while the other is a device tensor returned from cache length." + }, + { + "left": "issue:41720", + "right": "issue:45161", + "accept": false, + "reason": "Auto device mapping cudaErrorAssert on A800 is not the same as TP-only failure in GPT-OSS MoE." + }, + { + "left": "issue:28530", + "right": "issue:33666", + "accept": false, + "reason": "Early-stopping metric resolution and Qwen2-VL multi-GPU training are completely unrelated." + }, + { + "left": "issue:44671", + "right": "issue:44743", + "accept": false, + "reason": "CamemBERT MLM prediction drift is a different bug from Qwen3.5 recurrent-state reset with cache." + }, + { + "left": "issue:43531", + "right": "issue:43572", + "accept": false, + "reason": "Sliding-window behavior in Qwen3-MoE is unrelated to the missing pad_token_idx config regression in StableLm." + }, + { + "left": "issue:42915", + "right": "issue:46291", + "accept": false, + "reason": "FineGrainedFP8Config model loading failure is unrelated to importing transformers.pipelines with kernels version/revision constraints." + }, + { + "left": "issue:36331", + "right": "issue:41720", + "accept": false, + "reason": "Trainer compute_loss API mismatch is unrelated to Qwen3 auto device-map cudaErrorAssert." + }, + { + "left": "issue:44186", + "right": "issue:44361", + "accept": false, + "reason": "Both are tokenizer-related AttributeErrors, but they affect different tokenizers, inputs, and failure points." + }, + { + "left": "issue:43526", + "right": "issue:43881", + "accept": false, + "reason": "BeitImageProcessorFast label reduction bug does not match a glm-4v-9b loading failure." + }, + { + "left": "issue:43316", + "right": "issue:44913", + "accept": false, + "reason": "Config API discrepancy for Gemma3TextConfig is a different issue from GPTNeoX rotary_pct not persisting on reload." + }, + { + "left": "issue:43425", + "right": "issue:46821", + "accept": false, + "reason": "Torch 2.10 compatibility is a broad environment issue, not the same as a TimesFM window_size AttributeError." + }, + { + "left": "issue:42915", + "right": "issue:43322", + "accept": false, + "reason": "Qwen3Moe FineGrainedFP8Config loading failure is unrelated to Llava Next segmentation faults." + }, + { + "left": "issue:28530", + "right": "issue:30064", + "accept": false, + "reason": "Trainer early-stopping metric handling and void segmentation-map processing are different subsystems and bugs." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 16, + "review_unit_count": 61, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-502", + "nodes": [ + "issue:29159", + "issue:30064", + "issue:33357", + "issue:34567", + "issue:34634", + "issue:35707", + "issue:36032", + "issue:41720", + "issue:42490", + "issue:43329", + "issue:43388", + "issue:43531", + "issue:43540", + "issue:43550", + "issue:43606", + "issue:43644", + "issue:43653", + "issue:43720", + "issue:43746", + "issue:43782", + "issue:43873", + "issue:44062", + "issue:44077", + "issue:44387", + "issue:44451", + "issue:44492", + "issue:44509", + "issue:44534", + "issue:44589", + "issue:44625", + "issue:44898", + "issue:44987", + "issue:44991", + "issue:45042", + "issue:45059", + "issue:45072", + "issue:45357", + "issue:45406", + "issue:45464", + "issue:45479", + "issue:45701", + "issue:45704", + "issue:45759", + "issue:45859", + "issue:46402", + "issue:46665", + "issue:46821", + "issue:46858" + ], + "soft_pairs": [ + "issue:33357|issue:35707", + "issue:29159|issue:36032", + "issue:43540|issue:43782", + "issue:43550|issue:43653", + "issue:44387|issue:45701", + "issue:43873|issue:45704", + "issue:30064|issue:34634", + "issue:43606|issue:43720", + "issue:44492|issue:44509", + "issue:44534|issue:46665", + "issue:43531|issue:45759", + "issue:41720|issue:43746", + "issue:42490|issue:43329", + "issue:43388|issue:45059", + "issue:45406|issue:45464", + "issue:43606|issue:43653", + "issue:34567|issue:35707", + "issue:43531|issue:46821", + "issue:34634|issue:41720", + "issue:43540|issue:46858", + "issue:44451|issue:44991", + "issue:45759|issue:45859", + "issue:45357|issue:46402", + "issue:43531|issue:44589", + "issue:44077|issue:44625", + "issue:43644|issue:46665", + "issue:44062|issue:44451", + "issue:44898|issue:45072", + "issue:44987|issue:45042", + "issue:44625|issue:45479" + ], + "prepared_review_unit_hash": "3fb9efd8616dee1f442fe42f09f31252907ab136a4f7b059398de4fc8b5915c2", + "component_budget": { + "node_count": 502, + "item_count": 502, + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 30, + "serialized_chars": 36842, + "estimated_input_tokens": 9211, + "estimated_eval_tokens": 18678 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a semantic bag of unrelated Transformers issues. The soft-similarity links are largely false positives across different models, code paths, and bug classes. No pull requests are present, and no soft edge should be merged.", + "confidence": 0.93, + "canonical_issue_reason": "Issue 43644 is the best canonical anchor because it is the clearest concrete regression in the set with downstream references, and it matches the only near-duplicate theme here (the later 44534 report about non-persistent buffers).", + "canonical_pr_reason": null, + "best_issue_reason": "43644 is the strongest representative issue: specific, reproducible, and the only clear duplicate-style theme in the cluster. Most other items are unrelated tokenization, loading, multimodal, or training bugs.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:33357", + "right": "issue:35707", + "accept": false, + "reason": "Both are runtime failures, but one is a CLIP/MacOS bus error and the other is a generation bug with inputs_embeds/past_key_values; different code paths and causes." + }, + { + "left": "issue:29159", + "right": "issue:36032", + "accept": false, + "reason": "Both involve tokenizers, but one is slow-vs-fast inconsistency and the other is a T5Tokenizer loading/name conflict; not the same underlying bug." + }, + { + "left": "issue:43540", + "right": "issue:43782", + "accept": false, + "reason": "Different multimodal failures: video input processing in Qwen3OmniMoe vs weight_only loading in Qwen3VL; not mergeable as one fix." + }, + { + "left": "issue:43550", + "right": "issue:43653", + "accept": false, + "reason": "Unrelated subsystems: torch.compile/SDPA on Bamba vs BigBirdTokenizer special-token registration." + }, + { + "left": "issue:44387", + "right": "issue:45701", + "accept": false, + "reason": "One is int4 quantization memory growth, the other is tokenization changes across versions; same project area but not the same bug." + }, + { + "left": "issue:43873", + "right": "issue:45704", + "accept": false, + "reason": "Offloading/quantization behavior is unrelated to T5's apex.FusedRMSNorm memory leak." + }, + { + "left": "issue:30064", + "right": "issue:34634", + "accept": false, + "reason": "Image segmentation-map processing vs BarkProcessor voice_preset handling are separate features with different failures." + }, + { + "left": "issue:43606", + "right": "issue:43720", + "accept": false, + "reason": "Both touch loading/offloading, but Bark CPU offload device mismatch and BitNet packed-weight unpacking are distinct code paths." + }, + { + "left": "issue:44492", + "right": "issue:44509", + "accept": false, + "reason": "Both are docs/wording issues, but a cache-strategy typo is not the same as stale pipeline-task documentation." + }, + { + "left": "issue:44534", + "right": "issue:46665", + "accept": false, + "reason": "The non-persistent-buffer regression is unrelated to the from_pretrained behavior discrepancy inside Python classes." + }, + { + "left": "issue:43531", + "right": "issue:45759", + "accept": false, + "reason": "Qwen3-MoE sliding_window behavior and Qwen 3.5/3.6 config unwrapping are different bugs." + }, + { + "left": "issue:41720", + "right": "issue:43746", + "accept": false, + "reason": "CUDA assert under auto device mapping on Qwen3 is unrelated to GraniteSpeech PEFT checkpoint loading." + }, + { + "left": "issue:42490", + "right": "issue:43329", + "accept": false, + "reason": "Local-path slash warning and undefined video-token helper names in multimodal token counting are unrelated." + }, + { + "left": "issue:43388", + "right": "issue:45059", + "accept": false, + "reason": "A metrics-gathering last-batch truncation bug is not the same as SAM3 text/bbox behavior." + }, + { + "left": "issue:45406", + "right": "issue:45464", + "accept": false, + "reason": "Serve crashes due to missing _tokenizer and chat/completions streaming failures are different failure modes." + }, + { + "left": "issue:43606", + "right": "issue:43653", + "accept": false, + "reason": "Bark CPU offload mismatch and BigBirdTokenizer special-token registration share only a loose CI context." + }, + { + "left": "issue:34567", + "right": "issue:35707", + "accept": false, + "reason": "TrainerState token counting and progressive generation with cache are unrelated." + }, + { + "left": "issue:43531", + "right": "issue:46821", + "accept": false, + "reason": "Qwen3-MoE sliding-window handling is unrelated to TimesFM window_size attribute access." + }, + { + "left": "issue:34634", + "right": "issue:41720", + "accept": false, + "reason": "BarkProcessor voice_preset and Qwen3 device-mapping CUDA asserts are different components and symptoms." + }, + { + "left": "issue:43540", + "right": "issue:46858", + "accept": false, + "reason": "Video-input ValueError in Qwen3OmniMoe is unrelated to static-cache slice arithmetic on non-CUDA backends." + }, + { + "left": "issue:44451", + "right": "issue:44991", + "accept": false, + "reason": "Both are model/tokenizer loading failures, but different models and likely different serialization/tokenizer issues." + }, + { + "left": "issue:45759", + "right": "issue:45859", + "accept": false, + "reason": "Composite-config unwrapping and CPU-offload incompatibility in rotary embedding are separate Qwen 3.5/3.6 bugs." + }, + { + "left": "issue:45357", + "right": "issue:46402", + "accept": false, + "reason": "Incorrect visual-encoder save keys and missing text-encoder weight loading are different loading/saving problems." + }, + { + "left": "issue:43531", + "right": "issue:44589", + "accept": false, + "reason": "Sliding-window behavior in Qwen3-MoE does not match the Float8 storage lookup failure." + }, + { + "left": "issue:44077", + "right": "issue:44625", + "accept": false, + "reason": "A patchtsmixer post_init policy issue is unrelated to Qwen3.5 num_labels propagation." + }, + { + "left": "issue:43644", + "right": "issue:46665", + "accept": false, + "reason": "Non-persistent-buffer junk initialization is not the same as from_pretrained behavior inside Python classes." + }, + { + "left": "issue:44062", + "right": "issue:44451", + "accept": false, + "reason": "AddedToken special-argument collision and ScandiBERT loading failure are separate tokenizer/load issues." + }, + { + "left": "issue:44898", + "right": "issue:45072", + "accept": false, + "reason": "Perceiver interpolation failure and bfloat16 dtype mismatch in SwitchTransformers/TimmWrapperModel are unrelated." + }, + { + "left": "issue:44987", + "right": "issue:45042", + "accept": false, + "reason": "Physical-intelligence/fast loading failure and PIL backend torchvision requirement are different regressions." + }, + { + "left": "issue:44625", + "right": "issue:45479", + "accept": false, + "reason": "Qwen3.5 num_labels propagation and single-label zero-loss classification are not the same bug." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 17, + "review_unit_count": 61, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-502", + "nodes": [ + "issue:30064", + "issue:36010", + "issue:39692", + "issue:41720", + "issue:42886", + "issue:43295", + "issue:43381", + "issue:43421", + "issue:43493", + "issue:43550", + "issue:43653", + "issue:43720", + "issue:43746", + "issue:43782", + "issue:43819", + "issue:43844", + "issue:43873", + "issue:43927", + "issue:43994", + "issue:44186", + "issue:44361", + "issue:44368", + "issue:44514", + "issue:44545", + "issue:44610", + "issue:44671", + "issue:44843", + "issue:44918", + "issue:44936", + "issue:44995", + "issue:45081", + "issue:45356", + "issue:45357", + "issue:45381", + "issue:45704", + "issue:45759", + "issue:45834", + "issue:45907", + "issue:45923", + "issue:45987", + "issue:46032", + "issue:46050", + "issue:46463", + "issue:46655", + "issue:46693", + "issue:46829", + "issue:46832", + "issue:46858" + ], + "soft_pairs": [ + "issue:43381|issue:43844", + "issue:44368|issue:45357", + "issue:39692|issue:44610", + "issue:46655|issue:46693", + "issue:43421|issue:43927", + "issue:41720|issue:45704", + "issue:43746|issue:46050", + "issue:43381|issue:45704", + "issue:42886|issue:44843", + "issue:45704|issue:45923", + "issue:44918|issue:44936", + "issue:44514|issue:44545", + "issue:45759|issue:45923", + "issue:43421|issue:43653", + "issue:43746|issue:43873", + "issue:43994|issue:44671", + "issue:39692|issue:43493", + "issue:46032|issue:46463", + "issue:43421|issue:44186", + "issue:43782|issue:45381", + "issue:30064|issue:36010", + "issue:44995|issue:46858", + "issue:45081|issue:45356", + "issue:45356|issue:45907", + "issue:43720|issue:43819", + "issue:39692|issue:43873", + "issue:43653|issue:43720", + "issue:45987|issue:46829", + "issue:46829|issue:46832", + "issue:43295|issue:45834", + "issue:43550|issue:44361" + ], + "prepared_review_unit_hash": "8ec04d4f33e9a087cc7c78188aeaac8051fa76bac97ea12684dac08a75645026", + "component_budget": { + "node_count": 502, + "item_count": 502, + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 31, + "serialized_chars": 37468, + "estimated_input_tokens": 9367, + "estimated_eval_tokens": 18990 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly heterogeneous. I found two real duplicate subclusters: the Qwen2_5_VLProcessor batched `padding=False` crash, and an HF_HUB_OFFLINE tokenizer regression around `_patch_mistral_regex` / `model_info()`. The rest of the soft matches look like false positives across unrelated bugs.", + "confidence": 0.76, + "canonical_issue_reason": "issue:44545 is the best canonical issue here: it is an exact duplicate of issue:44514 with the same `Qwen2_5_VLProcessor.apply_chat_template` batched `padding=False` crash, and it has the stronger linkage signal.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:44545 is the clearest duplicate target in the set: precise title, same failure mode as its twin, and better reference signal than the alternative duplicates.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43381", + "right": "issue:43844", + "accept": false, + "reason": "Different bugs: gradient checkpointing eval-mode misuse vs ZeRO-3 gradient growth during training." + }, + { + "left": "issue:44368", + "right": "issue:45357", + "accept": false, + "reason": "Unrelated Qwen warnings/regressions: tie_word_embeddings warning vs bad visual encoder keys on save." + }, + { + "left": "issue:39692", + "right": "issue:44610", + "accept": false, + "reason": "Different model paths and failures: SigLIP2 doc example issues vs OmDet-Turbo image-size mismatch." + }, + { + "left": "issue:46655", + "right": "issue:46693", + "accept": false, + "reason": "Different areas: SmolVLM video padding bug vs FA2 long-context performance regression." + }, + { + "left": "issue:43421", + "right": "issue:43927", + "accept": false, + "reason": "Both involve tokenizer/config state, but the concrete bugs differ: runtime special-token updates vs save/load custom token IDs." + }, + { + "left": "issue:41720", + "right": "issue:45704", + "accept": false, + "reason": "Completely different problems: Qwen3 device-mapping CUDA assert vs T5 apex RMSNorm memory leak." + }, + { + "left": "issue:43746", + "right": "issue:46050", + "accept": false, + "reason": "Different load/save failures and different model families; not the same concrete code-path bug." + }, + { + "left": "issue:43381", + "right": "issue:45704", + "accept": false, + "reason": "No overlap beyond being model/runtime issues; the failure modes are unrelated." + }, + { + "left": "issue:42886", + "right": "issue:44843", + "accept": true, + "reason": "Same offline-tokenizer regression: 44843 identifies the concrete root cause (`model_info()` in `_patch_mistral_regex`) behind 42886's cache-load failure." + }, + { + "left": "issue:45704", + "right": "issue:45923", + "accept": false, + "reason": "Different issues in different models: T5 RMSNorm leak vs missing `supports_gradient_checkpointing` flag." + }, + { + "left": "issue:44918", + "right": "issue:44936", + "accept": false, + "reason": "Different trainer problems: unpacking Qwen3.5 embeddings in TRL vs evaluate failing after train." + }, + { + "left": "issue:44514", + "right": "issue:44545", + "accept": true, + "reason": "Exact duplicate: same `Qwen2_5_VLProcessor.apply_chat_template` batched `padding=False` crash." + }, + { + "left": "issue:45759", + "right": "issue:45923", + "accept": false, + "reason": "Unrelated multimodal config issue vs trust_remote_code capability flag issue." + }, + { + "left": "issue:43421", + "right": "issue:43653", + "accept": false, + "reason": "Both are tokenizer-related, but one is runtime post-processor updates and the other is BigBird mask-token registration; different bugs." + }, + { + "left": "issue:43746", + "right": "issue:43873", + "accept": false, + "reason": "Different loading/offloading problems with different model scopes; not the same underlying bug." + }, + { + "left": "issue:43994", + "right": "issue:44671", + "accept": false, + "reason": "Different models and symptoms: SigLIP2 nonsensical outputs vs CamemBERT masked-LM regression." + }, + { + "left": "issue:39692", + "right": "issue:43493", + "accept": false, + "reason": "Same broad SigLIP2 family, but different concrete problems: documentation/example failure vs HF/JAX implementation mismatch." + }, + { + "left": "issue:46032", + "right": "issue:46463", + "accept": false, + "reason": "Different code paths: Mamba2 cache misuse vs T5 decoder-input validation message wording." + }, + { + "left": "issue:43421", + "right": "issue:44186", + "accept": false, + "reason": "Both tokenizer issues, but one is post-processor updates and the other is LayoutLMv2 NER/padding crashes." + }, + { + "left": "issue:43782", + "right": "issue:45381", + "accept": false, + "reason": "Different Qwen VL failures: `weight_only=True` load error vs wrong `vision_position_ids` on video input." + }, + { + "left": "issue:30064", + "right": "issue:36010", + "accept": false, + "reason": "Unrelated issues: image processor segmentation-map handling vs GenerationMixin import error." + }, + { + "left": "issue:44995", + "right": "issue:46858", + "accept": false, + "reason": "Both mention cache/state, but the concrete bugs differ: stale indexer cache on second forward vs static-cache seq-length return type in compile paths." + }, + { + "left": "issue:45081", + "right": "issue:45356", + "accept": false, + "reason": "Related tokenizer patch area, but different failures: AttributeError in `_patch_mistral_regex` vs codec-handling regression and warning text." + }, + { + "left": "issue:45356", + "right": "issue:45907", + "accept": false, + "reason": "Different problems: Kimi-K2.5 tokenizer regression vs `list_repo_files` not forwarding `revision`." + }, + { + "left": "issue:43720", + "right": "issue:43819", + "accept": false, + "reason": "Different model families and bugs: packed-weight loading vs DAC latent/forward mismatch." + }, + { + "left": "issue:39692", + "right": "issue:43873", + "accept": false, + "reason": "No shared concrete failure mode; SigLIP2 doc/model mismatch is unrelated to quantization offloading behavior." + }, + { + "left": "issue:43653", + "right": "issue:43720", + "accept": false, + "reason": "Tokenizer special-token registration bug vs BitNet packed-weight loading bug; unrelated." + }, + { + "left": "issue:45987", + "right": "issue:46829", + "accept": false, + "reason": "Both are output-shape/interface issues, but one is cache seq-length type and the other is classifier-output hidden_states misassignment." + }, + { + "left": "issue:46829", + "right": "issue:46832", + "accept": false, + "reason": "Different model outputs and failure modes; no duplicate evidence." + }, + { + "left": "issue:43295", + "right": "issue:45834", + "accept": false, + "reason": "Both are multimodal processor regressions, but the concrete bugs differ: processor.tokenizer/images regression vs Kosmos2.5 long-OCR index error." + }, + { + "left": "issue:43550", + "right": "issue:44361", + "accept": false, + "reason": "Completely unrelated: torch.compile+SDPA on Bamba vs MLukeTokenizer AttributeError." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The analysis is grounded in the packet and stays conservative. The two accepted duplicate pairs are well supported by the titles and linkage signals, and the rejected soft matches are generally justified by clearly different failure modes or model paths. No pull-request pairing issues apply here." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 18, + "review_unit_count": 61, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-502", + "nodes": [ + "issue:30064", + "issue:34634", + "issue:36010", + "issue:36331", + "issue:39692", + "issue:41720", + "issue:43335", + "issue:43531", + "issue:43540", + "issue:43550", + "issue:43596", + "issue:43611", + "issue:43618", + "issue:43643", + "issue:43644", + "issue:43653", + "issue:43697", + "issue:43749", + "issue:43756", + "issue:43819", + "issue:43927", + "issue:44008", + "issue:44062", + "issue:44186", + "issue:44220", + "issue:44442", + "issue:44448", + "issue:44488", + "issue:44661", + "issue:44671", + "issue:44821", + "issue:44843", + "issue:44877", + "issue:44898", + "issue:44991", + "issue:45081", + "issue:45084", + "issue:45183", + "issue:45381", + "issue:45479", + "issue:45584", + "issue:45834", + "issue:45987", + "issue:46294", + "issue:46665", + "issue:46693", + "issue:46858", + "issue:46932" + ], + "soft_pairs": [ + "issue:44062|issue:44488", + "issue:43550|issue:43927", + "issue:43596|issue:44220", + "issue:45381|issue:46693", + "issue:43540|issue:46294", + "issue:43531|issue:43643", + "issue:44008|issue:46858", + "issue:43611|issue:46665", + "issue:43653|issue:43819", + "issue:44220|issue:44661", + "issue:45987|issue:46932", + "issue:44186|issue:44898", + "issue:41720|issue:44991", + "issue:43618|issue:43697", + "issue:30064|issue:36331", + "issue:43643|issue:44877", + "issue:43550|issue:43819", + "issue:43335|issue:43756", + "issue:44843|issue:45081", + "issue:44821|issue:45084", + "issue:44186|issue:44442", + "issue:39692|issue:45834", + "issue:44448|issue:44671", + "issue:43540|issue:44008", + "issue:43644|issue:43749", + "issue:34634|issue:36010", + "issue:45183|issue:45479", + "issue:45584|issue:45834" + ], + "prepared_review_unit_hash": "24909c2a8408db001127ec7614a3a5791741f9480c0202c6e4e5718a52433c01", + "component_budget": { + "node_count": 502, + "item_count": 502, + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 28, + "serialized_chars": 36286, + "estimated_input_tokens": 9072, + "estimated_eval_tokens": 18400 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is largely heterogeneous: most pairs are only superficially similar (same model family, same subsystem, or same version-regression theme) but describe different code paths and fixes. No soft edge should be merged as a duplicate.", + "confidence": 0.97, + "canonical_issue_reason": "Issue 44843 is the clearest concrete regression in the set and the most actionable representative of the tokenizer-loading/_patch_mistral_regex family, with a direct failure mode and downstream references.", + "canonical_pr_reason": null, + "best_issue_reason": "44843 is the strongest standalone issue to anchor the cluster because it has a specific repro, a single code path, and the clearest connection to a nearby soft-candidate issue family.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44062", + "right": "issue:44488", + "accept": false, + "reason": "Both involve tokenizers, but one is an AddedToken constructor error while the other is a model-loading failure for a specific checkpoint; different bugs and fixes." + }, + { + "left": "issue:43550", + "right": "issue:43927", + "accept": false, + "reason": "Unrelated failures: torch.compile/SDPA on Bamba versus save/load of DiaConfig token IDs." + }, + { + "left": "issue:43596", + "right": "issue:44220", + "accept": false, + "reason": "Different subsystems and symptoms: deepspeed zero3/BertModel initialization versus fbank feature extraction." + }, + { + "left": "issue:45381", + "right": "issue:46693", + "accept": false, + "reason": "One is a qwen2.5-vl video-position-id bug; the other is a long-context FA2 performance regression. Same model area only superficially." + }, + { + "left": "issue:43540", + "right": "issue:46294", + "accept": false, + "reason": "Both are video-related Qwen issues, but they fail in different stages with different root causes and are not the same bug." + }, + { + "left": "issue:43531", + "right": "issue:43643", + "accept": false, + "reason": "Sliding-window behavior in Qwen3-MoE is unrelated to AutoConfig missing fields with trust_remote_code." + }, + { + "left": "issue:44008", + "right": "issue:46858", + "accept": false, + "reason": "Gemma3n variable-name collision versus StaticCache sequence-length tensor behavior; different code paths." + }, + { + "left": "issue:43611", + "right": "issue:46665", + "accept": false, + "reason": "Base-model-prefix loading breakage is not the same as inconsistent from_pretrained behavior inside Python classes." + }, + { + "left": "issue:43653", + "right": "issue:43819", + "accept": false, + "reason": "Tokenizer special-token registration bug versus DAC.from_latents/STE mismatch; unrelated." + }, + { + "left": "issue:44220", + "right": "issue:44661", + "accept": false, + "reason": "Feature extraction failure versus tokenizer-mapping-name handling in add-new-model-like; not the same underlying bug." + }, + { + "left": "issue:45987", + "right": "issue:46932", + "accept": false, + "reason": "StaticCache.get_seq_length() returning a tensor is unrelated to Gemma4 quantized vision/audio embedder dtype casting." + }, + { + "left": "issue:44186", + "right": "issue:44898", + "accept": false, + "reason": "LayoutLMv2Tokenizer padding/NER crash is a different bug than Perceiver image classification on non-default resolution." + }, + { + "left": "issue:41720", + "right": "issue:44991", + "accept": false, + "reason": "CUDA assert with auto device mapping is unrelated to tokenizer loading for est-roberta." + }, + { + "left": "issue:43618", + "right": "issue:43697", + "accept": false, + "reason": "CLIPOutput attentions assignment and RTDetrV2 output drift are different model regressions." + }, + { + "left": "issue:30064", + "right": "issue:36331", + "accept": false, + "reason": "ImageProcessor void segmentation maps and CustomTrainer.compute_loss signature mismatch are unrelated." + }, + { + "left": "issue:43643", + "right": "issue:44877", + "accept": false, + "reason": "Both concern config loading, but one drops fields from trust_remote_code and the other rejects a strict config; different failure mechanisms." + }, + { + "left": "issue:43550", + "right": "issue:43819", + "accept": false, + "reason": "torch.compile/SDPA failure is unrelated to DAC.from_latents missing STE behavior." + }, + { + "left": "issue:43335", + "right": "issue:43756", + "accept": false, + "reason": "SwitchTransformers sparse-layer construction and Smollm3 RoPE-layer loss are unrelated model-architecture bugs." + }, + { + "left": "issue:44843", + "right": "issue:45081", + "accept": false, + "reason": "Same helper function, but different bugs: offline model_info() call versus missing backend_tokenizer attribute. Not the same underlying defect." + }, + { + "left": "issue:44821", + "right": "issue:45084", + "accept": false, + "reason": "AutoImageProcessor URL loading and Jinja/non-template compilation are unrelated." + }, + { + "left": "issue:44186", + "right": "issue:44442", + "accept": false, + "reason": "Tokenizer crash on NER/padding is not the same as AutoTokenizer failing to load FastSpeech2ConformerTokenizer." + }, + { + "left": "issue:39692", + "right": "issue:45834", + "accept": false, + "reason": "Documentation example errors and long OCR index errors are different problem classes." + }, + { + "left": "issue:44448", + "right": "issue:44671", + "accept": false, + "reason": "Both are version-regression output changes, but for different models and different code paths." + }, + { + "left": "issue:43540", + "right": "issue:44008", + "accept": false, + "reason": "Qwen video-input ValueError and Gemma3n audio_mel_mask collision are unrelated." + }, + { + "left": "issue:43644", + "right": "issue:43749", + "accept": false, + "reason": "Non-persistent buffer junk initialization and FSDP_CPU_RAM_EFFICIENT_LOADING are separate loading/runtime issues." + }, + { + "left": "issue:34634", + "right": "issue:36010", + "accept": false, + "reason": "BarkProcessor voice_preset and GenerationMixin import failure are unrelated." + }, + { + "left": "issue:45183", + "right": "issue:45479", + "accept": false, + "reason": "Input validation XOR messaging and single-label classification zero-loss are different bugs." + }, + { + "left": "issue:45584", + "right": "issue:45834", + "accept": false, + "reason": "Whisper empty transcription after align_special_tokens is unrelated to Kosmos2.5 long OCR input index errors." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 19, + "review_unit_count": 61, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-502", + "nodes": [ + "issue:28218", + "issue:29159", + "issue:42175", + "issue:43322", + "issue:43388", + "issue:43421", + "issue:43540", + "issue:43550", + "issue:43577", + "issue:43582", + "issue:43606", + "issue:43645", + "issue:43716", + "issue:43720", + "issue:43749", + "issue:43782", + "issue:43818", + "issue:43825", + "issue:43827", + "issue:43931", + "issue:43994", + "issue:44008", + "issue:44291", + "issue:44479", + "issue:44561", + "issue:44610", + "issue:44671", + "issue:44805", + "issue:44811", + "issue:44936", + "issue:44977", + "issue:44991", + "issue:45020", + "issue:45072", + "issue:45081", + "issue:45381", + "issue:45685", + "issue:45701", + "issue:45834", + "issue:46032", + "issue:46175", + "issue:46291", + "issue:46294", + "issue:46463", + "issue:46512", + "issue:46665", + "issue:46846", + "issue:46932" + ], + "soft_pairs": [ + "issue:43606|issue:45072", + "issue:43322|issue:45685", + "issue:43550|issue:45072", + "issue:43388|issue:46032", + "issue:43782|issue:44977", + "issue:44479|issue:45834", + "issue:43825|issue:43994", + "issue:44991|issue:46291", + "issue:43582|issue:46512", + "issue:43577|issue:43818", + "issue:43421|issue:44811", + "issue:44561|issue:45020", + "issue:43716|issue:44008", + "issue:43931|issue:46846", + "issue:46294|issue:46463", + "issue:28218|issue:29159", + "issue:44610|issue:45072", + "issue:44805|issue:44936", + "issue:43540|issue:45381", + "issue:43720|issue:45072", + "issue:43825|issue:43827", + "issue:43749|issue:44291", + "issue:43645|issue:46665", + "issue:44671|issue:45701", + "issue:43577|issue:43582", + "issue:42175|issue:45081", + "issue:46175|issue:46932" + ], + "prepared_review_unit_hash": "8579c11b2a6a0d3f8196b2b6f061f4c600107e371bbe2ea0804966c24f2f32ce", + "component_budget": { + "node_count": 502, + "item_count": 502, + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 27, + "serialized_chars": 36226, + "estimated_input_tokens": 9057, + "estimated_eval_tokens": 18370 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "No soft-edge pair looks like the same concrete bug/change; most are only loosely related by model family or subsystem. The strongest standalone issue here is the FSDP CPU RAM-efficient loading regression, which is concrete and well-discussed.", + "confidence": 0.59, + "canonical_issue_reason": "issue:43749 is a concrete, user-facing regression with high discussion activity and inbound references, making it a strong duplicate anchor.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:43749 is the most representative standalone bug report: specific failure mode, broad impact, and enough engagement to serve as a stable triage target.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43606", + "right": "issue:45072", + "accept": false, + "reason": "Different failures and code paths: CPU offload device mismatch vs bfloat16 dtype mismatch." + }, + { + "left": "issue:43322", + "right": "issue:45685", + "accept": false, + "reason": "Different subsystems and symptoms: Llava Next loading segfault vs an MPS histogram/int error." + }, + { + "left": "issue:43550", + "right": "issue:45072", + "accept": false, + "reason": "One is torch.compile + SDPA on Bamba-9B-v2; the other is a generic dtype mismatch regression." + }, + { + "left": "issue:43388", + "right": "issue:46032", + "accept": false, + "reason": "Different logic bugs: metric gathering truncation vs Mamba2 use_cache correctness." + }, + { + "left": "issue:43782", + "right": "issue:44977", + "accept": false, + "reason": "Different models and bugs: Qwen3VL weight_only loading error vs Qwen3.5 flash-attention generation failure." + }, + { + "left": "issue:44479", + "right": "issue:45834", + "accept": false, + "reason": "Both involve vision/video inputs, but the affected models and failure modes differ materially." + }, + { + "left": "issue:43825", + "right": "issue:43994", + "accept": false, + "reason": "Error-message wording about pipeline support is unrelated to SigLIP2 producing bad outputs." + }, + { + "left": "issue:44991", + "right": "issue:46291", + "accept": false, + "reason": "Tokenizer loading failure for a specific model vs an import-time kernels dependency error." + }, + { + "left": "issue:43582", + "right": "issue:46512", + "accept": false, + "reason": "AppleSilicon allocator TypeError and stale config.dtype after from_config are different bugs." + }, + { + "left": "issue:43577", + "right": "issue:43818", + "accept": false, + "reason": "Blip2 dtype propagation and Video-LLaVA architecture/state-shared issues are unrelated." + }, + { + "left": "issue:43421", + "right": "issue:44811", + "accept": false, + "reason": "Runtime special-token post-processor updates are not the same bug as Whisper batch_decode ignoring skip_special_tokens." + }, + { + "left": "issue:44561", + "right": "issue:45020", + "accept": false, + "reason": "Both mention remote_code/v5 breakage, but the concrete root causes are not shown to be the same." + }, + { + "left": "issue:43716", + "right": "issue:44008", + "accept": false, + "reason": "Different models and error mechanisms: dtype mismatch in Mistral-3 vs a Gemma3n variable-name collision." + }, + { + "left": "issue:43931", + "right": "issue:46846", + "accept": false, + "reason": "Qwen3-VL weight shape mismatch is not the same as missing linear_attn tensor-parallel plan entries." + }, + { + "left": "issue:46294", + "right": "issue:46463", + "accept": false, + "reason": "A video chat-template token/feature mismatch is unrelated to T5 decoder input validation messaging." + }, + { + "left": "issue:28218", + "right": "issue:29159", + "accept": false, + "reason": "Both concern tokenizer behavior, but one is a specific added-token spacing bug and the other is a broad slow/fast inconsistency report." + }, + { + "left": "issue:44610", + "right": "issue:45072", + "accept": false, + "reason": "OmDet-Turbo image-size mismatch is a different class of bug from bfloat16 dtype mismatches." + }, + { + "left": "issue:44805", + "right": "issue:44936", + "accept": false, + "reason": "Mask/tensor shape mismatch is not the same underlying issue as trainer.evaluate() failing after train()." + }, + { + "left": "issue:43540", + "right": "issue:45381", + "accept": false, + "reason": "Different Qwen video-input failures with different model families and error conditions." + }, + { + "left": "issue:43720", + "right": "issue:45072", + "accept": false, + "reason": "Packed-weights loading failure and dtype mismatch in inference are unrelated." + }, + { + "left": "issue:43825", + "right": "issue:43827", + "accept": false, + "reason": "One is an error message issue; the other is stale docs still referencing removed pipelines." + }, + { + "left": "issue:43749", + "right": "issue:44291", + "accept": false, + "reason": "Both are loading-related, but FSDP CPU RAM-efficient loading and init_empty_weights argument handling are distinct code paths." + }, + { + "left": "issue:43645", + "right": "issue:46665", + "accept": false, + "reason": "Jupyter custom-model initialization breakage is different from from_pretrained behavior inside Python classes." + }, + { + "left": "issue:44671", + "right": "issue:45701", + "accept": false, + "reason": "CamemBERT masked-LM regression and a generic tokenization-change report are not clearly the same defect." + }, + { + "left": "issue:43577", + "right": "issue:43582", + "accept": false, + "reason": "A dtype propagation bug in BLIP2 is unrelated to an AppleSilicon allocator TypeError." + }, + { + "left": "issue:42175", + "right": "issue:45081", + "accept": false, + "reason": "Packaging/backend dependency omission and a Mistral tokenizer regex crash are unrelated." + }, + { + "left": "issue:46175", + "right": "issue:46932", + "accept": false, + "reason": "Massive one-hot position embedding materialization is a different problem than quantized embedder dtype casting." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 20, + "review_unit_count": 61, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-502", + "nodes": [ + "issue:34634", + "issue:36010", + "issue:36331", + "issue:42175", + "issue:42490", + "issue:43257", + "issue:43262", + "issue:43329", + "issue:43452", + "issue:43526", + "issue:43531", + "issue:43550", + "issue:43582", + "issue:43638", + "issue:43645", + "issue:43653", + "issue:43697", + "issue:43720", + "issue:43742", + "issue:43782", + "issue:43819", + "issue:43906", + "issue:43927", + "issue:44060", + "issue:44368", + "issue:44554", + "issue:44589", + "issue:44610", + "issue:44857", + "issue:44912", + "issue:44936", + "issue:44991", + "issue:45003", + "issue:45020", + "issue:45042", + "issue:45081", + "issue:45084", + "issue:45325", + "issue:45372", + "issue:45406", + "issue:45538", + "issue:45600", + "issue:46209", + "issue:46285", + "issue:46291", + "issue:46326", + "issue:46697", + "issue:46728" + ], + "soft_pairs": [ + "issue:34634|issue:36331", + "issue:45003|issue:46285", + "issue:43653|issue:44912", + "issue:46697|issue:46728", + "issue:43782|issue:45325", + "issue:43262|issue:43819", + "issue:45081|issue:45406", + "issue:45372|issue:46326", + "issue:43582|issue:44857", + "issue:44589|issue:45042", + "issue:43550|issue:44912", + "issue:43742|issue:45020", + "issue:45084|issue:45538", + "issue:43329|issue:45600", + "issue:43452|issue:44991", + "issue:44060|issue:44368", + "issue:43526|issue:43697", + "issue:43257|issue:43329", + "issue:43526|issue:44554", + "issue:43906|issue:46209", + "issue:36010|issue:36331", + "issue:43819|issue:43927", + "issue:36010|issue:43645", + "issue:43531|issue:45042", + "issue:42175|issue:46291", + "issue:43638|issue:44936", + "issue:42490|issue:43257", + "issue:43720|issue:44610" + ], + "prepared_review_unit_hash": "85db1b9ba79c3c01ef722254e1e44b04e06c5405499b5f8b026ab69f810531ca", + "component_budget": { + "node_count": 502, + "item_count": 502, + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 28, + "serialized_chars": 36525, + "estimated_input_tokens": 9132, + "estimated_eval_tokens": 18520 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is mostly a loose bundle of unrelated Transformers bug reports that only share broad subsystem or version-regression themes. I do not see any true duplicate pairs among the soft edges, so all proposed merges are rejected.", + "confidence": 0.84, + "canonical_issue_reason": "Issue 45020 is the broadest representative of the cluster\u2019s general theme of Transformers version regressions affecting loading/remote-code models, but it is not a true duplicate umbrella for the whole set.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 45020 is the most central and broadly scoped report here; the others are mostly model- or component-specific regressions that should stay separate.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:34634", + "right": "issue:36331", + "accept": false, + "reason": "Different components and failure modes: BarkProcessor voice_preset vs CustomTrainer.compute_loss signature mismatch." + }, + { + "left": "issue:45003", + "right": "issue:46285", + "accept": false, + "reason": "Both are model-internals regressions, but one is unsafe sys.modules access and the other is CLIP class structure inconsistency." + }, + { + "left": "issue:43653", + "right": "issue:44912", + "accept": false, + "reason": "Tokenizer special-token decode bug vs MXFP4 quantization loading/fallback bug; unrelated issues." + }, + { + "left": "issue:46697", + "right": "issue:46728", + "accept": false, + "reason": "Both involve image-processing sizing logic, but they affect different models and different calculations." + }, + { + "left": "issue:43782", + "right": "issue:45325", + "accept": false, + "reason": "Both are Qwen VL-related, but one is a from_pretrained weight_only failure and the other is a rope-index temporal scaling bug." + }, + { + "left": "issue:43262", + "right": "issue:43819", + "accept": false, + "reason": "Audio processor sampling-rate default mismatch is unrelated to DAC latent/STE forward-pass mismatch." + }, + { + "left": "issue:45081", + "right": "issue:45406", + "accept": false, + "reason": "Tokenizer regex patch crash and Gemma4Processor missing _tokenizer are distinct load-path failures." + }, + { + "left": "issue:45372", + "right": "issue:46326", + "accept": false, + "reason": "Both mention mistral_common, but one is an import error and the other is a continue_final_message behavior bug." + }, + { + "left": "issue:43582", + "right": "issue:44857", + "accept": false, + "reason": "Apple Silicon allocator TypeError and CUDA AMP image-loss crash are unrelated runtime/platform bugs." + }, + { + "left": "issue:44589", + "right": "issue:45042", + "accept": false, + "reason": "Storage-object lookup failure is unrelated to PIL image processors incorrectly requiring torchvision." + }, + { + "left": "issue:43550", + "right": "issue:44912", + "accept": false, + "reason": "torch.compile/SDPA failure for Bamba is unrelated to MXFP4 quantization loading behavior." + }, + { + "left": "issue:43742", + "right": "issue:45020", + "accept": false, + "reason": "A model-specific key error when loading MobileLLM-125M is not the same as the broad remote_code regression." + }, + { + "left": "issue:45084", + "right": "issue:45538", + "accept": false, + "reason": "Template compilation TypeError and CLIPTokenizer max_length bug are unrelated." + }, + { + "left": "issue:43329", + "right": "issue:45600", + "accept": false, + "reason": "Multimodal token-count video-branch bug is unrelated to auto_mappings referencing removed configs." + }, + { + "left": "issue:43452", + "right": "issue:44991", + "accept": false, + "reason": "Both are from_pretrained loading problems, but the affected artifacts and root causes differ." + }, + { + "left": "issue:44060", + "right": "issue:44368", + "accept": false, + "reason": "One is an incorrect tied-weights warning bug; the other is a user-facing config warning about tie_word_embeddings." + }, + { + "left": "issue:43526", + "right": "issue:43697", + "accept": false, + "reason": "BeitImageProcessorFast label reduction bug is unrelated to RTDetrV2 output divergence." + }, + { + "left": "issue:43257", + "right": "issue:43329", + "accept": false, + "reason": "Qwen3-MoE weight conversion/loading issue is unrelated to multimodal token counting in the video branch." + }, + { + "left": "issue:43526", + "right": "issue:44554", + "accept": false, + "reason": "Different domains and code paths: image label reduction vs MPS attention correctness." + }, + { + "left": "issue:43906", + "right": "issue:46209", + "accept": false, + "reason": "An isolated reproduction of an older issue is unrelated to FP8 inference failure on B200." + }, + { + "left": "issue:36010", + "right": "issue:36331", + "accept": false, + "reason": "ImportError for GenerationMixin and compute_loss signature mismatch are separate regressions." + }, + { + "left": "issue:43819", + "right": "issue:43927", + "accept": false, + "reason": "DAC latent consistency bug and DiaConfig save/load token-id loss are different audio-model issues." + }, + { + "left": "issue:36010", + "right": "issue:43645", + "accept": false, + "reason": "Import path breakage and Jupyter custom-model initialization failure are not the same underlying bug." + }, + { + "left": "issue:43531", + "right": "issue:45042", + "accept": false, + "reason": "Qwen3-MoE sliding_window and PIL/torchvision backend requirements are unrelated." + }, + { + "left": "issue:42175", + "right": "issue:46291", + "accept": false, + "reason": "Packaging backend-selection issue and pipelines/kernels version requirement issue are unrelated." + }, + { + "left": "issue:43638", + "right": "issue:44936", + "accept": false, + "reason": "Zero3/non-pretrained Bert index error and train-then-evaluate failure are separate trainer/deepspeed bugs." + }, + { + "left": "issue:42490", + "right": "issue:43257", + "accept": false, + "reason": "Trailing-slash local path warning is unrelated to Qwen3 MoE weight conversion/loading." + }, + { + "left": "issue:43720", + "right": "issue:44610", + "accept": false, + "reason": "BitNet packed-weight loading and OmDet-Turbo processor input-size mismatch are unrelated." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 21, + "review_unit_count": 61, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-502", + "nodes": [ + "issue:28282", + "issue:29159", + "issue:29870", + "issue:34567", + "issue:38175", + "issue:39692", + "issue:40990", + "issue:42915", + "issue:43257", + "issue:43262", + "issue:43329", + "issue:43450", + "issue:43525", + "issue:43540", + "issue:43606", + "issue:43643", + "issue:43650", + "issue:43696", + "issue:43742", + "issue:43746", + "issue:43749", + "issue:43854", + "issue:43906", + "issue:43927", + "issue:43931", + "issue:44291", + "issue:44387", + "issue:44479", + "issue:44568", + "issue:44610", + "issue:44617", + "issue:44756", + "issue:44811", + "issue:44821", + "issue:44912", + "issue:44977", + "issue:44987", + "issue:45042", + "issue:45356", + "issue:45459", + "issue:45538", + "issue:45759", + "issue:45803", + "issue:45859", + "issue:46291", + "issue:46421", + "issue:46424", + "issue:46701" + ], + "soft_pairs": [ + "issue:39692|issue:44912", + "issue:43854|issue:44977", + "issue:43643|issue:45759", + "issue:43746|issue:44811", + "issue:43257|issue:43696", + "issue:38175|issue:44610", + "issue:44387|issue:44617", + "issue:43262|issue:44811", + "issue:43257|issue:43931", + "issue:45459|issue:45803", + "issue:43329|issue:43450", + "issue:46421|issue:46424", + "issue:42915|issue:43742", + "issue:43329|issue:44479", + "issue:43749|issue:44756", + "issue:34567|issue:40990", + "issue:43650|issue:43906", + "issue:44291|issue:45042", + "issue:42915|issue:44821", + "issue:29159|issue:34567", + "issue:43606|issue:45859", + "issue:43927|issue:44912", + "issue:28282|issue:29870", + "issue:43540|issue:46701", + "issue:43525|issue:46291", + "issue:44568|issue:45356", + "issue:44987|issue:45538" + ], + "prepared_review_unit_hash": "ca64c1a518ef71a46b0a9b7d3bfc1d1f3574d2970a32b7a224bfc66184dbc4f9", + "component_budget": { + "node_count": 502, + "item_count": 502, + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 27, + "serialized_chars": 35923, + "estimated_input_tokens": 8981, + "estimated_eval_tokens": 18218 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Most candidate pairs are only loosely related by subsystem or symptom and do not look like the same underlying bug. The only strong duplicate candidate is the Qwen video regression pair, where the broader regression report is the better anchor.", + "confidence": 0.73, + "canonical_issue_reason": "Issue 44479 is the best canonical issue: it is the broader user-facing regression report across multiple Qwen VL variants, while issue 43329 is a narrower root-cause/debugging report in the same video-token path.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 44479 is the most representative and actionable issue in the set because it describes the regression scope and affected models clearly.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:39692", + "right": "issue:44912", + "accept": false, + "reason": "Both are model/report quality problems, but they concern different models and different failure modes (docs example errors vs MXFP4 load/quantization fallback)." + }, + { + "left": "issue:43854", + "right": "issue:44977", + "accept": false, + "reason": "Different models and different code paths: one is a load-shape mismatch, the other is a flash-attention generation issue." + }, + { + "left": "issue:43643", + "right": "issue:45759", + "accept": false, + "reason": "Both involve config/model loading, but one is missing fields from trust_remote_code AutoConfig, the other is AutoModelForCausalLM.from_config not unwrapping text_config for composite multimodal configs." + }, + { + "left": "issue:43746", + "right": "issue:44811", + "accept": false, + "reason": "Unrelated bugs: PEFT adapter local checkpoint loading versus Whisper batch_decode skip_special_tokens handling." + }, + { + "left": "issue:43257", + "right": "issue:43696", + "accept": false, + "reason": "One is Qwen3 MoE weight conversion with accelerate/deepspeed; the other is a GPT-oss CUDA OOM. Same broad area, different bugs." + }, + { + "left": "issue:38175", + "right": "issue:44610", + "accept": false, + "reason": "Different models and failures: zero probabilities in SigLIP2 versus a processor/model input-size mismatch in OmDet-Turbo." + }, + { + "left": "issue:44387", + "right": "issue:44617", + "accept": false, + "reason": "Both mention memory pressure, but they are different issues: int4 quantization reserved-memory regression versus Sam3Video OOM." + }, + { + "left": "issue:43262", + "right": "issue:44811", + "accept": false, + "reason": "Audio chat-template sampling-rate default versus Whisper decode token skipping are separate processor behaviors." + }, + { + "left": "issue:43257", + "right": "issue:43931", + "accept": false, + "reason": "Both are Qwen-family loading problems, but the failures differ: MoE weight conversion under accelerate/deepspeed versus Qwen3-VL weight-shape mismatch." + }, + { + "left": "issue:45459", + "right": "issue:45803", + "accept": false, + "reason": "Same general anti-pattern (exception swallowing) but different components and bugs; not the same underlying issue." + }, + { + "left": "issue:43329", + "right": "issue:43450", + "accept": false, + "reason": "Both involve video processors, but one is an undefined-variable bug in multimodal token counting and the other is a batched-shape bug in video processing." + }, + { + "left": "issue:46421", + "right": "issue:46424", + "accept": false, + "reason": "Both are cache-related, but one is lazy-init recompilation with chunked prefill and the other is max_cache_length only being honored for Quantized Cache." + }, + { + "left": "issue:42915", + "right": "issue:43742", + "accept": false, + "reason": "Different models and different failure classes: FineGrainedFP8Config failure versus a MobileLLM loading key error." + }, + { + "left": "issue:43329", + "right": "issue:44479", + "accept": true, + "reason": "These describe the same video-input regression in the Qwen VL family; 43329 names the broken helper/undefined variables in the video token path, and 44479 reports the user-facing regression across affected models." + }, + { + "left": "issue:43749", + "right": "issue:44756", + "accept": false, + "reason": "Both can lead to memory issues, but one is FSDP CPU RAM efficient loading and the other is mmap on Strix Halo; different mechanisms and fixes." + }, + { + "left": "issue:34567", + "right": "issue:40990", + "accept": false, + "reason": "Trainer token accounting not updating is unrelated to GPT-oss perplexity on WikiText." + }, + { + "left": "issue:43650", + "right": "issue:43906", + "accept": false, + "reason": "The titles are not about the same bug; one is a placeholder/placeholder-like report, the other is an isolated reproduction of a separate upstream issue." + }, + { + "left": "issue:44291", + "right": "issue:45042", + "accept": false, + "reason": "Different regressions in different areas: init_empty_weights argument compatibility versus PIL backend requiring torchvision." + }, + { + "left": "issue:42915", + "right": "issue:44821", + "accept": false, + "reason": "Qwen3 MoE FP8 failure and AutoImageProcessor-from-URL loading failure are unrelated." + }, + { + "left": "issue:29159", + "right": "issue:34567", + "accept": false, + "reason": "Tokenizer slow/fast inconsistency is unrelated to TrainerState num_input_tokens_seen not updating." + }, + { + "left": "issue:43606", + "right": "issue:45859", + "accept": false, + "reason": "Both mention CPU offload, but they are different models and different code paths; not the same underlying bug." + }, + { + "left": "issue:43927", + "right": "issue:44912", + "accept": false, + "reason": "DiaConfig save/load token-ID loss causing generation IndexError is unrelated to MXFP4 quantization fallback on gpt-oss-20b." + }, + { + "left": "issue:28282", + "right": "issue:29870", + "accept": false, + "reason": "ImportError about missing PyTorch is unrelated to Bart generate() using decoder_attention_mask before assignment." + }, + { + "left": "issue:43540", + "right": "issue:46701", + "accept": false, + "reason": "Different multimodal pipelines and different failure modes: Qwen3OmniMoe video input ValueError versus DiffusionGemma image-text-to-text pipeline failure." + }, + { + "left": "issue:43525", + "right": "issue:46291", + "accept": false, + "reason": "Missing pad_token_id on Llama4Config is unrelated to the transformers.pipelines import failure with kernels v0.15." + }, + { + "left": "issue:44568", + "right": "issue:45356", + "accept": false, + "reason": "Both are tokenizer regressions, but they affect different models and different mechanisms (special-token insertion vs codec/warning behavior)." + }, + { + "left": "issue:44987", + "right": "issue:45538", + "accept": false, + "reason": "Model-loading failure for physical-intelligence/fast is unrelated to CLIPTokenizer model_max_length behavior." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet: the chosen canonical pair (43329, 44479) is supported by the shared Qwen video-token regression path, and the broader/narrower framing stays conservative. The other soft-edge verdicts mostly distinguish different models or failure modes without overclaiming duplication. No mergeability issues apply here since these are issue pairs, not PR pairs." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 22, + "review_unit_count": 61, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-502", + "nodes": [ + "issue:24643", + "issue:27957", + "issue:28282", + "issue:29127", + "issue:29159", + "issue:33357", + "issue:34567", + "issue:34689", + "issue:41720", + "issue:42175", + "issue:42915", + "issue:43404", + "issue:43425", + "issue:43452", + "issue:43454", + "issue:43540", + "issue:43577", + "issue:43582", + "issue:43697", + "issue:43742", + "issue:43749", + "issue:43824", + "issue:44062", + "issue:44291", + "issue:44479", + "issue:44488", + "issue:44568", + "issue:44792", + "issue:44849", + "issue:44857", + "issue:44936", + "issue:44964", + "issue:44987", + "issue:44991", + "issue:45127", + "issue:45137", + "issue:45357", + "issue:45405", + "issue:45488", + "issue:45538", + "issue:45701", + "issue:45907", + "issue:46291", + "issue:46512", + "issue:46693", + "issue:46821", + "issue:46832", + "issue:46869" + ], + "soft_pairs": [ + "issue:43454|issue:43577", + "issue:45538|issue:46869", + "issue:28282|issue:34689", + "issue:44857|issue:46832", + "issue:43697|issue:45357", + "issue:43425|issue:46291", + "issue:33357|issue:34567", + "issue:42915|issue:45538", + "issue:43404|issue:45127", + "issue:45137|issue:45405", + "issue:42175|issue:43824", + "issue:28282|issue:29127", + "issue:44062|issue:44991", + "issue:44479|issue:46693", + "issue:43742|issue:44291", + "issue:44936|issue:46821", + "issue:42915|issue:43749", + "issue:43577|issue:46512", + "issue:44849|issue:45701", + "issue:43452|issue:43582", + "issue:27957|issue:28282", + "issue:24643|issue:29159", + "issue:44987|issue:45907", + "issue:41720|issue:43540", + "issue:42175|issue:44488", + "issue:44568|issue:45488", + "issue:44792|issue:44964" + ], + "prepared_review_unit_hash": "b8f8fb07eb6f8dd870a6ceb6a8e702e982d5db5ac1329601d3c2202b4f55dcc5", + "component_budget": { + "node_count": 502, + "item_count": 502, + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 27, + "serialized_chars": 35732, + "estimated_input_tokens": 8933, + "estimated_eval_tokens": 18122 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is heterogeneous: the paired issues mostly concern different models, subsystems, and failure modes, so I do not see any confident duplicate matches. If a single representative is needed, issue 45488 is the strongest anchor because it is the most detailed and has inbound references.", + "confidence": 0.87, + "canonical_issue_reason": "No true canonical duplicate stands out; if forced to pick one representative, issue 45488 is the most substantial/open regression report with the best discussion signal.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 45488 is the most actionable and representative issue in this set: detailed repro, clear impact, and inbound references.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43454", + "right": "issue:43577", + "accept": false, + "reason": "Both are model-weight/dtype problems, but they affect different model families and different failure points; not the same bug." + }, + { + "left": "issue:45538", + "right": "issue:46869", + "accept": false, + "reason": "Unrelated failures: tokenizer max-length handling vs multi-GPU device-map crashes for YOLO." + }, + { + "left": "issue:28282", + "right": "issue:34689", + "accept": false, + "reason": "Both are loading errors, but one is a missing PyTorch import path issue and the other is a model-specific loading regression." + }, + { + "left": "issue:44857", + "right": "issue:46832", + "accept": false, + "reason": "Different detection-model crashes with different triggers and code paths; not a shared root cause." + }, + { + "left": "issue:43697", + "right": "issue:45357", + "accept": false, + "reason": "RT-DETR output mismatch and Qwen3.5 save_pretrained key regression are unrelated bugs." + }, + { + "left": "issue:43425", + "right": "issue:46291", + "accept": false, + "reason": "Torch version incompatibility and kernels revision/version import failure are separate dependency issues." + }, + { + "left": "issue:33357", + "right": "issue:34567", + "accept": false, + "reason": "MacOS bus error in CLIP loading vs TrainerState token counter not updating are unrelated." + }, + { + "left": "issue:42915", + "right": "issue:45538", + "accept": false, + "reason": "Fine-grained FP8 config failure and CLIP tokenizer max-length bug are different subsystems." + }, + { + "left": "issue:43404", + "right": "issue:45127", + "accept": false, + "reason": "Both involve tied weights, but they are different models and different scenarios; not the same concrete bug." + }, + { + "left": "issue:45137", + "right": "issue:45405", + "accept": false, + "reason": "DeepSpeed ZeRO3 deque error and PEFT version pin issue are unrelated." + }, + { + "left": "issue:42175", + "right": "issue:43824", + "accept": false, + "reason": "Package/backend installation issue vs missing Qwen2.5-VL export are separate problems." + }, + { + "left": "issue:28282", + "right": "issue:29127", + "accept": false, + "reason": "Generic import error and LayoutLMv3 box-validation message clarity are not the same bug." + }, + { + "left": "issue:44062", + "right": "issue:44991", + "accept": false, + "reason": "The tokenizer TypeError may be related in spirit, but the second issue is model-specific and the evidence is not enough to confirm the same root cause." + }, + { + "left": "issue:44479", + "right": "issue:46693", + "accept": false, + "reason": "Video-input regression and FA2 long-context performance regression are distinct failures." + }, + { + "left": "issue:43742", + "right": "issue:44291", + "accept": false, + "reason": "KeyError loading a model and init_empty_weights argument mismatch are different loading paths." + }, + { + "left": "issue:44936", + "right": "issue:46821", + "accept": false, + "reason": "Trainer lifecycle failure and TimesFM window_size AttributeError do not share a concrete code path." + }, + { + "left": "issue:42915", + "right": "issue:43749", + "accept": false, + "reason": "Qwen3Moe FP8 failure and FSDP CPU RAM efficient loading broken are separate distributed/memory-loading issues." + }, + { + "left": "issue:43577", + "right": "issue:46512", + "accept": false, + "reason": "Both mention dtype, but one is a component dtype bug in BLIP2 and the other is stale config dtype after from_config; not the same bug." + }, + { + "left": "issue:44849", + "right": "issue:45701", + "accept": false, + "reason": "Qwen3.5 hidden-states bug and a broad tokenization regression are unrelated." + }, + { + "left": "issue:43452", + "right": "issue:43582", + "accept": false, + "reason": "GGUF loading breakage and Apple Silicon caching_allocator_warmup TypeError are unrelated." + }, + { + "left": "issue:27957", + "right": "issue:28282", + "accept": false, + "reason": "XLMRoberta/FlashAttention2 and missing-PyTorch import error are different issues." + }, + { + "left": "issue:24643", + "right": "issue:29159", + "accept": false, + "reason": "DeepSpeed training weight-shape error and tokenizer slow/fast inconsistency are unrelated." + }, + { + "left": "issue:44987", + "right": "issue:45907", + "accept": false, + "reason": "Model loading regression and list_repo_files revision forwarding bug are different code paths." + }, + { + "left": "issue:41720", + "right": "issue:43540", + "accept": false, + "reason": "Auto device-map CUDA assert and Qwen3OmniMoe video input ValueError are unrelated." + }, + { + "left": "issue:42175", + "right": "issue:44488", + "accept": false, + "reason": "Backend install dependency issue and model loading failure for a specific checkpoint are not the same bug." + }, + { + "left": "issue:44568", + "right": "issue:45488", + "accept": false, + "reason": "Different tokenizer regressions: one is missing BOS/EOS addition, the other overrides tokenizer.json pre-tokenizer behavior." + }, + { + "left": "issue:44792", + "right": "issue:44964", + "accept": false, + "reason": "Failed image-generation test and Phi-4 multimodal loading regression are different model issues." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 23, + "review_unit_count": 61, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-502", + "nodes": [ + "issue:24643", + "issue:29870", + "issue:34567", + "issue:36331", + "issue:38175", + "issue:43257", + "issue:43452", + "issue:43531", + "issue:43540", + "issue:43577", + "issue:43582", + "issue:43697", + "issue:43716", + "issue:43742", + "issue:43756", + "issue:43906", + "issue:44062", + "issue:44291", + "issue:44479", + "issue:44488", + "issue:44560", + "issue:44568", + "issue:44589", + "issue:44625", + "issue:44737", + "issue:44821", + "issue:44877", + "issue:44898", + "issue:45042", + "issue:45072", + "issue:45081", + "issue:45325", + "issue:45381", + "issue:45405", + "issue:45600", + "issue:45663", + "issue:45672", + "issue:45674", + "issue:45823", + "issue:46399", + "issue:46463", + "issue:46468", + "issue:46491", + "issue:46693", + "issue:46701", + "issue:46821", + "issue:46829", + "issue:46846" + ], + "soft_pairs": [ + "issue:43540|issue:44560", + "issue:44291|issue:44821", + "issue:44479|issue:45381", + "issue:38175|issue:44898", + "issue:34567|issue:36331", + "issue:43257|issue:43756", + "issue:43540|issue:44479", + "issue:44062|issue:45081", + "issue:45042|issue:45081", + "issue:43742|issue:44589", + "issue:43582|issue:46821", + "issue:43756|issue:46399", + "issue:43756|issue:45325", + "issue:44568|issue:46491", + "issue:45405|issue:45600", + "issue:45672|issue:45674", + "issue:43531|issue:44821", + "issue:43716|issue:46701", + "issue:43577|issue:45072", + "issue:43906|issue:46468", + "issue:24643|issue:29870", + "issue:43906|issue:46693", + "issue:44737|issue:46463", + "issue:45823|issue:46463", + "issue:43452|issue:44488", + "issue:44625|issue:44877", + "issue:43697|issue:46829", + "issue:45663|issue:46846" + ], + "prepared_review_unit_hash": "22b256e4157719fa225ef6d047094f83815e1f6a6e8dcff7d9f7625690e17bf1", + "component_budget": { + "node_count": 502, + "item_count": 502, + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 28, + "serialized_chars": 36479, + "estimated_input_tokens": 9120, + "estimated_eval_tokens": 18496 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Most candidates are distinct one-off bugs in different model families or code paths. The only plausible duplicate edge is the Qwen2.5-VL 5.3.0 video-input position-id regression, where one issue is a broad umbrella report and the other is a specific symptom report.", + "confidence": 0.71, + "canonical_issue_reason": "Issue 44479 is the broadest and most representative report in the only strong duplicate family here: it captures the Transformers v5.3.0 Qwen video-input regression across multiple model variants, so it can serve as the umbrella canonical issue.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 44479 is the best global issue candidate because it describes the widest affected surface and the clearest shared root cause among the near-duplicate Qwen video-input reports.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43540", + "right": "issue:44560", + "accept": false, + "reason": "Different Qwen multimodal failure modes: one is Qwen3OmniMoe video-input processing, the other is a Qwen3-vl-embedding StopIteration during video handling." + }, + { + "left": "issue:44291", + "right": "issue:44821", + "accept": false, + "reason": "Unrelated loader bugs: one is init_empty_weights / _is_hf_initialized, the other is AutoImageProcessor loading from URL." + }, + { + "left": "issue:44479", + "right": "issue:45381", + "accept": true, + "reason": "Both point to the same Transformers 5.3.0 Qwen2.5-VL video-input position-id regression; 44479 is the broader umbrella report and 45381 is a specific reproduction." + }, + { + "left": "issue:38175", + "right": "issue:44898", + "accept": false, + "reason": "Different models and symptoms: SigLIP2 zero probabilities versus Perceiver non-default-resolution classification failure." + }, + { + "left": "issue:34567", + "right": "issue:36331", + "accept": false, + "reason": "These are separate Trainer API problems: token accounting not updating versus a custom compute_loss signature mismatch." + }, + { + "left": "issue:43257", + "right": "issue:43756", + "accept": false, + "reason": "Different model-specific bugs: Qwen3 MoE weight conversion with accelerate/deepspeed versus SmolLM3 RoPE-layer dropping." + }, + { + "left": "issue:43540", + "right": "issue:44479", + "accept": false, + "reason": "Both are video-related, but they target different model families and distinct failures, so they are not the same underlying bug." + }, + { + "left": "issue:44062", + "right": "issue:45081", + "accept": false, + "reason": "Tokenizer construction bugs in different places: AddedToken special-flag collision versus Mistral regex patching on a Tokenizer object." + }, + { + "left": "issue:45042", + "right": "issue:45081", + "accept": false, + "reason": "Image processor torchvision dependency regression is unrelated to the Mistral tokenizer backend_tokenizer AttributeError." + }, + { + "left": "issue:43742", + "right": "issue:44589", + "accept": false, + "reason": "Different loading errors: MobileLLM config/key failure versus a Float8 storage lookup error." + }, + { + "left": "issue:43582", + "right": "issue:46821", + "accept": false, + "reason": "Apple Silicon allocator warmup TypeError and TimesFM window_size AttributeError are unrelated." + }, + { + "left": "issue:43756", + "right": "issue:46399", + "accept": false, + "reason": "SmolLM3 RoPE-layer count bug and Qwen3.5 kernel registration crash are different code paths." + }, + { + "left": "issue:43756", + "right": "issue:45325", + "accept": false, + "reason": "These both touch Qwen-ish rotary logic, but one is SmolLM3 RoPE-layer dropping and the other is Qwen2.5-VL image temporal scaling; not the same bug." + }, + { + "left": "issue:44568", + "right": "issue:46491", + "accept": false, + "reason": "Different tokenizer regressions: missing BOS/EOS on add_special_tokens versus leading-space loss on CodeLlama round-trip." + }, + { + "left": "issue:45405", + "right": "issue:45600", + "accept": false, + "reason": "One is a version-pin problem for PEFT, the other is stale auto_mappings references causing CI breakage." + }, + { + "left": "issue:45672", + "right": "issue:45674", + "accept": false, + "reason": "Both involve quantized inference, but one is a Gemma4 uint8/torch.finfo audio-module failure and the other is a BitsAndBytesConfig skip-module exclusion bug." + }, + { + "left": "issue:43531", + "right": "issue:44821", + "accept": false, + "reason": "Qwen3-MoE sliding_window behavior and AutoImageProcessor-from-URL loading are unrelated." + }, + { + "left": "issue:43716", + "right": "issue:46701", + "accept": false, + "reason": "Different multimodal issues: Mistral-3 dtype mismatch in preprocessing versus DiffusionGemma image-text-to-text pipeline breakage." + }, + { + "left": "issue:43577", + "right": "issue:45072", + "accept": false, + "reason": "Both mention dtype, but one is BLIP2 model/qformer dtype loading and the other is inference mismatches in SwitchTransformers/TimmWrapperModel." + }, + { + "left": "issue:43906", + "right": "issue:46468", + "accept": false, + "reason": "An isolated reproduction of another issue is not the same as a broken official Nemotron example." + }, + { + "left": "issue:24643", + "right": "issue:29870", + "accept": false, + "reason": "DeepSpeed training weight-shape RuntimeError and BART generate decoder_attention_mask unbound variable are separate failures." + }, + { + "left": "issue:43906", + "right": "issue:46693", + "accept": false, + "reason": "Reproduction of an older issue is unrelated to the FA2 long-context performance regression." + }, + { + "left": "issue:44737", + "right": "issue:46463", + "accept": false, + "reason": "XLNet CPU-device placement bug and T5 decoder-input validation error are unrelated." + }, + { + "left": "issue:45823", + "right": "issue:46463", + "accept": false, + "reason": "Gemma4 PLE device mismatch during forward is unrelated to the T5 decoder validation message issue." + }, + { + "left": "issue:43452", + "right": "issue:44488", + "accept": false, + "reason": "Tokenizer/model loading breakage around gguf_file is unrelated to the cjvt/sleng-bert load failure." + }, + { + "left": "issue:44625", + "right": "issue:44877", + "accept": false, + "reason": "Qwen3.5 num_labels propagation and strict granite_speech config loading are different config-handling bugs." + }, + { + "left": "issue:43697", + "right": "issue:46829", + "accept": false, + "reason": "RTDetrV2 output changes and VideoPrism hidden_states packaging are separate model-output issues." + }, + { + "left": "issue:45663", + "right": "issue:46846", + "accept": false, + "reason": "Gemma4 FSDP2 shared_kv_states KeyError and Qwen3.5 tensor-parallel plan omissions are different training/distribution bugs." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays appropriately conservative. The only accepted soft edge, issue:44479 with issue:45381, is well supported by the titles as the same Qwen2.5-VL 5.3.0 video-input regression. The remaining rejected pairs are phrased cautiously and do not overclaim duplicate status from the available evidence." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 24, + "review_unit_count": 61, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-502", + "nodes": [ + "issue:28282", + "issue:32090", + "issue:38175", + "issue:41762", + "issue:43295", + "issue:43452", + "issue:43525", + "issue:43526", + "issue:43540", + "issue:43643", + "issue:43653", + "issue:43688", + "issue:43697", + "issue:43720", + "issue:43749", + "issue:43761", + "issue:43819", + "issue:43994", + "issue:44361", + "issue:44442", + "issue:44466", + "issue:44493", + "issue:44737", + "issue:44811", + "issue:44843", + "issue:44898", + "issue:44913", + "issue:44936", + "issue:45005", + "issue:45200", + "issue:45216", + "issue:45357", + "issue:45372", + "issue:45375", + "issue:45479", + "issue:45588", + "issue:45920", + "issue:46050", + "issue:46121", + "issue:46170", + "issue:46463", + "issue:46693", + "issue:46701", + "issue:46829", + "issue:46832", + "issue:46899", + "issue:46932" + ], + "soft_pairs": [ + "issue:45357|issue:45920", + "issue:43540|issue:46693", + "issue:44913|issue:46121", + "issue:44737|issue:45588", + "issue:28282|issue:32090", + "issue:43697|issue:45216", + "issue:46463|issue:46899", + "issue:43526|issue:45479", + "issue:43720|issue:44898", + "issue:43643|issue:45375", + "issue:45200|issue:46463", + "issue:43761|issue:46829", + "issue:44466|issue:45005", + "issue:44361|issue:44442", + "issue:44493|issue:46050", + "issue:43653|issue:44361", + "issue:41762|issue:44936", + "issue:43452|issue:44843", + "issue:43819|issue:44811", + "issue:38175|issue:43295", + "issue:43688|issue:43749", + "issue:43994|issue:44898", + "issue:46832|issue:46932", + "issue:43525|issue:46701", + "issue:45372|issue:46170" + ], + "prepared_review_unit_hash": "0f9cae0ea6da7b42a487965945f9e877910785663b96b8af120231c3748ab1ad", + "component_budget": { + "node_count": 502, + "item_count": 502, + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 25, + "serialized_chars": 35255, + "estimated_input_tokens": 8814, + "estimated_eval_tokens": 17884 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is highly heterogeneous: the soft pairs mostly share broad wording but not the same concrete bug or code path. I did not find any true duplicate/mergeable pairs. I used issue #45920 as the representative issue because it is the broadest, most actionable regression report in the set.", + "confidence": 0.76, + "canonical_issue_reason": "#45920 is the best representative issue: it describes a concrete tokenizer regression with multiple affected model families and a clear failure mode.", + "canonical_pr_reason": null, + "best_issue_reason": "#45920 is the strongest global issue candidate because it is broad, reproducible, and tied to a specific tokenizer path rather than a model-specific edge case.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:45357", + "right": "issue:45920", + "accept": false, + "reason": "Different bugs: visual-encoder key serialization in save_pretrained vs wrong token IDs from AutoTokenizer." + }, + { + "left": "issue:43540", + "right": "issue:46693", + "accept": false, + "reason": "Different subsystems and symptoms: video input processing error vs FA2 long-context performance regression." + }, + { + "left": "issue:44913", + "right": "issue:46121", + "accept": false, + "reason": "Both involve config/rope-related plumbing, but one is a reload default bug and the other is a TypeError on JSON-loaded lists." + }, + { + "left": "issue:44737", + "right": "issue:45588", + "accept": false, + "reason": "Unrelated code paths: XLNet CPU device placement vs flash-attention crashing on s_aux=None." + }, + { + "left": "issue:28282", + "right": "issue:32090", + "accept": false, + "reason": "Missing PyTorch import vs Trainer GPU broadcast type error; no shared underlying bug." + }, + { + "left": "issue:43697", + "right": "issue:45216", + "accept": false, + "reason": "Different model families and failures: RT-DETR output mismatch vs Qwen3.5 save_pretrained checkpoint corruption." + }, + { + "left": "issue:46463", + "right": "issue:46899", + "accept": false, + "reason": "A T5 validation error message bug is unrelated to Gemma4 quantized multimodal dtype casting." + }, + { + "left": "issue:43526", + "right": "issue:45479", + "accept": false, + "reason": "Different logic errors: BEiT label reduction bug vs sequence-classification zero-loss behavior across models." + }, + { + "left": "issue:43720", + "right": "issue:44898", + "accept": false, + "reason": "BitNet packed-weight loading bug is not the same as Perceiver interpolation failure." + }, + { + "left": "issue:43643", + "right": "issue:45375", + "accept": false, + "reason": "AutoConfig trust_remote_code missing-fields issue is distinct from Qwen vision config field dropping under @strict." + }, + { + "left": "issue:45200", + "right": "issue:46463", + "accept": false, + "reason": "Gemma4 token-type defaults and T5 decoder validation are different bugs." + }, + { + "left": "issue:43761", + "right": "issue:46829", + "accept": false, + "reason": "Both mention hidden_states, but the failures differ: missing hidden_states vs misassigned last_hidden_state in another model." + }, + { + "left": "issue:44466", + "right": "issue:45005", + "accept": false, + "reason": "Related theme of tied weights, but not enough evidence of the same concrete serialization bug or mergeable fix." + }, + { + "left": "issue:44361", + "right": "issue:44442", + "accept": false, + "reason": "Different tokenizer classes and different failure modes; not the same underlying issue." + }, + { + "left": "issue:44493", + "right": "issue:46050", + "accept": false, + "reason": "Unexpected position-id keys and sinq quantized save/load breakage are unrelated." + }, + { + "left": "issue:43653", + "right": "issue:44361", + "accept": false, + "reason": "BigBird special-token decode bug is unrelated to MLuke task AttributeError." + }, + { + "left": "issue:41762", + "right": "issue:44936", + "accept": false, + "reason": "ZeRO-3 loading index error is not the same as trainer.evaluate() failing after training." + }, + { + "left": "issue:43452", + "right": "issue:44843", + "accept": false, + "reason": "Both touch AutoTokenizer/from_pretrained, but one is a gguf_file breakage and the other is an offline-mode regex patch bug." + }, + { + "left": "issue:43819", + "right": "issue:44811", + "accept": false, + "reason": "DAC latent reconstruction mismatch is unrelated to Whisper batch_decode skip_special_tokens handling." + }, + { + "left": "issue:38175", + "right": "issue:43295", + "accept": false, + "reason": "Zero-probability outputs for siglip2 are not the same as processor.tokenizer/image-input regression." + }, + { + "left": "issue:43688", + "right": "issue:43749", + "accept": false, + "reason": "Auxiliary-loss normalization and FSDP CPU RAM efficient loading are different problems." + }, + { + "left": "issue:43994", + "right": "issue:44898", + "accept": false, + "reason": "Different models and symptoms: nonsensical SigLIP2 outputs vs Perceiver interpolation failure." + }, + { + "left": "issue:46832", + "right": "issue:46932", + "accept": false, + "reason": "RT-DETR feature-level mismatch is unrelated to Gemma4 quantized embedder dtype casting." + }, + { + "left": "issue:43525", + "right": "issue:46701", + "accept": false, + "reason": "Missing pad_token_id in Llama4Config is a different bug from the DiffusionGemma pipeline failure." + }, + { + "left": "issue:45372", + "right": "issue:46170", + "accept": false, + "reason": "Gemma4 processor import failure and Windows ROCm torch.distributed dependency issues are distinct." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 25, + "review_unit_count": 61, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-502", + "nodes": [ + "issue:29159", + "issue:32090", + "issue:33357", + "issue:33666", + "issue:36010", + "issue:41720", + "issue:42898", + "issue:43295", + "issue:43452", + "issue:43504", + "issue:43531", + "issue:43550", + "issue:43577", + "issue:43646", + "issue:43653", + "issue:43688", + "issue:43720", + "issue:43742", + "issue:43749", + "issue:43756", + "issue:43927", + "issue:44464", + "issue:44466", + "issue:44479", + "issue:44484", + "issue:44568", + "issue:44625", + "issue:44704", + "issue:44734", + "issue:44743", + "issue:44821", + "issue:44843", + "issue:44977", + "issue:44991", + "issue:45356", + "issue:45440", + "issue:45491", + "issue:45542", + "issue:45701", + "issue:45704", + "issue:46002", + "issue:46050", + "issue:46178", + "issue:46402", + "issue:46491", + "issue:46612", + "issue:46693", + "issue:46821" + ], + "soft_pairs": [ + "issue:45542|issue:46178", + "issue:43295|issue:45356", + "issue:43531|issue:43742", + "issue:29159|issue:33357", + "issue:42898|issue:46491", + "issue:43653|issue:44568", + "issue:43646|issue:43749", + "issue:43756|issue:46402", + "issue:44843|issue:44991", + "issue:43653|issue:43927", + "issue:43688|issue:45704", + "issue:43452|issue:43577", + "issue:44464|issue:44743", + "issue:44484|issue:44704", + "issue:43688|issue:45701", + "issue:45491|issue:46002", + "issue:44479|issue:44625", + "issue:44821|issue:46821", + "issue:43688|issue:45440", + "issue:33666|issue:36010", + "issue:32090|issue:33666", + "issue:44466|issue:46050", + "issue:44734|issue:46612", + "issue:43504|issue:43720", + "issue:44977|issue:46693", + "issue:43504|issue:43550", + "issue:44464|issue:44977", + "issue:41720|issue:44464" + ], + "prepared_review_unit_hash": "3ecc479a82f19e883147fa8954d856c80483d265a13ce1645ee385beb948c008", + "component_budget": { + "node_count": 502, + "item_count": 502, + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 28, + "serialized_chars": 36209, + "estimated_input_tokens": 9053, + "estimated_eval_tokens": 18362 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a loose theme of tokenizer/model-loading regressions in Transformers, but the soft pairs are not the same underlying bug or change. None of the candidate pairs are strong enough to merge as duplicates.", + "confidence": 0.89, + "canonical_issue_reason": "issue:45701 is the best umbrella issue in this set: it is broad, version-regression oriented, and sits closest to the repeated tokenizer-behavior complaints without being overly model-specific.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:45701 is the most suitable representative issue because it describes a general version-dependent tokenization regression and can serve as the broadest anchor for related tokenizer reports.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:45542", + "right": "issue:46178", + "accept": false, + "reason": "Both are backend/configuration import errors, but they involve different missing-backend paths: TensorFlow/tensorboard vs tensorflow_text." + }, + { + "left": "issue:43295", + "right": "issue:45356", + "accept": false, + "reason": "Different regressions: processor.tokenizer/image-handling breakage versus Kimi-K2.5 codec handling and mistral-regex warning behavior." + }, + { + "left": "issue:43531", + "right": "issue:43742", + "accept": false, + "reason": "Unrelated failures in different models: Qwen3-MoE sliding-window behavior versus a KeyError loading MobileLLM-125M." + }, + { + "left": "issue:29159", + "right": "issue:33357", + "accept": false, + "reason": "Both mention tokenizers, but one is slow/fast inconsistency and the other is a MacOS bus error with a CLIP model." + }, + { + "left": "issue:42898", + "right": "issue:46491", + "accept": false, + "reason": "Both are tokenization regressions, but one is a general v5 behavior change and the other is a CodeLlama leading-space round-trip bug." + }, + { + "left": "issue:43653", + "right": "issue:44568", + "accept": false, + "reason": "Special-token handling is the broad theme, but these are different tokenizers and different symptoms (mask decode output vs BOS/EOS insertion)." + }, + { + "left": "issue:43646", + "right": "issue:43749", + "accept": false, + "reason": "Both concern loading/runtime breakage, but one is custom model initialization in v5 and the other is FSDP CPU RAM-efficient loading." + }, + { + "left": "issue:43756", + "right": "issue:46402", + "accept": false, + "reason": "Different model-loading problems: Smollm3 RoPE layer loading versus Sam3 text-encoder weights not being loaded." + }, + { + "left": "issue:44843", + "right": "issue:44991", + "accept": false, + "reason": "Both are tokenizer loading issues, but one is offline-model_info patch behavior and the other is a specific tokenizer failure for EMBEDDIA/est-roberta." + }, + { + "left": "issue:43653", + "right": "issue:43927", + "accept": false, + "reason": "Different bugs: BigBird special-token registration versus DiaConfig losing custom token IDs after save/load." + }, + { + "left": "issue:43688", + "right": "issue:45704", + "accept": false, + "reason": "These affect entirely different code paths: auxiliary-loss normalization versus a T5 RMSNorm memory leak." + }, + { + "left": "issue:43452", + "right": "issue:43577", + "accept": false, + "reason": "Loading gguf files and BLIP2 dtype propagation are unrelated issues." + }, + { + "left": "issue:44464", + "right": "issue:44743", + "accept": false, + "reason": "Both touch generation/cache, but one is chunked generation with compiled forward and the other is recurrent-state reset in Qwen3.5 cache handling." + }, + { + "left": "issue:44484", + "right": "issue:44704", + "accept": false, + "reason": "Unrelated APIs: save_pretrained shard sizing versus AutoProcessor kwargs not being passed to cached_file." + }, + { + "left": "issue:43688", + "right": "issue:45701", + "accept": false, + "reason": "Aaux-loss normalization is unrelated to the broad tokenization regression described by issue:45701." + }, + { + "left": "issue:45491", + "right": "issue:46002", + "accept": false, + "reason": "Both mention NaNs, but they arise in different subsystems: Gemma3 sliding-window embeddings versus classification heads during checkout/init." + }, + { + "left": "issue:44479", + "right": "issue:44625", + "accept": false, + "reason": "One is a video-input regression across Qwen VL variants; the other is num_labels propagation in Qwen3.5 config." + }, + { + "left": "issue:44821", + "right": "issue:46821", + "accept": false, + "reason": "Different failures in different components: AutoImageProcessor URL loading versus TimesFM window_size AttributeError." + }, + { + "left": "issue:43688", + "right": "issue:45440", + "accept": false, + "reason": "Unrelated model-specific issues: auxiliary-loss normalization versus DeepseekV3MoE divergence from remote implementation." + }, + { + "left": "issue:33666", + "right": "issue:36010", + "accept": false, + "reason": "Multi-GPU training and a GenerationMixin import error are not the same bug." + }, + { + "left": "issue:32090", + "right": "issue:33666", + "accept": false, + "reason": "Both affect training, but the NoneType GPU broadcast error is distinct from Qwen2-VL multi-GPU training behavior." + }, + { + "left": "issue:44466", + "right": "issue:46050", + "accept": false, + "reason": "Different serialization/load issues: tied-weight serialization depending on device versus quantized sinq save/load breakage." + }, + { + "left": "issue:44734", + "right": "issue:46612", + "accept": false, + "reason": "Both are cache/generation related, but server-side KV continuation indexing is not the same as beam-search cache reorder being skipped." + }, + { + "left": "issue:43504", + "right": "issue:43720", + "accept": false, + "reason": "Legacy-field loading for Beit is unrelated to BitNet packed-weight unpacking during accelerate loading." + }, + { + "left": "issue:44977", + "right": "issue:46693", + "accept": false, + "reason": "One is a correctness regression for Qwen3.5 flash-attention generation; the other is a performance regression with FA2 and long-context generation." + }, + { + "left": "issue:43504", + "right": "issue:43550", + "accept": false, + "reason": "Both are model-loading/runtime problems, but one is a pretrained preset legacy-field issue and the other is a torch.compile + SDPA failure." + }, + { + "left": "issue:44464", + "right": "issue:44977", + "accept": false, + "reason": "Chunked compiled-forward generation inconsistency and Qwen3.5 flash-attention generation failure are different code paths and symptoms." + }, + { + "left": "issue:41720", + "right": "issue:44464", + "accept": false, + "reason": "Auto device mapping cudaErrorAssert on A800 is unrelated to compiled-forward chunked generation output inconsistency." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 26, + "review_unit_count": 61, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-502", + "nodes": [ + "issue:28218", + "issue:33357", + "issue:33666", + "issue:36331", + "issue:39692", + "issue:42796", + "issue:43262", + "issue:43278", + "issue:43295", + "issue:43452", + "issue:43493", + "issue:43504", + "issue:43525", + "issue:43572", + "issue:43577", + "issue:43606", + "issue:43611", + "issue:43643", + "issue:43646", + "issue:43688", + "issue:43720", + "issue:43782", + "issue:43825", + "issue:43844", + "issue:43927", + "issue:43994", + "issue:44361", + "issue:44451", + "issue:44466", + "issue:44514", + "issue:44554", + "issue:44561", + "issue:44610", + "issue:44625", + "issue:44849", + "issue:44964", + "issue:44977", + "issue:45290", + "issue:45405", + "issue:45584", + "issue:45823", + "issue:46050", + "issue:46192", + "issue:46209", + "issue:46365", + "issue:46693", + "issue:46710", + "issue:46846" + ], + "soft_pairs": [ + "issue:43643|issue:44561", + "issue:43577|issue:44554", + "issue:43611|issue:46846", + "issue:43720|issue:43927", + "issue:43504|issue:44610", + "issue:43606|issue:43927", + "issue:33666|issue:36331", + "issue:43278|issue:43688", + "issue:43782|issue:46693", + "issue:43262|issue:43295", + "issue:43720|issue:44361", + "issue:44514|issue:45290", + "issue:44964|issue:45405", + "issue:43493|issue:43825", + "issue:44554|issue:46365", + "issue:44977|issue:46209", + "issue:43525|issue:43572", + "issue:44625|issue:45584", + "issue:43452|issue:44451", + "issue:28218|issue:33357", + "issue:45823|issue:46050", + "issue:42796|issue:44977", + "issue:43994|issue:46192", + "issue:43493|issue:43844", + "issue:39692|issue:43295", + "issue:43278|issue:43844", + "issue:43646|issue:44466", + "issue:46209|issue:46710", + "issue:44849|issue:46209", + "issue:33357|issue:33666" + ], + "prepared_review_unit_hash": "e5fbcef063db724c09eec8711707e885c4de79388e1c950c18114b1e3cf79ab6", + "component_budget": { + "node_count": 502, + "item_count": 502, + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 30, + "serialized_chars": 36825, + "estimated_input_tokens": 9207, + "estimated_eval_tokens": 18670 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a loose collection of unrelated Transformers bug reports, mostly across different models and regressions. The suggested soft edges share only broad subsystem similarity, not the same concrete bug, so none should be merged as duplicates.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43643", + "right": "issue:44561", + "accept": false, + "reason": "Both mention trust_remote_code, but one is missing config fields while the other is a v5 regression from removing is_torch_fx_available; different root causes." + }, + { + "left": "issue:43577", + "right": "issue:44554", + "accept": false, + "reason": "BLIP2 dtype propagation and an MPS attention correctness issue are unrelated bugs." + }, + { + "left": "issue:43611", + "right": "issue:46846", + "accept": false, + "reason": "One is a base_model_prefix loading regression; the other is a Qwen3.5 TP-plan OOM/shape issue." + }, + { + "left": "issue:43720", + "right": "issue:43927", + "accept": false, + "reason": "Packed-weight loading failure and DiaConfig token-ID loss are different failure modes in different code paths." + }, + { + "left": "issue:43504", + "right": "issue:44610", + "accept": false, + "reason": "Legacy preset loading and processor/model image-size mismatch are unrelated." + }, + { + "left": "issue:43606", + "right": "issue:43927", + "accept": false, + "reason": "CPU offload device mismatch is not the same bug as DiaConfig save/load token corruption." + }, + { + "left": "issue:33666", + "right": "issue:36331", + "accept": false, + "reason": "Multi-GPU training and a custom compute_loss signature break are unrelated." + }, + { + "left": "issue:43278", + "right": "issue:43688", + "accept": false, + "reason": "Embedding dtype drift and auxiliary-loss normalization are different bugs." + }, + { + "left": "issue:43782", + "right": "issue:46693", + "accept": false, + "reason": "Qwen3VL weight_only loading and FA2 long-context performance regression are unrelated." + }, + { + "left": "issue:43262", + "right": "issue:43295", + "accept": false, + "reason": "Audio apply_chat_template sampling-rate defaulting and processor.tokenizer/image handling regression are different issues." + }, + { + "left": "issue:43720", + "right": "issue:44361", + "accept": false, + "reason": "BitNet packed-weight loading and MLukeTokenizer attribute failure are unrelated." + }, + { + "left": "issue:44514", + "right": "issue:45290", + "accept": false, + "reason": "Both crash in apply_chat_template, but on different inputs and different edge cases." + }, + { + "left": "issue:44964", + "right": "issue:45405", + "accept": false, + "reason": "Phi-4 multimodal loading and an unreleased PEFT version bump are unrelated." + }, + { + "left": "issue:43493", + "right": "issue:43825", + "accept": false, + "reason": "A SigLIP2 implementation discrepancy is not the same as a pipeline error-message bug." + }, + { + "left": "issue:44554", + "right": "issue:46365", + "accept": false, + "reason": "An attention correctness issue and a Trainer import conflict are unrelated." + }, + { + "left": "issue:44977", + "right": "issue:46209", + "accept": false, + "reason": "Qwen3.5 flash-attention generation and FP8/B200 inference failures are different bugs." + }, + { + "left": "issue:43525", + "right": "issue:43572", + "accept": false, + "reason": "Missing pad_token_id and missing pad_token_idx are similar in theme but on different configs and attributes." + }, + { + "left": "issue:44625", + "right": "issue:45584", + "accept": false, + "reason": "Qwen3.5 num_labels propagation and Whisper empty-transcription failure are unrelated." + }, + { + "left": "issue:43452", + "right": "issue:44451", + "accept": false, + "reason": "gguf_file loading regression and a specific model load failure are not the same concrete bug." + }, + { + "left": "issue:28218", + "right": "issue:33357", + "accept": false, + "reason": "Tokenizer spacing after added tokens is unrelated to the MacOS CLIP bus error." + }, + { + "left": "issue:45823", + "right": "issue:46050", + "accept": false, + "reason": "A device_map auto forward mismatch and a sinq save/load quantization break are different issues." + }, + { + "left": "issue:42796", + "right": "issue:44977", + "accept": false, + "reason": "Ministral BPE marker handling and Qwen3.5 flash-attention generation are unrelated." + }, + { + "left": "issue:43994", + "right": "issue:46192", + "accept": false, + "reason": "SigLIP2 model/pipeline correctness and return_tensors handling are different bugs." + }, + { + "left": "issue:43493", + "right": "issue:43844", + "accept": false, + "reason": "SigLIP2 implementation mismatch and a ZeRO-3 gradient regression are unrelated." + }, + { + "left": "issue:39692", + "right": "issue:43295", + "accept": false, + "reason": "Doc-example mismatches/quantization errors are not the same as processor.tokenizer and image-tokenizer regressions." + }, + { + "left": "issue:43278", + "right": "issue:43844", + "accept": false, + "reason": "Embedding dtype changes and abnormal gradient growth under ZeRO-3 are different problems." + }, + { + "left": "issue:43646", + "right": "issue:44466", + "accept": false, + "reason": "Custom model initialization breakage and lm_head serialization inconsistency are not the same bug." + }, + { + "left": "issue:46209", + "right": "issue:46710", + "accept": false, + "reason": "FP8 inference on B200 and a DeepSeek accuracy regression after upgrade are unrelated." + }, + { + "left": "issue:44849", + "right": "issue:46209", + "accept": false, + "reason": "Qwen3.5 hidden-states bug and FP8 inference failure are different issues." + }, + { + "left": "issue:33357", + "right": "issue:33666", + "accept": false, + "reason": "A MacOS bus error in CLIP loading is unrelated to Qwen2-VL multi-GPU training." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 27, + "review_unit_count": 61, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-502", + "nodes": [ + "issue:29942", + "issue:30333", + "issue:43257", + "issue:43278", + "issue:43299", + "issue:43334", + "issue:43504", + "issue:43572", + "issue:43575", + "issue:43606", + "issue:43653", + "issue:43701", + "issue:43824", + "issue:43844", + "issue:43873", + "issue:43927", + "issue:43937", + "issue:44112", + "issue:44291", + "issue:44361", + "issue:44464", + "issue:44488", + "issue:44493", + "issue:44568", + "issue:44589", + "issue:44792", + "issue:44843", + "issue:44849", + "issue:44964", + "issue:45072", + "issue:45200", + "issue:45229", + "issue:45245", + "issue:45357", + "issue:45405", + "issue:45520", + "issue:45759", + "issue:45812", + "issue:46050", + "issue:46285", + "issue:46399", + "issue:46424", + "issue:46489", + "issue:46701", + "issue:46710", + "issue:46821", + "issue:46832", + "issue:46846" + ], + "soft_pairs": [ + "issue:44568|issue:46489", + "issue:43278|issue:43701", + "issue:44488|issue:44843", + "issue:43927|issue:44361", + "issue:44792|issue:45405", + "issue:44568|issue:45812", + "issue:43824|issue:44589", + "issue:43575|issue:45405", + "issue:44849|issue:45357", + "issue:43504|issue:43606", + "issue:43257|issue:43299", + "issue:44589|issue:45520", + "issue:43606|issue:44112", + "issue:46399|issue:46846", + "issue:43334|issue:43572", + "issue:45245|issue:46710", + "issue:44493|issue:46285", + "issue:46821|issue:46832", + "issue:43334|issue:45759", + "issue:43504|issue:43653", + "issue:43572|issue:44291", + "issue:45072|issue:46832", + "issue:44464|issue:46424", + "issue:43937|issue:46701", + "issue:43844|issue:44964", + "issue:43873|issue:46050", + "issue:45200|issue:45229", + "issue:29942|issue:30333" + ], + "prepared_review_unit_hash": "f687c56af1acc1542a575618dc67b24542b573422b5d7e8568ca54559d898bab", + "component_budget": { + "node_count": 502, + "item_count": 502, + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 28, + "serialized_chars": 36202, + "estimated_input_tokens": 9051, + "estimated_eval_tokens": 18358 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is highly heterogeneous: tokenizer regressions, model-loading failures, training/offload issues, generation bugs, and unrelated CI/test failures. No pair looks like the same concrete bug/change, so all soft edges should be rejected.", + "confidence": 0.97, + "canonical_issue_reason": "No single canonical issue fits this cluster because the items cover unrelated failures across many subsystems and models.", + "canonical_pr_reason": null, + "best_issue_reason": "No global best issue: none of the issues is representative of the others, and several are clearly unrelated.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44568", + "right": "issue:46489", + "accept": false, + "reason": "Both are tokenizer regressions, but they affect different models and symptoms; not the same underlying bug." + }, + { + "left": "issue:43278", + "right": "issue:43701", + "accept": false, + "reason": "One is a dtype drift in evaluate, the other a checkpoint key mismatch; unrelated failures." + }, + { + "left": "issue:44488", + "right": "issue:44843", + "accept": false, + "reason": "Different subsystems: model load failure for a specific tokenizer/model vs offline-mode hub access bug." + }, + { + "left": "issue:43927", + "right": "issue:44361", + "accept": false, + "reason": "Both involve tokenizers/configs, but the affected models and failure modes are different." + }, + { + "left": "issue:44792", + "right": "issue:45405", + "accept": false, + "reason": "Image generation test failure vs dependency version bump; no shared bug." + }, + { + "left": "issue:44568", + "right": "issue:45812", + "accept": false, + "reason": "Both are tokenizer-related regressions, but they are distinct model-specific issues with different outputs." + }, + { + "left": "issue:43824", + "right": "issue:44589", + "accept": false, + "reason": "Completely unrelated: import error for a Qwen class vs float8 storage lookup failure." + }, + { + "left": "issue:43575", + "right": "issue:45405", + "accept": false, + "reason": "OOM during tensor-parallel model load is unrelated to the PEFT minimum-version bump." + }, + { + "left": "issue:44849", + "right": "issue:45357", + "accept": false, + "reason": "Different bugs in different Qwen models: hidden-states output handling vs save_pretrained key serialization." + }, + { + "left": "issue:43504", + "right": "issue:43606", + "accept": false, + "reason": "Legacy-field load bug for BEiT is unrelated to CPU offload device mismatch in bark-small." + }, + { + "left": "issue:43257", + "right": "issue:43299", + "accept": false, + "reason": "Both mention Qwen MOE loading, but one is accelerate/deepspeed conversion and the other is a general transformers-dev load break; not clearly the same fix." + }, + { + "left": "issue:44589", + "right": "issue:45520", + "accept": false, + "reason": "Distinct low-level import/storage problems with no shared code path." + }, + { + "left": "issue:43606", + "right": "issue:44112", + "accept": false, + "reason": "One is a real device mismatch bug, the other is a stale CI test; not the same underlying issue." + }, + { + "left": "issue:46399", + "right": "issue:46846", + "accept": false, + "reason": "Both are Qwen3.5-related, but one is a kernel registration error and the other is a TP plan omission causing OOM/shape issues." + }, + { + "left": "issue:43334", + "right": "issue:43572", + "accept": false, + "reason": "Both are post-v5 config-field problems, but they concern different models and different missing fields." + }, + { + "left": "issue:45245", + "right": "issue:46710", + "accept": false, + "reason": "Category-count limit error is unrelated to an accuracy regression after upgrade." + }, + { + "left": "issue:44493", + "right": "issue:46285", + "accept": false, + "reason": "Both are structure/keys regressions, but they affect different model families and different attributes." + }, + { + "left": "issue:46821", + "right": "issue:46832", + "accept": false, + "reason": "Unrelated model errors: TimesFM window_size attribute issue vs RT-DETR forward TypeError." + }, + { + "left": "issue:43334", + "right": "issue:45759", + "accept": false, + "reason": "Both touch multimodal Qwen configs, but one is a missing pad_token_id and the other is text_config unwrapping; not the same bug." + }, + { + "left": "issue:43504", + "right": "issue:43653", + "accept": false, + "reason": "Different components and symptoms: BEiT pretrained loading vs BigBirdTokenizer special-token registration." + }, + { + "left": "issue:43572", + "right": "issue:44291", + "accept": false, + "reason": "Missing pad_token_idx and unexpected _is_hf_initialized arg are separate config/init compatibility issues." + }, + { + "left": "issue:45072", + "right": "issue:46832", + "accept": false, + "reason": "Different inference bugs in different models; no shared code path." + }, + { + "left": "issue:44464", + "right": "issue:46424", + "accept": false, + "reason": "Both are generation/cache-related, but one is chunked generation with compiled forward and the other is cache_length handling for quantized cache only." + }, + { + "left": "issue:43937", + "right": "issue:46701", + "accept": false, + "reason": "Different generation/pipeline failures with no evidence of a common cause." + }, + { + "left": "issue:43844", + "right": "issue:44964", + "accept": false, + "reason": "ZeRO-3 gradient behavior and multimodal model loading are unrelated." + }, + { + "left": "issue:43873", + "right": "issue:46050", + "accept": false, + "reason": "Quantization/offloading themes overlap, but save/load breakage in sinq is not the same bug as offloading not working." + }, + { + "left": "issue:45200", + "right": "issue:45229", + "accept": false, + "reason": "Text-only fine-tuning defaults vs multi-GPU inference OOM are unrelated." + }, + { + "left": "issue:29942", + "right": "issue:30333", + "accept": false, + "reason": "Flash Attention test failures and MLflow job-reporting issues are unrelated." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 28, + "review_unit_count": 61, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-502", + "nodes": [ + "issue:15354", + "issue:28218", + "issue:29159", + "issue:32090", + "issue:34567", + "issue:37428", + "issue:38175", + "issue:41720", + "issue:42490", + "issue:43065", + "issue:43295", + "issue:43450", + "issue:43493", + "issue:43540", + "issue:43572", + "issue:43696", + "issue:43701", + "issue:43873", + "issue:43957", + "issue:43994", + "issue:44186", + "issue:44423", + "issue:44493", + "issue:44554", + "issue:44610", + "issue:44734", + "issue:44869", + "issue:44977", + "issue:45072", + "issue:45084", + "issue:45198", + "issue:45200", + "issue:45229", + "issue:45406", + "issue:45446", + "issue:45520", + "issue:45588", + "issue:45759", + "issue:45823", + "issue:46050", + "issue:46121", + "issue:46192", + "issue:46369", + "issue:46701", + "issue:46710", + "issue:46752", + "issue:46821" + ], + "soft_pairs": [ + "issue:44869|issue:46752", + "issue:43994|issue:44493", + "issue:45229|issue:45823", + "issue:45588|issue:46050", + "issue:42490|issue:43696", + "issue:29159|issue:32090", + "issue:44977|issue:46701", + "issue:45084|issue:45520", + "issue:28218|issue:34567", + "issue:43065|issue:46369", + "issue:44554|issue:45446", + "issue:15354|issue:38175", + "issue:44423|issue:44734", + "issue:43493|issue:43701", + "issue:41720|issue:46710", + "issue:45406|issue:46121", + "issue:44977|issue:46710", + "issue:45072|issue:45198", + "issue:43295|issue:43450", + "issue:37428|issue:43957", + "issue:45588|issue:45759", + "issue:43540|issue:45200", + "issue:43295|issue:43873", + "issue:44186|issue:46192", + "issue:43572|issue:46821", + "issue:43295|issue:44610", + "issue:45823|issue:46192" + ], + "prepared_review_unit_hash": "7d300313cced9717d3ef40dfbce810226e0ca378179ab48aa0bb545f01be7985", + "component_budget": { + "node_count": 502, + "item_count": 502, + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 27, + "serialized_chars": 35267, + "estimated_input_tokens": 8817, + "estimated_eval_tokens": 17890 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is heterogeneous: the soft-linked issues share broad library areas but not the same concrete bug, model, or failing code path. I would not merge any of the suggested pairs as duplicates.", + "confidence": 0.94, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": "Issue 43295 is the strongest standalone anchor: it describes a clear, user-facing regression with a concrete breakage pattern and broad impact on custom code using processor/tokenizer access.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44869", + "right": "issue:46752", + "accept": false, + "reason": "Different failure modes in different subsystems: Whisper word-timestamp decoding vs chat template handling on empty input." + }, + { + "left": "issue:43994", + "right": "issue:44493", + "accept": false, + "reason": "One is a SigLIP2 inference/output quality problem; the other is a position-id/key regression affecting many models. No shared code-path bug." + }, + { + "left": "issue:45229", + "right": "issue:45823", + "accept": false, + "reason": "Both are Gemma4-related but the bugs differ: multi-GPU CUDA OOM versus a device mismatch during forward with auto device mapping." + }, + { + "left": "issue:45588", + "right": "issue:46050", + "accept": false, + "reason": "Unrelated: flash-attention sinkless-model attribute error versus quantized save/load corruption after sinq. Different code paths and symptoms." + }, + { + "left": "issue:42490", + "right": "issue:43696", + "accept": false, + "reason": "Path parsing/import warning is unrelated to a CUDA out-of-memory report for GPT-oss-20b." + }, + { + "left": "issue:29159", + "right": "issue:32090", + "accept": false, + "reason": "Tokenizer slow/fast inconsistency is unrelated to a Trainer distributed broadcast TypeError." + }, + { + "left": "issue:44977", + "right": "issue:46701", + "accept": false, + "reason": "Different models and bugs: flash-attention generation issue versus an image-text-to-text pipeline failure for DiffusionGemmaGenerationOutput." + }, + { + "left": "issue:45084", + "right": "issue:45520", + "accept": false, + "reason": "A template-node compilation TypeError is not the same as the Python 3.13 flash_attn import KeyError." + }, + { + "left": "issue:28218", + "right": "issue:34567", + "accept": false, + "reason": "Tokenizer spacing behavior and TrainerState token-count tracking are unrelated." + }, + { + "left": "issue:43065", + "right": "issue:46369", + "accept": false, + "reason": "Both mention SAM3, but one is a dummy Conv2d implementation detail and the other is a memory overflow in sam3-litetext. Not the same bug." + }, + { + "left": "issue:44554", + "right": "issue:45446", + "accept": false, + "reason": "MPS attention correctness with mismatched head dims is unrelated to a flex_attention version-check import bug." + }, + { + "left": "issue:15354", + "right": "issue:38175", + "accept": false, + "reason": "TorchScript GeneratorExp export failure is unrelated to SigLIP2 zero-probability inference behavior." + }, + { + "left": "issue:44423", + "right": "issue:44734", + "accept": false, + "reason": "Continuous-batching multimodal crash from string tensor handling is different from KV-cache continuation tensor indexing." + }, + { + "left": "issue:43493", + "right": "issue:43701", + "accept": false, + "reason": "SigLIP2 implementation discrepancy is unrelated to resume_from_checkpoint key mismatch." + }, + { + "left": "issue:41720", + "right": "issue:46710", + "accept": false, + "reason": "CUDA assert on auto device mapping for Qwen3 is not the same as an accuracy regression after a Transformers upgrade." + }, + { + "left": "issue:45406", + "right": "issue:46121", + "accept": false, + "reason": "Gemma4Processor _tokenizer attribute crash is unrelated to rope-parameter list type handling." + }, + { + "left": "issue:44977", + "right": "issue:46710", + "accept": false, + "reason": "Flash-attention generation failure for Qwen3.5 is a different issue from a model accuracy regression after upgrade." + }, + { + "left": "issue:45072", + "right": "issue:45198", + "accept": false, + "reason": "dtype mismatch in CI inference and Wav2Vec2 save_pretrained/tokenization failure are unrelated." + }, + { + "left": "issue:43295", + "right": "issue:43450", + "accept": false, + "reason": "Both involve processors, but one is a regression in processor.tokenizer access and image passing; the other is batched video shape handling. Different bugs." + }, + { + "left": "issue:37428", + "right": "issue:43957", + "accept": false, + "reason": "Flash-attention utility import failure and meta-device model loading breakage are separate loading/runtime issues." + }, + { + "left": "issue:45588", + "right": "issue:45759", + "accept": false, + "reason": "AttributeError on s_aux=None in flash_attention integration is unrelated to AutoModelForCausalLM.from_config not unwrapping text_config." + }, + { + "left": "issue:43540", + "right": "issue:45200", + "accept": false, + "reason": "Video input validation in Qwen3OmniMoe is unrelated to Gemma4 mm_token_type_ids defaults for text-only fine-tuning." + }, + { + "left": "issue:43295", + "right": "issue:43873", + "accept": false, + "reason": "Processor/tokenizer regression and offloading with quantization are different failure modes and code paths." + }, + { + "left": "issue:44186", + "right": "issue:46192", + "accept": false, + "reason": "LayoutLMv2Tokenizer NER/padding crash is unrelated to return_tensors being ignored when text_kwargs is passed." + }, + { + "left": "issue:43572", + "right": "issue:46821", + "accept": false, + "reason": "Missing StableLmConfig pad_token_idx after v5 update is unrelated to TimesFM window_size AttributeError." + }, + { + "left": "issue:43295", + "right": "issue:44610", + "accept": false, + "reason": "Both are processor-related, but one is a tokenizer-access regression and the other is an OmDet-Turbo image size mismatch. Not the same underlying bug." + }, + { + "left": "issue:45823", + "right": "issue:46192", + "accept": false, + "reason": "Gemma4 device mismatch with auto device mapping is unrelated to return_tensors being ignored in multimodal kwargs handling." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 29, + "review_unit_count": 61, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-502", + "nodes": [ + "issue:28218", + "issue:29870", + "issue:29942", + "issue:35141", + "issue:37428", + "issue:39692", + "issue:42490", + "issue:42915", + "issue:43232", + "issue:43262", + "issue:43278", + "issue:43295", + "issue:43334", + "issue:43450", + "issue:43452", + "issue:43493", + "issue:43525", + "issue:43550", + "issue:43572", + "issue:43575", + "issue:43653", + "issue:43824", + "issue:43844", + "issue:43856", + "issue:43950", + "issue:43975", + "issue:44112", + "issue:44442", + "issue:44466", + "issue:44521", + "issue:44554", + "issue:44737", + "issue:44849", + "issue:44987", + "issue:45127", + "issue:45198", + "issue:45200", + "issue:45290", + "issue:45356", + "issue:45381", + "issue:45520", + "issue:45823", + "issue:46178", + "issue:46192", + "issue:46209", + "issue:46285", + "issue:46512", + "issue:46894" + ], + "soft_pairs": [ + "issue:29870|issue:29942", + "issue:43550|issue:43844", + "issue:45520|issue:46178", + "issue:45200|issue:45823", + "issue:43450|issue:43575", + "issue:43572|issue:44987", + "issue:44554|issue:45356", + "issue:37428|issue:43824", + "issue:42490|issue:43493", + "issue:43232|issue:44849", + "issue:43295|issue:43493", + "issue:44466|issue:45127", + "issue:39692|issue:46192", + "issue:44521|issue:45290", + "issue:43653|issue:44112", + "issue:45381|issue:46209", + "issue:28218|issue:35141", + "issue:43278|issue:43856", + "issue:44737|issue:46894", + "issue:42915|issue:43824", + "issue:43950|issue:46285", + "issue:43824|issue:43975", + "issue:43452|issue:46512", + "issue:43334|issue:43525", + "issue:43262|issue:43450", + "issue:44442|issue:45198" + ], + "prepared_review_unit_hash": "b695e4e7af99b7139fb30aad8b05060c40f5a6edcff52fa5dd78a998a0a45fa4", + "component_budget": { + "node_count": 502, + "item_count": 502, + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 26, + "serialized_chars": 35883, + "estimated_input_tokens": 8971, + "estimated_eval_tokens": 18198 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Most pairs are only superficially related by subsystem or keyword and do not look like the same underlying bug. I would keep issue 43950 as the strongest representative issue: it describes a clear, core regression with a direct reproduction path and broad impact.", + "confidence": 0.71, + "canonical_issue_reason": "Issue 43950 is the most concrete, regression-framed core bug in the set, with a clear failure mode and strong user impact.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 43950 is the best overall issue to keep as the representative triage target because it is specific, actionable, and clearly describes a real regression.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:29870", + "right": "issue:29942", + "accept": false, + "reason": "Different failures: generation-time decoder_attention_mask bug vs Flash Attention 2 test failures. Same broad area, not the same bug." + }, + { + "left": "issue:43550", + "right": "issue:43844", + "accept": false, + "reason": "One is a torch.compile/SDPA model failure, the other is a DeepSpeed ZeRO-3 gradient anomaly. Distinct code paths and symptoms." + }, + { + "left": "issue:45520", + "right": "issue:46178", + "accept": false, + "reason": "Import error for flash_attn on Python 3.13 is unrelated to a backend mapping ValueError for tensorflow_text." + }, + { + "left": "issue:45200", + "right": "issue:45823", + "accept": false, + "reason": "Gemma 4 token type ID defaulting bug and PLE device mismatch are different issues in different parts of the stack." + }, + { + "left": "issue:43450", + "right": "issue:43575", + "accept": false, + "reason": "Video processor batch-shape bug vs tensor-parallel OOM while loading Qwen2-57B. Not the same underlying problem." + }, + { + "left": "issue:43572", + "right": "issue:44987", + "accept": false, + "reason": "Both are load-time regressions, but the titles point to different models and different root causes." + }, + { + "left": "issue:44554", + "right": "issue:45356", + "accept": false, + "reason": "MPS attention correctness issue and Kimi tokenizer regression are unrelated." + }, + { + "left": "issue:37428", + "right": "issue:43824", + "accept": false, + "reason": "Separate import errors for different symbols/models; no shared code-path bug is evident." + }, + { + "left": "issue:42490", + "right": "issue:43493", + "accept": false, + "reason": "Path-with-slash import warning is unrelated to SigLIP2 implementation discrepancy." + }, + { + "left": "issue:43232", + "right": "issue:44849", + "accept": false, + "reason": "Generation kwargs sync issue and Qwen3.5 hidden-states bug are distinct generation behaviors." + }, + { + "left": "issue:43295", + "right": "issue:43493", + "accept": false, + "reason": "Processor/tokenizer regression is not the same as SigLIP2 JAX-vs-HF discrepancy." + }, + { + "left": "issue:44466", + "right": "issue:45127", + "accept": false, + "reason": "Both involve weights and vocabulary/tie behavior, but one is serialization inconsistency and the other is LoRA merge collapse. Different bugs." + }, + { + "left": "issue:39692", + "right": "issue:46192", + "accept": false, + "reason": "Documentation example errors and return_tensors being ignored are unrelated." + }, + { + "left": "issue:44521", + "right": "issue:45290", + "accept": false, + "reason": "Both touch apply_chat_template, but one is incorrect assistant masks for multimodal inputs and the other is a crash on tool-call messages without content." + }, + { + "left": "issue:43653", + "right": "issue:44112", + "accept": false, + "reason": "Tokenizer special-token registration bug vs CI failure in a stale device override test. Not a duplicate." + }, + { + "left": "issue:45381", + "right": "issue:46209", + "accept": false, + "reason": "Wrong vision_position_ids for qwen2.5-vl video input is unrelated to FP8 inference failure on B200." + }, + { + "left": "issue:28218", + "right": "issue:35141", + "accept": false, + "reason": "Tokenizer spacing after added tokens and embedding reinitialization after resize are separate behaviors." + }, + { + "left": "issue:43278", + "right": "issue:43856", + "accept": false, + "reason": "Dtype drift during evaluate and inefficient Qwen3 MoE training memory use are different problems." + }, + { + "left": "issue:44737", + "right": "issue:46894", + "accept": false, + "reason": "CPU-device placement bug in XLNet positional encoding is unrelated to a training-loss double-shift bug." + }, + { + "left": "issue:42915", + "right": "issue:43824", + "accept": false, + "reason": "FineGrainedFP8Config failure in Qwen3MoE and a Qwen2_5_VL import error do not share the same root cause." + }, + { + "left": "issue:43950", + "right": "issue:46285", + "accept": false, + "reason": "Non-persistent buffer corruption on from_pretrained is unrelated to CLIPTextModel structure inconsistency." + }, + { + "left": "issue:43824", + "right": "issue:43975", + "accept": false, + "reason": "Importing a missing Qwen2_5_VL class and incorrect detokenization in DeepSeek-Coder are unrelated." + }, + { + "left": "issue:43452", + "right": "issue:46512", + "accept": false, + "reason": "Tokenizer/model loading failure with gguf_file is not the same as stale config dtype after from_config(dtype=...)." + }, + { + "left": "issue:43334", + "right": "issue:43525", + "accept": false, + "reason": "Both mention missing pad_token_id, but they are model-specific load failures for different configs; not enough evidence of one underlying bug." + }, + { + "left": "issue:43262", + "right": "issue:43450", + "accept": false, + "reason": "Audio chat-template sample-rate defaulting and video batch-shape handling are different processor bugs." + }, + { + "left": "issue:44442", + "right": "issue:45198", + "accept": false, + "reason": "FastSpeech2ConformerTokenizer loading and Wav2Vec2 save/tokenization failure are different tokenizer/model issues." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 30, + "review_unit_count": 61, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-502", + "nodes": [ + "issue:28282", + "issue:29942", + "issue:32090", + "issue:33357", + "issue:37428", + "issue:38175", + "issue:41720", + "issue:43335", + "issue:43388", + "issue:43425", + "issue:43526", + "issue:43550", + "issue:43606", + "issue:43611", + "issue:43653", + "issue:43819", + "issue:43827", + "issue:43901", + "issue:43931", + "issue:44112", + "issue:44206", + "issue:44448", + "issue:44561", + "issue:44610", + "issue:44743", + "issue:44792", + "issue:44857", + "issue:44898", + "issue:45083", + "issue:45200", + "issue:45325", + "issue:45412", + "issue:45520", + "issue:45563", + "issue:45584", + "issue:45593", + "issue:45759", + "issue:46170", + "issue:46192", + "issue:46209", + "issue:46234", + "issue:46285", + "issue:46424", + "issue:46512", + "issue:46821", + "issue:46858", + "issue:46899", + "issue:46932" + ], + "soft_pairs": [ + "issue:45200|issue:46932", + "issue:44792|issue:45563", + "issue:45200|issue:46899", + "issue:43819|issue:44112", + "issue:44898|issue:45412", + "issue:43526|issue:46192", + "issue:45200|issue:45325", + "issue:43827|issue:43901", + "issue:41720|issue:46209", + "issue:43335|issue:43550", + "issue:44561|issue:46234", + "issue:43388|issue:43526", + "issue:44857|issue:45593", + "issue:43819|issue:45083", + "issue:44610|issue:44857", + "issue:46424|issue:46858", + "issue:45759|issue:46285", + "issue:45520|issue:46170", + "issue:43606|issue:45083", + "issue:46512|issue:46821", + "issue:44206|issue:45584", + "issue:38175|issue:45412", + "issue:43335|issue:43653", + "issue:37428|issue:43425", + "issue:43550|issue:44112", + "issue:29942|issue:32090", + "issue:28282|issue:33357", + "issue:43611|issue:43931", + "issue:44448|issue:44743" + ], + "prepared_review_unit_hash": "e961155c6b2d15856b65cfd48bda0e33471af7ec9298881509a26f8ef9cd7b76", + "component_budget": { + "node_count": 502, + "item_count": 502, + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 29, + "serialized_chars": 36683, + "estimated_input_tokens": 9171, + "estimated_eval_tokens": 18598 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a grab-bag of unrelated issues, with only a small Gemma4 quantization-related subgroup. None of the soft pairs look like the same underlying bug/change, so I reject all soft edges.", + "confidence": 0.29, + "canonical_issue_reason": "issue:46899 is the clearest, most concrete bug report in the only closely related subgroup (Gemma4 multimodal quantization). It describes an actionable failing code path more directly than the neighboring Gemma4 items.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:46899 is the best representative issue here because it is specific, reproducible, and tightly scoped; the other items are either different bugs or broader/less directly actionable.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:45200", + "right": "issue:46932", + "accept": false, + "reason": "Different bugs: mm_token_type_ids defaults for text-only fine-tuning vs quantized vision/audio embedder dtype casting." + }, + { + "left": "issue:44792", + "right": "issue:45563", + "accept": false, + "reason": "Different failures: janus image generation test vs stale num_return_sequences warning in paged generate()." + }, + { + "left": "issue:45200", + "right": "issue:46899", + "accept": false, + "reason": "Different bugs in the same model family, but one is about token type ids and the other about 4-bit multimodal embedder casting." + }, + { + "left": "issue:43819", + "right": "issue:44112", + "accept": false, + "reason": "Different code paths: DAC latent forward mismatch vs GraniteSpeech stale device-override CI test." + }, + { + "left": "issue:44898", + "right": "issue:45412", + "accept": false, + "reason": "Different subsystems: Perceiver resolution handling vs RT-DETR memory release/GC." + }, + { + "left": "issue:43526", + "right": "issue:46192", + "accept": false, + "reason": "Different bugs: BeitImageProcessorFast label reduction vs return_tensors being ignored with explicit text_kwargs." + }, + { + "left": "issue:45200", + "right": "issue:45325", + "accept": false, + "reason": "Different multimodal bugs: Gemma4 token-type defaults vs Qwen2.5-VL rope index scaling." + }, + { + "left": "issue:43827", + "right": "issue:43901", + "accept": false, + "reason": "Both are docs issues, but they concern different pipeline behaviors and are not the same change." + }, + { + "left": "issue:41720", + "right": "issue:46209", + "accept": false, + "reason": "Different hardware/model failures: Qwen3 auto device mapping cudaErrorAssert vs FP8 inference failure on B200." + }, + { + "left": "issue:43335", + "right": "issue:43550", + "accept": false, + "reason": "Different problems in different models: SwitchTransformers sparse layer construction vs Bamba torch.compile SDPA failure." + }, + { + "left": "issue:44561", + "right": "issue:46234", + "accept": false, + "reason": "Different missing-helper/import problems; same utility module does not make them the same bug." + }, + { + "left": "issue:43388", + "right": "issue:43526", + "accept": false, + "reason": "Different label-handling bugs in different components and failure modes." + }, + { + "left": "issue:44857", + "right": "issue:45593", + "accept": false, + "reason": "Different object-detection issues: AMP/CUDA crash in LwDetrImageLoss vs missing auxiliary losses in D-FINE." + }, + { + "left": "issue:43819", + "right": "issue:45083", + "accept": false, + "reason": "Different models and failure modes: DAC latent mismatch vs qwen3_omni_moe length helper behavior." + }, + { + "left": "issue:44610", + "right": "issue:44857", + "accept": false, + "reason": "Different bugs: OmDet-Turbo preprocessing size mismatch vs LwDetrImageLoss AMP crash." + }, + { + "left": "issue:46424", + "right": "issue:46858", + "accept": false, + "reason": "Both touch cache internals, but one is max_cache_length semantics and the other is a device-tensor seq_length arithmetic bug." + }, + { + "left": "issue:45759", + "right": "issue:46285", + "accept": false, + "reason": "Different model-loading regressions: composite Qwen config unwrapping vs CLIPTextModel structure change." + }, + { + "left": "issue:45520", + "right": "issue:46170", + "accept": false, + "reason": "Different import/platform issues: flash_attn lookup on Python 3.13 vs Windows/ROCm torch.distributed dependency problem." + }, + { + "left": "issue:43606", + "right": "issue:45083", + "accept": false, + "reason": "Different bugs: CPU offload device mismatch in Bark vs qwen3_omni_moe feature-length helper behavior." + }, + { + "left": "issue:46512", + "right": "issue:46821", + "accept": false, + "reason": "Different issues: stale config dtype after from_config vs TimesFM window_size AttributeError." + }, + { + "left": "issue:44206", + "right": "issue:45584", + "accept": false, + "reason": "Different regressions in unrelated components: feature extractor center arg vs Whisper empty transcription handling." + }, + { + "left": "issue:38175", + "right": "issue:45412", + "accept": false, + "reason": "Different issues: unexpected zero probabilities in SigLIP2 vs RT-DETR memory not released." + }, + { + "left": "issue:43335", + "right": "issue:43653", + "accept": false, + "reason": "Different bugs: SwitchTransformers sparse-layer creation vs BigBirdTokenizer special-token registration." + }, + { + "left": "issue:37428", + "right": "issue:43425", + "accept": false, + "reason": "One is a missing flash-attention helper import; the other is broad Torch 2.10 incompatibility." + }, + { + "left": "issue:43550", + "right": "issue:44112", + "accept": false, + "reason": "Different code paths and symptoms: Bamba torch.compile/SDPA failure vs GraniteSpeech CI stale device override test." + }, + { + "left": "issue:29942", + "right": "issue:32090", + "accept": false, + "reason": "Different failures: Flash Attention 2 test breakage vs Trainer broadcast TypeError on NoneType." + }, + { + "left": "issue:28282", + "right": "issue:33357", + "accept": false, + "reason": "Different environments and bugs: AutoModel/PyTorch missing import vs MacOS bus error on community CLIP model." + }, + { + "left": "issue:43611", + "right": "issue:43931", + "accept": false, + "reason": "Different model-loading regressions: base_model_prefix breakage vs Qwen3-VL weight-shape mismatch." + }, + { + "left": "issue:44448", + "right": "issue:44743", + "accept": false, + "reason": "Different bugs: output drift across v4/v5 for Pegasus vs recurrent state reset in qwen3_5 cache handling." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 31, + "review_unit_count": 61, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-502", + "nodes": [ + "issue:26413", + "issue:28218", + "issue:29127", + "issue:31501", + "issue:32090", + "issue:33357", + "issue:34689", + "issue:35141", + "issue:37428", + "issue:39692", + "issue:42490", + "issue:43278", + "issue:43335", + "issue:43504", + "issue:43526", + "issue:43606", + "issue:43701", + "issue:43720", + "issue:43819", + "issue:43825", + "issue:43854", + "issue:44112", + "issue:44206", + "issue:44351", + "issue:44448", + "issue:44493", + "issue:44514", + "issue:44521", + "issue:44617", + "issue:44779", + "issue:44855", + "issue:44857", + "issue:45083", + "issue:45127", + "issue:45245", + "issue:45290", + "issue:45310", + "issue:45459", + "issue:45478", + "issue:45588", + "issue:45593", + "issue:45823", + "issue:45859", + "issue:46234", + "issue:46326", + "issue:46431", + "issue:46566", + "issue:46697" + ], + "soft_pairs": [ + "issue:43335|issue:43504", + "issue:44855|issue:45083", + "issue:43278|issue:43825", + "issue:29127|issue:32090", + "issue:43720|issue:44112", + "issue:43335|issue:45083", + "issue:37428|issue:44351", + "issue:43335|issue:45593", + "issue:43335|issue:44112", + "issue:44514|issue:44521", + "issue:45290|issue:46326", + "issue:34689|issue:35141", + "issue:45588|issue:45859", + "issue:42490|issue:43854", + "issue:46431|issue:46566", + "issue:45310|issue:45478", + "issue:45127|issue:45245", + "issue:44448|issue:44779", + "issue:43819|issue:44493", + "issue:39692|issue:43526", + "issue:43606|issue:44857", + "issue:26413|issue:31501", + "issue:28218|issue:32090", + "issue:45459|issue:46234", + "issue:44617|issue:46697", + "issue:43701|issue:45083", + "issue:39692|issue:44493", + "issue:44206|issue:45823", + "issue:32090|issue:33357" + ], + "prepared_review_unit_hash": "b97a8e6b3173156e5cbcaa14633dc3c99a624a1a8ea6b33cb8b37efd9ed47f42", + "component_budget": { + "node_count": 502, + "item_count": 502, + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 29, + "serialized_chars": 36595, + "estimated_input_tokens": 9149, + "estimated_eval_tokens": 18554 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly heterogeneous and does not look like one duplicate family. The only clear duplicate pair is the Qwen3.5 MoE from_pretrained error reports; most other soft edges are just broad subsystem overlap or shared regression themes, not the same concrete bug.", + "confidence": 0.93, + "canonical_issue_reason": "issue:45310 is the best canonical anchor because it is the earlier, clearer report of the exact same Qwen3.5 MoE from_pretrained failure; issue:45478 repeats the same title and appears to be a later duplicate.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:45310 is the strongest issue to keep as the representative bug report among the items because it has the clearest duplicate counterpart and a precise, concrete failure description.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43335", + "right": "issue:43504", + "accept": false, + "reason": "Different models and failures: SwitchTransformers config bug versus BEiT legacy-field load failure." + }, + { + "left": "issue:44855", + "right": "issue:45083", + "accept": false, + "reason": "Unrelated bugs: a Python import/IndentationError versus a qwen3_omni_moe helper-function behavior issue." + }, + { + "left": "issue:43278", + "right": "issue:43825", + "accept": false, + "reason": "Different code paths and symptoms: dtype drift in evaluate versus an incorrect pipeline error message." + }, + { + "left": "issue:29127", + "right": "issue:32090", + "accept": false, + "reason": "One is a LayoutLMv3 error-message clarity issue; the other is a Trainer/NCCL broadcast type error." + }, + { + "left": "issue:43720", + "right": "issue:44112", + "accept": false, + "reason": "Different underlying problems: packed weights not unpacked during Accelerate loading versus a stale CI test in GraniteSpeech." + }, + { + "left": "issue:43335", + "right": "issue:45083", + "accept": false, + "reason": "No shared underlying bug; they concern different architectures and different failure modes." + }, + { + "left": "issue:37428", + "right": "issue:44351", + "accept": false, + "reason": "Both are import failures, but for different symbols and different modules, so they are not the same bug." + }, + { + "left": "issue:43335", + "right": "issue:45593", + "accept": false, + "reason": "Different model/config issues: sparse-layer config creation versus auxiliary-loss behavior when denoising is off." + }, + { + "left": "issue:43335", + "right": "issue:44112", + "accept": false, + "reason": "No evidence they share the same code path; one is config construction, the other is a CI override regression." + }, + { + "left": "issue:44514", + "right": "issue:44521", + "accept": false, + "reason": "Both involve Qwen2.5 VL chat templating, but one is a padding-related crash and the other is incorrect assistant masks; not the same concrete bug." + }, + { + "left": "issue:45290", + "right": "issue:46326", + "accept": false, + "reason": "Both are chat-template related, but they target different tokenizer features and different failure modes." + }, + { + "left": "issue:34689", + "right": "issue:35141", + "accept": false, + "reason": "Different problems: model loading regression for Llama 3.2 Vision versus embedding reinitialization after resizing." + }, + { + "left": "issue:45588", + "right": "issue:45859", + "accept": false, + "reason": "Different subsystems and failures: flash_attention integration attribute error versus CPU-offload incompatibility in a rotary embedding implementation." + }, + { + "left": "issue:42490", + "right": "issue:43854", + "accept": false, + "reason": "Both involve loading behavior, but the local-path import warning and the GLM-4.7-Flash unit-test load failure are not the same bug." + }, + { + "left": "issue:46431", + "right": "issue:46566", + "accept": false, + "reason": "Chat-template absence and meta-device loading are unrelated issues." + }, + { + "left": "issue:45310", + "right": "issue:45478", + "accept": true, + "reason": "Exact same Qwen3.5 MoE from_pretrained error; the titles match and the later issue appears to be a duplicate report." + }, + { + "left": "issue:45127", + "right": "issue:45245", + "accept": false, + "reason": "Different failures: LoRA merge collapse with tied embeddings versus a category-count runtime limit." + }, + { + "left": "issue:44448", + "right": "issue:44779", + "accept": false, + "reason": "Both are v5 regressions, but one is Pegasus output drift and the other is Deepseek tokenizer correctness; not the same bug." + }, + { + "left": "issue:43819", + "right": "issue:44493", + "accept": false, + "reason": "Different code paths: DAC from_latents mismatch versus unexpected position-id keys across models." + }, + { + "left": "issue:39692", + "right": "issue:43526", + "accept": false, + "reason": "Documentation example errors are not the same issue as BeitImageProcessorFast reduce_labels returning the wrong shape/result." + }, + { + "left": "issue:43606", + "right": "issue:44857", + "accept": false, + "reason": "Different runtime failures in different models/components: CPU-offload device mismatch versus AMP/CUDA loss crash." + }, + { + "left": "issue:26413", + "right": "issue:31501", + "accept": false, + "reason": "Both mention training startup/resume, but one is a checkpoint-resume iterator failure and the other is a FineWeb training start problem; too vague to merge." + }, + { + "left": "issue:28218", + "right": "issue:32090", + "accept": false, + "reason": "Tokenizer spacing after added tokens is unrelated to a Trainer GPU broadcast NoneType error." + }, + { + "left": "issue:45459", + "right": "issue:46234", + "accept": false, + "reason": "Tokenizer error masking and a missing import_utils helper are unrelated issues." + }, + { + "left": "issue:44617", + "right": "issue:46697", + "accept": false, + "reason": "CUDA OOM in Sam3Video and over-padding in Swin2SR image processing are unrelated." + }, + { + "left": "issue:43701", + "right": "issue:45083", + "accept": false, + "reason": "Resume-from-checkpoint key mismatch and qwen3_omni_moe feature-length behavior are unrelated." + }, + { + "left": "issue:44206", + "right": "issue:45823", + "accept": false, + "reason": "Different failures in different models: unsupported feature-extractor arg versus device-map forward mismatch." + }, + { + "left": "issue:32090", + "right": "issue:33357", + "accept": false, + "reason": "A NoneType GPU broadcast error and a MacOS bus error in CLIP loading are not the same underlying bug." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is well grounded in the packet. The representative/canonical choice for issue:45310 over issue:45478 is supported by identical titles and earlier creation time, and the cluster summary stays conservative by treating the rest as non-duplicates unless they clearly share the same concrete bug. No overstatement stands out." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 32, + "review_unit_count": 61, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-502", + "nodes": [ + "issue:10000", + "issue:15354", + "issue:25251", + "issue:27957", + "issue:29870", + "issue:29942", + "issue:32090", + "issue:33290", + "issue:34689", + "issue:35141", + "issue:37428", + "issue:38175", + "issue:40444", + "issue:42175", + "issue:42490", + "issue:43278", + "issue:43329", + "issue:43381", + "issue:43526", + "issue:43550", + "issue:43575", + "issue:43873", + "issue:43927", + "issue:43976", + "issue:44360", + "issue:44448", + "issue:44466", + "issue:44493", + "issue:44561", + "issue:44779", + "issue:44855", + "issue:44857", + "issue:44977", + "issue:45072", + "issue:45083", + "issue:45229", + "issue:45290", + "issue:45381", + "issue:45440", + "issue:45588", + "issue:45698", + "issue:45920", + "issue:46192", + "issue:46399", + "issue:46421", + "issue:46655", + "issue:46858" + ], + "soft_pairs": [ + "issue:42490|issue:43575", + "issue:44360|issue:44561", + "issue:44857|issue:45072", + "issue:43550|issue:45083", + "issue:27957|issue:34689", + "issue:29942|issue:35141", + "issue:38175|issue:44493", + "issue:29870|issue:32090", + "issue:44448|issue:44977", + "issue:43278|issue:43329", + "issue:45920|issue:46192", + "issue:45290|issue:45381", + "issue:10000|issue:15354", + "issue:44977|issue:45229", + "issue:29870|issue:33290", + "issue:43278|issue:43381", + "issue:44855|issue:46399", + "issue:43927|issue:45698", + "issue:45290|issue:46655", + "issue:40444|issue:43526", + "issue:37428|issue:42175", + "issue:44857|issue:45588", + "issue:32090|issue:33290", + "issue:46421|issue:46858", + "issue:44561|issue:44779", + "issue:44448|issue:44466", + "issue:43873|issue:45229", + "issue:25251|issue:32090", + "issue:43976|issue:45440", + "issue:42490|issue:43873" + ], + "prepared_review_unit_hash": "dd52108c39aeaa880081bdf8638eaa2926a922b44630661e2fb5d03d277037f2", + "component_budget": { + "node_count": 502, + "item_count": 502, + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 30, + "serialized_chars": 36350, + "estimated_input_tokens": 9088, + "estimated_eval_tokens": 18432 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The soft pairs are mostly false positives from broad semantic similarity: they mention shared themes like loading, generation, Flash Attention, dtype, or compile issues, but they point to different models, different code paths, or different failure modes. I would not merge any of these pairs as duplicates.", + "confidence": 0.95, + "canonical_issue_reason": "No clear canonical issue: this set is heterogeneous and the apparent matches are not the same underlying bug.", + "canonical_pr_reason": null, + "best_issue_reason": "No single issue is a safe representative for the cluster; the items span unrelated regressions and model-specific failures.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:42490", + "right": "issue:43575", + "accept": false, + "reason": "Both involve loading, but one is a local path trailing-slash warning and the other is a tensor-parallel OOM on a specific Qwen model; different failures and fixes." + }, + { + "left": "issue:44360", + "right": "issue:44561", + "accept": false, + "reason": "DSA indexer/ReLU discussion is unrelated to removal of `is_torch_fx_available` and `trust_remote_code` breakage." + }, + { + "left": "issue:44857", + "right": "issue:45072", + "accept": false, + "reason": "Both mention dtype/AMP-style failures, but one is a loss crash on CUDA fp16 and the other is a bfloat16 inference mismatch in different models; not the same bug." + }, + { + "left": "issue:43550", + "right": "issue:45083", + "accept": false, + "reason": "`torch.compile`/SDPA on Bamba-9B-v2 is a model/runtime bug; `_get_feat_extract_output_lengths` in qwen3_omni_moe is a separate helper-function behavior issue." + }, + { + "left": "issue:27957", + "right": "issue:34689", + "accept": false, + "reason": "Both concern model loading/attention support, but they are different models with different compatibility problems." + }, + { + "left": "issue:29942", + "right": "issue:35141", + "accept": false, + "reason": "Flash Attention test failures are unrelated to token-embedding resizing and `post_init` reinitialization." + }, + { + "left": "issue:38175", + "right": "issue:44493", + "accept": false, + "reason": "Unexpected zero probabilities in SigLIP2 is a model-specific output issue; unexpected position-id keys is a different serialization/config problem." + }, + { + "left": "issue:29870", + "right": "issue:32090", + "accept": false, + "reason": "BART `generate()` decoder mask bug and Trainer `_gpu_broadcast_one` NoneType error are separate runtime paths." + }, + { + "left": "issue:44448", + "right": "issue:44977", + "accept": false, + "reason": "Pegasus v4/v5 output divergence and Qwen3.5 flash-attention generation failure are different model regressions." + }, + { + "left": "issue:43278", + "right": "issue:43329", + "accept": false, + "reason": "Embedding dtype drift in evaluate is unrelated to multimodal token counting bugs in the video branch." + }, + { + "left": "issue:45920", + "right": "issue:46192", + "accept": false, + "reason": "AutoTokenizer wrong token IDs is a tokenizer regression; `return_tensors` being ignored with explicit `text_kwargs` is an input-processing API bug." + }, + { + "left": "issue:45290", + "right": "issue:45381", + "accept": false, + "reason": "Chat template crashes on tool-call assistant messages; the other is incorrect video `vision_position_ids` in Qwen2.5-VL." + }, + { + "left": "issue:10000", + "right": "issue:15354", + "accept": false, + "reason": "German DistilBertModel and a GeneratorExp/torch.jit.script export error are unrelated model/export failures." + }, + { + "left": "issue:44977", + "right": "issue:45229", + "accept": false, + "reason": "Flash-attention generation issue for Qwen3.5 is not the same as multi-GPU CUDA OOM for Gemma4 31B." + }, + { + "left": "issue:29870", + "right": "issue:33290", + "accept": false, + "reason": "Decoder attention mask generation bug is unrelated to Adafactor/deepspeed OOM." + }, + { + "left": "issue:43278", + "right": "issue:43381", + "accept": false, + "reason": "BF16/FP32 dtype drift in evaluate is unrelated to gradient checkpointing being unusable in eval mode." + }, + { + "left": "issue:44855", + "right": "issue:46399", + "accept": false, + "reason": "Python 3.13 `IndentationError`/`torch.jit.script` parsing issue is unrelated to Qwen3.5 rotary-pos-emb registration failure." + }, + { + "left": "issue:43927", + "right": "issue:45698", + "accept": false, + "reason": "Custom token IDs lost after save/load causing generation IndexError is a different persistence bug from loading the wrong custom module after `save_pretrained`." + }, + { + "left": "issue:45290", + "right": "issue:46655", + "accept": false, + "reason": "Chat-template tool-call crash and SmolVLM video padding miscalculation affect different subsystems." + }, + { + "left": "issue:40444", + "right": "issue:43526", + "accept": false, + "reason": "IterableDataset multi-image finetuning failure and BeitImageProcessorFast label reduction bug are unrelated multimodal preprocessing issues." + }, + { + "left": "issue:37428", + "right": "issue:42175", + "accept": false, + "reason": "Flash-support import regression and missing TensorFlow in a torch-only install are different packaging/import problems." + }, + { + "left": "issue:44857", + "right": "issue:45588", + "accept": false, + "reason": "These are both crashes, but on different components: LwDetrImageLoss AMP/CUDA vs flash_attention integration with `s_aux=None`." + }, + { + "left": "issue:32090", + "right": "issue:33290", + "accept": false, + "reason": "Trainer broadcast NoneType error and deepspeed Adafactor OOM are distinct training/runtime issues." + }, + { + "left": "issue:46421", + "right": "issue:46858", + "accept": false, + "reason": "Lazy cache initialization causing recompilation and `_prefill` seq-length device-tensor arithmetic are different cache/compile bugs." + }, + { + "left": "issue:44561", + "right": "issue:44779", + "accept": false, + "reason": "Removal of `is_torch_fx_available` breaks trust-remote-code models, while DeepSeek tokenizer regression is a separate tokenizer behavior change." + }, + { + "left": "issue:44448", + "right": "issue:44466", + "accept": false, + "reason": "Model output divergence across versions and inconsistent `lm_head.weight` serialization are both v5 regressions, but they concern different mechanisms and are not the same bug." + }, + { + "left": "issue:43873", + "right": "issue:45229", + "accept": false, + "reason": "Quantization offloading behavior and Gemma4 multi-GPU OOM are unrelated memory-management issues." + }, + { + "left": "issue:25251", + "right": "issue:32090", + "accept": false, + "reason": "Pipeline `top_k` nesting behavior and Trainer broadcast NoneType error are unrelated API/runtime bugs." + }, + { + "left": "issue:43976", + "right": "issue:45440", + "accept": false, + "reason": "Python version compatibility for Transformers 5.1.0 is unrelated to native DeepseekV3MoE diverging from the remote implementation." + }, + { + "left": "issue:42490", + "right": "issue:43873", + "accept": false, + "reason": "Local-path import warning and quantization offloading behavior are different loading/offload concerns with no shared code-path evidence." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 33, + "review_unit_count": 61, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-502", + "nodes": [ + "issue:29127", + "issue:29870", + "issue:29942", + "issue:30064", + "issue:33357", + "issue:34567", + "issue:34689", + "issue:39692", + "issue:41720", + "issue:42915", + "issue:42994", + "issue:43065", + "issue:43278", + "issue:43381", + "issue:43412", + "issue:43575", + "issue:43611", + "issue:43653", + "issue:43701", + "issue:43756", + "issue:43994", + "issue:44186", + "issue:44206", + "issue:44265", + "issue:44361", + "issue:44442", + "issue:44479", + "issue:44568", + "issue:44743", + "issue:44871", + "issue:44977", + "issue:45198", + "issue:45229", + "issue:45245", + "issue:45372", + "issue:45406", + "issue:45412", + "issue:45440", + "issue:45459", + "issue:45491", + "issue:45563", + "issue:45588", + "issue:45712", + "issue:45803", + "issue:45823", + "issue:46121", + "issue:46693" + ], + "soft_pairs": [ + "issue:44568|issue:45491", + "issue:43278|issue:43575", + "issue:44871|issue:44977", + "issue:43381|issue:43701", + "issue:41720|issue:45229", + "issue:45588|issue:45823", + "issue:33357|issue:39692", + "issue:29127|issue:34567", + "issue:44977|issue:45563", + "issue:45198|issue:45803", + "issue:44361|issue:45198", + "issue:44186|issue:45198", + "issue:43653|issue:44743", + "issue:43611|issue:46121", + "issue:29870|issue:34689", + "issue:45372|issue:45588", + "issue:42915|issue:42994", + "issue:43756|issue:45440", + "issue:43653|issue:44442", + "issue:29942|issue:30064", + "issue:44206|issue:44479", + "issue:45372|issue:45406", + "issue:43412|issue:45588", + "issue:45459|issue:45712", + "issue:45491|issue:46693", + "issue:44265|issue:45198", + "issue:29127|issue:33357", + "issue:43065|issue:43994", + "issue:43065|issue:45412", + "issue:45245|issue:45563", + "issue:43575|issue:43701" + ], + "prepared_review_unit_hash": "8a94729b7886b94a8f6ff418a569b9ef4bd76d3a418c310ab3e1f52f6a9222cb", + "component_budget": { + "node_count": 502, + "item_count": 502, + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 31, + "serialized_chars": 36562, + "estimated_input_tokens": 9141, + "estimated_eval_tokens": 18538 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a grab-bag of unrelated bug reports that only share broad wording around model loading, generation, tokenizers, or attention. None of the suggested soft links look like true duplicates; they differ in model family, failing code path, or symptom.", + "confidence": 0.68, + "canonical_issue_reason": "issue:46693 is the most broadly scoped and currently active report in the set, with the clearest regression framing and enough detail to serve as an anchor issue, though the cluster itself is heterogeneous.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:46693 is the strongest standalone issue here: it is specific enough to reproduce, broad enough to represent a real regression class, and has the most obvious tracking value among the listed issues.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44568", + "right": "issue:45491", + "accept": false, + "reason": "Different failures: BOS/EOS token insertion for mdeberta-v3 vs Gemma3 NaN embeddings with mixed-length GPU batching." + }, + { + "left": "issue:43278", + "right": "issue:43575", + "accept": false, + "reason": "Unrelated bugs: BF16->FP32 dtype drift in eval vs tensor-parallel loading OOM for Qwen2-57B." + }, + { + "left": "issue:44871", + "right": "issue:44977", + "accept": false, + "reason": "Gemma eos_token_id config mismatch is not the same as Qwen3.5 flash-attention generation failure." + }, + { + "left": "issue:43381", + "right": "issue:43701", + "accept": false, + "reason": "Gradient checkpointing in eval mode and resume_from_checkpoint key mismatch are different code paths and symptoms." + }, + { + "left": "issue:41720", + "right": "issue:45229", + "accept": false, + "reason": "Both involve large-model inference, but one is a cudaErrorAssert on Qwen3 auto device mapping and the other is a CUDA OOM on Gemma4 multi-GPU." + }, + { + "left": "issue:45588", + "right": "issue:45823", + "accept": false, + "reason": "Flash-attention crash on s_aux=None is distinct from Gemma4 PLE device mismatch during forward." + }, + { + "left": "issue:33357", + "right": "issue:39692", + "accept": false, + "reason": "MacOS bus error for a community CLIP model is not the same as SigLIP2 documentation/example errors." + }, + { + "left": "issue:29127", + "right": "issue:34567", + "accept": false, + "reason": "LayoutLMv3 error-message clarity and TrainerState num_input_tokens_seen not updating are unrelated issues." + }, + { + "left": "issue:44977", + "right": "issue:45563", + "accept": false, + "reason": "Qwen3.5 flash-attention generation breakage is unrelated to a stale warning in paged generate()." + }, + { + "left": "issue:45198", + "right": "issue:45803", + "accept": false, + "reason": "Wav2Vec2 save/tokenization failures and a bare-except swallowing errors in FuyuBatchFeature are different bugs." + }, + { + "left": "issue:44361", + "right": "issue:45198", + "accept": false, + "reason": "MLukeTokenizer AttributeError is unrelated to Wav2Vec2 save_pretrained/tokenization failure." + }, + { + "left": "issue:44186", + "right": "issue:45198", + "accept": false, + "reason": "LayoutLMv2Tokenizer NER/padding crash does not match the Wav2Vec2 save/tokenization bug." + }, + { + "left": "issue:43653", + "right": "issue:44743", + "accept": false, + "reason": "BigBirdTokenizer special-token registration and Qwen3.5 recurrent-state reset are separate components with different root causes." + }, + { + "left": "issue:43611", + "right": "issue:46121", + "accept": false, + "reason": "base_model_prefix loading regression and rope-parameter JSON list type error are different failures." + }, + { + "left": "issue:29870", + "right": "issue:34689", + "accept": false, + "reason": "Bart generate decoder_attention_mask bug and Llama 3.2 Vision loading regression are unrelated." + }, + { + "left": "issue:45372", + "right": "issue:45588", + "accept": false, + "reason": "Both touch Gemma4, but one is an import error for ReasoningEffort during processor loading and the other is a flash_attention.py AttributeError on s_aux=None." + }, + { + "left": "issue:42915", + "right": "issue:42994", + "accept": false, + "reason": "Qwen3Moe FineGrainedFP8Config failure and quantized model saving failure are different bugs." + }, + { + "left": "issue:43756", + "right": "issue:45440", + "accept": false, + "reason": "Smollm3 RoPE-layer mismatch and DeepseekV3MoE divergence from remote implementation are not the same issue." + }, + { + "left": "issue:43653", + "right": "issue:44442", + "accept": false, + "reason": "BigBirdTokenizer special-token bug and AutoTokenizer loading FastSpeech2ConformerTokenizer are unrelated." + }, + { + "left": "issue:29942", + "right": "issue:30064", + "accept": false, + "reason": "Flash Attention 2 test failures and void segmentation-map processing are different subsystems and symptoms." + }, + { + "left": "issue:44206", + "right": "issue:44479", + "accept": false, + "reason": "Unsupported center arg crash in LasrFeatureExtractor is not the same as the Qwen video-input regression." + }, + { + "left": "issue:45372", + "right": "issue:45406", + "accept": false, + "reason": "Both are Gemma4-related, but the first is a processor import/loading problem while the second is a missing _tokenizer attribute in transformers serve." + }, + { + "left": "issue:43412", + "right": "issue:45588", + "accept": false, + "reason": "Gemma3n export/training-guard and erfinv support issues are unrelated to the flash_attention s_aux=None crash." + }, + { + "left": "issue:45459", + "right": "issue:45712", + "accept": false, + "reason": "Tokenizer exception masking when protobuf is missing and leftover dummy classes in dummy_pt_objects are different maintenance bugs." + }, + { + "left": "issue:45491", + "right": "issue:46693", + "accept": false, + "reason": "These are both FlashAttention/Gemma-adjacent, but one is a NaN-embedding correctness bug and the other is a long-context performance regression; not the same underlying defect." + }, + { + "left": "issue:44265", + "right": "issue:45198", + "accept": false, + "reason": "torch.export failure with torch_compilable_check is unrelated to Wav2Vec2 save/tokenization failure." + }, + { + "left": "issue:29127", + "right": "issue:33357", + "accept": false, + "reason": "LayoutLMv3 error-message clarity and CLIP bus error on MacOS are unrelated." + }, + { + "left": "issue:43065", + "right": "issue:43994", + "accept": false, + "reason": "Dummy Conv2d in Sam3PixelDecoder and SigLIP2 nonsensical AutoModel/pipeline outputs are different model issues." + }, + { + "left": "issue:43065", + "right": "issue:45412", + "accept": false, + "reason": "Sam3PixelDecoder dummy-layer cleanup is unrelated to RT-DETR memory not being released." + }, + { + "left": "issue:45245", + "right": "issue:45563", + "accept": false, + "reason": "Category-count runtime error and paged-generate stale warning are different bug classes." + }, + { + "left": "issue:43575", + "right": "issue:43701", + "accept": false, + "reason": "Tensor-parallel OOM when loading Qwen2-57B and resume_from_checkpoint key mismatch are unrelated." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 34, + "review_unit_count": 61, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-502", + "nodes": [ + "issue:27957", + "issue:29127", + "issue:33290", + "issue:34689", + "issue:35141", + "issue:42490", + "issue:42915", + "issue:43066", + "issue:43334", + "issue:43381", + "issue:43421", + "issue:43526", + "issue:43531", + "issue:43575", + "issue:43646", + "issue:43761", + "issue:43824", + "issue:43873", + "issue:43931", + "issue:43994", + "issue:44077", + "issue:44186", + "issue:44279", + "issue:44387", + "issue:44442", + "issue:44462", + "issue:44464", + "issue:44514", + "issue:45072", + "issue:45250", + "issue:45372", + "issue:45381", + "issue:45412", + "issue:45440", + "issue:45459", + "issue:45491", + "issue:45584", + "issue:45588", + "issue:45593", + "issue:45600", + "issue:45834", + "issue:45859", + "issue:46121", + "issue:46369", + "issue:46399", + "issue:46640", + "issue:46693" + ], + "soft_pairs": [ + "issue:43066|issue:44462", + "issue:45440|issue:45593", + "issue:45372|issue:45459", + "issue:42490|issue:43381", + "issue:29127|issue:35141", + "issue:43421|issue:45491", + "issue:43994|issue:45412", + "issue:45412|issue:46369", + "issue:43575|issue:44387", + "issue:44442|issue:45072", + "issue:44464|issue:45584", + "issue:42915|issue:43931", + "issue:43526|issue:45834", + "issue:43824|issue:44279", + "issue:43526|issue:46693", + "issue:27957|issue:29127", + "issue:27957|issue:33290", + "issue:45250|issue:46640", + "issue:43994|issue:46369", + "issue:33290|issue:34689", + "issue:33290|issue:35141", + "issue:43526|issue:44186", + "issue:43334|issue:43531", + "issue:46121|issue:46399", + "issue:45588|issue:45600", + "issue:43381|issue:43873", + "issue:44514|issue:45381", + "issue:43646|issue:44077", + "issue:43761|issue:45859" + ], + "prepared_review_unit_hash": "9f918b078dd5f69f359e788795648790464e4debd3f2477ee1d8d0abd4ea2cd9", + "component_budget": { + "node_count": 502, + "item_count": 502, + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 29, + "serialized_chars": 35677, + "estimated_input_tokens": 8920, + "estimated_eval_tokens": 18096 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The set is a grab bag of unrelated Transformers issues with only superficial overlap in terms like Flash Attention, OOM, or loading errors, so I rejected all soft pairs as duplicates. If one issue must serve as the representative entry, 43873 is the strongest standalone report, but it is not a duplicate of the others.", + "confidence": 0.94, + "canonical_issue_reason": "Issue 43873 is the best standalone representative here: it is detailed, actively discussed, and has inbound references, but the overall set is not a true duplicate cluster.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 43873 has the strongest combination of specificity, discussion activity, and external references, making it the most usable issue entry in this mixed set.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43066", + "right": "issue:44462", + "accept": false, + "reason": "Different tokenizer bugs: v5 decoder-type mismatch vs AutoTokenizer ignoring repo tokenizer.json." + }, + { + "left": "issue:45440", + "right": "issue:45593", + "accept": false, + "reason": "Different model-specific failures: DeepSeekV3 implementation divergence vs D-FINE auxiliary-loss behavior." + }, + { + "left": "issue:45372", + "right": "issue:45459", + "accept": false, + "reason": "Import/dependency error vs tokenizer error masking; same area, but not the same underlying bug." + }, + { + "left": "issue:42490", + "right": "issue:43381", + "accept": false, + "reason": "Local-path import warning and eval-mode gradient checkpointing are unrelated." + }, + { + "left": "issue:29127", + "right": "issue:35141", + "accept": false, + "reason": "LayoutLMv3 box-validation messaging and embedding reinitialization are different code paths." + }, + { + "left": "issue:43421", + "right": "issue:45491", + "accept": false, + "reason": "Tokenizer post-processor sync bug and Gemma3 NaNs are unrelated." + }, + { + "left": "issue:43994", + "right": "issue:45412", + "accept": false, + "reason": "Different symptoms and models: wrong outputs vs memory not being released." + }, + { + "left": "issue:45412", + "right": "issue:46369", + "accept": false, + "reason": "Both are memory-related, but they affect different models and appear to be different bugs." + }, + { + "left": "issue:43575", + "right": "issue:44387", + "accept": false, + "reason": "Both mention OOM, but one is tensor-parallel loading and the other is int4 quantization memory growth." + }, + { + "left": "issue:44442", + "right": "issue:45072", + "accept": false, + "reason": "Tokenizer loading failure and bf16 dtype mismatch are not the same issue." + }, + { + "left": "issue:44464", + "right": "issue:45584", + "accept": false, + "reason": "Compiled-forward chunked generation inconsistency is unrelated to Whisper empty-transcription handling." + }, + { + "left": "issue:42915", + "right": "issue:43931", + "accept": false, + "reason": "FineGrainedFP8Config failure and Qwen3-VL weight-shape mismatch are distinct bugs." + }, + { + "left": "issue:43526", + "right": "issue:45834", + "accept": false, + "reason": "BEiT label reduction and Kosmos2.5 OCR index errors are different processor bugs." + }, + { + "left": "issue:43824", + "right": "issue:44279", + "accept": false, + "reason": "Specific missing-class ImportError and a generic dependency issue are too different to merge." + }, + { + "left": "issue:43526", + "right": "issue:46693", + "accept": false, + "reason": "Image-processor label handling and FA2 long-context regression are unrelated." + }, + { + "left": "issue:27957", + "right": "issue:29127", + "accept": false, + "reason": "Flash Attention 2 support and LayoutLMv3 box-error clarity do not share the same underlying bug." + }, + { + "left": "issue:27957", + "right": "issue:33290", + "accept": false, + "reason": "FA2 support issue vs DeepSpeed+AdaFactor OOM are unrelated." + }, + { + "left": "issue:45250", + "right": "issue:46640", + "accept": false, + "reason": "Both are generic Flash Attention 2 titles, but there is no concrete evidence they describe the same bug." + }, + { + "left": "issue:43994", + "right": "issue:46369", + "accept": false, + "reason": "Different model families and failure modes: nonsensical outputs vs memory overflow." + }, + { + "left": "issue:33290", + "right": "issue:34689", + "accept": false, + "reason": "Optimizer OOM and model-loading regression are unrelated." + }, + { + "left": "issue:33290", + "right": "issue:35141", + "accept": false, + "reason": "Optimizer memory blowup and embedding reinitialization are different problems." + }, + { + "left": "issue:43526", + "right": "issue:44186", + "accept": false, + "reason": "BEiT fast processor label reduction is unrelated to LayoutLMv2 tokenizer padding/NER crashes." + }, + { + "left": "issue:43334", + "right": "issue:43531", + "accept": false, + "reason": "Missing pad_token_id in Qwen3-VL and Qwen3-MoE sliding-window issues are different failures." + }, + { + "left": "issue:46121", + "right": "issue:46399", + "accept": false, + "reason": "Rope-validation TypeError and missing-decorator registration error are unrelated." + }, + { + "left": "issue:45588", + "right": "issue:45600", + "accept": false, + "reason": "Flash-attention s_aux crash and Sam3LiteText auto-mapping cleanup are not the same bug." + }, + { + "left": "issue:43381", + "right": "issue:43873", + "accept": false, + "reason": "Eval-mode gradient checkpointing and quantization/offloading behavior are different issues." + }, + { + "left": "issue:44514", + "right": "issue:45381", + "accept": false, + "reason": "Both are Qwen2.5-VL bugs, but batched chat-template padding and video vision_position_ids are distinct code paths." + }, + { + "left": "issue:43646", + "right": "issue:44077", + "accept": false, + "reason": "Custom model initialization breakage and patchtsmixer post_init constraints are not the same concrete defect." + }, + { + "left": "issue:43761", + "right": "issue:45859", + "accept": false, + "reason": "CLIPVisionModel hidden_states regression and Qwen3.5 MoE CPU-offload incompatibility are unrelated." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 35, + "review_unit_count": 61, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-502", + "nodes": [ + "issue:27957", + "issue:29870", + "issue:42175", + "issue:42490", + "issue:43232", + "issue:43262", + "issue:43316", + "issue:43388", + "issue:43493", + "issue:43525", + "issue:43643", + "issue:43720", + "issue:43749", + "issue:43761", + "issue:43827", + "issue:43873", + "issue:43931", + "issue:44062", + "issue:44077", + "issue:44079", + "issue:44188", + "issue:44190", + "issue:44206", + "issue:44442", + "issue:44479", + "issue:44496", + "issue:44743", + "issue:44811", + "issue:44855", + "issue:44871", + "issue:44898", + "issue:44913", + "issue:44977", + "issue:44987", + "issue:45003", + "issue:45137", + "issue:45216", + "issue:45245", + "issue:45325", + "issue:45440", + "issue:45759", + "issue:45803", + "issue:45907", + "issue:46178", + "issue:46234", + "issue:46697", + "issue:46829", + "issue:46932" + ], + "soft_pairs": [ + "issue:44977|issue:45245", + "issue:44743|issue:44811", + "issue:44442|issue:44898", + "issue:43262|issue:44206", + "issue:42490|issue:44743", + "issue:45003|issue:45440", + "issue:43720|issue:44442", + "issue:42175|issue:46234", + "issue:27957|issue:29870", + "issue:43316|issue:43525", + "issue:43493|issue:43827", + "issue:44062|issue:45907", + "issue:44077|issue:44479", + "issue:45325|issue:46697", + "issue:43873|issue:44871", + "issue:44855|issue:45137", + "issue:43643|issue:44913", + "issue:43749|issue:43931", + "issue:43316|issue:45759", + "issue:44190|issue:44496", + "issue:46829|issue:46932", + "issue:44987|issue:46178", + "issue:43761|issue:44188", + "issue:44442|issue:45803", + "issue:43761|issue:44079", + "issue:43931|issue:45216", + "issue:43232|issue:43388" + ], + "prepared_review_unit_hash": "28fda421d669eb707715e7ea3eb7689eff203bb3765afe84264bb31e23c02b3f", + "component_budget": { + "node_count": 502, + "item_count": 502, + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 27, + "serialized_chars": 36183, + "estimated_input_tokens": 9046, + "estimated_eval_tokens": 18348 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This set is mostly a collection of unrelated bugs across different models, generation paths, loading code, and docs/backend issues. None of the soft pairs looks like the same underlying defect, so I would not collapse any of them as duplicates.", + "confidence": 0.98, + "canonical_issue_reason": "Issue 43873 is the most active open bug in the set and the best available representative, even though the cluster itself does not form a true duplicate group.", + "canonical_pr_reason": null, + "best_issue_reason": "43873 has the strongest ongoing signal (open, high discussion, inbound references) and is the most suitable issue to keep as the placeholder canonical record.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44977", + "right": "issue:45245", + "accept": false, + "reason": "Flash-attention generation failure and a save_pretrained checkpoint regression are different bugs with different code paths." + }, + { + "left": "issue:44743", + "right": "issue:44811", + "accept": false, + "reason": "Recurrent-state reset with cache is unrelated to Whisper batch_decode skipping special tokens." + }, + { + "left": "issue:44442", + "right": "issue:44898", + "accept": false, + "reason": "Tokenizer loading failure and Perceiver image interpolation failure affect different subsystems." + }, + { + "left": "issue:43262", + "right": "issue:44206", + "accept": false, + "reason": "Audio chat-template sampling-rate defaults and a feature extractor argument crash are separate issues." + }, + { + "left": "issue:42490", + "right": "issue:44743", + "accept": false, + "reason": "Local-path trailing-slash import warning is not the same as the cache/recurrent-state bug." + }, + { + "left": "issue:45003", + "right": "issue:45440", + "accept": false, + "reason": "Unsafe sys.modules access and DeepSeekV3MoE divergence are different problems in different areas." + }, + { + "left": "issue:43720", + "right": "issue:44442", + "accept": false, + "reason": "BitNet packed-weight loading and FastSpeech tokenizer loading failure do not share the same code path." + }, + { + "left": "issue:42175", + "right": "issue:46234", + "accept": false, + "reason": "Missing tensorflow backend in extras and a missing import_utils availability helper are distinct packaging/API issues." + }, + { + "left": "issue:27957", + "right": "issue:29870", + "accept": false, + "reason": "XLMRoberta Flash Attention 2 support and BART generate decoder_attention_mask assignment are unrelated." + }, + { + "left": "issue:43316", + "right": "issue:43525", + "accept": false, + "reason": "Gemma3TextConfig API mismatch and Llama4Config missing pad_token_id are different config-level bugs." + }, + { + "left": "issue:43493", + "right": "issue:43827", + "accept": false, + "reason": "SigLIP2 implementation discrepancy and a docs reference to pipeline() are not the same change." + }, + { + "left": "issue:44062", + "right": "issue:45907", + "accept": false, + "reason": "AddedToken special-argument duplication and list_repo_files revision forwarding are unrelated API bugs." + }, + { + "left": "issue:44077", + "right": "issue:44479", + "accept": false, + "reason": "PatchTSMixer post_init validation and Qwen video-input regression are different model-specific problems." + }, + { + "left": "issue:45325", + "right": "issue:46697", + "accept": false, + "reason": "Qwen2.5-VL rope-index scaling and Swin2SR over-padding are separate preprocessing/positioning bugs." + }, + { + "left": "issue:43873", + "right": "issue:44871", + "accept": false, + "reason": "Quantized offloading failure and Gemma-3 eos_token_id inconsistency are unrelated." + }, + { + "left": "issue:44855", + "right": "issue:45137", + "accept": false, + "reason": "Python 3.13 import parsing error and DeepSpeed ZeRO3 deque underflow are different failure modes." + }, + { + "left": "issue:43643", + "right": "issue:44913", + "accept": false, + "reason": "trust_remote_code field loss and GPTNeoX rotary_pct reload fallback are different serialization/config bugs." + }, + { + "left": "issue:43749", + "right": "issue:43931", + "accept": false, + "reason": "FSDP CPU RAM efficient loading breakage and Qwen3-VL weight-shape mismatch are not the same issue." + }, + { + "left": "issue:44190", + "right": "issue:44496", + "accept": false, + "reason": "Local dataset loading in a training script and an unrecognized model config error are unrelated." + }, + { + "left": "issue:46829", + "right": "issue:46932", + "accept": false, + "reason": "VideoPrism hidden_states output bug and Gemma4 quantized embedder dtype casting are different model paths." + }, + { + "left": "issue:44987", + "right": "issue:46178", + "accept": false, + "reason": "Model loading failure for a specific repo and tensorflow_text backend mapping are unrelated problems." + }, + { + "left": "issue:43761", + "right": "issue:44188", + "accept": false, + "reason": "CLIPVisionModel hidden_states regression and torch.compile attention-kernel divergence are different bugs." + }, + { + "left": "issue:44442", + "right": "issue:45803", + "accept": false, + "reason": "FastSpeech2 tokenizer loading and Fuyu batch feature exception handling are unrelated." + }, + { + "left": "issue:43761", + "right": "issue:44079", + "accept": false, + "reason": "CLIP hidden_states regression and ModelOutput None-key assignment are not the same underlying defect." + }, + { + "left": "issue:43931", + "right": "issue:45216", + "accept": false, + "reason": "Qwen3-VL loading shape mismatch and Qwen3.5 save_pretrained regression are separate issues at different stages." + }, + { + "left": "issue:43232", + "right": "issue:43388", + "accept": false, + "reason": "Generation kwarg handling after sync_gpus and gather_for_metrics label trimming are unrelated code paths." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 36, + "review_unit_count": 61, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-502", + "nodes": [ + "issue:16998", + "issue:22355", + "issue:25251", + "issue:27957", + "issue:28282", + "issue:32090", + "issue:42175", + "issue:42617", + "issue:42907", + "issue:43295", + "issue:43316", + "issue:43404", + "issue:43441", + "issue:43452", + "issue:43550", + "issue:43575", + "issue:43606", + "issue:43646", + "issue:43650", + "issue:43701", + "issue:43716", + "issue:43742", + "issue:43825", + "issue:44077", + "issue:44162", + "issue:44242", + "issue:44464", + "issue:44496", + "issue:44534", + "issue:44610", + "issue:44792", + "issue:44871", + "issue:44913", + "issue:45584", + "issue:45632", + "issue:45698", + "issue:45803", + "issue:46050", + "issue:46170", + "issue:46202", + "issue:46234", + "issue:46291", + "issue:46365", + "issue:46369", + "issue:46421", + "issue:46424", + "issue:46506", + "issue:46894" + ], + "soft_pairs": [ + "issue:43701|issue:46506", + "issue:42907|issue:45698", + "issue:42617|issue:43575", + "issue:27957|issue:32090", + "issue:43575|issue:43742", + "issue:43550|issue:43701", + "issue:43606|issue:43701", + "issue:25251|issue:28282", + "issue:46234|issue:46291", + "issue:43295|issue:43825", + "issue:16998|issue:22355", + "issue:43295|issue:44871", + "issue:44464|issue:46421", + "issue:44871|issue:45584", + "issue:43316|issue:43716", + "issue:42175|issue:43316", + "issue:44077|issue:44534", + "issue:44871|issue:46050", + "issue:43646|issue:44913", + "issue:45803|issue:46365", + "issue:44162|issue:44792", + "issue:43452|issue:44496", + "issue:44242|issue:46894", + "issue:43404|issue:43441", + "issue:45632|issue:46424", + "issue:45803|issue:46170", + "issue:43650|issue:46202", + "issue:44610|issue:46369" + ], + "prepared_review_unit_hash": "f52e09d5fb9164115067c3e57d1b5af581d35b16ea7dd73b56a446bfff737208", + "component_budget": { + "node_count": 502, + "item_count": 502, + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 28, + "serialized_chars": 35731, + "estimated_input_tokens": 8933, + "estimated_eval_tokens": 18122 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a grab bag of unrelated Transformers issues; the soft-similarity links are generally too weak to be true duplicates. I\u2019d treat issue 45698 as the best hub candidate only because it is the broadest active custom-module loading regression, not because it matches the rest closely.", + "confidence": 0.61, + "canonical_issue_reason": "Issue 45698 is the broadest, most active open report here (custom module loading after save_pretrained/from_pretrained) and is the closest thing to a hub for nearby reports.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 45698 is the strongest triage anchor: actionable, recently active, and centered on a core loading path that could plausibly gather related custom-model reports.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43701", + "right": "issue:46506", + "accept": false, + "reason": "Checkpoint-key mismatch in training resume is unrelated to PPDocLayoutV3 CPU/CUDA output divergence." + }, + { + "left": "issue:42907", + "right": "issue:45698", + "accept": false, + "reason": "Both involve save/load, but one is dequantized-model saving and the other is custom-module resolution after reload." + }, + { + "left": "issue:42617", + "right": "issue:43575", + "accept": false, + "reason": "Different failures: running 3d_parallel.py vs Qwen2 tensor-parallel OOM." + }, + { + "left": "issue:27957", + "right": "issue:32090", + "accept": false, + "reason": "Flash Attention support issue for XLMRoberta is unrelated to the Trainer NoneType broadcast error." + }, + { + "left": "issue:43575", + "right": "issue:43742", + "accept": false, + "reason": "OOM during tensor-parallel load is not the same bug as a KeyError while loading MobileLLM-125M." + }, + { + "left": "issue:43550", + "right": "issue:43701", + "accept": false, + "reason": "torch.compile/SDPA failure and resume-from-checkpoint key mismatch affect different code paths." + }, + { + "left": "issue:43606", + "right": "issue:43701", + "accept": false, + "reason": "CPU offload device mismatch is unrelated to checkpoint key-name mismatch." + }, + { + "left": "issue:25251", + "right": "issue:28282", + "accept": false, + "reason": "Pipeline top_k output shape regression is unrelated to missing PyTorch import in AutoModel." + }, + { + "left": "issue:46234", + "right": "issue:46291", + "accept": false, + "reason": "Missing helper symbol in import_utils and kernels dependency-version validation are different import-time problems." + }, + { + "left": "issue:43295", + "right": "issue:43825", + "accept": false, + "reason": "Processor/tokenizer regression and a pipeline error-message/support regression are distinct v5 issues." + }, + { + "left": "issue:16998", + "right": "issue:22355", + "accept": false, + "reason": "Model_max_length behavior for DeBERTa-V3 has nothing to do with a missing transformers.onnx module." + }, + { + "left": "issue:43295", + "right": "issue:44871", + "accept": false, + "reason": "Both touch config regression, but one is processor.tokenizer access and the other is eos_token_id mismatch in Gemma." + }, + { + "left": "issue:44464", + "right": "issue:46421", + "accept": false, + "reason": "Chunked generation inconsistency with compiled forward is not the same as lazy cache initialization causing recompilation." + }, + { + "left": "issue:44871", + "right": "issue:45584", + "accept": false, + "reason": "Gemma eos_token_id mismatch and Whisper empty-transcription behavior are different model-specific token handling bugs." + }, + { + "left": "issue:43316", + "right": "issue:43716", + "accept": false, + "reason": "Gemma3TextConfig API discrepancy and Mistral-3 dtype mismatch are unrelated model-specific issues." + }, + { + "left": "issue:42175", + "right": "issue:43316", + "accept": false, + "reason": "Packaging/backend install behavior is unrelated to a Gemma3TextConfig API discrepancy." + }, + { + "left": "issue:44077", + "right": "issue:44534", + "accept": false, + "reason": "Optional post_init deprecation and non-persistent buffer junk values are different initialization problems." + }, + { + "left": "issue:44871", + "right": "issue:46050", + "accept": false, + "reason": "Token ID config mismatch and quantized save/load breakage are separate serialization issues." + }, + { + "left": "issue:43646", + "right": "issue:44913", + "accept": false, + "reason": "Custom model initialization breakage and GPTNeoX rotary_pct reload loss are not the same underlying bug." + }, + { + "left": "issue:45803", + "right": "issue:46365", + "accept": false, + "reason": "Bare except swallowing interrupts in tensor conversion is unrelated to importing Trainer being blocked by datasets." + }, + { + "left": "issue:44162", + "right": "issue:44792", + "accept": false, + "reason": "ESM2 regression and janus generate-images test failure are different model-specific failures." + }, + { + "left": "issue:43452", + "right": "issue:44496", + "accept": false, + "reason": "gguf_file argument handling and missing model_type config validation are both load-path issues but not the same bug." + }, + { + "left": "issue:44242", + "right": "issue:46894", + "accept": false, + "reason": "These are both loss-related, but they concern different models and different loss-calculation mistakes, so they should not be merged." + }, + { + "left": "issue:43404", + "right": "issue:43441", + "accept": false, + "reason": "Untied lm_head weights and FlashAttention failures are distinct Mistral-family issues with different code paths." + }, + { + "left": "issue:45632", + "right": "issue:46424", + "accept": false, + "reason": "Cache-path collision for local trust_remote_code models is unrelated to max_cache_length semantics." + }, + { + "left": "issue:45803", + "right": "issue:46170", + "accept": false, + "reason": "A bare except in Fuyu batch conversion is unrelated to Windows ROCm/torch.distributed import errors." + }, + { + "left": "issue:43650", + "right": "issue:46202", + "accept": false, + "reason": "ADD THE DATA is non-technical noise and not a duplicate of the Gemma pipeline roundtrip bug." + }, + { + "left": "issue:44610", + "right": "issue:46369", + "accept": false, + "reason": "OmDet-Turbo input-size mismatch and sam3-litetext memory overflow are different model/runtime issues." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 37, + "review_unit_count": 61, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-502", + "nodes": [ + "issue:35141", + "issue:41950", + "issue:42490", + "issue:43065", + "issue:43352", + "issue:43388", + "issue:43475", + "issue:43504", + "issue:43525", + "issue:43531", + "issue:43540", + "issue:43572", + "issue:43673", + "issue:43688", + "issue:43761", + "issue:43827", + "issue:43994", + "issue:44190", + "issue:44514", + "issue:44534", + "issue:44617", + "issue:44734", + "issue:44743", + "issue:44792", + "issue:44855", + "issue:44869", + "issue:44913", + "issue:44964", + "issue:44977", + "issue:45072", + "issue:45127", + "issue:45200", + "issue:45245", + "issue:45406", + "issue:45542", + "issue:45588", + "issue:45600", + "issue:45632", + "issue:45698", + "issue:46082", + "issue:46178", + "issue:46234", + "issue:46617", + "issue:46640", + "issue:46697", + "issue:46832", + "issue:46858", + "issue:46932" + ], + "soft_pairs": [ + "issue:43761|issue:46932", + "issue:45542|issue:46234", + "issue:43504|issue:45698", + "issue:43540|issue:44514", + "issue:43572|issue:46178", + "issue:43352|issue:46640", + "issue:43065|issue:44617", + "issue:44792|issue:45600", + "issue:44792|issue:44977", + "issue:43388|issue:43688", + "issue:45127|issue:45200", + "issue:43673|issue:44743", + "issue:41950|issue:43475", + "issue:44190|issue:44964", + "issue:43525|issue:43531", + "issue:35141|issue:42490", + "issue:44534|issue:44913", + "issue:44734|issue:45406", + "issue:46082|issue:46697", + "issue:45632|issue:46617", + "issue:45072|issue:46932", + "issue:44855|issue:45588", + "issue:45245|issue:46858", + "issue:44869|issue:46832", + "issue:43827|issue:43994" + ], + "prepared_review_unit_hash": "6560d6a7aa3be34975d9d2f38ec76cc6df6e1b4134dc5f3af2a0fc14048e47bf", + "component_budget": { + "node_count": 502, + "item_count": 502, + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 35402, + "estimated_input_tokens": 8851, + "estimated_eval_tokens": 17958 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a loose collection of unrelated Transformers issues rather than a duplicate set. None of the soft pairs look like the same underlying bug/change, so all soft edges should be rejected. Issue 45698 is the best standalone representative because it has the strongest discussion signal and a broadly impactful loading bug.", + "confidence": 0.92, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": "issue:45698 is the strongest standalone issue in the set: it has active discussion, inbound references, and a clear user-facing regression in `from_pretrained`/`save_pretrained` custom module loading.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43761", + "right": "issue:46932", + "accept": false, + "reason": "Different regressions: CLIPVision hidden states vs Gemma4 quantized embedder dtype casting." + }, + { + "left": "issue:45542", + "right": "issue:46234", + "accept": false, + "reason": "Unrelated backend/import problems in different modules." + }, + { + "left": "issue:43504", + "right": "issue:45698", + "accept": false, + "reason": "Different loading failures: legacy preset field vs wrong custom module after save/load." + }, + { + "left": "issue:43540", + "right": "issue:44514", + "accept": false, + "reason": "Both are multimodal processor bugs, but they affect different models and code paths." + }, + { + "left": "issue:43572", + "right": "issue:46178", + "accept": false, + "reason": "Different concerns: config field migration vs backend mapping validation." + }, + { + "left": "issue:43352", + "right": "issue:46640", + "accept": false, + "reason": "Too broad; one is Nemotron-specific FlashAttention support, the other is a generic flash-attention issue." + }, + { + "left": "issue:43065", + "right": "issue:44617", + "accept": false, + "reason": "Different SAM3 bugs: a dummy layer vs CUDA out-of-memory." + }, + { + "left": "issue:44792", + "right": "issue:45600", + "accept": false, + "reason": "Different subsystems: janus test failure vs auto_mappings config reference breakage." + }, + { + "left": "issue:44792", + "right": "issue:44977", + "accept": false, + "reason": "Different model failures and symptoms; only share that they involve generation." + }, + { + "left": "issue:43388", + "right": "issue:43688", + "accept": false, + "reason": "Unrelated training/evaluation bugs with no shared code path." + }, + { + "left": "issue:45127", + "right": "issue:45200", + "accept": false, + "reason": "Different model behaviors: LoRA/extended vocab collapse vs missing mm_token_type_ids default." + }, + { + "left": "issue:43673", + "right": "issue:44743", + "accept": false, + "reason": "Both mention cache, but they are different bugs in different model families and logic." + }, + { + "left": "issue:41950", + "right": "issue:43475", + "accept": false, + "reason": "Pipeline processor lookup bug is unrelated to Sam3 vision output attributes." + }, + { + "left": "issue:44190", + "right": "issue:44964", + "accept": false, + "reason": "Different loading paths and failure modes; no evidence of the same bug." + }, + { + "left": "issue:43525", + "right": "issue:43531", + "accept": false, + "reason": "Different config issues: missing pad_token_id vs sliding_window handling." + }, + { + "left": "issue:35141", + "right": "issue:42490", + "accept": false, + "reason": "Embedding resize reinit bug is unrelated to trailing-slash local model warning/import behavior." + }, + { + "left": "issue:44534", + "right": "issue:44913", + "accept": false, + "reason": "Different persistence/regression issues: non-persistent buffers vs config reload defaults." + }, + { + "left": "issue:44734", + "right": "issue:45406", + "accept": false, + "reason": "Serve KV-cache indexing crash and Gemma4Processor missing attribute are different defects." + }, + { + "left": "issue:46082", + "right": "issue:46697", + "accept": false, + "reason": "Different validation domains: Llama attention config vs Swin2SR image padding." + }, + { + "left": "issue:45632", + "right": "issue:46617", + "accept": false, + "reason": "Both involve trust_remote_code caching, but one is a leaf-dir collision and the other is a missing cached file; not the same bug." + }, + { + "left": "issue:45072", + "right": "issue:46932", + "accept": false, + "reason": "Both are dtype-related, but they affect different models and inference paths." + }, + { + "left": "issue:44855", + "right": "issue:45588", + "accept": false, + "reason": "Python parsing/import error vs flash-attention None handling are unrelated." + }, + { + "left": "issue:45245", + "right": "issue:46858", + "accept": false, + "reason": "No shared underlying problem: category-count limit vs cache sequence-length arithmetic." + }, + { + "left": "issue:44869", + "right": "issue:46832", + "accept": false, + "reason": "Different crashes in different models and code paths." + }, + { + "left": "issue:43827", + "right": "issue:43994", + "accept": false, + "reason": "Docs referring to removed pipeline APIs is unrelated to SigLIP2 inference quality." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 38, + "review_unit_count": 61, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-502", + "nodes": [ + "issue:25251", + "issue:27957", + "issue:29870", + "issue:35141", + "issue:39692", + "issue:41720", + "issue:43065", + "issue:43295", + "issue:43335", + "issue:43550", + "issue:43644", + "issue:43673", + "issue:43931", + "issue:43976", + "issue:44060", + "issue:44077", + "issue:44112", + "issue:44190", + "issue:44242", + "issue:44462", + "issue:44464", + "issue:44479", + "issue:44610", + "issue:44617", + "issue:44734", + "issue:44779", + "issue:44855", + "issue:44877", + "issue:44913", + "issue:44933", + "issue:45127", + "issue:45245", + "issue:45276", + "issue:45290", + "issue:45357", + "issue:45362", + "issue:45381", + "issue:45563", + "issue:45584", + "issue:45600", + "issue:45907", + "issue:46369", + "issue:46421", + "issue:46489", + "issue:46655", + "issue:46752", + "issue:46897" + ], + "soft_pairs": [ + "issue:43673|issue:46421", + "issue:43931|issue:44060", + "issue:44242|issue:46897", + "issue:29870|issue:35141", + "issue:44617|issue:45584", + "issue:43644|issue:44913", + "issue:44933|issue:45362", + "issue:25251|issue:27957", + "issue:43976|issue:44855", + "issue:45563|issue:45584", + "issue:43550|issue:44190", + "issue:44077|issue:44877", + "issue:45381|issue:46369", + "issue:41720|issue:45245", + "issue:44479|issue:44734", + "issue:43335|issue:44190", + "issue:43065|issue:44610", + "issue:44464|issue:45563", + "issue:39692|issue:44190", + "issue:45127|issue:45357", + "issue:45381|issue:46655", + "issue:44112|issue:45600", + "issue:44779|issue:46489", + "issue:44617|issue:46369", + "issue:43295|issue:45276", + "issue:44462|issue:45907", + "issue:45290|issue:46752" + ], + "prepared_review_unit_hash": "5159dc4fca32783f2f86fcd75d1e5c1445f98cd63c208c109abab273809f609a", + "component_budget": { + "node_count": 502, + "item_count": 502, + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 27, + "serialized_chars": 35238, + "estimated_input_tokens": 8810, + "estimated_eval_tokens": 17876 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "These items are not duplicates of one another; they span many unrelated Transformers regressions and model-specific bugs. I did not find any soft-edge pair that looks like the same underlying issue.", + "confidence": 0.91, + "canonical_issue_reason": "issue:43644 is the broadest and most central bug report in the set, with inbound references and a clear, high-impact regression description, making it the best anchor if one issue must represent this cluster.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:43644 is the strongest standalone issue: it describes a concrete framework-level regression, is broadly applicable, and has evidence of being referenced by others.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43673", + "right": "issue:46421", + "accept": false, + "reason": "Both mention cache/lazy initialization and chunked prefill, but they describe different failure modes: missing GenerationMixin cache in v5.0.0 vs spurious recompilation. Not the same bug." + }, + { + "left": "issue:43931", + "right": "issue:44060", + "accept": false, + "reason": "These are unrelated model-specific bugs: Qwen3-VL weight shape mismatch vs Qwen3-Next tied-weights warning. Different code paths and symptoms." + }, + { + "left": "issue:44242", + "right": "issue:46897", + "accept": false, + "reason": "Both involve training loss behavior, but one is about MoE load-balancing loss gating and the other is about a Florence2 training-loss double-shift bug. Not the same underlying defect." + }, + { + "left": "issue:29870", + "right": "issue:35141", + "accept": false, + "reason": "One is a BART generate() decoder_attention_mask assignment error; the other is token embedding reinitialization after resize when tie_word_embedding is False. Different mechanisms." + }, + { + "left": "issue:44617", + "right": "issue:45584", + "accept": false, + "reason": "CUDA OOM in Sam3Video and Whisper empty-transcription failure are completely different problems with different models and failure modes." + }, + { + "left": "issue:43644", + "right": "issue:44913", + "accept": false, + "reason": "Both are reload/init-state related, but 43644 is about junk-filled non-persistent buffers while 44913 is a GPTNeoXConfig field not persisting on reload. Different bug classes." + }, + { + "left": "issue:44933", + "right": "issue:45362", + "accept": false, + "reason": "Both concern failures around chat/image-related APIs, but one is a missing import from image_utils and the other is Qwen3.5 chat crashing. Not the same issue." + }, + { + "left": "issue:25251", + "right": "issue:27957", + "accept": false, + "reason": "Pipeline top_k nesting bug and XLMRoberta Flash Attention 2 support are unrelated." + }, + { + "left": "issue:43976", + "right": "issue:44855", + "accept": false, + "reason": "These are both Python-version-related issues, but one is a package support/version mismatch and the other is a TorchScript parsing IndentationError. Different root causes." + }, + { + "left": "issue:45563", + "right": "issue:45584", + "accept": false, + "reason": "Stale num_return_sequences warning in paged generate() is unrelated to Whisper failing on empty transcription after token alignment." + }, + { + "left": "issue:43550", + "right": "issue:44190", + "accept": false, + "reason": "torch.compile + SDPA failure in Bamba-9B-v2 and local dataset loading failure in an example script are unrelated." + }, + { + "left": "issue:44077", + "right": "issue:44877", + "accept": false, + "reason": "Optional post_init in patchtsmixer vs strict config blocking granite_speech loading are different configuration issues." + }, + { + "left": "issue:45381", + "right": "issue:46369", + "accept": false, + "reason": "Both involve video/image models, but one is a Qwen2.5-VL vision_position_ids regression and the other is Sam3-LiteText memory overflow. Not the same bug." + }, + { + "left": "issue:41720", + "right": "issue:45245", + "accept": false, + "reason": "Qwen3 auto device mapping cudaErrorAssert and a categories-count limit runtime error are unrelated." + }, + { + "left": "issue:44479", + "right": "issue:44734", + "accept": false, + "reason": "Video input regression for Qwen VL models and serve KV-cache continuation tensor indexing crash are different code paths." + }, + { + "left": "issue:43335", + "right": "issue:44190", + "accept": false, + "reason": "A SwitchTransformers sparse-layer config bug and a dataset-loading example bug are unrelated." + }, + { + "left": "issue:43065", + "right": "issue:44610", + "accept": false, + "reason": "Sam3PixelDecoder dummy Conv2d and OmDet-Turbo processor/model input-size mismatch are different model implementation issues." + }, + { + "left": "issue:44464", + "right": "issue:45563", + "accept": false, + "reason": "Chunked generation inconsistency with compiled forward and a stale paged-generate warning are not the same defect." + }, + { + "left": "issue:39692", + "right": "issue:44190", + "accept": false, + "reason": "SigLIP2 doc example errors and local dataset loading in a training script are unrelated documentation/example failures." + }, + { + "left": "issue:45127", + "right": "issue:45357", + "accept": false, + "reason": "Both involve embedding/weight handling after LoRA or saving, but one is model collapse after merging LoRA and the other is incorrect visual encoder keys on save. Different bugs." + }, + { + "left": "issue:45381", + "right": "issue:46655", + "accept": false, + "reason": "Both are video-processing related, but one is a Qwen VL vision_position_ids regression and the other is SmolVLM padding calculation. Different root causes." + }, + { + "left": "issue:44112", + "right": "issue:45600", + "accept": false, + "reason": "CI stale device override test and auto_mappings references to removed configs are separate CI breakages." + }, + { + "left": "issue:44779", + "right": "issue:46489", + "accept": false, + "reason": "Deepseek tokenizer regression in v5 and DeepSeek-Coder v1 tokenizer gap in PR #44801 are related only by tokenizer area, not the same bug." + }, + { + "left": "issue:44617", + "right": "issue:46369", + "accept": false, + "reason": "CUDA OOM in Sam3Video and memory overflow in sam3-litetext are both memory issues but for different models and causes." + }, + { + "left": "issue:43295", + "right": "issue:45276", + "accept": false, + "reason": "Processor/tokenizer API regression in v4.57.5 and gemma4 resize_token_embeddings not updating output embeddings are different issues." + }, + { + "left": "issue:44462", + "right": "issue:45907", + "accept": false, + "reason": "Tokenizer.json being ignored by AutoTokenizer and list_repo_files missing revision forwarding are unrelated repository/file-loading bugs." + }, + { + "left": "issue:45290", + "right": "issue:46752", + "accept": false, + "reason": "Both are chat-template crashes on edge cases, but one is assistant messages with tool calls and no content, the other is an empty list IndexError. Different inputs and failure paths." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 39, + "review_unit_count": 61, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-502", + "nodes": [ + "issue:25251", + "issue:27957", + "issue:29870", + "issue:32090", + "issue:35141", + "issue:36246", + "issue:38617", + "issue:41628", + "issue:41950", + "issue:42175", + "issue:42796", + "issue:43404", + "issue:43526", + "issue:43531", + "issue:43540", + "issue:43643", + "issue:43650", + "issue:43673", + "issue:43854", + "issue:44016", + "issue:44112", + "issue:44220", + "issue:44246", + "issue:44279", + "issue:44295", + "issue:44496", + "issue:44521", + "issue:44734", + "issue:44737", + "issue:44938", + "issue:44964", + "issue:45003", + "issue:45216", + "issue:45276", + "issue:45381", + "issue:45542", + "issue:45584", + "issue:45600", + "issue:45632", + "issue:45803", + "issue:46082", + "issue:46121", + "issue:46234", + "issue:46424", + "issue:46506", + "issue:46697", + "issue:46858" + ], + "soft_pairs": [ + "issue:43526|issue:46697", + "issue:45276|issue:45584", + "issue:44938|issue:45542", + "issue:44737|issue:45803", + "issue:43540|issue:44734", + "issue:25251|issue:29870", + "issue:44521|issue:45381", + "issue:43673|issue:46858", + "issue:43854|issue:44964", + "issue:44246|issue:45003", + "issue:45632|issue:46234", + "issue:44964|issue:45600", + "issue:43643|issue:44496", + "issue:44112|issue:44964", + "issue:27957|issue:35141", + "issue:42175|issue:44279", + "issue:46082|issue:46121", + "issue:36246|issue:41628", + "issue:42796|issue:44521", + "issue:36246|issue:38617", + "issue:43404|issue:45216", + "issue:32090|issue:35141", + "issue:44295|issue:46506", + "issue:41950|issue:44016", + "issue:43531|issue:45600", + "issue:43650|issue:44220", + "issue:43673|issue:46424" + ], + "prepared_review_unit_hash": "7ef8e6db4aafa145a106bfdfbc205fe90adcfff933bd828869db81bdca066233", + "component_budget": { + "node_count": 502, + "item_count": 502, + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 27, + "serialized_chars": 35413, + "estimated_input_tokens": 8854, + "estimated_eval_tokens": 17964 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is highly heterogeneous: it groups many unrelated single-issue reports across pipelines, model loading, generation/cache, multimodal processing, packaging, and notebook/docs. I would not merge any of the soft pairs; they are only superficially similar. If one issue must anchor the cluster, issue 44521 is the strongest representative because it has the most ongoing discussion and inbound references, but even that is only a loose anchor.", + "confidence": 0.93, + "canonical_issue_reason": "issue:44521 is the most active, well-scoped, and externally referenced report in the set, so it is the best available anchor even though the cluster is not a true duplicate set.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:44521 has the clearest actionable bug description and the strongest signal (discussion activity and inbound references), making it the best representative issue in this heterogeneous cluster.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43526", + "right": "issue:46697", + "accept": false, + "reason": "Both are image-processor bugs, but one is about reduce_labels collapsing labels and the other is about over-padding divisible images; different code paths and fixes." + }, + { + "left": "issue:45276", + "right": "issue:45584", + "accept": false, + "reason": "Embedding resize/tie-word-embedding behavior is unrelated to Whisper empty-transcription generation after align_special_tokens." + }, + { + "left": "issue:44938", + "right": "issue:45542", + "accept": false, + "reason": "Python 3.14 import loading failure is a different problem from the missing TensorFlow backend in a torch-only install." + }, + { + "left": "issue:44737", + "right": "issue:45803", + "accept": false, + "reason": "XLNet CPU tensor creation in forward pass and Fuyu's bare except swallowing errors are unrelated bugs." + }, + { + "left": "issue:43540", + "right": "issue:44734", + "accept": false, + "reason": "Qwen3OmniMoe video-input validation and transformers serve KV-cache continuation indexing are different failure modes in different paths." + }, + { + "left": "issue:25251", + "right": "issue:29870", + "accept": false, + "reason": "A pipeline returning a nested list and BART's decoder_attention_mask unbound error are not the same bug." + }, + { + "left": "issue:44521", + "right": "issue:45381", + "accept": false, + "reason": "Both are multimodal, but assistant_masks all-zero and wrong vision_position_ids are separate bugs with different internals and fixes." + }, + { + "left": "issue:43673", + "right": "issue:46858", + "accept": false, + "reason": "Both involve cache handling, but one is a missing cache in chunked_prefill and the other is a device-tensor seq_length arithmetic issue for static cache; not mergeable as one fix." + }, + { + "left": "issue:43854", + "right": "issue:44964", + "accept": false, + "reason": "These are load failures for different models with different symptoms; they do not look like the same underlying bug." + }, + { + "left": "issue:44246", + "right": "issue:45003", + "accept": false, + "reason": "Slow import and unsafe sys.modules access are distinct import-time concerns with different fixes." + }, + { + "left": "issue:45632", + "right": "issue:46234", + "accept": false, + "reason": "A trust_remote_code cache-path collision is unrelated to a missing import_utils availability helper." + }, + { + "left": "issue:44964", + "right": "issue:45600", + "accept": false, + "reason": "Model loading failure and stale auto_mappings references are different problems." + }, + { + "left": "issue:43643", + "right": "issue:44496", + "accept": false, + "reason": "Missing fields when trust_remote_code=True and unrecognized model_type in config.json are both config-loading issues, but they are not the same concrete bug." + }, + { + "left": "issue:44112", + "right": "issue:44964", + "accept": false, + "reason": "A stale CI test in GraniteSpeech is unrelated to Phi-4 multimodal model loading." + }, + { + "left": "issue:27957", + "right": "issue:35141", + "accept": false, + "reason": "Flash Attention 2 support for XLMRoberta and output embedding reinitialization after resize_token_embeddings are unrelated." + }, + { + "left": "issue:42175", + "right": "issue:44279", + "accept": false, + "reason": "The first is specifically about TensorFlow not being included with pip install '.[torch]'; the second is a generic dependency complaint." + }, + { + "left": "issue:46082", + "right": "issue:46121", + "accept": false, + "reason": "LlamaConfig head_dim validation and convert_rope_params_to_dict type handling are different configuration-validation bugs." + }, + { + "left": "issue:36246", + "right": "issue:41628", + "accept": false, + "reason": "Both are import errors, but they involve different symbols and modules; not the same missing-export bug." + }, + { + "left": "issue:42796", + "right": "issue:44521", + "accept": false, + "reason": "Tokenizer BPE-marker handling and multimodal assistant_masks are separate bugs in different subsystems." + }, + { + "left": "issue:36246", + "right": "issue:38617", + "accept": false, + "reason": "Different missing imports from different modules; no clear shared root cause." + }, + { + "left": "issue:43404", + "right": "issue:45216", + "accept": false, + "reason": "Mistral3 weight-tying and a Qwen3.5 save_pretrained regression are different model-specific issues." + }, + { + "left": "issue:32090", + "right": "issue:35141", + "accept": false, + "reason": "Trainer broadcast of NoneType and embedding reinitialization after resize are unrelated." + }, + { + "left": "issue:44295", + "right": "issue:46506", + "accept": false, + "reason": "Registering position_ids as a buffer causing read errors is not the same as CPU/CUDA output divergence." + }, + { + "left": "issue:41950", + "right": "issue:44016", + "accept": false, + "reason": "A video-classification pipeline looking for image processors and a notebook syntax error are unrelated." + }, + { + "left": "issue:43531", + "right": "issue:45600", + "accept": false, + "reason": "Qwen3-MoE sliding_window behavior and stale auto_mappings config references are distinct bugs." + }, + { + "left": "issue:43650", + "right": "issue:44220", + "accept": false, + "reason": "A placeholder issue title and a _torch_extract_fbank_features bug are not the same underlying report." + }, + { + "left": "issue:43673", + "right": "issue:46424", + "accept": false, + "reason": "Both mention cache config, but one is a chunked_prefill cache regression and the other is about max_cache_length only applying to Quantized Cache; too different to merge." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 40, + "review_unit_count": 61, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-502", + "nodes": [ + "issue:42907", + "issue:43295", + "issue:43299", + "issue:43381", + "issue:43441", + "issue:43479", + "issue:43526", + "issue:43582", + "issue:43673", + "issue:43824", + "issue:43901", + "issue:43976", + "issue:44016", + "issue:44062", + "issue:44075", + "issue:44220", + "issue:44279", + "issue:44327", + "issue:44336", + "issue:44368", + "issue:44496", + "issue:44521", + "issue:44683", + "issue:44779", + "issue:44792", + "issue:44933", + "issue:44987", + "issue:45042", + "issue:45083", + "issue:45216", + "issue:45335", + "issue:45375", + "issue:45405", + "issue:45446", + "issue:45584", + "issue:45588", + "issue:45593", + "issue:45759", + "issue:46082", + "issue:46123", + "issue:46369", + "issue:46399", + "issue:46424", + "issue:46506", + "issue:46515", + "issue:46693", + "issue:46752", + "issue:46932" + ], + "soft_pairs": [ + "issue:43976|issue:44987", + "issue:43582|issue:46932", + "issue:44016|issue:44220", + "issue:43295|issue:45335", + "issue:45083|issue:46506", + "issue:42907|issue:43441", + "issue:44683|issue:45446", + "issue:43976|issue:44279", + "issue:43526|issue:43901", + "issue:43381|issue:45593", + "issue:43479|issue:45375", + "issue:44496|issue:45042", + "issue:43901|issue:44496", + "issue:43673|issue:44336", + "issue:44792|issue:46424", + "issue:44016|issue:44062", + "issue:44496|issue:45759", + "issue:43824|issue:44933", + "issue:43441|issue:45588", + "issue:46123|issue:46369", + "issue:44327|issue:46752", + "issue:44521|issue:46693", + "issue:44279|issue:46515", + "issue:45335|issue:45584", + "issue:43299|issue:45216", + "issue:44279|issue:45405", + "issue:46082|issue:46399", + "issue:43295|issue:44779", + "issue:44075|issue:44368", + "issue:43976|issue:45405" + ], + "prepared_review_unit_hash": "815a07a11faa3bfa7a280c3519f577b7660bd236d349d5634a27fc1ecc5e3253", + "component_budget": { + "node_count": 502, + "item_count": 502, + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 30, + "serialized_chars": 36682, + "estimated_input_tokens": 9171, + "estimated_eval_tokens": 18598 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is heterogeneous: the issues span unrelated bugs across models, configs, tokenizers, pipelines, and environment compatibility. None of the soft-similarity edges look like true duplicates, so no PR candidate applies and only one issue can serve as a loose anchor.", + "confidence": 0.94, + "canonical_issue_reason": "Issue 44521 is the strongest anchor by discussion/inbound refs and is a concrete user-facing multimodal regression; however, the surrounding items are not true duplicates of it.", + "canonical_pr_reason": null, + "best_issue_reason": "44521 is the best single issue to represent the cluster because it has the clearest active impact signal, but it should not be treated as subsuming the rest of the set.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43976", + "right": "issue:44987", + "accept": false, + "reason": "Both are version/loading failures, but one is Python compatibility and the other is a specific model load failure; different bugs." + }, + { + "left": "issue:43582", + "right": "issue:46932", + "accept": false, + "reason": "Apple Silicon TypeError in warmup vs Gemma4 quantized vision/audio dtype casting; unrelated code paths." + }, + { + "left": "issue:44016", + "right": "issue:44220", + "accept": false, + "reason": "Notebook syntax error and audio feature-extraction helper issue are distinct." + }, + { + "left": "issue:43295", + "right": "issue:45335", + "accept": false, + "reason": "Processor/tokenizer image-handling regression is unrelated to t5gemma embedding resize behavior." + }, + { + "left": "issue:45083", + "right": "issue:46506", + "accept": false, + "reason": "Different bugs: qwen3_omni_moe feature-length helper vs PPDocLayoutV3 CPU/CUDA result mismatch." + }, + { + "left": "issue:42907", + "right": "issue:43441", + "accept": false, + "reason": "Same model family, but save_pretrained dequantization failure is not the FlashAttention runtime failure described here." + }, + { + "left": "issue:44683", + "right": "issue:45446", + "accept": false, + "reason": "Both involve flex_attention, but one is compiled-kernel failure on torch>=2.9 and the other is a version-check/import guard bug." + }, + { + "left": "issue:43976", + "right": "issue:44279", + "accept": false, + "reason": "General dependency/version complaints, but they refer to different failures and causes." + }, + { + "left": "issue:43526", + "right": "issue:43901", + "accept": false, + "reason": "BeitImageProcessorFast label reduction bug vs TextClassificationPipeline docs mismatch; unrelated." + }, + { + "left": "issue:43381", + "right": "issue:45593", + "accept": false, + "reason": "Gradient checkpointing eval-mode restriction is unrelated to D-FINE auxiliary-loss behavior." + }, + { + "left": "issue:43479", + "right": "issue:45375", + "accept": false, + "reason": "Both are config issues, but different models and different missing/default fields." + }, + { + "left": "issue:44496", + "right": "issue:45042", + "accept": false, + "reason": "Unrecognized model/config import error vs torchvision requirement in PIL image processors; different problems." + }, + { + "left": "issue:43901", + "right": "issue:44496", + "accept": false, + "reason": "Docs mismatch in a pipeline vs model/config recognition failure; no shared bug." + }, + { + "left": "issue:43673", + "right": "issue:44336", + "accept": false, + "reason": "Generation cache issue and ANSI logging bug are unrelated." + }, + { + "left": "issue:44792", + "right": "issue:46424", + "accept": false, + "reason": "Failed janus image-generation test vs cache max length semantics; different code paths." + }, + { + "left": "issue:44016", + "right": "issue:44062", + "accept": false, + "reason": "Notebook syntax error is unrelated to AddedToken keyword collision." + }, + { + "left": "issue:44496", + "right": "issue:45759", + "accept": false, + "reason": "Unrecognized model error vs AutoModelForCausalLM text_config unwrap bug; distinct loading failures." + }, + { + "left": "issue:43824", + "right": "issue:44933", + "accept": false, + "reason": "Both are import errors, but for different symbols/modules and different root causes." + }, + { + "left": "issue:43441", + "right": "issue:45588", + "accept": false, + "reason": "Both touch flash attention, but one is model-specific failure and the other is a sink-less s_aux=None crash; not the same bug." + }, + { + "left": "issue:46123", + "right": "issue:46369", + "accept": false, + "reason": "Partial-batch result dropping vs memory overflow; unrelated issues." + }, + { + "left": "issue:44327", + "right": "issue:46752", + "accept": false, + "reason": "QA span decoding top-k bounds error and empty apply_chat_template IndexError are different failures." + }, + { + "left": "issue:44521", + "right": "issue:46693", + "accept": false, + "reason": "Assistant-mask generation bug is unrelated to FA2 long-context performance regression." + }, + { + "left": "issue:44279", + "right": "issue:46515", + "accept": false, + "reason": "Generic dependency issue vs torch 2.4.1 incompatibility; different causes." + }, + { + "left": "issue:45335", + "right": "issue:45584", + "accept": false, + "reason": "Embedding resize bug in t5gemma is unrelated to Whisper empty transcription after token alignment." + }, + { + "left": "issue:43299", + "right": "issue:45216", + "accept": false, + "reason": "Qwen3VL Moe load failure and Qwen3.5 save_pretrained regression are related only by family, not the same concrete bug." + }, + { + "left": "issue:44279", + "right": "issue:45405", + "accept": false, + "reason": "Unrelated dependency problems: package dependency issue vs unreleased MIN_PEFT_VERSION bump." + }, + { + "left": "issue:46082", + "right": "issue:46399", + "accept": false, + "reason": "LlamaConfig head_dim validation and Qwen3.5 kernel registration crash are different issues." + }, + { + "left": "issue:43295", + "right": "issue:44779", + "accept": false, + "reason": "Custom processor.tokenizer/image handling regression is not the same as Deepseek tokenizer incorrect outputs." + }, + { + "left": "issue:44075", + "right": "issue:44368", + "accept": false, + "reason": "SGD argument handling bug and Qwen3.5 warning about tie_word_embeddings are unrelated." + }, + { + "left": "issue:43976", + "right": "issue:45405", + "accept": false, + "reason": "Python-version support issue is unrelated to the PEFT minimum-version bump." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 41, + "review_unit_count": 61, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-502", + "nodes": [ + "issue:33290", + "issue:39692", + "issue:41950", + "issue:42796", + "issue:42994", + "issue:43299", + "issue:43317", + "issue:43441", + "issue:43479", + "issue:43525", + "issue:43577", + "issue:43650", + "issue:43653", + "issue:43854", + "issue:43901", + "issue:44016", + "issue:44242", + "issue:44261", + "issue:44263", + "issue:44291", + "issue:44297", + "issue:44336", + "issue:44351", + "issue:44361", + "issue:44462", + "issue:44466", + "issue:44623", + "issue:44683", + "issue:44749", + "issue:44779", + "issue:45020", + "issue:45081", + "issue:45200", + "issue:45216", + "issue:45276", + "issue:45278", + "issue:45357", + "issue:45542", + "issue:45593", + "issue:45656", + "issue:45812", + "issue:45823", + "issue:45834", + "issue:45907", + "issue:46082", + "issue:46209", + "issue:46512", + "issue:46851" + ], + "soft_pairs": [ + "issue:39692|issue:43901", + "issue:43577|issue:45216", + "issue:44351|issue:45278", + "issue:42796|issue:43650", + "issue:46082|issue:46512", + "issue:44779|issue:45812", + "issue:41950|issue:45020", + "issue:43299|issue:45357", + "issue:44466|issue:45276", + "issue:43299|issue:43854", + "issue:43653|issue:45593", + "issue:41950|issue:45823", + "issue:44242|issue:45200", + "issue:43650|issue:46209", + "issue:44297|issue:44462", + "issue:44683|issue:45542", + "issue:43479|issue:43525", + "issue:44016|issue:44291", + "issue:44361|issue:45593", + "issue:45081|issue:45907", + "issue:33290|issue:45656", + "issue:43441|issue:45834", + "issue:42994|issue:43317", + "issue:44623|issue:44749", + "issue:44261|issue:44263", + "issue:44336|issue:46851" + ], + "prepared_review_unit_hash": "635ec453bacdeedb0e087b9513560c1bdca623ef49a4704c79b6c352313028c6", + "component_budget": { + "node_count": 502, + "item_count": 502, + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 26, + "serialized_chars": 35517, + "estimated_input_tokens": 8880, + "estimated_eval_tokens": 18016 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is heterogeneous: most soft pairs only overlap at a broad subsystem or regression level, but they point to different bugs, models, or code paths. I found no safe duplicate merges.", + "confidence": 0.84, + "canonical_issue_reason": "issue:45278 is the broadest umbrella report in the set (multiple import errors after an upgrade), so it is the best cluster anchor even though it is not a true duplicate of the others.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:45278 has the widest blast radius and the clearest version-regression framing, making it the most suitable representative issue in this mixed cluster.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:39692", + "right": "issue:43901", + "accept": false, + "reason": "Both are documentation-related, but they describe different docs and different failures (SigLIP2 example errors vs TextClassificationPipeline return_all_scores mismatch)." + }, + { + "left": "issue:43577", + "right": "issue:45216", + "accept": false, + "reason": "These are distinct regression reports: one is a Blip2 dtype propagation issue, the other is Qwen3.5 save_pretrained writing incorrect checkpoint keys." + }, + { + "left": "issue:44351", + "right": "issue:45278", + "accept": false, + "reason": "45278 is a broad import-regression report, while 44351 is one specific missing export; not enough evidence they are the same underlying defect." + }, + { + "left": "issue:42796", + "right": "issue:43650", + "accept": false, + "reason": "Completely different problems: tokenizer BPE marker handling vs an unrelated placeholder issue titled 'ADD THE DATA'." + }, + { + "left": "issue:46082", + "right": "issue:46512", + "accept": false, + "reason": "One concerns LlamaConfig head_dim validation, the other stale model.config.dtype after from_config(dtype=...). Different code paths and symptoms." + }, + { + "left": "issue:44779", + "right": "issue:45812", + "accept": false, + "reason": "Both are tokenizer regressions, but for different model families and likely different loading logic; not the same concrete bug." + }, + { + "left": "issue:41950", + "right": "issue:45020", + "accept": false, + "reason": "Video-classification processor lookup and remote_code breakage are unrelated failure modes." + }, + { + "left": "issue:43299", + "right": "issue:45357", + "accept": false, + "reason": "Different Qwen model/version regressions affecting different behaviors: Moe model loading vs save_pretrained visual encoder key serialization." + }, + { + "left": "issue:44466", + "right": "issue:45276", + "accept": false, + "reason": "Serialization of tied weights depending on device is not the same bug as resize_token_embeddings failing to update embeddings in Gemma4." + }, + { + "left": "issue:43299", + "right": "issue:43854", + "accept": false, + "reason": "Both mention loading issues, but one is a Qwen3VL MoE regression and the other is a unit-test load failure for GLM-4.7-Flash." + }, + { + "left": "issue:43653", + "right": "issue:45593", + "accept": false, + "reason": "Tokenizer special-token registration and D-FINE auxiliary-loss behavior are unrelated." + }, + { + "left": "issue:41950", + "right": "issue:45823", + "accept": false, + "reason": "Video-classification processor lookup and Gemma4 device-mapping mismatch are different bugs." + }, + { + "left": "issue:44242", + "right": "issue:45200", + "accept": false, + "reason": "MoE load-balancing loss gating and Gemma4 mm_token_type_ids defaults are separate model-training issues." + }, + { + "left": "issue:43650", + "right": "issue:46209", + "accept": false, + "reason": "No shared bug: one is a vague data request issue, the other is an FP8 inference failure on B200 GPUs." + }, + { + "left": "issue:44297", + "right": "issue:44462", + "accept": false, + "reason": "One is about save_pretrained writing the wrong tokenizer_class, the other about AutoTokenizer ignoring tokenizer.json when loading." + }, + { + "left": "issue:44683", + "right": "issue:45542", + "accept": false, + "reason": "Compiled flex_attention on torch>=2.9 and a TensorFlow-backend import issue are unrelated." + }, + { + "left": "issue:43479", + "right": "issue:43525", + "accept": false, + "reason": "Different config bugs: default multimodal subconfigs being initialized from None versus missing pad_token_id on Llama4Config." + }, + { + "left": "issue:44016", + "right": "issue:44291", + "accept": false, + "reason": "A notebook syntax error is not the same as the init_empty_weights unexpected _is_hf_initialized argument regression." + }, + { + "left": "issue:44361", + "right": "issue:45593", + "accept": false, + "reason": "MLukeTokenizer task failures and D-FINE auxiliary-loss behavior are unrelated." + }, + { + "left": "issue:45081", + "right": "issue:45907", + "accept": false, + "reason": "Mistral regex patching crashes and list_repo_files not forwarding revision are separate utility bugs." + }, + { + "left": "issue:33290", + "right": "issue:45656", + "accept": false, + "reason": "Both involve DeepSpeed and optimizers, but one is OOM with Adafactor while the other is double optimizer stepping; not the same bug." + }, + { + "left": "issue:43441", + "right": "issue:45834", + "accept": false, + "reason": "FlashAttention failure for Ministral-3 and long-ocr index errors in Kosmos2.5 are unrelated." + }, + { + "left": "issue:42994", + "right": "issue:43317", + "accept": false, + "reason": "Saving a quantized model and loading a dequantized model with device_map=auto are different phases and different failures." + }, + { + "left": "issue:44623", + "right": "issue:44749", + "accept": false, + "reason": "processor.save_pretrained missing files is unrelated to the reported slowdown after upgrading Transformers." + }, + { + "left": "issue:44261", + "right": "issue:44263", + "accept": false, + "reason": "These affect different GLM/MoE internals: rms_norm_eps precision behavior versus torch.split return values." + }, + { + "left": "issue:44336", + "right": "issue:46851", + "accept": false, + "reason": "ANSI code emission in loading_report is unrelated to Qwen3.5 variable-length position_ids behavior." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 42, + "review_unit_count": 61, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-502", + "nodes": [ + "issue:41628", + "issue:42175", + "issue:42796", + "issue:43441", + "issue:43525", + "issue:43526", + "issue:43643", + "issue:43673", + "issue:43723", + "issue:43761", + "issue:43784", + "issue:43824", + "issue:43881", + "issue:43901", + "issue:43976", + "issue:44016", + "issue:44162", + "issue:44188", + "issue:44246", + "issue:44297", + "issue:44336", + "issue:44373", + "issue:44496", + "issue:44589", + "issue:44623", + "issue:44671", + "issue:44779", + "issue:45003", + "issue:45083", + "issue:45103", + "issue:45325", + "issue:45362", + "issue:45372", + "issue:45397", + "issue:45542", + "issue:45561", + "issue:45563", + "issue:45600", + "issue:45712", + "issue:45800", + "issue:45823", + "issue:45859", + "issue:46291", + "issue:46399", + "issue:46655", + "issue:46726", + "issue:46851" + ], + "soft_pairs": [ + "issue:45103|issue:45372", + "issue:45600|issue:45712", + "issue:45859|issue:46851", + "issue:42175|issue:43976", + "issue:44623|issue:45397", + "issue:44671|issue:44779", + "issue:43723|issue:44162", + "issue:43901|issue:44246", + "issue:43673|issue:43761", + "issue:43824|issue:43976", + "issue:45542|issue:45800", + "issue:42796|issue:43441", + "issue:43525|issue:43881", + "issue:46291|issue:46399", + "issue:44188|issue:44336", + "issue:44016|issue:45823", + "issue:44336|issue:44373", + "issue:44589|issue:45362", + "issue:43525|issue:43643", + "issue:46655|issue:46726", + "issue:45325|issue:46851", + "issue:41628|issue:43976", + "issue:44297|issue:45397", + "issue:43784|issue:45542", + "issue:45083|issue:46851", + "issue:44246|issue:44496", + "issue:46399|issue:46851", + "issue:45561|issue:45563", + "issue:45103|issue:46399", + "issue:43526|issue:45003" + ], + "prepared_review_unit_hash": "e70a0a213fd943aa329f9382ae73c4c63ec910090da111e7e5243488ec039a06", + "component_budget": { + "node_count": 502, + "item_count": 502, + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 30, + "serialized_chars": 36124, + "estimated_input_tokens": 9031, + "estimated_eval_tokens": 18318 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is heterogeneous: import/load failures, tokenizer/model-specific regressions, docs issues, and packaging/backend problems. The soft pairs share broad vocabulary or model families, but none look like the same underlying bug or mergeable change.", + "confidence": 0.91, + "canonical_issue_reason": "No single canonical issue: the items are not duplicates of one bug, but a mixed set of unrelated Transformers regressions and maintenance issues.", + "canonical_pr_reason": null, + "best_issue_reason": "No single best issue to represent the cluster; each candidate is too specific to its own model, backend, or subsystem.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:45103", + "right": "issue:45372", + "accept": false, + "reason": "Both are import-time failures, but one is auto-docstring/annotations handling and the other is a missing `ReasoningEffort` dependency for Gemma 4 processor loading." + }, + { + "left": "issue:45600", + "right": "issue:45712", + "accept": false, + "reason": "Both touch repo hygiene, but one is stale `auto_mappings.py` config references and the other is leftover dummy PT classes; different files and fixes." + }, + { + "left": "issue:45859", + "right": "issue:46851", + "accept": false, + "reason": "Same Qwen3.5 family, but one is CPU offload compatibility in rotary embeddings and the other is variable-length `position_ids` handling." + }, + { + "left": "issue:42175", + "right": "issue:43976", + "accept": false, + "reason": "One is missing TensorFlow from `.[torch]` install extras; the other is a Python 3.9/3.10 compatibility issue. Not the same problem." + }, + { + "left": "issue:44623", + "right": "issue:45397", + "accept": false, + "reason": "`processor.save_pretrained()` missing files versus a `from_pretrained()` zero3 loading failure are different code paths and failure modes." + }, + { + "left": "issue:44671", + "right": "issue:44779", + "accept": false, + "reason": "Both are output-correctness bugs, but on different models and tokenization/LM behavior; no shared concrete code path." + }, + { + "left": "issue:43723", + "right": "issue:44162", + "accept": false, + "reason": "Generic tokenizer-loading regression versus ESM2 model breakage; they are model-specific and not the same underlying bug." + }, + { + "left": "issue:43901", + "right": "issue:44246", + "accept": false, + "reason": "Documentation mismatch about `return_all_scores` versus intermittent import slowness are unrelated." + }, + { + "left": "issue:43673", + "right": "issue:43761", + "accept": false, + "reason": "Chunked prefill cache regression in generation versus CLIP hidden-states regression; both are v5 issues, but different model/runtime paths." + }, + { + "left": "issue:43824", + "right": "issue:43976", + "accept": false, + "reason": "Missing Qwen2.5-VL class export versus a Python version compatibility break; no shared bug or fix." + }, + { + "left": "issue:45542", + "right": "issue:45800", + "accept": false, + "reason": "Both mention installation/runtime compatibility, but one is TensorFlow backend detection and the other is a torch/transformers version mismatch." + }, + { + "left": "issue:42796", + "right": "issue:43441", + "accept": false, + "reason": "Same Ministral family, but tokenizer BPE-marker handling and FlashAttention failure are different subsystems." + }, + { + "left": "issue:43525", + "right": "issue:43881", + "accept": false, + "reason": "Llama4 config missing `pad_token_id` and glm-4v-9b loading failure are unrelated model-loading bugs." + }, + { + "left": "issue:46291", + "right": "issue:46399", + "accept": false, + "reason": "Both involve kernels, but one is a dependency/version requirement on importing pipelines and the other is a missing decorator registration in Qwen3.5 code." + }, + { + "left": "issue:44188", + "right": "issue:44336", + "accept": false, + "reason": "Attention-kernel divergence under `torch.compile` versus ANSI codes in a loading report are unrelated." + }, + { + "left": "issue:44016", + "right": "issue:45823", + "accept": false, + "reason": "Notebook syntax error and Gemma4 device-mismatch during forward are completely different issues." + }, + { + "left": "issue:44336", + "right": "issue:44373", + "accept": false, + "reason": "One is terminal-color output behavior, the other is a wrong docstring; no shared underlying bug." + }, + { + "left": "issue:44589", + "right": "issue:45362", + "accept": false, + "reason": "Float8 storage lookup failure versus Qwen3.5 chat crash are unrelated runtime/model problems." + }, + { + "left": "issue:43525", + "right": "issue:43643", + "accept": false, + "reason": "Both mention config fields, but one is Llama4-specific `pad_token_id` and the other is `trust_remote_code` returning incomplete AutoConfig objects." + }, + { + "left": "issue:46655", + "right": "issue:46726", + "accept": false, + "reason": "Both are processor/image-related, but one is a padding-calculation bug and the other is an explicit missing-vision-backend error path." + }, + { + "left": "issue:45325", + "right": "issue:46851", + "accept": false, + "reason": "Both concern Qwen position IDs, but one is a Qwen2.5-VL rope scaling bug for still images and the other is Qwen3.5 variable-length input handling." + }, + { + "left": "issue:41628", + "right": "issue:43976", + "accept": false, + "reason": "`AutoImageProcessor` import failure and Python version incompatibility are unrelated." + }, + { + "left": "issue:44297", + "right": "issue:45397", + "accept": false, + "reason": "Tokenizer save metadata bug versus Gemma4 zero3 loading failure are different subsystems." + }, + { + "left": "issue:43784", + "right": "issue:45542", + "accept": false, + "reason": "Sentence-transformers import NameError and TensorFlow backend absence are unrelated packaging/runtime issues." + }, + { + "left": "issue:45083", + "right": "issue:46851", + "accept": false, + "reason": "Both are Qwen-related position/length logic, but they affect different models and helper functions." + }, + { + "left": "issue:44246", + "right": "issue:44496", + "accept": false, + "reason": "Intermittent import slowness and unrecognized model config loading are not the same bug." + }, + { + "left": "issue:46399", + "right": "issue:46851", + "accept": false, + "reason": "Both are Qwen3.5 regressions, but one is kernel registration failure and the other is variable-length `position_ids`; distinct code paths." + }, + { + "left": "issue:45561", + "right": "issue:45563", + "accept": false, + "reason": "A pytest-xdist file race and a stale generation warning are unrelated test/runtime issues." + }, + { + "left": "issue:45103", + "right": "issue:46399", + "accept": false, + "reason": "Auto-docstring annotation handling and Qwen3.5 kernel registration failure have no concrete overlap." + }, + { + "left": "issue:43526", + "right": "issue:45003", + "accept": false, + "reason": "BEiTImageProcessorFast label reduction and unsafe `sys.modules` access in modeling_utils are unrelated bugs." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 43, + "review_unit_count": 61, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-502", + "nodes": [ + "issue:36246", + "issue:41950", + "issue:42491", + "issue:42757", + "issue:42886", + "issue:43257", + "issue:43262", + "issue:43299", + "issue:43531", + "issue:43540", + "issue:43550", + "issue:43901", + "issue:44162", + "issue:44220", + "issue:44242", + "issue:44246", + "issue:44295", + "issue:44336", + "issue:44351", + "issue:44451", + "issue:44479", + "issue:44485", + "issue:44496", + "issue:44749", + "issue:44779", + "issue:44871", + "issue:45003", + "issue:45072", + "issue:45137", + "issue:45200", + "issue:45292", + "issue:45362", + "issue:45375", + "issue:45397", + "issue:45561", + "issue:45588", + "issue:45593", + "issue:45759", + "issue:46132", + "issue:46506", + "issue:46614", + "issue:46655", + "issue:46665", + "issue:46752", + "issue:46821", + "issue:46829", + "issue:46851" + ], + "soft_pairs": [ + "issue:43540|issue:46752", + "issue:46614|issue:46665", + "issue:45072|issue:46829", + "issue:44295|issue:45397", + "issue:43531|issue:45362", + "issue:42491|issue:43299", + "issue:44246|issue:44871", + "issue:44242|issue:45593", + "issue:44336|issue:44485", + "issue:45588|issue:46851", + "issue:45003|issue:46821", + "issue:42886|issue:46132", + "issue:44451|issue:46752", + "issue:43550|issue:46506", + "issue:44779|issue:45292", + "issue:36246|issue:42757", + "issue:36246|issue:44351", + "issue:41950|issue:44220", + "issue:43901|issue:45003", + "issue:43257|issue:45137", + "issue:45003|issue:45561", + "issue:44496|issue:45003", + "issue:43540|issue:46655", + "issue:44162|issue:45200", + "issue:45375|issue:45759", + "issue:43262|issue:44246", + "issue:44295|issue:44749", + "issue:44479|issue:46655" + ], + "prepared_review_unit_hash": "6f9ff847ee63b51f42b71a5cf7bfaf5a18f406adc19217819b6d5023ccbba978", + "component_budget": { + "node_count": 502, + "item_count": 502, + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 28, + "serialized_chars": 35502, + "estimated_input_tokens": 8876, + "estimated_eval_tokens": 18008 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly heterogeneous issue reports. The best anchor is the broad v5.3.0 video regression report (44479); the soft pairs mostly share a subsystem or model family but describe different concrete failures, so none should be merged as duplicates.", + "confidence": 0.72, + "canonical_issue_reason": "issue:44479 is the strongest representative because it describes a concrete, version-specific regression affecting multiple Qwen video/model variants, with a clear shared code path and broad enough scope to anchor related reports.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:44479 is the best standalone issue: it is specific, impactful, and frames a regression that plausibly subsumes adjacent video-input complaints better than the narrower model-specific reports.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43540", + "right": "issue:46752", + "accept": false, + "reason": "Different failures: Qwen3OmniMoe video input ValueError vs empty-list apply_chat_template IndexError." + }, + { + "left": "issue:46614", + "right": "issue:46665", + "accept": false, + "reason": "Stale phonemizer backend state and from_pretrained behavior are unrelated code paths." + }, + { + "left": "issue:45072", + "right": "issue:46829", + "accept": false, + "reason": "One is a bfloat16 dtype mismatch; the other is a bad hidden_states field in an output object." + }, + { + "left": "issue:44295", + "right": "issue:45397", + "accept": false, + "reason": "Registered buffer access error and ZeRO3 from_pretrained crash do not point to the same bug." + }, + { + "left": "issue:43531", + "right": "issue:45362", + "accept": false, + "reason": "Sliding-window handling and a Qwen3.5 chat crash are different model issues." + }, + { + "left": "issue:42491", + "right": "issue:43299", + "accept": false, + "reason": "Both involve Qwen MoE loading, but one is LoRA compatibility across HF versions and the other is Qwen3VL MoE loading; not the same concrete defect." + }, + { + "left": "issue:44246", + "right": "issue:44871", + "accept": false, + "reason": "Load-balancing loss omission and eos_token_id config inconsistency are unrelated." + }, + { + "left": "issue:44242", + "right": "issue:45593", + "accept": false, + "reason": "Both mention auxiliary losses, but they affect different models and triggers; not mergeable as one bug." + }, + { + "left": "issue:44336", + "right": "issue:44485", + "accept": false, + "reason": "ANSI output formatting and RoPE implementation are unrelated." + }, + { + "left": "issue:45588", + "right": "issue:46851", + "accept": false, + "reason": "Flash-attention sinkless-model crash and variable-length position_ids failure are different code paths." + }, + { + "left": "issue:45003", + "right": "issue:46821", + "accept": false, + "reason": "Unsafe sys.modules access and TimesFM window_size attribute error are unrelated." + }, + { + "left": "issue:42886", + "right": "issue:46132", + "accept": false, + "reason": "Offline cache loading and AttentionInterface.register behavior changes are different subsystems." + }, + { + "left": "issue:44451", + "right": "issue:46752", + "accept": false, + "reason": "Model loading failure and empty-list chat template crash are not the same issue." + }, + { + "left": "issue:43550", + "right": "issue:46506", + "accept": false, + "reason": "torch.compile/SDPA failure and CPU-vs-CUDA output mismatch are different defects." + }, + { + "left": "issue:44779", + "right": "issue:45292", + "accept": false, + "reason": "Tokenizer correctness regression and resize_token_embeddings output-sync bug are distinct." + }, + { + "left": "issue:36246", + "right": "issue:42757", + "accept": false, + "reason": "Import errors for different symbols from different modules; no shared root cause is evident." + }, + { + "left": "issue:36246", + "right": "issue:44351", + "accept": false, + "reason": "Qwen2_5_VLImageProcessor export failure and HybridCache export failure are separate import problems." + }, + { + "left": "issue:41950", + "right": "issue:44220", + "accept": false, + "reason": "Pipeline image-processor lookup and _torch_extract_fbank_features() are different bugs." + }, + { + "left": "issue:43901", + "right": "issue:45003", + "accept": false, + "reason": "Docs/behavior mismatch in TextClassificationPipeline and sys.modules access are unrelated." + }, + { + "left": "issue:43257", + "right": "issue:45137", + "accept": false, + "reason": "DeepSpeed weight conversion and ZeRO3 deque underflow are both distributed-training related but not the same bug." + }, + { + "left": "issue:45003", + "right": "issue:45561", + "accept": false, + "reason": "sys.modules access and pytest-xdist captured_info.txt race are unrelated." + }, + { + "left": "issue:44496", + "right": "issue:45003", + "accept": false, + "reason": "Unrecognized model config and unsafe sys.modules access are different failure modes." + }, + { + "left": "issue:43540", + "right": "issue:46655", + "accept": false, + "reason": "Both are video-processing reports, but they target different models and distinct processing bugs." + }, + { + "left": "issue:44162", + "right": "issue:45200", + "accept": false, + "reason": "ESM2 breakage and Gemma 4 mm_token_type_ids default are unrelated." + }, + { + "left": "issue:45375", + "right": "issue:45759", + "accept": false, + "reason": "Both touch Qwen 3.5 multimodal config handling, but one is strict-field dropping and the other is from_config text_config unwrapping; not the same concrete change." + }, + { + "left": "issue:43262", + "right": "issue:44246", + "accept": false, + "reason": "Audio chat-template sampling-rate default and router-logits load-balancing loss are unrelated." + }, + { + "left": "issue:44295", + "right": "issue:44749", + "accept": false, + "reason": "A buffer-related runtime error and a Chinese performance regression report are not duplicates." + }, + { + "left": "issue:44479", + "right": "issue:46655", + "accept": false, + "reason": "Both concern video regressions, but they affect different model families and likely different padding/processor code paths." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 44, + "review_unit_count": 61, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-502", + "nodes": [ + "issue:29870", + "issue:30333", + "issue:36246", + "issue:42199", + "issue:42886", + "issue:43065", + "issue:43295", + "issue:43352", + "issue:43408", + "issue:43475", + "issue:43526", + "issue:43596", + "issue:43653", + "issue:44038", + "issue:44060", + "issue:44077", + "issue:44162", + "issue:44242", + "issue:44368", + "issue:44466", + "issue:44479", + "issue:44623", + "issue:44857", + "issue:44877", + "issue:44977", + "issue:44987", + "issue:45137", + "issue:45229", + "issue:45250", + "issue:45292", + "issue:45335", + "issue:45375", + "issue:45397", + "issue:45412", + "issue:45459", + "issue:45584", + "issue:45600", + "issue:45656", + "issue:45859", + "issue:46123", + "issue:46170", + "issue:46369", + "issue:46399", + "issue:46506", + "issue:46726", + "issue:46851", + "issue:46894", + "issue:46897" + ], + "soft_pairs": [ + "issue:43475|issue:46369", + "issue:45859|issue:46399", + "issue:45137|issue:45600", + "issue:36246|issue:43596", + "issue:44877|issue:45375", + "issue:43295|issue:45292", + "issue:45292|issue:45584", + "issue:44466|issue:45335", + "issue:44623|issue:46726", + "issue:44242|issue:44857", + "issue:43526|issue:46897", + "issue:44038|issue:46506", + "issue:44987|issue:45459", + "issue:29870|issue:30333", + "issue:43352|issue:45250", + "issue:42199|issue:46851", + "issue:43526|issue:46894", + "issue:44060|issue:44077", + "issue:44368|issue:44977", + "issue:43653|issue:44162", + "issue:45412|issue:46123", + "issue:44479|issue:46170", + "issue:42886|issue:45229", + "issue:45397|issue:45656", + "issue:43065|issue:43408" + ], + "prepared_review_unit_hash": "a01bf47f5c52e0ff5a1763246cf7a1c8aaf2f87a49615b50d11aa7eede42da6e", + "component_budget": { + "node_count": 502, + "item_count": 502, + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 35314, + "estimated_input_tokens": 8829, + "estimated_eval_tokens": 17914 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a loose collection of unrelated Transformers bug reports. The soft pairs mostly share broad themes (SAM3, Qwen, DeepSpeed, tied embeddings, tokenizer/config issues) but not the same concrete defect or fix path, so none should be merged as duplicates.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43475", + "right": "issue:46369", + "accept": false, + "reason": "Same SAM3 family, but one is an attribute error in vision encoder outputs and the other is a memory overflow in litetext; different failures and fixes." + }, + { + "left": "issue:45859", + "right": "issue:46399", + "accept": false, + "reason": "Both touch Qwen3.5 rotary/kernel code, but one is CPU offload incompatibility and the other is a missing decorator causing kernel registration failure." + }, + { + "left": "issue:45137", + "right": "issue:45600", + "accept": false, + "reason": "DeepSpeed ZeRO3 deque error vs a broken auto_mappings CI reference; no shared bug path." + }, + { + "left": "issue:36246", + "right": "issue:43596", + "accept": false, + "reason": "ImportError for a missing Qwen2.5-VL image processor is unrelated to the BertModel/Zero3 index error." + }, + { + "left": "issue:44877", + "right": "issue:45375", + "accept": false, + "reason": "Both involve strict config handling, but they concern different models and different missing fields, so not the same defect." + }, + { + "left": "issue:43295", + "right": "issue:45292", + "accept": false, + "reason": "Processor/tokenizer regression and resize_token_embeddings propagation are distinct code paths." + }, + { + "left": "issue:45292", + "right": "issue:45584", + "accept": false, + "reason": "Both mention embeddings/tokens, but one is a resize propagation bug while the other is Whisper empty-transcription handling after special-token alignment." + }, + { + "left": "issue:44466", + "right": "issue:45335", + "accept": false, + "reason": "Both are tied-embedding related, but one is inconsistent serialization and the other is resize_token_embeddings not updating decoder embeddings." + }, + { + "left": "issue:44623", + "right": "issue:46726", + "accept": false, + "reason": "Missing saved files and missing vision backend are different save/load problems." + }, + { + "left": "issue:44242", + "right": "issue:44857", + "accept": false, + "reason": "Load-balancing loss omission in an MoE model is unrelated to an AMP/CUDA crash in LwDetrImageLoss." + }, + { + "left": "issue:43526", + "right": "issue:46897", + "accept": false, + "reason": "BEiT label reduction and Florence2 training-loss double-shift are different model-specific bugs." + }, + { + "left": "issue:44038", + "right": "issue:46506", + "accept": false, + "reason": "Qwen3-VL-Moe compatibility with Transformers 5 is unrelated to PPDocLayoutV3 CPU/CUDA result differences." + }, + { + "left": "issue:44987", + "right": "issue:45459", + "accept": false, + "reason": "Loading failures on transformers>=5.1.0 and tokenizer decode-error masking are separate issues." + }, + { + "left": "issue:29870", + "right": "issue:30333", + "accept": false, + "reason": "BART generate() decoder mask bug and MLFlow job-status reporting are unrelated." + }, + { + "left": "issue:43352", + "right": "issue:45250", + "accept": false, + "reason": "One is a specific model not supporting Flash Attention 2.0; the other is only a generic Flash Attention topic." + }, + { + "left": "issue:42199", + "right": "issue:46851", + "accept": false, + "reason": "DETR cardinality/background-class handling and Qwen3.5 variable-length position_ids are unrelated." + }, + { + "left": "issue:44060", + "right": "issue:44077", + "accept": false, + "reason": "Tied-weights warning in Qwen3-Next and patchtsmixer post_init policy are different issues." + }, + { + "left": "issue:44368", + "right": "issue:44977", + "accept": false, + "reason": "One is a tied-embeddings warning during LoRA fine-tuning; the other is a flash-attention generation regression." + }, + { + "left": "issue:43653", + "right": "issue:44162", + "accept": false, + "reason": "BigBirdTokenizer special-token registration and ESM2 breakage do not share the same code path." + }, + { + "left": "issue:45412", + "right": "issue:46123", + "accept": false, + "reason": "Memory not released on deletion and final partial-batch results being dropped are different lifecycle bugs." + }, + { + "left": "issue:44479", + "right": "issue:46170", + "accept": false, + "reason": "Video-input regression and Windows ROCm torch.distributed import/type errors are unrelated." + }, + { + "left": "issue:42886", + "right": "issue:45229", + "accept": false, + "reason": "Offline tokenizer cache loading and Gemma4 CUDA OOM are not the same underlying defect." + }, + { + "left": "issue:45397", + "right": "issue:45656", + "accept": false, + "reason": "Gemma-4 ZeRO3 from_pretrained and a double optimizer step under DeepSpeed are different bugs despite both mentioning DeepSpeed." + }, + { + "left": "issue:43065", + "right": "issue:43408", + "accept": false, + "reason": "Dummy Conv2d in Sam3PixelDecoder and a sam3_video/sam3_tracker config mismatch are separate SAM3 issues." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 45, + "review_unit_count": 61, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-502", + "nodes": [ + "issue:42994", + "issue:43065", + "issue:43232", + "issue:43441", + "issue:43475", + "issue:43482", + "issue:43540", + "issue:43643", + "issue:43646", + "issue:43673", + "issue:43701", + "issue:43723", + "issue:43906", + "issue:43931", + "issue:43994", + "issue:44162", + "issue:44188", + "issue:44297", + "issue:44361", + "issue:44368", + "issue:44483", + "issue:44623", + "issue:44749", + "issue:44871", + "issue:44877", + "issue:44995", + "issue:45103", + "issue:45200", + "issue:45375", + "issue:45478", + "issue:45563", + "issue:45588", + "issue:45656", + "issue:45685", + "issue:45987", + "issue:46082", + "issue:46123", + "issue:46132", + "issue:46170", + "issue:46202", + "issue:46421", + "issue:46431", + "issue:46710", + "issue:46726", + "issue:46752", + "issue:46832", + "issue:46851", + "issue:46858" + ], + "soft_pairs": [ + "issue:43475|issue:43540", + "issue:44877|issue:46170", + "issue:46170|issue:46832", + "issue:43723|issue:46710", + "issue:44368|issue:45563", + "issue:44995|issue:45200", + "issue:43931|issue:44368", + "issue:44188|issue:46202", + "issue:44483|issue:46202", + "issue:43906|issue:46202", + "issue:45987|issue:46082", + "issue:44995|issue:45588", + "issue:44188|issue:46421", + "issue:45685|issue:46431", + "issue:45103|issue:46858", + "issue:44297|issue:44623", + "issue:43065|issue:46123", + "issue:44361|issue:44749", + "issue:43994|issue:46726", + "issue:44749|issue:45656", + "issue:43701|issue:44749", + "issue:44871|issue:46132", + "issue:44623|issue:45656", + "issue:43646|issue:44162", + "issue:43441|issue:46752", + "issue:45375|issue:45478", + "issue:42994|issue:43482", + "issue:43232|issue:43673", + "issue:43475|issue:43643", + "issue:44188|issue:46851" + ], + "prepared_review_unit_hash": "9b131206d965e0f5605fb4d5c0fa6448ddc18d13c9b12e7f726de968f22a68b8", + "component_budget": { + "node_count": 502, + "item_count": 502, + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 30, + "serialized_chars": 36815, + "estimated_input_tokens": 9204, + "estimated_eval_tokens": 18664 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Most pairs are only superficially similar; there is one clear duplicate pair around Qwen3.5 MoE config loading. Everything else appears to be separate bugs or regressions in different code paths.", + "confidence": 0.81, + "canonical_issue_reason": "issue:45375 is the most specific root-cause report in the only credible duplicate pair: it names the missing `deepstack_visual_indexes` field and the `@strict` drop behavior directly.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:45375 is the best representative issue for this cluster because it captures the concrete underlying config-loading bug rather than the broader downstream symptom.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43475", + "right": "issue:43540", + "accept": false, + "reason": "Different models and failures: missing `fpn_position_embeddings` in SAM 3 video vs a video-input `ValueError` in Qwen3OmniMoe." + }, + { + "left": "issue:44877", + "right": "issue:46170", + "accept": false, + "reason": "Unrelated problems: strict config loading for `granite_speech` vs a Windows/ROCm `torch.distributed` import/type issue." + }, + { + "left": "issue:46170", + "right": "issue:46832", + "accept": false, + "reason": "Different subsystems and symptoms: platform import dependency issue vs RT-DETR feature-level mismatch in forward." + }, + { + "left": "issue:43723", + "right": "issue:46710", + "accept": false, + "reason": "Tokenizer loading failure and an accuracy regression are not the same bug or code path." + }, + { + "left": "issue:44368", + "right": "issue:45563", + "accept": false, + "reason": "Both mention warnings, but they are different warnings in different generation/tokenization paths." + }, + { + "left": "issue:44995", + "right": "issue:45200", + "accept": false, + "reason": "Stale cache crash in GlmMoeDsa is unrelated to Gemma 4 text-only fine-tuning requiring `mm_token_type_ids`." + }, + { + "left": "issue:43931", + "right": "issue:44368", + "accept": false, + "reason": "Shape-mismatch loading failure for Qwen3-VL is unrelated to a tie-embeddings warning during Qwen3.5 fine-tuning." + }, + { + "left": "issue:44188", + "right": "issue:46202", + "accept": false, + "reason": "Torch.compile attention-kernel divergence is unrelated to Gemma pipeline roundtrip token handling." + }, + { + "left": "issue:44483", + "right": "issue:46202", + "accept": false, + "reason": "API request rejection in `/v1/chat/completions` is a different bug from Gemma token roundtripping." + }, + { + "left": "issue:43906", + "right": "issue:46202", + "accept": false, + "reason": "An isolated reproduction of an older issue is not the same as the Gemma pipeline roundtrip failure." + }, + { + "left": "issue:45987", + "right": "issue:46082", + "accept": false, + "reason": "`StaticCache.get_seq_length()` return-type bug and a `LlamaConfig` validation error are distinct issues." + }, + { + "left": "issue:44995", + "right": "issue:45588", + "accept": false, + "reason": "Different failure modes: stale indexer cache crash vs `flash_attention.py` `AttributeError` on `s_aux=None`." + }, + { + "left": "issue:44188", + "right": "issue:46421", + "accept": false, + "reason": "Both involve compile/cache themes, but one is attention-kernel divergence and the other is spurious recompilation from lazy cache init." + }, + { + "left": "issue:45685", + "right": "issue:46431", + "accept": false, + "reason": "MPS histogram kernel limitation is unrelated to missing chat template support in `apply_chat_template`." + }, + { + "left": "issue:45103", + "right": "issue:46858", + "accept": false, + "reason": "Docstring annotation handling crash is unrelated to static-cache slice arithmetic failing on non-CUDA backends." + }, + { + "left": "issue:44297", + "right": "issue:44623", + "accept": false, + "reason": "Both are save-pretrained problems, but tokenizer config mismatch and missing processor files are different bugs." + }, + { + "left": "issue:43065", + "right": "issue:46123", + "accept": false, + "reason": "A dummy Conv2d in SAM 3 pixel decoding is unrelated to a masking pipeline partial-batch result drop." + }, + { + "left": "issue:44361", + "right": "issue:44749", + "accept": false, + "reason": "Tokenizer AttributeError and transformer-upgrade slowdown are unrelated." + }, + { + "left": "issue:43994", + "right": "issue:46726", + "accept": false, + "reason": "Wrong outputs from siglip2 and missing-vision-backend processor detection are different loading/runtime issues." + }, + { + "left": "issue:44749", + "right": "issue:45656", + "accept": false, + "reason": "Data filtering slowdown after upgrade is unrelated to DeepSpeed double optimizer stepping." + }, + { + "left": "issue:43701", + "right": "issue:44749", + "accept": false, + "reason": "Checkpoint key mismatch and performance regression are not the same underlying bug." + }, + { + "left": "issue:44871", + "right": "issue:46132", + "accept": false, + "reason": "Gemma eos-token-id config inconsistency is unrelated to AttentionInterface registration behavior changes." + }, + { + "left": "issue:44623", + "right": "issue:45656", + "accept": false, + "reason": "Missing files on `processor.save_pretrained` is unrelated to the optimizer being stepped twice under DeepSpeed." + }, + { + "left": "issue:43646", + "right": "issue:44162", + "accept": false, + "reason": "Generic v5 custom-model initialization breakage and ESM2 breakage are different reports with different concrete causes." + }, + { + "left": "issue:43441", + "right": "issue:46752", + "accept": false, + "reason": "FlashAttention failure for Ministral-3 and empty-chat-template `IndexError` are unrelated." + }, + { + "left": "issue:45375", + "right": "issue:45478", + "accept": true, + "reason": "Both describe the same Qwen3.5 MoE config-loading bug: a required field is dropped/ignored under strict loading, producing downstream `from_pretrained` errors." + }, + { + "left": "issue:42994", + "right": "issue:43482", + "accept": false, + "reason": "Quantized model saving failure and GGUF loading failure are different directions and different model formats." + }, + { + "left": "issue:43232", + "right": "issue:43673", + "accept": false, + "reason": "These are separate generation-related regressions with different triggers: `sync_gpus`/`_update_model_kwargs_for_generation` vs chunked-prefill cache handling." + }, + { + "left": "issue:43475", + "right": "issue:43643", + "accept": false, + "reason": "Missing SAM3 vision encoder field and missing fields from `AutoConfig.from_pretrained` are different config-loading issues." + }, + { + "left": "issue:44188", + "right": "issue:46851", + "accept": false, + "reason": "Attention-kernel divergence under compile is unrelated to Qwen3.5 variable-length position-id handling." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is well grounded in the packet metadata. The single accepted pair (45375/45478) is a plausible duplicate rooted in the same Qwen3.5 MoE config-loading failure, and the other soft-edge rejections are conservative based on distinct titles/symptoms." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 46, + "review_unit_count": 61, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-502", + "nodes": [ + "issue:28282", + "issue:30990", + "issue:36010", + "issue:39692", + "issue:41628", + "issue:42371", + "issue:42673", + "issue:42994", + "issue:43278", + "issue:43475", + "issue:43525", + "issue:43704", + "issue:43723", + "issue:43784", + "issue:43824", + "issue:43873", + "issue:43986", + "issue:44038", + "issue:44222", + "issue:44351", + "issue:44373", + "issue:44623", + "issue:44857", + "issue:44869", + "issue:44933", + "issue:44945", + "issue:44964", + "issue:45081", + "issue:45092", + "issue:45103", + "issue:45137", + "issue:45230", + "issue:45310", + "issue:45356", + "issue:45375", + "issue:45397", + "issue:45406", + "issue:45593", + "issue:45753", + "issue:46202", + "issue:46326", + "issue:46399", + "issue:46463", + "issue:46701", + "issue:46726", + "issue:46821", + "issue:46851", + "issue:46877" + ], + "soft_pairs": [ + "issue:28282|issue:30990", + "issue:41628|issue:44933", + "issue:45230|issue:45397", + "issue:45310|issue:45375", + "issue:43475|issue:46821", + "issue:44964|issue:45137", + "issue:43986|issue:46726", + "issue:44869|issue:45356", + "issue:44038|issue:44623", + "issue:39692|issue:46726", + "issue:46463|issue:46877", + "issue:44373|issue:46851", + "issue:42994|issue:43525", + "issue:42371|issue:45092", + "issue:43723|issue:46326", + "issue:45406|issue:46202", + "issue:43873|issue:44945", + "issue:44945|issue:45753", + "issue:41628|issue:44351", + "issue:43784|issue:46399", + "issue:44857|issue:46877", + "issue:45593|issue:46877", + "issue:42673|issue:43704", + "issue:44038|issue:45310", + "issue:44222|issue:45310", + "issue:36010|issue:41628", + "issue:43278|issue:45137", + "issue:42994|issue:43824", + "issue:46202|issue:46701", + "issue:45081|issue:45103" + ], + "prepared_review_unit_hash": "b541748974b90acd393daaaf0e8fb4e0def7fa148c3f1b54b188e6c31c0decfc", + "component_budget": { + "node_count": 502, + "item_count": 502, + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 30, + "serialized_chars": 36407, + "estimated_input_tokens": 9102, + "estimated_eval_tokens": 18460 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a loose grab-bag of unrelated transformer/model bugs, not a tight duplicate set. The only clear duplicate is the repeated Qwen3ForCausalLM VRAM leak report.", + "confidence": 0.95, + "canonical_issue_reason": "issue:42673 is the earliest and more complete report of the exact same VRAM leak; issue:43704 has the same title and appears to be a duplicate.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:42673 is the cleanest anchor for the cluster: specific, reproducible, and duplicated verbatim by a later report.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:28282", + "right": "issue:30990", + "accept": false, + "reason": "Different failures: PyTorch missing import vs a loading hang in Sentence Transformers." + }, + { + "left": "issue:41628", + "right": "issue:44933", + "accept": false, + "reason": "Both are import-related, but they concern different missing symbols/modules and different code paths." + }, + { + "left": "issue:45230", + "right": "issue:45397", + "accept": false, + "reason": "Unrelated bugs: a placeholder bug report vs Gemma-4 ZeRO3 from_pretrained failure." + }, + { + "left": "issue:45310", + "right": "issue:45375", + "accept": false, + "reason": "Related model family, but not the same bug: generic Qwen3.5 Moe from_pretrained error vs a missing config field in the vision config." + }, + { + "left": "issue:43475", + "right": "issue:46821", + "accept": false, + "reason": "Different model-specific attribute errors on different components; no shared underlying bug." + }, + { + "left": "issue:44964", + "right": "issue:45137", + "accept": false, + "reason": "Different subsystems and symptoms: model load regression vs DeepSpeed ZeRO3 deque underflow." + }, + { + "left": "issue:43986", + "right": "issue:46726", + "accept": false, + "reason": "Both mention missing vision backend handling, but one is a video-model AutoProcessor crash and the other is an unrecognized image processor error." + }, + { + "left": "issue:44869", + "right": "issue:45356", + "accept": false, + "reason": "Different tokenizer regressions affecting different models and failure modes." + }, + { + "left": "issue:44038", + "right": "issue:44623", + "accept": false, + "reason": "Qwen3-VL-Moe loading error is unrelated to processor.save_pretrained missing files." + }, + { + "left": "issue:39692", + "right": "issue:46726", + "accept": false, + "reason": "Doc example errors for SigLIP2 are not the same as the missing-vision-backend processor error." + }, + { + "left": "issue:46463", + "right": "issue:46877", + "accept": false, + "reason": "Incorrect validation error message is unrelated to auto_docstring stdout diagnostics." + }, + { + "left": "issue:44373", + "right": "issue:46851", + "accept": false, + "reason": "One is a docstring issue; the other is a runtime bug with variable-length position_ids." + }, + { + "left": "issue:42994", + "right": "issue:43525", + "accept": false, + "reason": "Quantized model saving failure is unrelated to a missing pad_token_id attribute on Llama4Config." + }, + { + "left": "issue:42371", + "right": "issue:45092", + "accept": false, + "reason": "TF32 settings guidance is unrelated to InternVL2 meta-initialization incompatibility." + }, + { + "left": "issue:43723", + "right": "issue:46326", + "accept": false, + "reason": "Both are tokenizer-related, but they describe different regressions and different code paths." + }, + { + "left": "issue:45406", + "right": "issue:46202", + "accept": false, + "reason": "Gemma processor missing _tokenizer is not the same as the Gemma pipeline roundtrip failure." + }, + { + "left": "issue:43873", + "right": "issue:44945", + "accept": false, + "reason": "Quantization/offloading behavior is unrelated to pipeline-parallel incorrect outputs." + }, + { + "left": "issue:44945", + "right": "issue:45753", + "accept": false, + "reason": "Pipeline-parallel output corruption is not the same as a Qwen3.5 image-specific infinite loop." + }, + { + "left": "issue:41628", + "right": "issue:44351", + "accept": false, + "reason": "Different missing public imports: AutoImageProcessor vs HybridCache." + }, + { + "left": "issue:43784", + "right": "issue:46399", + "accept": false, + "reason": "Import-time NameError in sentence-transformers is unrelated to a rotary-embedding registration error." + }, + { + "left": "issue:44857", + "right": "issue:46877", + "accept": false, + "reason": "Training loss AMP crash is unrelated to auto_docstring emitting diagnostics at import time." + }, + { + "left": "issue:45593", + "right": "issue:46877", + "accept": false, + "reason": "D-FINE auxiliary-loss behavior is unrelated to docstring lint output on import." + }, + { + "left": "issue:42673", + "right": "issue:43704", + "accept": true, + "reason": "Exact duplicate: same title and same Qwen3ForCausalLM VRAM leak symptom." + }, + { + "left": "issue:44038", + "right": "issue:45310", + "accept": false, + "reason": "Different Qwen model families and different loading errors; not the same concrete bug." + }, + { + "left": "issue:44222", + "right": "issue:45310", + "accept": false, + "reason": "FP8 save_pretrained failure is distinct from a from_pretrained load error." + }, + { + "left": "issue:36010", + "right": "issue:41628", + "accept": false, + "reason": "Both are import errors, but they involve different missing API symbols and likely different regressions." + }, + { + "left": "issue:43278", + "right": "issue:45137", + "accept": false, + "reason": "Dtype drift between train/eval is unrelated to the ZeRO3 deque error." + }, + { + "left": "issue:42994", + "right": "issue:43824", + "accept": false, + "reason": "Quantized saving failure is unrelated to a missing Qwen2.5-VL import." + }, + { + "left": "issue:46202", + "right": "issue:46701", + "accept": false, + "reason": "Both mention Gemma, but one is a tokenizer roundtrip problem and the other is a pipeline integration issue." + }, + { + "left": "issue:45081", + "right": "issue:45103", + "accept": false, + "reason": "Mistral tokenizer backend_tokenizer crash is unrelated to the future-annotations docstring crash." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays conservative. The Qwen3ForCausalLM VRAM leak pair is a clear duplicate, and the remaining soft-edge verdicts are appropriately cautious rather than overstated." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 47, + "review_unit_count": 61, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-502", + "nodes": [ + "issue:27957", + "issue:30333", + "issue:30990", + "issue:33290", + "issue:34689", + "issue:35141", + "issue:39692", + "issue:42175", + "issue:42222", + "issue:42673", + "issue:42831", + "issue:42898", + "issue:43232", + "issue:43479", + "issue:43504", + "issue:43723", + "issue:43784", + "issue:43824", + "issue:43874", + "issue:43976", + "issue:44016", + "issue:44079", + "issue:44222", + "issue:44496", + "issue:44568", + "issue:44623", + "issue:44625", + "issue:44734", + "issue:44779", + "issue:44792", + "issue:44869", + "issue:44933", + "issue:45092", + "issue:45103", + "issue:45200", + "issue:45310", + "issue:45405", + "issue:45561", + "issue:45600", + "issue:45712", + "issue:45812", + "issue:46077", + "issue:46123", + "issue:46231", + "issue:46693", + "issue:46829", + "issue:46877", + "issue:46909" + ], + "soft_pairs": [ + "issue:42831|issue:44222", + "issue:43232|issue:44792", + "issue:42175|issue:44933", + "issue:43504|issue:45561", + "issue:45561|issue:45600", + "issue:43976|issue:44933", + "issue:44496|issue:45561", + "issue:30333|issue:33290", + "issue:45405|issue:46877", + "issue:30990|issue:35141", + "issue:42898|issue:44568", + "issue:42175|issue:45712", + "issue:43479|issue:46829", + "issue:46077|issue:46123", + "issue:44623|issue:45310", + "issue:42673|issue:46693", + "issue:43479|issue:44869", + "issue:43723|issue:44568", + "issue:42898|issue:45812", + "issue:44016|issue:46231", + "issue:44625|issue:46877", + "issue:42898|issue:44779", + "issue:39692|issue:46909", + "issue:43479|issue:44079", + "issue:43874|issue:45092", + "issue:43784|issue:43824", + "issue:27957|issue:30333", + "issue:44734|issue:45103", + "issue:30333|issue:34689", + "issue:42222|issue:45200" + ], + "prepared_review_unit_hash": "d7839f6071f88e05676feebc157e07c99618fdbc26ba9a5a74e76ecc68f42597", + "component_budget": { + "node_count": 502, + "item_count": 502, + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 30, + "serialized_chars": 36795, + "estimated_input_tokens": 9199, + "estimated_eval_tokens": 18654 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is heterogeneous: the soft edges mostly connect superficially similar issues (tokenizers, multimodal loading, docs, FP8) but not the same underlying bug. No PRs are present.", + "confidence": 0.94, + "canonical_issue_reason": "Issue 45561 is the best representative by triage value: it is open, highly referenced, and has clear reproduction details. That said, the set is not a true duplicate family, so the choice is pragmatic rather than semantic.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 45561 is the strongest standalone issue to surface because it is active, well-scoped, and has multiple inbound references; it is more actionable than the other candidates.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:42831", + "right": "issue:44222", + "accept": false, + "reason": "Both mention FP8, but one is an accuracy regression in FineGrainedFP8 and the other is a save_pretrained/MoE serialization bug; different code paths." + }, + { + "left": "issue:43232", + "right": "issue:44792", + "accept": false, + "reason": "Generation-state update after sync_gpus is unrelated to a Janus image-generation test failure; no shared underlying bug." + }, + { + "left": "issue:42175", + "right": "issue:44933", + "accept": false, + "reason": "Backend packaging/install omission and a missing image_utils import are different failures, even if both are import/setup related." + }, + { + "left": "issue:43504", + "right": "issue:45561", + "accept": false, + "reason": "Beit legacy-field loading and an xdist captured-info race are unrelated subsystems and symptoms." + }, + { + "left": "issue:45561", + "right": "issue:45600", + "accept": false, + "reason": "A test-file race in patched utils is not the same bug as stale auto_mappings references to removed configs." + }, + { + "left": "issue:43976", + "right": "issue:44933", + "accept": false, + "reason": "Python version compatibility and a nonexistent import export are distinct issues." + }, + { + "left": "issue:44496", + "right": "issue:45561", + "accept": false, + "reason": "Model config recognition failure and a testing-utils race do not share a code path." + }, + { + "left": "issue:30333", + "right": "issue:33290", + "accept": false, + "reason": "MLflow reporting after failed jobs is unrelated to an Adafactor/deepspeed OOM." + }, + { + "left": "issue:45405", + "right": "issue:46877", + "accept": false, + "reason": "A dependency version pin problem and docstring lint output at import time are unrelated." + }, + { + "left": "issue:30990", + "right": "issue:35141", + "accept": false, + "reason": "A SentenceTransformers loading hang is different from output-embedding reinitialization during resize." + }, + { + "left": "issue:42898", + "right": "issue:44568", + "accept": false, + "reason": "Both are tokenizer regressions, but clean_up_tokenization_spaces behavior and add_special_tokens/BOS-EOS handling are different bugs." + }, + { + "left": "issue:42175", + "right": "issue:45712", + "accept": false, + "reason": "Packaging backend omission and leftover dummy classes leaking into dir() are unrelated maintenance issues." + }, + { + "left": "issue:43479", + "right": "issue:46829", + "accept": false, + "reason": "Multimodal config default initialization and VideoPrism output field misassignment are different model bugs." + }, + { + "left": "issue:46077", + "right": "issue:46123", + "accept": false, + "reason": "Attention-mask construction for Gemma 4 and final-partial-batch handling in MaskGenerationPipeline are separate problems." + }, + { + "left": "issue:44623", + "right": "issue:45310", + "accept": false, + "reason": "Processor.save_pretrained missing files and Qwen3.5 MoE from_pretrained errors are both loading-related but not the same failure." + }, + { + "left": "issue:42673", + "right": "issue:46693", + "accept": false, + "reason": "A VRAM leak in multi-threaded dataloaders is not the same as a long-context FA2 performance regression." + }, + { + "left": "issue:43479", + "right": "issue:44869", + "accept": false, + "reason": "Config defaults for Phi4Multimodal and Whisper timestamp decode crashes are unrelated code paths." + }, + { + "left": "issue:43723", + "right": "issue:44568", + "accept": false, + "reason": "AutoTokenizer.from_pretrained failure and missing BOS/EOS on add_special_tokens are different tokenizer regressions." + }, + { + "left": "issue:42898", + "right": "issue:45812", + "accept": false, + "reason": "Both are tokenizer regressions, but one concerns cleanup-space behavior and the other wrong Granite token IDs; not the same bug." + }, + { + "left": "issue:44016", + "right": "issue:46231", + "accept": false, + "reason": "A notebook syntax error and a documentation parameter-name mismatch are different doc issues." + }, + { + "left": "issue:44625", + "right": "issue:46877", + "accept": false, + "reason": "Qwen3.5 num_labels propagation and auto_docstring stdout noise have no shared underlying code path." + }, + { + "left": "issue:42898", + "right": "issue:44779", + "accept": false, + "reason": "Both are v5 tokenizer regressions, but they affect different behaviors: cleanup spacing versus DeepSeek token output correctness." + }, + { + "left": "issue:39692", + "right": "issue:46909", + "accept": false, + "reason": "These are both broken docs examples, but for different models/examples and different failure modes." + }, + { + "left": "issue:43479", + "right": "issue:44079", + "accept": false, + "reason": "Multimodal config initialization and ModelOutput key assignment are separate core issues." + }, + { + "left": "issue:43874", + "right": "issue:45092", + "accept": false, + "reason": "A missing GLM46V processor method and an InternVL2 meta-init incompatibility are different multimodal-loading bugs." + }, + { + "left": "issue:43784", + "right": "issue:43824", + "accept": false, + "reason": "Both are import-related, but one is a NameError in sentence-transformers and the other is a missing Qwen2.5-VL class export." + }, + { + "left": "issue:27957", + "right": "issue:30333", + "accept": false, + "reason": "Flash Attention 2 support for XLMRoberta is unrelated to MLflow reporting after failed training." + }, + { + "left": "issue:44734", + "right": "issue:45103", + "accept": false, + "reason": "A serving KV-cache indexing crash and an auto_docstring annotations crash are different subsystems." + }, + { + "left": "issue:30333", + "right": "issue:34689", + "accept": false, + "reason": "MLflow reporting and Llama 3.2 Vision model loading regressions are unrelated." + }, + { + "left": "issue:42222", + "right": "issue:45200", + "accept": false, + "reason": "VitPose model breakage and Gemma 4 mm_token_type_ids defaulting are different model-family bugs." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 48, + "review_unit_count": 61, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-502", + "nodes": [ + "issue:22355", + "issue:28282", + "issue:30333", + "issue:35141", + "issue:41628", + "issue:42757", + "issue:42796", + "issue:42831", + "issue:42898", + "issue:43116", + "issue:43122", + "issue:43295", + "issue:43479", + "issue:43644", + "issue:43723", + "issue:44060", + "issue:44075", + "issue:44190", + "issue:44222", + "issue:44263", + "issue:44295", + "issue:44297", + "issue:44360", + "issue:44373", + "issue:44448", + "issue:44484", + "issue:44485", + "issue:44623", + "issue:44933", + "issue:44977", + "issue:45071", + "issue:45092", + "issue:45125", + "issue:45245", + "issue:45310", + "issue:45356", + "issue:45405", + "issue:45656", + "issue:45712", + "issue:46082", + "issue:46132", + "issue:46209", + "issue:46231", + "issue:46614", + "issue:46710", + "issue:46762", + "issue:46932" + ], + "soft_pairs": [ + "issue:42898|issue:43295", + "issue:42898|issue:45356", + "issue:44295|issue:45656", + "issue:43116|issue:44190", + "issue:30333|issue:35141", + "issue:44222|issue:44297", + "issue:45405|issue:45712", + "issue:22355|issue:28282", + "issue:44060|issue:45125", + "issue:44075|issue:46231", + "issue:44933|issue:45712", + "issue:44360|issue:46762", + "issue:42831|issue:44623", + "issue:44297|issue:45656", + "issue:42796|issue:43723", + "issue:44222|issue:44623", + "issue:44448|issue:46132", + "issue:45245|issue:46132", + "issue:43479|issue:43644", + "issue:41628|issue:42757", + "issue:44373|issue:44485", + "issue:44297|issue:45310", + "issue:46132|issue:46710", + "issue:45071|issue:46082", + "issue:44222|issue:44484", + "issue:43723|issue:44977", + "issue:46614|issue:46932", + "issue:43122|issue:46209", + "issue:44263|issue:45092", + "issue:44190|issue:46082" + ], + "prepared_review_unit_hash": "ee1cbba10a4aa09f4dce9177175f13c7a8b36b092248bf142f5faced20f071ba", + "component_budget": { + "node_count": 502, + "item_count": 502, + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 30, + "serialized_chars": 35893, + "estimated_input_tokens": 8974, + "estimated_eval_tokens": 18204 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "These soft pairs are all false positives: they share broad themes like tokenization, imports, or save_pretrained, but each points to a different concrete bug or code path.", + "confidence": 0.97, + "canonical_issue_reason": "No canonical issue: the items span unrelated failures, so there is no duplicate hub to collapse into one issue.", + "canonical_pr_reason": null, + "best_issue_reason": "No single best issue overall; if a representative were forced, 42898 is one of the broadest regression reports, but it does not subsume the rest.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:42898", + "right": "issue:43295", + "accept": false, + "reason": "Both are v5 regressions, but one is about clean_up_tokenization_spaces and the other about processor.tokenizer/image passing; different bugs." + }, + { + "left": "issue:42898", + "right": "issue:45356", + "accept": false, + "reason": "Tokenizer regressions in general, but 45356 is a Kimi-K2.5 codec/regex warning issue, not the same behavior change as 42898." + }, + { + "left": "issue:44295", + "right": "issue:45656", + "accept": false, + "reason": "Registering position_ids as a buffer causing read errors is unrelated to DeepSpeed making optimizer.step run twice." + }, + { + "left": "issue:43116", + "right": "issue:44190", + "accept": false, + "reason": "Both hit example scripts, but one is multi-label classification returning empty results and the other is local dataset loading failure in a different script." + }, + { + "left": "issue:30333", + "right": "issue:35141", + "accept": false, + "reason": "MLflow not reporting failed training is unrelated to output embeddings being reinitialized after resize." + }, + { + "left": "issue:44222", + "right": "issue:44297", + "accept": false, + "reason": "FP8 MoE save_pretrained problems and tokenizer_class mismatch in tokenizer_config.json are different save paths and failures." + }, + { + "left": "issue:45405", + "right": "issue:45712", + "accept": false, + "reason": "A dependency minimum-version bump is unrelated to leftover dummy classes leaking into the package namespace." + }, + { + "left": "issue:22355", + "right": "issue:28282", + "accept": false, + "reason": "Both are import errors, but one is missing transformers.onnx and the other is missing PyTorch for AutoModel; different causes." + }, + { + "left": "issue:44060", + "right": "issue:45125", + "accept": false, + "reason": "An incorrect tied-weights warning is not the same as a missing _tp_plan for tensor parallelism." + }, + { + "left": "issue:44075", + "right": "issue:46231", + "accept": false, + "reason": "Unused SGD arguments in the optimizer are a code bug, while the set_training issue is just a documentation parameter-name mismatch." + }, + { + "left": "issue:44933", + "right": "issue:45712", + "accept": false, + "reason": "A nonexistent image_utils import and dummy classes leaking into dir(transformers) are separate packaging issues." + }, + { + "left": "issue:44360", + "right": "issue:46762", + "accept": false, + "reason": "The DSA indexer missing a ReLU is an architecture/design discussion, not the same as Minimax attention numerical differences." + }, + { + "left": "issue:42831", + "right": "issue:44623", + "accept": false, + "reason": "FineGrainedFP8 accuracy loss is unrelated to processor.save_pretrained omitting files." + }, + { + "left": "issue:44297", + "right": "issue:45656", + "accept": false, + "reason": "Tokenizer metadata mismatch on save is unrelated to an optimizer stepping twice under DeepSpeed." + }, + { + "left": "issue:42796", + "right": "issue:43723", + "accept": false, + "reason": "Ministral BPE-marker handling and AutoTokenizer.from_pretrained loading failures are different tokenizer bugs." + }, + { + "left": "issue:44222", + "right": "issue:44623", + "accept": false, + "reason": "FP8 save_pretrained for MoE and processor save_pretrained missing files are different components and failure modes." + }, + { + "left": "issue:44448", + "right": "issue:46132", + "accept": false, + "reason": "Pegasus output regression across versions is unrelated to AttentionInterface.register changing callback behavior." + }, + { + "left": "issue:45245", + "right": "issue:46132", + "accept": false, + "reason": "Category-count runtime limits and AttentionInterface.register semantics are unrelated." + }, + { + "left": "issue:43479", + "right": "issue:43644", + "accept": false, + "reason": "Bad default vision/audio config initialization is unrelated to non-persistent buffers being filled with junk." + }, + { + "left": "issue:41628", + "right": "issue:42757", + "accept": false, + "reason": "Both are missing-import reports, but for different symbols in different packages with different fixes." + }, + { + "left": "issue:44373", + "right": "issue:44485", + "accept": false, + "reason": "A wrong docstring for position_ids is not the same as a GLM-5 RoPE implementation bug." + }, + { + "left": "issue:44297", + "right": "issue:45310", + "accept": false, + "reason": "Tokenizer save metadata mismatch and Qwen3.5 MoE from_pretrained failure are different issues." + }, + { + "left": "issue:46132", + "right": "issue:46710", + "accept": false, + "reason": "AttentionInterface.register behavior changes do not match a model accuracy regression after upgrading Transformers." + }, + { + "left": "issue:45071", + "right": "issue:46082", + "accept": false, + "reason": "PretrainedConfig type-checking regression is unrelated to LlamaConfig rejecting explicit head_dim under divisibility constraints." + }, + { + "left": "issue:44222", + "right": "issue:44484", + "accept": false, + "reason": "FP8 save_pretrained bugs and the default max_shard_size question are not the same problem." + }, + { + "left": "issue:43723", + "right": "issue:44977", + "accept": false, + "reason": "Tokenizer loading issues in v5 are unrelated to Qwen3.5 flash-attention generation failures." + }, + { + "left": "issue:46614", + "right": "issue:46932", + "accept": false, + "reason": "Stale phonemizer language backend state is unrelated to Gemma4 quantized vision/audio embedder dtype casting." + }, + { + "left": "issue:43122", + "right": "issue:46209", + "accept": false, + "reason": "Tokenization differences across versions and FP8 inference failing on B200 are unrelated bugs." + }, + { + "left": "issue:44263", + "right": "issue:45092", + "accept": false, + "reason": "torch.split return handling in GlmMoeDsaIndexer is unrelated to remote-code checkpoint incompatibility with v5 meta initialization." + }, + { + "left": "issue:44190", + "right": "issue:46082", + "accept": false, + "reason": "Local dataset loading in an example script is unrelated to LlamaConfig head_dim validation." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 49, + "review_unit_count": 61, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-502", + "nodes": [ + "issue:22355", + "issue:27957", + "issue:28218", + "issue:30990", + "issue:38175", + "issue:39401", + "issue:41628", + "issue:42757", + "issue:42831", + "issue:42886", + "issue:42907", + "issue:42994", + "issue:43066", + "issue:43232", + "issue:43408", + "issue:43475", + "issue:43644", + "issue:43824", + "issue:44038", + "issue:44261", + "issue:44263", + "issue:44279", + "issue:44295", + "issue:44360", + "issue:44373", + "issue:44483", + "issue:44568", + "issue:44623", + "issue:44829", + "issue:45092", + "issue:45278", + "issue:45310", + "issue:45357", + "issue:45381", + "issue:45406", + "issue:45753", + "issue:45834", + "issue:46002", + "issue:46050", + "issue:46209", + "issue:46305", + "issue:46328", + "issue:46369", + "issue:46402", + "issue:46424", + "issue:46431", + "issue:46655", + "issue:46909" + ], + "soft_pairs": [ + "issue:43232|issue:46424", + "issue:43232|issue:45834", + "issue:44295|issue:44623", + "issue:43475|issue:46305", + "issue:46209|issue:46431", + "issue:44263|issue:44360", + "issue:46305|issue:46655", + "issue:44263|issue:44373", + "issue:30990|issue:38175", + "issue:42831|issue:45310", + "issue:45092|issue:45357", + "issue:46369|issue:46909", + "issue:42907|issue:46050", + "issue:44279|issue:46328", + "issue:45381|issue:46305", + "issue:44261|issue:45092", + "issue:22355|issue:27957", + "issue:44483|issue:45406", + "issue:43644|issue:46305", + "issue:42831|issue:44038", + "issue:44829|issue:46002", + "issue:41628|issue:43824", + "issue:43408|issue:46402", + "issue:42831|issue:45753", + "issue:22355|issue:28218", + "issue:39401|issue:44568", + "issue:42886|issue:43066", + "issue:42886|issue:42994", + "issue:42757|issue:45278", + "issue:27957|issue:30990" + ], + "prepared_review_unit_hash": "5e4b41f9f37573d035a6069c0268053acae84c75c85fe86d9e91021d52943b75", + "component_budget": { + "node_count": 502, + "item_count": 502, + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 30, + "serialized_chars": 36161, + "estimated_input_tokens": 9041, + "estimated_eval_tokens": 18338 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This set is not a single duplicate cluster; it spans many unrelated bugs across tokenization, imports, quantization, SAM3/Qwen models, cache/offline behavior, and docs. None of the soft pairs are the same concrete underlying issue.", + "confidence": 0.97, + "canonical_issue_reason": "No clear canonical issue: the items are heterogeneous and do not share one underlying bug or change to consolidate around.", + "canonical_pr_reason": null, + "best_issue_reason": "No issue is a strong global canonical representative because the set is too diverse; any single issue would be overly narrow and misleading.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43232", + "right": "issue:46424", + "accept": false, + "reason": "Different code paths: generation kwargs after sync_gpus vs cache max length handling." + }, + { + "left": "issue:43232", + "right": "issue:45834", + "accept": false, + "reason": "Unrelated generation sync bug vs Kosmos2.5 OCR index error." + }, + { + "left": "issue:44295", + "right": "issue:44623", + "accept": false, + "reason": "Position_ids buffer access error is unrelated to missing processor save files." + }, + { + "left": "issue:43475", + "right": "issue:46305", + "accept": false, + "reason": "SAM3 attribute error and MoonshineStreaming mask off-by-one are different models and bugs." + }, + { + "left": "issue:46209", + "right": "issue:46431", + "accept": false, + "reason": "FP8 B200 inference failure is unrelated to processor chat-template/save_pretrained behavior." + }, + { + "left": "issue:44263", + "right": "issue:44360", + "accept": false, + "reason": "Both touch DSA indexer, but one is split output behavior and the other is missing ReLU; not the same bug." + }, + { + "left": "issue:46305", + "right": "issue:46655", + "accept": false, + "reason": "Different processors/models: streaming mask logic vs video padding calculation." + }, + { + "left": "issue:44263", + "right": "issue:44373", + "accept": false, + "reason": "Indexer tensor split bug vs docstring issue for position_ids." + }, + { + "left": "issue:30990", + "right": "issue:38175", + "accept": false, + "reason": "Sentence Transformers loading hang is unrelated to SigLIP2 zero probabilities." + }, + { + "left": "issue:42831", + "right": "issue:45310", + "accept": false, + "reason": "FineGrainedFP8 accuracy issue is unrelated to Qwen3.5 MoE from_pretrained failure." + }, + { + "left": "issue:45092", + "right": "issue:45357", + "accept": false, + "reason": "Different models and failures: InternVL2 meta-init incompatibility vs Qwen3.5 save_pretrained key regression." + }, + { + "left": "issue:46369", + "right": "issue:46909", + "accept": false, + "reason": "SAM3 LiteText memory overflow is unrelated to a broken DIA documentation example." + }, + { + "left": "issue:42907", + "right": "issue:46050", + "accept": false, + "reason": "Saving dequantized Ministrals/Devstrals differs from sinq quantized save/load corruption." + }, + { + "left": "issue:44279", + "right": "issue:46328", + "accept": false, + "reason": "General dependency issue is unrelated to SequenceFeatureExtractor.pad performance regression." + }, + { + "left": "issue:45381", + "right": "issue:46305", + "accept": false, + "reason": "Qwen2.5-VL video position_ids bug is unrelated to MoonshineStreaming window masking." + }, + { + "left": "issue:44261", + "right": "issue:45092", + "accept": false, + "reason": "Layernorm epsilon precision mismatch is unrelated to remote-code checkpoint meta initialization." + }, + { + "left": "issue:22355", + "right": "issue:27957", + "accept": false, + "reason": "ONNX import failure and Flash Attention 2 behavior are unrelated." + }, + { + "left": "issue:44483", + "right": "issue:45406", + "accept": false, + "reason": "v5.3 chat/completions request rejection is unrelated to Gemma4Processor missing _tokenizer." + }, + { + "left": "issue:43644", + "right": "issue:46305", + "accept": false, + "reason": "Non-persistent buffer junk in v5.0 is unrelated to streaming mask off-by-one." + }, + { + "left": "issue:42831", + "right": "issue:44038", + "accept": false, + "reason": "FP8 accuracy issue is unrelated to Qwen3-VL-Moe from_pretrained errors." + }, + { + "left": "issue:44829", + "right": "issue:46002", + "accept": false, + "reason": "Flash Attention 3 training degeneration is unrelated to NaNs on checkout/init." + }, + { + "left": "issue:41628", + "right": "issue:43824", + "accept": false, + "reason": "Both are import errors, but different missing symbols and likely different causes." + }, + { + "left": "issue:43408", + "right": "issue:46402", + "accept": false, + "reason": "SAM3 tracker/model-type warning is unrelated to missing text encoder weights in SAM3 model loading." + }, + { + "left": "issue:42831", + "right": "issue:45753", + "accept": false, + "reason": "FP8 accuracy issue and Qwen3_5 infinite loop are different symptoms and code paths." + }, + { + "left": "issue:22355", + "right": "issue:28218", + "accept": false, + "reason": "ONNX module import failure is unrelated to tokenizer added-token spacing." + }, + { + "left": "issue:39401", + "right": "issue:44568", + "accept": false, + "reason": "Both are tokenizer bugs, but offset_mapping and add_special_tokens/BOS-EOS are different defects." + }, + { + "left": "issue:42886", + "right": "issue:43066", + "accept": false, + "reason": "Offline cache loading failure is unrelated to wrong tokenizer decoder type in v5." + }, + { + "left": "issue:42886", + "right": "issue:42994", + "accept": false, + "reason": "Tokenizer cache loading and quantized model saving are separate issues." + }, + { + "left": "issue:42757", + "right": "issue:45278", + "accept": false, + "reason": "A single huggingface_hub import error is not the same as many import errors after a transformers upgrade." + }, + { + "left": "issue:27957", + "right": "issue:30990", + "accept": false, + "reason": "XLMRoberta Flash Attention 2 issue is unrelated to Sentence Transformers loading hang." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 50, + "review_unit_count": 61, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-502", + "nodes": [ + "issue:30990", + "issue:33357", + "issue:39401", + "issue:41628", + "issue:42371", + "issue:42796", + "issue:42907", + "issue:43122", + "issue:43232", + "issue:43408", + "issue:43452", + "issue:43475", + "issue:43540", + "issue:43576", + "issue:43646", + "issue:43756", + "issue:43784", + "issue:43937", + "issue:43976", + "issue:44038", + "issue:44060", + "issue:44117", + "issue:44188", + "issue:44263", + "issue:44295", + "issue:44297", + "issue:44360", + "issue:44373", + "issue:44485", + "issue:44488", + "issue:44521", + "issue:44625", + "issue:44704", + "issue:44829", + "issue:44945", + "issue:44964", + "issue:45278", + "issue:45290", + "issue:45381", + "issue:45593", + "issue:45600", + "issue:46002", + "issue:46172", + "issue:46326", + "issue:46369", + "issue:46762", + "issue:46909" + ], + "soft_pairs": [ + "issue:44188|issue:44373", + "issue:43408|issue:46369", + "issue:43232|issue:45593", + "issue:42796|issue:43122", + "issue:43646|issue:46002", + "issue:45381|issue:46909", + "issue:43540|issue:46909", + "issue:30990|issue:33357", + "issue:42907|issue:44945", + "issue:44295|issue:44297", + "issue:43576|issue:43784", + "issue:42371|issue:44263", + "issue:39401|issue:44488", + "issue:42907|issue:44964", + "issue:43976|issue:44945", + "issue:46172|issue:46762", + "issue:42907|issue:43937", + "issue:43756|issue:44060", + "issue:41628|issue:45278", + "issue:43784|issue:45600", + "issue:44117|issue:44704", + "issue:39401|issue:44521", + "issue:44704|issue:45290", + "issue:44625|issue:44829", + "issue:44360|issue:44485", + "issue:43408|issue:43475", + "issue:43122|issue:44521", + "issue:44038|issue:44297", + "issue:43122|issue:46326", + "issue:42907|issue:43452", + "issue:44038|issue:44295" + ], + "prepared_review_unit_hash": "6b6e8c69bc4451e4095ad1266da23e4e1f739e25e3be204bce4d66542fc05b45", + "component_budget": { + "node_count": 502, + "item_count": 502, + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 31, + "serialized_chars": 35942, + "estimated_input_tokens": 8986, + "estimated_eval_tokens": 18228 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is broad and mostly heterogeneous: a few issues relate to v5 upgrade/import regressions, but most are model-specific tokenizer, generation, or multimodal bugs. None of the soft pairs look like the same underlying bug or change.", + "confidence": 0.42, + "canonical_issue_reason": "issue:45278 is the broadest representative of the cluster\u2019s main theme: upgrade regressions in v5 causing import failures and API breakage.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:45278 is the best global representative because it captures the widest, most reusable symptom in the cluster (many import errors after upgrading).", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44188", + "right": "issue:44373", + "accept": false, + "reason": "Different bugs: torch.compile attention-kernel divergence vs a docstring issue for position_ids." + }, + { + "left": "issue:43408", + "right": "issue:46369", + "accept": false, + "reason": "Different SAM 3 problems: model-type warning on import/config vs memory overflow in sam3-litetext." + }, + { + "left": "issue:43232", + "right": "issue:45593", + "accept": false, + "reason": "Unrelated: generation kwargs after sync_gpus vs D-FINE auxiliary-loss behavior." + }, + { + "left": "issue:42796", + "right": "issue:43122", + "accept": false, + "reason": "Both tokenizer-related, but one is BPE marker handling for a specific model and the other is a cross-version tokenization drift report; not the same bug." + }, + { + "left": "issue:43646", + "right": "issue:46002", + "accept": false, + "reason": "Custom model initialization breakage is distinct from NaNs in classification heads after checkout/init." + }, + { + "left": "issue:45381", + "right": "issue:46909", + "accept": false, + "reason": "Multimodal video position-id bug vs broken documentation example; unrelated." + }, + { + "left": "issue:43540", + "right": "issue:46909", + "accept": false, + "reason": "Qwen3OmniMoe video-processing ValueError is unrelated to a DIA docs example." + }, + { + "left": "issue:30990", + "right": "issue:33357", + "accept": false, + "reason": "Different loading failures: Sentence Transformers getting stuck vs MacOS bus error with a CLIP model." + }, + { + "left": "issue:42907", + "right": "issue:44945", + "accept": false, + "reason": "Dequantized Ministrals save failure is unrelated to incorrect pipeline-parallel LLM outputs." + }, + { + "left": "issue:44295", + "right": "issue:44297", + "accept": false, + "reason": "Position_ids buffer access error and tokenizer_class mismatch are different tokenizer serialization bugs." + }, + { + "left": "issue:43576", + "right": "issue:43784", + "accept": false, + "reason": "Both are v5 breakages, but one is the env command and the other is an import NameError; different code paths." + }, + { + "left": "issue:42371", + "right": "issue:44263", + "accept": false, + "reason": "TF32 API guidance and torch.split return semantics in GlmMoeDsaIndexer are unrelated." + }, + { + "left": "issue:39401", + "right": "issue:44488", + "accept": false, + "reason": "Qwen3 offset_mapping bug vs inability to load a separate model; not the same issue." + }, + { + "left": "issue:42907", + "right": "issue:44964", + "accept": false, + "reason": "Saving dequantized Ministrals/Devstrals is unrelated to loading Phi-4 multimodal." + }, + { + "left": "issue:43976", + "right": "issue:44945", + "accept": false, + "reason": "Python version support regression vs pipeline-parallel output corruption; unrelated." + }, + { + "left": "issue:46172", + "right": "issue:46762", + "accept": false, + "reason": "Both involve numerical behavior, but they are different models and different attention implementations; not mergeable as one fix." + }, + { + "left": "issue:42907", + "right": "issue:43937", + "accept": false, + "reason": "Dequantized model saving and invalid GenerationConfig for GLM-5 are separate problems." + }, + { + "left": "issue:43756", + "right": "issue:44060", + "accept": false, + "reason": "RoPE layer selection in Smollm3 is unrelated to tied-weights warnings in Qwen3-Next." + }, + { + "left": "issue:41628", + "right": "issue:45278", + "accept": false, + "reason": "Both mention import failures, but one is a specific missing AutoImageProcessor symbol and the other is broad post-upgrade import breakage; not enough evidence of the same root cause." + }, + { + "left": "issue:43784", + "right": "issue:45600", + "accept": false, + "reason": "Import NameError in sentence-transformers is unrelated to stale auto_mappings references breaking CI." + }, + { + "left": "issue:44117", + "right": "issue:44704", + "accept": false, + "reason": "TOKENIZER_MAPPING_NAMES returning None and AutoProcessor not forwarding kwargs are different from_pretrained plumbing bugs." + }, + { + "left": "issue:39401", + "right": "issue:44521", + "accept": false, + "reason": "Tokenizer offset mapping and multimodal assistant masks are distinct tokenizer/template issues." + }, + { + "left": "issue:44704", + "right": "issue:45290", + "accept": false, + "reason": "AutoProcessor cached_file kwargs and apply_chat_template tool-call crash are unrelated APIs." + }, + { + "left": "issue:44625", + "right": "issue:44829", + "accept": false, + "reason": "Qwen3.5 num_labels propagation and flash_attention_3 training degeneration are unrelated." + }, + { + "left": "issue:44360", + "right": "issue:44485", + "accept": false, + "reason": "DSA indexer ReLU discussion and GLM-5 RoPE implementation are different model internals." + }, + { + "left": "issue:43408", + "right": "issue:43475", + "accept": false, + "reason": "Both are SAM 3 Video, but one is a config/type warning and the other is a missing attribute in the encoder output; different bugs." + }, + { + "left": "issue:43122", + "right": "issue:44521", + "accept": false, + "reason": "Cross-version tokenization drift and all-zero assistant masks are different tokenizer/template failures." + }, + { + "left": "issue:44038", + "right": "issue:44297", + "accept": false, + "reason": "Qwen3-VL-Moe bug and tokenizer_class mismatch are unrelated." + }, + { + "left": "issue:43122", + "right": "issue:46326", + "accept": false, + "reason": "Tokenization differences across versions and mistral_common continue_final_message behavior are not the same underlying bug." + }, + { + "left": "issue:42907", + "right": "issue:43452", + "accept": false, + "reason": "Dequantized model saving and gguf_file loading for AutoTokenizer/AutoModelForCausalLM are unrelated." + }, + { + "left": "issue:44038", + "right": "issue:44295", + "accept": false, + "reason": "Qwen3-VL-Moe issue and position_ids buffer read problem are different code paths." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 51, + "review_unit_count": 61, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-502", + "nodes": [ + "issue:22355", + "issue:29127", + "issue:29159", + "issue:36296", + "issue:38617", + "issue:39401", + "issue:42175", + "issue:42371", + "issue:42673", + "issue:42757", + "issue:42898", + "issue:43066", + "issue:43116", + "issue:43122", + "issue:43278", + "issue:43493", + "issue:43611", + "issue:43824", + "issue:43874", + "issue:43937", + "issue:43976", + "issue:44038", + "issue:44188", + "issue:44222", + "issue:44351", + "issue:44355", + "issue:44485", + "issue:44610", + "issue:44625", + "issue:44779", + "issue:44938", + "issue:44945", + "issue:45072", + "issue:45083", + "issue:45200", + "issue:45216", + "issue:45278", + "issue:45357", + "issue:45446", + "issue:45478", + "issue:45753", + "issue:46002", + "issue:46231", + "issue:46328", + "issue:46614", + "issue:46877", + "issue:46909" + ], + "soft_pairs": [ + "issue:45216|issue:45478", + "issue:22355|issue:29159", + "issue:39401|issue:46614", + "issue:44038|issue:44222", + "issue:43824|issue:44351", + "issue:43116|issue:45200", + "issue:43122|issue:44779", + "issue:42757|issue:44351", + "issue:36296|issue:43116", + "issue:43278|issue:44945", + "issue:42175|issue:44938", + "issue:22355|issue:29127", + "issue:43493|issue:45753", + "issue:46231|issue:46877", + "issue:44188|issue:46328", + "issue:43937|issue:44485", + "issue:38617|issue:45278", + "issue:42898|issue:46614", + "issue:43066|issue:44779", + "issue:44038|issue:45478", + "issue:43611|issue:46002", + "issue:43976|issue:45446", + "issue:44355|issue:46231", + "issue:43116|issue:43493", + "issue:42673|issue:45072", + "issue:45357|issue:45478", + "issue:44625|issue:46002", + "issue:42371|issue:43874", + "issue:45083|issue:45446", + "issue:44610|issue:46909" + ], + "prepared_review_unit_hash": "3787419c5da0287b38cdab3c4836b075d1067a326b1419041edbe5174afb59a3", + "component_budget": { + "node_count": 502, + "item_count": 502, + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 30, + "serialized_chars": 35708, + "estimated_input_tokens": 8927, + "estimated_eval_tokens": 18110 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is mostly heterogeneous and does not contain a clear duplicate set. A few items revolve around Qwen3.5 save/load/tokenizer regressions, but they still look like separate bugs rather than one mergeable fix.", + "confidence": 0.87, + "canonical_issue_reason": "Issue 45216 is the strongest anchor among the related reports: it is a concrete, version-scoped Qwen3.5 save_pretrained regression with inbound references and a precise failure mode.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 45216 is the best representative issue if one must be chosen, because it is specific, actionable, and closest to the small Qwen3.5 serialization subgroup.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:45216", + "right": "issue:45478", + "accept": false, + "reason": "Same model family and release window, but one is a save_pretrained regression and the other is a from_pretrained import/load error; too different to treat as the same bug." + }, + { + "left": "issue:22355", + "right": "issue:29159", + "accept": false, + "reason": "Unrelated topics: ONNX module import failure vs tokenizer slow/fast inconsistency." + }, + { + "left": "issue:39401", + "right": "issue:46614", + "accept": false, + "reason": "Different subsystems and symptoms: tokenizer offset mapping vs phonemizer language backend state." + }, + { + "left": "issue:44038", + "right": "issue:44222", + "accept": false, + "reason": "Both mention Qwen/MoE, but one is a general transformers 5.0 compatibility bug and the other is FP8 save_pretrained behavior; not the same concrete failure." + }, + { + "left": "issue:43824", + "right": "issue:44351", + "accept": false, + "reason": "Both are import errors, but for different missing symbols and likely different fixes." + }, + { + "left": "issue:43116", + "right": "issue:45200", + "accept": false, + "reason": "Classification example script returns empty results vs Gemma 4 multimodal token type IDs defaulting; unrelated bugs." + }, + { + "left": "issue:43122", + "right": "issue:44779", + "accept": false, + "reason": "Both are tokenizer regressions, but they affect different model families and different observed failures." + }, + { + "left": "issue:42757", + "right": "issue:44351", + "accept": false, + "reason": "Different import sources and symbols: huggingface_hub import error vs transformers HybridCache import error." + }, + { + "left": "issue:36296", + "right": "issue:43116", + "accept": false, + "reason": "Tensor parallel training bug is unrelated to the multi-label classification example issue." + }, + { + "left": "issue:43278", + "right": "issue:44945", + "accept": false, + "reason": "Embedding dtype drift in eval is not the same as incorrect output under pipeline parallelism." + }, + { + "left": "issue:42175", + "right": "issue:44938", + "accept": false, + "reason": "Package backend dependency issue vs Python 3.14 load failure; no shared bug." + }, + { + "left": "issue:22355", + "right": "issue:29127", + "accept": false, + "reason": "ONNX import failure and LayoutLMv3 error-message clarity are unrelated." + }, + { + "left": "issue:43493", + "right": "issue:45753", + "accept": false, + "reason": "Both touch Qwen3_5 vision, but one is a HF-vs-JAX implementation discrepancy and the other is an infinite loop on one image; too different to merge." + }, + { + "left": "issue:46231", + "right": "issue:46877", + "accept": false, + "reason": "Documentation parameter name mismatch vs auto_docstring printing diagnostics at import time are unrelated." + }, + { + "left": "issue:44188", + "right": "issue:46328", + "accept": false, + "reason": "Attention-kernel divergence under torch.compile is unrelated to numpy/list padding performance." + }, + { + "left": "issue:43937", + "right": "issue:44485", + "accept": false, + "reason": "GenerationConfig validation failure and RoPE implementation discussion are different GLM-5 issues." + }, + { + "left": "issue:38617", + "right": "issue:45278", + "accept": false, + "reason": "Specific missing import vs broad post-upgrade import breakage; too coarse to call duplicate." + }, + { + "left": "issue:42898", + "right": "issue:46614", + "accept": false, + "reason": "Tokenization cleanup behavior changes and stale phonemizer language backend are unrelated." + }, + { + "left": "issue:43066", + "right": "issue:44779", + "accept": false, + "reason": "Both are v5 tokenizer regressions, but on different models and likely different code paths." + }, + { + "left": "issue:44038", + "right": "issue:45478", + "accept": false, + "reason": "Qwen3-VL-Moe compatibility bug vs Qwen3.5 Moe from_pretrained error; related area, but not the same underlying bug." + }, + { + "left": "issue:43611", + "right": "issue:46002", + "accept": false, + "reason": "Base model prefix loading regression is unrelated to NaNs in classification heads after init/checkout." + }, + { + "left": "issue:43976", + "right": "issue:45446", + "accept": false, + "reason": "Python version support regression is unrelated to the flex_attention AuxRequest version check." + }, + { + "left": "issue:44355", + "right": "issue:46231", + "accept": false, + "reason": "Compiled Python file import errors are unrelated to a documentation typo." + }, + { + "left": "issue:43116", + "right": "issue:43493", + "accept": false, + "reason": "Classification example bug is unrelated to a SigLIP2 implementation discrepancy." + }, + { + "left": "issue:42673", + "right": "issue:45072", + "accept": false, + "reason": "VRAM leak in multi-threaded dataloader use is unrelated to dtype mismatch in bfloat16 inference." + }, + { + "left": "issue:45357", + "right": "issue:45478", + "accept": false, + "reason": "Both are Qwen3.5 save/load regressions, but one is incorrect visual encoder key saving and the other is a from_pretrained error; similar area, distinct bug." + }, + { + "left": "issue:44625", + "right": "issue:46002", + "accept": false, + "reason": "Qwen3.5 num_labels propagation and NaNs in classification heads are not the same failure." + }, + { + "left": "issue:42371", + "right": "issue:43874", + "accept": false, + "reason": "TF32 behavior guidance and missing image-patch API are unrelated." + }, + { + "left": "issue:45083", + "right": "issue:45446", + "accept": false, + "reason": "Feat-extract output length helper behavior in qwen3_omni_moe is unrelated to flex_attention PyTorch version checking." + }, + { + "left": "issue:44610", + "right": "issue:46909", + "accept": false, + "reason": "Processor/model input-size mismatch is unrelated to a broken documentation example." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 52, + "review_unit_count": 61, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-502", + "nodes": [ + "issue:28348", + "issue:33453", + "issue:39401", + "issue:42222", + "issue:42673", + "issue:42796", + "issue:43064", + "issue:43066", + "issue:43116", + "issue:43441", + "issue:43450", + "issue:43618", + "issue:43673", + "issue:43825", + "issue:43827", + "issue:43937", + "issue:43986", + "issue:43992", + "issue:44075", + "issue:44117", + "issue:44263", + "issue:44291", + "issue:44380", + "issue:44410", + "issue:44466", + "issue:44704", + "issue:44779", + "issue:44936", + "issue:44977", + "issue:44995", + "issue:45071", + "issue:45092", + "issue:45310", + "issue:45405", + "issue:45446", + "issue:45563", + "issue:45588", + "issue:46002", + "issue:46077", + "issue:46328", + "issue:46614", + "issue:46640", + "issue:46655", + "issue:46697", + "issue:46710", + "issue:46752", + "issue:46762", + "issue:46829" + ], + "soft_pairs": [ + "issue:43992|issue:44704", + "issue:44075|issue:44936", + "issue:43618|issue:46829", + "issue:43116|issue:43825", + "issue:43066|issue:46710", + "issue:28348|issue:46640", + "issue:44263|issue:46762", + "issue:46614|issue:46752", + "issue:44995|issue:46697", + "issue:45092|issue:46762", + "issue:42222|issue:43441", + "issue:43827|issue:45446", + "issue:33453|issue:39401", + "issue:43064|issue:43673", + "issue:42673|issue:42796", + "issue:43937|issue:45563", + "issue:44380|issue:46077", + "issue:43673|issue:46328", + "issue:42673|issue:46655", + "issue:42222|issue:46752", + "issue:44291|issue:45071", + "issue:45405|issue:45446", + "issue:44117|issue:45588", + "issue:39401|issue:44779", + "issue:44410|issue:45310", + "issue:44466|issue:46002", + "issue:43937|issue:44977", + "issue:43450|issue:43986" + ], + "prepared_review_unit_hash": "6fdcd21c6b4eb6bf474436b9b764152d93a510dd3cf0232af0c2cfd709da0b6f", + "component_budget": { + "node_count": 502, + "item_count": 502, + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 28, + "serialized_chars": 35842, + "estimated_input_tokens": 8961, + "estimated_eval_tokens": 18178 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is highly heterogeneous: most pairs are only loosely similar by subsystem or wording and do not look like the same underlying bug. No PRs are present; the only plausible anchor issue is a broad tokenizer-regression report, but even that is a weak fit for the whole cluster.", + "confidence": 0.43, + "canonical_issue_reason": "issue:33453 is the broadest and most generic report in the set (tokenizer loading regression), so it is the least-specific issue to use as an anchor. That said, the cluster spans many unrelated bugs, so no issue is a strong canonical fit.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:33453 is the most umbrella-like issue and could subsume nearby tokenizer regressions better than the other, more specialized reports. It is still not a great global representative because the cluster is not a true duplicate set.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43992", + "right": "issue:44704", + "accept": false, + "reason": "Different code paths: encoder weight loading vs AutoProcessor/cached_file argument passing." + }, + { + "left": "issue:44075", + "right": "issue:44936", + "accept": false, + "reason": "Unrelated bugs: SGD argument handling vs trainer.evaluate() failure after train()." + }, + { + "left": "issue:43618", + "right": "issue:46829", + "accept": false, + "reason": "Both are output-field issues, but they affect different model classes and different dataclass fields." + }, + { + "left": "issue:43116", + "right": "issue:43825", + "accept": false, + "reason": "Example-script multi-label output bug vs pipeline() error-message/docs mismatch; not the same failure." + }, + { + "left": "issue:43066", + "right": "issue:46710", + "accept": false, + "reason": "Tokenizer decoder typing issue vs model accuracy regression; too far apart to be the same bug." + }, + { + "left": "issue:28348", + "right": "issue:46640", + "accept": false, + "reason": "Enhancement request for GPT2 FlashAttention 2 support vs a separate Flash Attention bug report." + }, + { + "left": "issue:44263", + "right": "issue:46762", + "accept": false, + "reason": "GlmMoeDsa indexing bug vs Minimax attention numerical mismatch; different models and symptoms." + }, + { + "left": "issue:46614", + "right": "issue:46752", + "accept": false, + "reason": "Tokenizer phonemizer language-state bug vs empty-chat-template IndexError; unrelated." + }, + { + "left": "issue:44995", + "right": "issue:46697", + "accept": false, + "reason": "Second-forward stale cache crash vs image over-padding; different subsystems entirely." + }, + { + "left": "issue:45092", + "right": "issue:46762", + "accept": false, + "reason": "Remote-code/meta-init incompatibility vs attention numerical differences; no shared code-path." + }, + { + "left": "issue:42222", + "right": "issue:43441", + "accept": false, + "reason": "VitPose model breakage and Ministral FlashAttention failure are unrelated model-specific issues." + }, + { + "left": "issue:43827", + "right": "issue:45446", + "accept": false, + "reason": "Docs still mention pipeline() vs PyTorch version check in flex_attention; not the same bug." + }, + { + "left": "issue:33453", + "right": "issue:39401", + "accept": false, + "reason": "Both are tokenizer-related, but one is loading regression and the other is offset_mapping correctness for a specific model." + }, + { + "left": "issue:43064", + "right": "issue:43673", + "accept": false, + "reason": "Training/FSDP optimizer-state corruption vs chunked_prefill cache missing during generation." + }, + { + "left": "issue:42673", + "right": "issue:42796", + "accept": false, + "reason": "VRAM leak in threaded dataloader use vs tokenizer BPE-marker handling; different layers." + }, + { + "left": "issue:43937", + "right": "issue:45563", + "accept": false, + "reason": "GenerationConfig validation error vs stale warning in paged generate(); not the same defect." + }, + { + "left": "issue:44380", + "right": "issue:46077", + "accept": false, + "reason": "Different attention bugs: ignored scaling config vs causal mask construction for bidirectional attention." + }, + { + "left": "issue:43673", + "right": "issue:46328", + "accept": false, + "reason": "Generation cache bug vs feature extractor padding performance issue; unrelated." + }, + { + "left": "issue:42673", + "right": "issue:46655", + "accept": false, + "reason": "Threaded VRAM leak and video padding calculation are unrelated." + }, + { + "left": "issue:42222", + "right": "issue:46752", + "accept": false, + "reason": "VitPose breakage vs apply_chat_template empty-input crash; different components." + }, + { + "left": "issue:44291", + "right": "issue:45071", + "accept": false, + "reason": "Meta-init argument TypeError vs PretrainedConfig type-checking regression; similar release area, but not the same bug." + }, + { + "left": "issue:45405", + "right": "issue:45446", + "accept": false, + "reason": "Dependency version bump problem vs flex_attention import version check; unrelated." + }, + { + "left": "issue:44117", + "right": "issue:45588", + "accept": false, + "reason": "Tokenizer mapping None-handling vs flash_attention sinkless-model AttributeError; different paths." + }, + { + "left": "issue:39401", + "right": "issue:44779", + "accept": false, + "reason": "Both are tokenizer regressions, but model-specific offset mapping and DeepSeek output errors are not shown to share a root cause." + }, + { + "left": "issue:44410", + "right": "issue:45310", + "accept": false, + "reason": "Qwen3Next missing projections vs Qwen3.5 MoE from_pretrained failure; related family, but insufficient evidence of same underlying bug." + }, + { + "left": "issue:44466", + "right": "issue:46002", + "accept": false, + "reason": "Serialization inconsistency of tied weights vs NaNs after checkout/init; different failure modes." + }, + { + "left": "issue:43937", + "right": "issue:44977", + "accept": false, + "reason": "GenerationConfig invalid vs flash-attention generation failure; no clear shared code-path." + }, + { + "left": "issue:43450", + "right": "issue:43986", + "accept": false, + "reason": "Batched video processor shape bug vs AutoProcessor crash without torchvision; different causes." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 53, + "review_unit_count": 61, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-502", + "nodes": [ + "issue:38617", + "issue:39401", + "issue:41628", + "issue:42222", + "issue:42371", + "issue:42617", + "issue:43066", + "issue:43097", + "issue:43122", + "issue:43208", + "issue:43408", + "issue:43673", + "issue:43742", + "issue:43931", + "issue:43937", + "issue:44075", + "issue:44117", + "issue:44220", + "issue:44261", + "issue:44263", + "issue:44410", + "issue:44484", + "issue:44485", + "issue:44493", + "issue:44496", + "issue:44829", + "issue:44877", + "issue:44933", + "issue:45071", + "issue:45081", + "issue:45216", + "issue:45278", + "issue:45310", + "issue:45341", + "issue:45362", + "issue:45478", + "issue:45542", + "issue:45561", + "issue:45656", + "issue:46178", + "issue:46209", + "issue:46231", + "issue:46298", + "issue:46328", + "issue:46512", + "issue:46617", + "issue:46762", + "issue:46894" + ], + "soft_pairs": [ + "issue:45341|issue:46328", + "issue:45341|issue:45561", + "issue:44829|issue:45216", + "issue:45071|issue:46512", + "issue:44117|issue:45081", + "issue:38617|issue:41628", + "issue:43408|issue:43742", + "issue:43122|issue:46298", + "issue:42617|issue:45341", + "issue:45362|issue:46328", + "issue:42222|issue:46178", + "issue:44484|issue:45310", + "issue:42371|issue:44261", + "issue:44877|issue:46231", + "issue:43937|issue:44496", + "issue:39401|issue:46209", + "issue:44410|issue:45478", + "issue:42371|issue:46617", + "issue:44261|issue:46762", + "issue:44933|issue:45278", + "issue:43208|issue:46894", + "issue:38617|issue:44220", + "issue:44075|issue:45656", + "issue:43066|issue:43122", + "issue:44263|issue:44485", + "issue:43097|issue:44493", + "issue:43931|issue:44410", + "issue:43673|issue:45341", + "issue:41628|issue:45071", + "issue:42222|issue:45542", + "issue:45071|issue:45542", + "issue:44484|issue:45478" + ], + "prepared_review_unit_hash": "eb626806da11ff3737f8d444d6d529bdeeb6593e4b2760445d923a93af103cd2", + "component_budget": { + "node_count": 502, + "item_count": 502, + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 32, + "serialized_chars": 36865, + "estimated_input_tokens": 9217, + "estimated_eval_tokens": 18690 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a loose collection of distinct Transformers regression/bug reports, mostly around imports, tokenizers, model loading, and training behavior. I do not see any true duplicate pair among the provided soft edges.", + "confidence": 0.87, + "canonical_issue_reason": "issue:45278 is the broadest and most representative post-upgrade regression report; it best serves as an umbrella for the cluster\u2019s import/API-breakage theme, even though the other items are mostly separate bugs.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:45278 has the clearest cross-cutting impact and a general regression framing, making it the most useful anchor issue in an otherwise heterogeneous set.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:45341", + "right": "issue:46328", + "accept": false, + "reason": "One is a small testing_utils bug, the other is a SequenceFeatureExtractor.pad performance issue; different code paths and symptoms." + }, + { + "left": "issue:45341", + "right": "issue:45561", + "accept": false, + "reason": "Both mention testing utils, but one is a generic bug report and the other is a specific pytest-xdist race on captured_info.txt; not the same defect." + }, + { + "left": "issue:44829", + "right": "issue:45216", + "accept": false, + "reason": "Flash-attention training degeneration and save_pretrained checkpoint corruption are different failure modes in different paths." + }, + { + "left": "issue:45071", + "right": "issue:46512", + "accept": false, + "reason": "PretrainedConfig type-checking breakage is unrelated to from_config(dtype) leaving config.dtype stale." + }, + { + "left": "issue:44117", + "right": "issue:45081", + "accept": false, + "reason": "A TOKENIZER_MAPPING_NAMES None-handling bug is not the same as the Mistral regex patch crash in tokenizer backend access." + }, + { + "left": "issue:38617", + "right": "issue:41628", + "accept": false, + "reason": "Both are import errors, but they target different symbols and modules; no evidence of a shared root cause." + }, + { + "left": "issue:43408", + "right": "issue:43742", + "accept": false, + "reason": "Different model-loading failures for different models, with unrelated error surfaces." + }, + { + "left": "issue:43122", + "right": "issue:46298", + "accept": false, + "reason": "Tokenizer regression versus a noisy logging message; unrelated issues." + }, + { + "left": "issue:42617", + "right": "issue:45341", + "accept": false, + "reason": "Training script failure and a testing_utils bug are separate concerns." + }, + { + "left": "issue:45362", + "right": "issue:46328", + "accept": false, + "reason": "Chat crashes for Qwen3.5-35B and a padding performance issue are unrelated." + }, + { + "left": "issue:42222", + "right": "issue:46178", + "accept": false, + "reason": "VitPose model breakage and tensorflow_text backend mapping error affect different subsystems." + }, + { + "left": "issue:44484", + "right": "issue:45310", + "accept": false, + "reason": "A question about shard size defaults is not the same as a Qwen3.5 MoE from_pretrained error." + }, + { + "left": "issue:42371", + "right": "issue:44261", + "accept": false, + "reason": "TF32 API guidance and MLA rms_norm_eps precision drift are unrelated." + }, + { + "left": "issue:44877", + "right": "issue:46231", + "accept": false, + "reason": "Strict config loading for granite_speech is unrelated to a documentation parameter-name mismatch." + }, + { + "left": "issue:43937", + "right": "issue:44496", + "accept": false, + "reason": "GenerationConfig validation on GLM-5 and missing model_type in config.json are different loading problems." + }, + { + "left": "issue:39401", + "right": "issue:46209", + "accept": false, + "reason": "Qwen3 offset_mapping bug and FP8 inference failure on B200 are unrelated." + }, + { + "left": "issue:44410", + "right": "issue:45478", + "accept": false, + "reason": "Missing qwen3next projections and Qwen3.5 MoE from_pretrained errors are separate model-specific defects." + }, + { + "left": "issue:42371", + "right": "issue:46617", + "accept": false, + "reason": "TF32 configuration advice is unrelated to a cache FileNotFoundError for trust_remote_code loading." + }, + { + "left": "issue:44261", + "right": "issue:46762", + "accept": false, + "reason": "Both involve numerical differences, but they are different models and different attention implementations; not the same bug." + }, + { + "left": "issue:44933", + "right": "issue:45278", + "accept": false, + "reason": "A specific missing image_utils import is not enough to equate it with the broader post-upgrade import-error report." + }, + { + "left": "issue:43208", + "right": "issue:46894", + "accept": false, + "reason": "xLSTM training blockers and CohereASR double-shift loss bug are distinct training issues." + }, + { + "left": "issue:38617", + "right": "issue:44220", + "accept": false, + "reason": "An import name error and an audio feature extraction bug are unrelated." + }, + { + "left": "issue:44075", + "right": "issue:45656", + "accept": false, + "reason": "SGD args being ignored and optimizer stepping twice under deepspeed are different optimizer bugs." + }, + { + "left": "issue:43066", + "right": "issue:43122", + "accept": false, + "reason": "Tokenizer decoder type regression and tokenization differences across versions are related thematically but not shown to be the same underlying defect." + }, + { + "left": "issue:44263", + "right": "issue:44485", + "accept": false, + "reason": "A torch.split return-value issue and GLM-5 RoPE implementation concern different code paths." + }, + { + "left": "issue:43097", + "right": "issue:44493", + "accept": false, + "reason": "Removal of tie_embeddings_and_encoder_decoder and unexpected position-id keys are distinct API/serialization issues." + }, + { + "left": "issue:43931", + "right": "issue:44410", + "accept": false, + "reason": "Weight-shape mismatch for Qwen3-VL and missing projections in qwen3next are unrelated model-loading problems." + }, + { + "left": "issue:43673", + "right": "issue:45341", + "accept": false, + "reason": "GenerationMixin cache handling during chunked_prefill is unrelated to the testing_utils bug." + }, + { + "left": "issue:41628", + "right": "issue:45071", + "accept": false, + "reason": "Missing AutoImageProcessor import and PretrainedConfig type-checking regression are separate API breakages." + }, + { + "left": "issue:42222", + "right": "issue:45542", + "accept": false, + "reason": "VitPose model breakage and the tensorflow/tensorboard backend issue do not share a concrete root cause." + }, + { + "left": "issue:45071", + "right": "issue:45542", + "accept": false, + "reason": "PretrainedConfig type checking and the undefined tf backend error are unrelated." + }, + { + "left": "issue:44484", + "right": "issue:45478", + "accept": false, + "reason": "max_shard_size default behavior has nothing to do with the Qwen3.5 MoE from_pretrained error." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 54, + "review_unit_count": 61, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-502", + "nodes": [ + "issue:27957", + "issue:38617", + "issue:42757", + "issue:42890", + "issue:43012", + "issue:43064", + "issue:43097", + "issue:43208", + "issue:43352", + "issue:43441", + "issue:43452", + "issue:43531", + "issue:43577", + "issue:43673", + "issue:43874", + "issue:43931", + "issue:43986", + "issue:44038", + "issue:44117", + "issue:44261", + "issue:44263", + "issue:44315", + "issue:44351", + "issue:44360", + "issue:44410", + "issue:44483", + "issue:44484", + "issue:44485", + "issue:44829", + "issue:44987", + "issue:45003", + "issue:45341", + "issue:45372", + "issue:45375", + "issue:45440", + "issue:45464", + "issue:45588", + "issue:45753", + "issue:46050", + "issue:46123", + "issue:46172", + "issue:46298", + "issue:46369", + "issue:46655", + "issue:46664", + "issue:46710", + "issue:46752", + "issue:46897" + ], + "soft_pairs": [ + "issue:44483|issue:45464", + "issue:43874|issue:44261", + "issue:44263|issue:46172", + "issue:44360|issue:46172", + "issue:27957|issue:43352", + "issue:43931|issue:44038", + "issue:43012|issue:43577", + "issue:43441|issue:44829", + "issue:43097|issue:43452", + "issue:45003|issue:45341", + "issue:43986|issue:45341", + "issue:44261|issue:44485", + "issue:45753|issue:46369", + "issue:43874|issue:44263", + "issue:44117|issue:44987", + "issue:45440|issue:46172", + "issue:43531|issue:45341", + "issue:43673|issue:46123", + "issue:43097|issue:45588", + "issue:38617|issue:42757", + "issue:44038|issue:44410", + "issue:44410|issue:45375", + "issue:43208|issue:46897", + "issue:43064|issue:44315", + "issue:43986|issue:45372", + "issue:43097|issue:46050", + "issue:43986|issue:44484", + "issue:46655|issue:46664", + "issue:46172|issue:46710", + "issue:38617|issue:44351", + "issue:42890|issue:46298", + "issue:46298|issue:46752" + ], + "prepared_review_unit_hash": "ec4a09d90f14b6e4e5f4657d2ff147216ab28d5af594a777d72a14f601585488", + "component_budget": { + "node_count": 502, + "item_count": 502, + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 32, + "serialized_chars": 37139, + "estimated_input_tokens": 9285, + "estimated_eval_tokens": 18826 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This set is mostly a grab-bag of unrelated issue reports across different models and subsystems. No pull requests are present. None of the soft pairs look like the same underlying defect, so all are rejected.", + "confidence": 0.66, + "canonical_issue_reason": "If one issue has to anchor the cluster, issue 43097 is the broadest and most systemic regression report here; the others are mostly model-specific or unrelated. That said, this is not a true duplicate cluster.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 43097 is the strongest single representative because it describes a clear, reusable v5 regression with wider impact than the other entries, and is more central than the highly model-specific reports.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44483", + "right": "issue:45464", + "accept": false, + "reason": "Same API area, but different failures: a generic /v1/chat/completions rejection vs a Qwen3.5 streaming inference problem." + }, + { + "left": "issue:43874", + "right": "issue:44261", + "accept": false, + "reason": "Different models and different bugs: missing image-patch method vs missing rms_norm_eps config." + }, + { + "left": "issue:44263", + "right": "issue:46172", + "accept": false, + "reason": "Both are DeepSeek/GLM-adjacent math bugs, but torch.split return values and comb.to(dtype).transpose(-1, -2) are different code paths." + }, + { + "left": "issue:44360", + "right": "issue:46172", + "accept": false, + "reason": "DSA indexer lacking ReLU is unrelated to the DeepSeek-v4 dtype/transpose issue." + }, + { + "left": "issue:27957", + "right": "issue:43352", + "accept": false, + "reason": "Both mention Flash Attention 2 support, but they are separate model-specific compatibility gaps, not one shared bug." + }, + { + "left": "issue:43931", + "right": "issue:44038", + "accept": false, + "reason": "Both are Qwen3-VL loading failures, but shape mismatch and VL-Moe/version behavior are distinct defects." + }, + { + "left": "issue:43012", + "right": "issue:43577", + "accept": false, + "reason": "One is a compile-time warning about bf16 precision; the other is a dtype propagation/loading bug in BLIP2 submodules." + }, + { + "left": "issue:43441", + "right": "issue:44829", + "accept": false, + "reason": "Both involve FlashAttention, but one is a Ministral-3 compatibility crash and the other is a FlashAttention-3 training regression." + }, + { + "left": "issue:43097", + "right": "issue:43452", + "accept": false, + "reason": "Removed config behavior and gguf_file loading breakage are different issues and would not be fixed by one PR." + }, + { + "left": "issue:45003", + "right": "issue:45341", + "accept": false, + "reason": "Unrelated modules and bug classes: unsafe sys.modules access vs a testing_utils bug." + }, + { + "left": "issue:43986", + "right": "issue:45341", + "accept": false, + "reason": "Video AutoProcessor crash without torchvision is unrelated to the testing_utils issue." + }, + { + "left": "issue:44261", + "right": "issue:44485", + "accept": false, + "reason": "Both are GLM-family reports, but missing rms_norm_eps and RoPE implementation are separate defects." + }, + { + "left": "issue:45753", + "right": "issue:46369", + "accept": false, + "reason": "Different multimodal models and failure modes: one infinite loop, the other memory overflow." + }, + { + "left": "issue:43874", + "right": "issue:44263", + "accept": false, + "reason": "Both touch GLM internals, but the missing image-patch method and torch.split contract issue are not the same bug." + }, + { + "left": "issue:44117", + "right": "issue:44987", + "accept": false, + "reason": "Generic tokenizer mapping None handling is not the same as a specific physical-intelligence/fast load failure." + }, + { + "left": "issue:45440", + "right": "issue:46172", + "accept": false, + "reason": "Different DeepSeek generations and different tensor logic; not the same underlying bug." + }, + { + "left": "issue:43531", + "right": "issue:45341", + "accept": false, + "reason": "Qwen3-MoE sliding_window behavior is unrelated to a testing utility bug." + }, + { + "left": "issue:43673", + "right": "issue:46123", + "accept": false, + "reason": "Chunked-prefill cache handling and MaskGenerationPipeline partial-batch logic are different code paths." + }, + { + "left": "issue:43097", + "right": "issue:45588", + "accept": false, + "reason": "Tie-embeddings removal and the s_aux=None flash_attention crash are unrelated regressions." + }, + { + "left": "issue:38617", + "right": "issue:42757", + "accept": false, + "reason": "Both are import errors, but they involve different symbols, modules, and likely different dependency breakages." + }, + { + "left": "issue:44038", + "right": "issue:44410", + "accept": false, + "reason": "Different model families and different missing pieces: Qwen3-VL-Moe vs qwen3next layer projections." + }, + { + "left": "issue:44410", + "right": "issue:45375", + "accept": false, + "reason": "Both are Qwen3-family issues, but missing projections and a silently dropped config field are not the same defect." + }, + { + "left": "issue:43208", + "right": "issue:46897", + "accept": false, + "reason": "xLSTM training blockers are unrelated to Florence2's training-loss double-shift bug." + }, + { + "left": "issue:43064", + "right": "issue:44315", + "accept": false, + "reason": "FSDP2/PEFT optimizer-state corruption is not the same as Liger Kernel not being applied under model_init." + }, + { + "left": "issue:43986", + "right": "issue:45372", + "accept": false, + "reason": "Both are processor-loading ImportErrors, but they stem from different missing optional symbols/dependencies." + }, + { + "left": "issue:43097", + "right": "issue:46050", + "accept": false, + "reason": "Quantized save/load breakage is unrelated to the tie_embeddings_and_encoder_decoder removal." + }, + { + "left": "issue:43986", + "right": "issue:44484", + "accept": false, + "reason": "A torchvision-dependent video-model crash is not the same as a max_shard_size policy question." + }, + { + "left": "issue:46655", + "right": "issue:46664", + "accept": false, + "reason": "Same component area, but padding calculation and num_frames semantics are separate video-processor bugs." + }, + { + "left": "issue:46172", + "right": "issue:46710", + "accept": false, + "reason": "The DeepSeek-v4 tensor bug is not the same as a broader accuracy regression after upgrading Transformers." + }, + { + "left": "issue:38617", + "right": "issue:44351", + "accept": false, + "reason": "Both are import-name failures, but they are for different symbols in different modules." + }, + { + "left": "issue:42890", + "right": "issue:46298", + "accept": false, + "reason": "A missing set_seed in tests is unrelated to a noisy logging message." + }, + { + "left": "issue:46298", + "right": "issue:46752", + "accept": false, + "reason": "Noisy logging and empty-list chat template IndexError are unrelated bugs." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 55, + "review_unit_count": 61, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-502", + "nodes": [ + "issue:36683", + "issue:42175", + "issue:42490", + "issue:42757", + "issue:42907", + "issue:43012", + "issue:43064", + "issue:43066", + "issue:43097", + "issue:43208", + "issue:43352", + "issue:43502", + "issue:43526", + "issue:43550", + "issue:43582", + "issue:43824", + "issue:43825", + "issue:43873", + "issue:43906", + "issue:43976", + "issue:43986", + "issue:43992", + "issue:44062", + "issue:44077", + "issue:44117", + "issue:44162", + "issue:44222", + "issue:44368", + "issue:44380", + "issue:44484", + "issue:44655", + "issue:44704", + "issue:44936", + "issue:44938", + "issue:45103", + "issue:45478", + "issue:45479", + "issue:45584", + "issue:45712", + "issue:45753", + "issue:46132", + "issue:46172", + "issue:46178", + "issue:46298", + "issue:46365", + "issue:46421", + "issue:46701", + "issue:46821" + ], + "soft_pairs": [ + "issue:44062|issue:44117", + "issue:43097|issue:43873", + "issue:43012|issue:46821", + "issue:44117|issue:46701", + "issue:43502|issue:43992", + "issue:44936|issue:46365", + "issue:44704|issue:45103", + "issue:44380|issue:46421", + "issue:44938|issue:46178", + "issue:43986|issue:44117", + "issue:42907|issue:44655", + "issue:42907|issue:45753", + "issue:43825|issue:44655", + "issue:43012|issue:43550", + "issue:45712|issue:46298", + "issue:43208|issue:43550", + "issue:43012|issue:44368", + "issue:43992|issue:44484", + "issue:43352|issue:43976", + "issue:43352|issue:44380", + "issue:43066|issue:43906", + "issue:44162|issue:46172", + "issue:43097|issue:46298", + "issue:44380|issue:46132", + "issue:45584|issue:46365", + "issue:44077|issue:46365", + "issue:44222|issue:45478", + "issue:36683|issue:43582", + "issue:44380|issue:45479", + "issue:42175|issue:46365", + "issue:42757|issue:43824", + "issue:43064|issue:44380", + "issue:42490|issue:43012", + "issue:43208|issue:43526" + ], + "prepared_review_unit_hash": "3d929e79e4817e941f47c463a9d983ad819593ed05ef418b5cf163a1ec5a1b54", + "component_budget": { + "node_count": 502, + "item_count": 502, + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 34, + "serialized_chars": 37723, + "estimated_input_tokens": 9431, + "estimated_eval_tokens": 19118 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "These items are largely unrelated transformer bugs and regressions; none of the soft pairs looks like the same underlying defect, so I would not merge them into duplicate clusters.", + "confidence": 0.9, + "canonical_issue_reason": "Issue 43824 is the best single representative only because it is a concrete, well-documented runtime import failure with strong discussion and inbound references; the overall set is otherwise heterogeneous.", + "canonical_pr_reason": "No PRs are present in this cluster.", + "best_issue_reason": "Issue 43824 is the strongest anchor if one issue must represent the set: clear repro, specific failure mode, and high signal compared with the rest.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44062", + "right": "issue:44117", + "accept": false, + "reason": "Both involve tokenizers, but one is an AddedToken constructor bug and the other is a mapping/lookup assumption bug." + }, + { + "left": "issue:43097", + "right": "issue:43873", + "accept": false, + "reason": "Unrelated areas: v5 API removal vs quantization/offloading behavior." + }, + { + "left": "issue:43012", + "right": "issue:46821", + "accept": false, + "reason": "Different subsystems and symptoms: PyTorch precision warning vs TimesFM config AttributeError." + }, + { + "left": "issue:44117", + "right": "issue:46701", + "accept": false, + "reason": "Tokenizer mapping failure and diffusion pipeline failure are separate code paths." + }, + { + "left": "issue:43502", + "right": "issue:43992", + "accept": false, + "reason": "Local-files-only network leak vs missing checkpoint weight are distinct bugs." + }, + { + "left": "issue:44936", + "right": "issue:46365", + "accept": false, + "reason": "Training/evaluation state issue vs import-time dependency conflict." + }, + { + "left": "issue:44704", + "right": "issue:45103", + "accept": false, + "reason": "Pretrained kwargs forwarding and docstring annotation parsing are unrelated." + }, + { + "left": "issue:44380", + "right": "issue:46421", + "accept": false, + "reason": "Attention scaling/backend dispatch vs chunked-prefill cache initialization." + }, + { + "left": "issue:44938", + "right": "issue:46178", + "accept": false, + "reason": "Python version load failure is unrelated to a tensorflow_text backend mapping error." + }, + { + "left": "issue:43986", + "right": "issue:44117", + "accept": false, + "reason": "Video-model torchvision fallback issue vs tokenizer mapping None issue." + }, + { + "left": "issue:42907", + "right": "issue:44655", + "accept": false, + "reason": "Different save_pretrained paths: dequantized model save vs pipeline object serialization." + }, + { + "left": "issue:42907", + "right": "issue:45753", + "accept": false, + "reason": "One is a save failure; the other is an inference infinite loop on a specific image." + }, + { + "left": "issue:43825", + "right": "issue:44655", + "accept": false, + "reason": "Pipeline task messaging bug is not the same as pipeline save_pretrained failure." + }, + { + "left": "issue:43012", + "right": "issue:43550", + "accept": false, + "reason": "Precision warning during compilation is not the same as Bamba SDPA compile failure." + }, + { + "left": "issue:45712", + "right": "issue:46298", + "accept": false, + "reason": "Repository cleanup of dummy classes vs a noisy log message." + }, + { + "left": "issue:43208", + "right": "issue:43550", + "accept": false, + "reason": "xLSTM training blockers are unrelated to Bamba torch.compile/SDPA behavior." + }, + { + "left": "issue:43012", + "right": "issue:44368", + "accept": false, + "reason": "General precision warning vs tied-embedding config warning; different triggers and fixes." + }, + { + "left": "issue:43992", + "right": "issue:44484", + "accept": false, + "reason": "Missing encoder weights is unrelated to shard-size defaults." + }, + { + "left": "issue:43352", + "right": "issue:43976", + "accept": false, + "reason": "Model-specific FlashAttention support issue vs Python compatibility issue." + }, + { + "left": "issue:43352", + "right": "issue:44380", + "accept": false, + "reason": "Nemotron FlashAttention support and GPT2 attention scaling are different bugs." + }, + { + "left": "issue:43066", + "right": "issue:43906", + "accept": false, + "reason": "Tokenizer decoder typing regression is not the same as an isolated reproduction issue." + }, + { + "left": "issue:44162", + "right": "issue:46172", + "accept": false, + "reason": "ESM2 model breakage and a deepseek tensor operation error are unrelated." + }, + { + "left": "issue:43097", + "right": "issue:46298", + "accept": false, + "reason": "API removal and a noisy logging message do not share a code path." + }, + { + "left": "issue:44380", + "right": "issue:46132", + "accept": false, + "reason": "Both are attention-related, but one is backend-specific scaling and the other is register-side behavior change." + }, + { + "left": "issue:45584", + "right": "issue:46365", + "accept": false, + "reason": "Whisper generation edge case and Trainer import conflict are separate problems." + }, + { + "left": "issue:44077", + "right": "issue:46365", + "accept": false, + "reason": "Model initialization policy issue vs import dependency conflict." + }, + { + "left": "issue:44222", + "right": "issue:45478", + "accept": false, + "reason": "FP8 save_pretrained failure and Qwen3.5 MoE from_pretrained error are different directions of failure." + }, + { + "left": "issue:36683", + "right": "issue:43582", + "accept": false, + "reason": "Gemma config attribute error and Apple Silicon allocator TypeError are unrelated." + }, + { + "left": "issue:44380", + "right": "issue:45479", + "accept": false, + "reason": "Attention backend config bug vs classification loss degeneracy." + }, + { + "left": "issue:42175", + "right": "issue:46365", + "accept": false, + "reason": "Backend packaging issue is not the same as Trainer import failure." + }, + { + "left": "issue:42757", + "right": "issue:43824", + "accept": false, + "reason": "Both are import errors, but for different symbols and different packages." + }, + { + "left": "issue:43064", + "right": "issue:44380", + "accept": false, + "reason": "Distributed optimizer-state corruption and GPT2 attention scaling are unrelated." + }, + { + "left": "issue:42490", + "right": "issue:43012", + "accept": false, + "reason": "Trailing-slash local model warning and bf16 compile warning are separate behaviors." + }, + { + "left": "issue:43208", + "right": "issue:43526", + "accept": false, + "reason": "xLSTM training blockers and BeitImageProcessorFast label reduction bug do not match." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 56, + "review_unit_count": 61, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-502", + "nodes": [ + "issue:29870", + "issue:36010", + "issue:36296", + "issue:36683", + "issue:38617", + "issue:42757", + "issue:42994", + "issue:43010", + "issue:43064", + "issue:43208", + "issue:43299", + "issue:43352", + "issue:43475", + "issue:43502", + "issue:43525", + "issue:43611", + "issue:43618", + "issue:43756", + "issue:43761", + "issue:43867", + "issue:43901", + "issue:43937", + "issue:43976", + "issue:43992", + "issue:44038", + "issue:44075", + "issue:44077", + "issue:44230", + "issue:44355", + "issue:44380", + "issue:44483", + "issue:44485", + "issue:44492", + "issue:44514", + "issue:44625", + "issue:44655", + "issue:44938", + "issue:45216", + "issue:45276", + "issue:45278", + "issue:45290", + "issue:45310", + "issue:46077", + "issue:46082", + "issue:46365", + "issue:46829", + "issue:46877" + ], + "soft_pairs": [ + "issue:44483|issue:44514", + "issue:43618|issue:44380", + "issue:38617|issue:43502", + "issue:44038|issue:44230", + "issue:43756|issue:44485", + "issue:42994|issue:44655", + "issue:43618|issue:43761", + "issue:43299|issue:43867", + "issue:36010|issue:42757", + "issue:44625|issue:46365", + "issue:43352|issue:43937", + "issue:44355|issue:46877", + "issue:45276|issue:46077", + "issue:44483|issue:45290", + "issue:36683|issue:43475", + "issue:43611|issue:44655", + "issue:43901|issue:44655", + "issue:43867|issue:45216", + "issue:43976|issue:44938", + "issue:36683|issue:44077", + "issue:44625|issue:46077", + "issue:46077|issue:46082", + "issue:43618|issue:45216", + "issue:43208|issue:46829", + "issue:36683|issue:43525", + "issue:36683|issue:44625", + "issue:29870|issue:36296", + "issue:43992|issue:45310", + "issue:43064|issue:44492", + "issue:43010|issue:45278", + "issue:44075|issue:46365" + ], + "prepared_review_unit_hash": "3f1b467e74265e0a250735c115ff120b3113afb80ff2bf35763bd696bfe7bd42", + "component_budget": { + "node_count": 502, + "item_count": 502, + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 31, + "serialized_chars": 36073, + "estimated_input_tokens": 9019, + "estimated_eval_tokens": 18294 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is highly heterogeneous: most pairs are only loosely related by subsystem keywords (imports, loading, saving, attention, chat templates) but describe different concrete bugs. I would not merge any of the soft edges as duplicates.", + "confidence": 0.21, + "canonical_issue_reason": "issue:45278 is the broadest umbrella-style report in the set ('many import errors after update'), so it is the closest thing to a cluster anchor, though it is still not a true duplicate target for most others.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:45278 is the best global issue candidate only because it is the most general regression report; however, the set is too mixed for a strong canonical duplicate choice.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44483", + "right": "issue:44514", + "accept": false, + "reason": "Both involve chat/completions behavior, but one is the v5.3 /v1/chat/completions request acceptance path and the other is apply_chat_template batching with padding=False; different code paths and failure modes." + }, + { + "left": "issue:43618", + "right": "issue:44380", + "accept": false, + "reason": "Both mention attention behavior, but one is CLIPOutput losing attentions and the other is GPT-2 attention scaling being ignored under SDPA/FlashAttention; not the same bug." + }, + { + "left": "issue:38617", + "right": "issue:43502", + "accept": false, + "reason": "One is an import symbol breakage from transformers.configuration_utils; the other is offline/local_files_only still triggering API calls. Different root causes." + }, + { + "left": "issue:44038", + "right": "issue:44230", + "accept": false, + "reason": "Both reference Qwen3-VL-Moe, but one is a loading regression in transformers 5.0 and the other is fp8 inference support; not mergeable as one fix." + }, + { + "left": "issue:43756", + "right": "issue:44485", + "accept": false, + "reason": "Both mention RoPE, but one is a SmolLM3 layer-dropping bug and the other is a GLM-5 RoPE implementation discussion; different model-specific issues." + }, + { + "left": "issue:42994", + "right": "issue:44655", + "accept": false, + "reason": "Saving is the only overlap: quantized model serialization failure vs Pipeline.save_pretrained support. Different objects and failure modes." + }, + { + "left": "issue:43618", + "right": "issue:43761", + "accept": false, + "reason": "Both are CLIP v5 regressions, but one is attentions missing from output objects and the other is hidden_states not returned; separate return-field bugs." + }, + { + "left": "issue:43299", + "right": "issue:43867", + "accept": false, + "reason": "Both are loading problems, but one is Qwen3-VL-Moe model loading and the other is a sorted state_dict load error; too different to treat as the same bug." + }, + { + "left": "issue:36010", + "right": "issue:42757", + "accept": false, + "reason": "Both are ImportErrors, but for different symbols and different packages (transformers.generation vs huggingface_hub)." + }, + { + "left": "issue:44625", + "right": "issue:46365", + "accept": false, + "reason": "One is Qwen3.5 config propagation, the other is an import-time Trainer problem tied to datasets; unrelated." + }, + { + "left": "issue:43352", + "right": "issue:43937", + "accept": false, + "reason": "Different models and different failures: Flash Attention 2 unsupported vs invalid GenerationConfig." + }, + { + "left": "issue:44355", + "right": "issue:46877", + "accept": false, + "reason": "Running compiled Python files vs auto_docstring stdout noise at import time are unrelated issues." + }, + { + "left": "issue:45276", + "right": "issue:46077", + "accept": false, + "reason": "Both are Gemma4-related, but one is token embedding/output resizing and the other is bidirectional attention mask construction; different code paths." + }, + { + "left": "issue:44483", + "right": "issue:45290", + "accept": false, + "reason": "Both touch chat template/completions, but one is API request acceptance and the other is assistant tool-call templating with empty content; not the same concrete bug." + }, + { + "left": "issue:36683", + "right": "issue:43475", + "accept": false, + "reason": "Both are missing-attribute errors, but in different model families and unrelated fields (vocab_size vs fpn_position_embeddings)." + }, + { + "left": "issue:43611", + "right": "issue:44655", + "accept": false, + "reason": "One is loading models with base_model_prefix under v5; the other is saving Pipeline objects. Different operations and fixes." + }, + { + "left": "issue:43901", + "right": "issue:44655", + "accept": false, + "reason": "Docs mismatch for TextClassificationPipeline vs inability to save Pipeline objects; unrelated." + }, + { + "left": "issue:43867", + "right": "issue:45216", + "accept": false, + "reason": "Both involve model serialization/loading, but one is a sorted state_dict load failure and the other is a bad saved checkpoint for Qwen3.5; not clearly the same root cause." + }, + { + "left": "issue:43976", + "right": "issue:44938", + "accept": false, + "reason": "Both are Python-version compatibility reports, but for different versions and likely different breakpoints; not the same bug." + }, + { + "left": "issue:36683", + "right": "issue:44077", + "accept": false, + "reason": "Gemma3 config attribute missing vs patchtsmixer post_init policy; no substantive overlap." + }, + { + "left": "issue:44625", + "right": "issue:46077", + "accept": false, + "reason": "Qwen3.5 num_labels propagation vs Gemma4 bidirectional attention masking are unrelated model-specific bugs." + }, + { + "left": "issue:46077", + "right": "issue:46082", + "accept": false, + "reason": "Both are config/behavior issues, but for different models and different constraints (Gemma4 attention masking vs LlamaConfig head_dim validation)." + }, + { + "left": "issue:43618", + "right": "issue:45216", + "accept": false, + "reason": "Attention-output regression in CLIP vs corrupted Qwen3.5 save_pretrained checkpoint; different components." + }, + { + "left": "issue:43208", + "right": "issue:46829", + "accept": false, + "reason": "xLSTM training bugs and VideoPrism output-field confusion are unrelated." + }, + { + "left": "issue:36683", + "right": "issue:43525", + "accept": false, + "reason": "Both are config attribute errors, but for different attributes and different models; no shared code-path bug is evident." + }, + { + "left": "issue:36683", + "right": "issue:44625", + "accept": false, + "reason": "Gemma3 missing vocab_size and Qwen3.5 num_labels propagation are different configuration issues." + }, + { + "left": "issue:29870", + "right": "issue:36296", + "accept": false, + "reason": "BART generate() decoder_attention_mask initialization bug vs tensor parallel training bug; unrelated." + }, + { + "left": "issue:43992", + "right": "issue:45310", + "accept": false, + "reason": "Both are from_pretrained/loading regressions, but they affect different models and appear to stem from different missing keys or version issues." + }, + { + "left": "issue:43064", + "right": "issue:44492", + "accept": false, + "reason": "FSDP2/PEFT optimizer-state corruption vs a cache-strategy typo; no overlap." + }, + { + "left": "issue:43010", + "right": "issue:45278", + "accept": false, + "reason": "Cache/update no_grad behavior vs broad import errors after upgrade; entirely different bugs." + }, + { + "left": "issue:44075", + "right": "issue:46365", + "accept": false, + "reason": "SGD optimizer args ignored vs Trainer import failure tied to datasets installation; unrelated." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 57, + "review_unit_count": 61, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-502", + "nodes": [ + "issue:28282", + "issue:29870", + "issue:33453", + "issue:36296", + "issue:36683", + "issue:41669", + "issue:42890", + "issue:42898", + "issue:43010", + "issue:43064", + "issue:43502", + "issue:43575", + "issue:43618", + "issue:43723", + "issue:43824", + "issue:43867", + "issue:43874", + "issue:43976", + "issue:43992", + "issue:44062", + "issue:44077", + "issue:44112", + "issue:44222", + "issue:44246", + "issue:44355", + "issue:44360", + "issue:44387", + "issue:44485", + "issue:44556", + "issue:44734", + "issue:44855", + "issue:44908", + "issue:44938", + "issue:44964", + "issue:45071", + "issue:45125", + "issue:45357", + "issue:45375", + "issue:45440", + "issue:45478", + "issue:45561", + "issue:45600", + "issue:45850", + "issue:46202", + "issue:46285", + "issue:46617", + "issue:46664", + "issue:46851" + ], + "soft_pairs": [ + "issue:44908|issue:46851", + "issue:29870|issue:33453", + "issue:44938|issue:45071", + "issue:44355|issue:44855", + "issue:43867|issue:46202", + "issue:36683|issue:43064", + "issue:43867|issue:45375", + "issue:42890|issue:45561", + "issue:43010|issue:45850", + "issue:41669|issue:44246", + "issue:36296|issue:43575", + "issue:43010|issue:43976", + "issue:43618|issue:44077", + "issue:43992|issue:45478", + "issue:43867|issue:45357", + "issue:28282|issue:33453", + "issue:43618|issue:46285", + "issue:44360|issue:45440", + "issue:42890|issue:44964", + "issue:42890|issue:44112", + "issue:42890|issue:45600", + "issue:42898|issue:46664", + "issue:43502|issue:44062", + "issue:44734|issue:46617", + "issue:43723|issue:44556", + "issue:44222|issue:45125", + "issue:44387|issue:44556", + "issue:43824|issue:46617", + "issue:43502|issue:44485", + "issue:43874|issue:46617" + ], + "prepared_review_unit_hash": "798ae6b9e0c779257a5d0d0c57f9e46754f90f74cca85060bcbbdf7c055dd67f", + "component_budget": { + "node_count": 502, + "item_count": 502, + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 30, + "serialized_chars": 36468, + "estimated_input_tokens": 9117, + "estimated_eval_tokens": 18490 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Mostly unrelated issues grouped by superficial similarity (v5 regressions, loading problems, test flakiness). Only one pair looks like a likely true duplicate around slow `transformers` import time.", + "confidence": 0.78, + "canonical_issue_reason": "issue:41669 is the clearest, most specific root-cause report in the set: it identifies the import-time slowdown and points to the `import *` usage in models as the likely cause.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:41669 is the best representative issue because it is actionable, has a concrete symptom/cause pairing, and appears to subsume the broader import-slowdown report.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44908", + "right": "issue:46851", + "accept": false, + "reason": "Different bugs: scheduler kwargs propagation vs Qwen3.5 variable-length position_ids handling." + }, + { + "left": "issue:29870", + "right": "issue:33453", + "accept": false, + "reason": "Bart generate-time decoder attention mask error and tokenizer-loading regression are unrelated symptoms/code paths." + }, + { + "left": "issue:44938", + "right": "issue:45071", + "accept": false, + "reason": "Both mention version compatibility, but one is Python 3.14 import failure and the other is `PretrainedConfig` type checking; not the same bug." + }, + { + "left": "issue:44355", + "right": "issue:44855", + "accept": false, + "reason": "One is a generic compiled-file import error; the other is a specific Python 3.13 `torch.jit.script` parsing issue in DebertaV2Model." + }, + { + "left": "issue:43867", + "right": "issue:46202", + "accept": false, + "reason": "A sorted-state_dict loading error is not the same as Gemma pipeline roundtrip failure on the underline token." + }, + { + "left": "issue:36683", + "right": "issue:43064", + "accept": false, + "reason": "Gemma3Config missing `vocab_size` and FSDP2/PEFT optimizer state corruption are different failures." + }, + { + "left": "issue:43867", + "right": "issue:45375", + "accept": false, + "reason": "Different models and different failure modes: state_dict loading vs missing vision config field." + }, + { + "left": "issue:42890", + "right": "issue:45561", + "accept": false, + "reason": "Both are flaky-test related, but one is missing `set_seed()` in SAM HQ tests and the other is an xdist file race." + }, + { + "left": "issue:43010", + "right": "issue:45850", + "accept": false, + "reason": "Decoratoring cache `update()` with `torch.no_grad` is a specific fix; \u201cbreaks too many models\u201d is a broad complaint, not the same issue." + }, + { + "left": "issue:41669", + "right": "issue:44246", + "accept": true, + "reason": "Both point to the same underlying problem: `transformers` import is unexpectedly slow, and 41669 gives the likely root cause in model-side `import *` usage." + }, + { + "left": "issue:36296", + "right": "issue:43575", + "accept": false, + "reason": "Tensor-parallel training bug and Qwen2-57B-A14B-Instruct OOM under tp are related only at a broad subsystem level, not the same concrete bug." + }, + { + "left": "issue:43010", + "right": "issue:43976", + "accept": false, + "reason": "A no-grad cache update change and a Python version support issue are unrelated." + }, + { + "left": "issue:43618", + "right": "issue:44077", + "accept": false, + "reason": "CLIP attentions assignment and patchtsmixer `post_init` policy are different model-specific issues." + }, + { + "left": "issue:43992", + "right": "issue:45478", + "accept": false, + "reason": "UMT5Encoder missing weights and Qwen3.5 Moe `from_pretrained` error are different model-loading problems." + }, + { + "left": "issue:43867", + "right": "issue:45357", + "accept": false, + "reason": "State-dict loading failure and incorrect visual encoder keys on save are different sides of different bugs." + }, + { + "left": "issue:28282", + "right": "issue:33453", + "accept": false, + "reason": "PyTorch missing ImportError and tokenizer-loading regression are unrelated." + }, + { + "left": "issue:43618", + "right": "issue:46285", + "accept": false, + "reason": "CLIP attentions assignment and CLIP text-model structure inconsistency are distinct changes." + }, + { + "left": "issue:44360", + "right": "issue:45440", + "accept": false, + "reason": "DSA indexer ReLU discussion and DeepSeekV3 MoE divergence are unrelated." + }, + { + "left": "issue:42890", + "right": "issue:44964", + "accept": false, + "reason": "SAM HQ test seeding flakiness is unrelated to Phi-4 multimodal model loading." + }, + { + "left": "issue:42890", + "right": "issue:44112", + "accept": false, + "reason": "Different CI/test failures: missing seeds vs stale device override in GraniteSpeech." + }, + { + "left": "issue:42890", + "right": "issue:45600", + "accept": false, + "reason": "Test seeding flakiness and auto-mappings referencing removed configs are not the same bug." + }, + { + "left": "issue:42898", + "right": "issue:46664", + "accept": false, + "reason": "Tokenization cleanup behavior changes and SmolVLM `num_frames` semantics are unrelated." + }, + { + "left": "issue:43502", + "right": "issue:44062", + "accept": false, + "reason": "Local-files-only network leakage and AddedToken keyword duplication are unrelated tokenizer/loading issues." + }, + { + "left": "issue:44734", + "right": "issue:46617", + "accept": false, + "reason": "KV-cache continuation indexing crash and trust_remote_code cache FileNotFoundError are different server/loading failures." + }, + { + "left": "issue:43723", + "right": "issue:44556", + "accept": false, + "reason": "Tokenizer `from_pretrained` regression and checkpoint reload incompatibility across v4/v5 are not the same concrete bug." + }, + { + "left": "issue:44222", + "right": "issue:45125", + "accept": false, + "reason": "FP8 save_pretrained for MoE and missing tensor-parallel plan are separate features/bugs." + }, + { + "left": "issue:44387", + "right": "issue:44556", + "accept": false, + "reason": "Int4 quantization reserved-memory/OOM and checkpoint reload incompatibility are different regression classes." + }, + { + "left": "issue:43824", + "right": "issue:46617", + "accept": false, + "reason": "Missing exported Qwen2_5_VL class and cache FileNotFoundError are unrelated." + }, + { + "left": "issue:43502", + "right": "issue:44485", + "accept": false, + "reason": "API requests despite `local_files_only=True` and GLM-5 RoPE implementation are unrelated." + }, + { + "left": "issue:43874", + "right": "issue:46617", + "accept": false, + "reason": "Missing image-patch helper and cache FileNotFoundError are different issues." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is well grounded in the packet and stays conservative: it identifies only one likely duplicate pair (41669 vs 44246) and the rest of the soft-edge rejections are appropriately cautious. I don\u2019t see unsupported escalation or overstatement." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 58, + "review_unit_count": 61, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-502", + "nodes": [ + "issue:29870", + "issue:30333", + "issue:30990", + "issue:31795", + "issue:33453", + "issue:34689", + "issue:36296", + "issue:38617", + "issue:41628", + "issue:42548", + "issue:42757", + "issue:43010", + "issue:43065", + "issue:43232", + "issue:43299", + "issue:43475", + "issue:43502", + "issue:43576", + "issue:43673", + "issue:43824", + "issue:43827", + "issue:43901", + "issue:44222", + "issue:44230", + "issue:44315", + "issue:44351", + "issue:44355", + "issue:44360", + "issue:44485", + "issue:44492", + "issue:44556", + "issue:44749", + "issue:44908", + "issue:44938", + "issue:44961", + "issue:45071", + "issue:45125", + "issue:45375", + "issue:45468", + "issue:45478", + "issue:46132", + "issue:46143", + "issue:46144", + "issue:46209", + "issue:46617", + "issue:46664", + "issue:46726" + ], + "soft_pairs": [ + "issue:43576|issue:46132", + "issue:43010|issue:43065", + "issue:45125|issue:45478", + "issue:30333|issue:33453", + "issue:44355|issue:44749", + "issue:42548|issue:44351", + "issue:45125|issue:45375", + "issue:46664|issue:46726", + "issue:43502|issue:44961", + "issue:43502|issue:44222", + "issue:43010|issue:43232", + "issue:30990|issue:31795", + "issue:43827|issue:44961", + "issue:30333|issue:36296", + "issue:43299|issue:44556", + "issue:29870|issue:31795", + "issue:44355|issue:45071", + "issue:33453|issue:34689", + "issue:42757|issue:43576", + "issue:43576|issue:45071", + "issue:44315|issue:44360", + "issue:31795|issue:38617", + "issue:44961|issue:45468", + "issue:43576|issue:43824", + "issue:44485|issue:44961", + "issue:44351|issue:46617", + "issue:44908|issue:44961", + "issue:43576|issue:44938", + "issue:36296|issue:43901", + "issue:46143|issue:46144", + "issue:43475|issue:46617", + "issue:44222|issue:44230", + "issue:41628|issue:42548", + "issue:44315|issue:46209", + "issue:44485|issue:46664", + "issue:43673|issue:44492" + ], + "prepared_review_unit_hash": "822bff15e8b00b9e0d044cd004a7bd90be32f93dc4a24fe92e6f1954d491f7c1", + "component_budget": { + "node_count": 502, + "item_count": 502, + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 36, + "serialized_chars": 37273, + "estimated_input_tokens": 9319, + "estimated_eval_tokens": 18894 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is not a true duplicate set; it mixes unrelated Transformers v5-era regressions, import errors, generation bugs, docs issues, and model-specific loading failures. None of the soft pairs look like the same underlying bug/change closely enough to merge.", + "confidence": 0.79, + "canonical_issue_reason": "Issue 45478 is the most concrete, actionable representative: it describes a specific versioned `from_pretrained` regression for Qwen3.5 MoE with a clear failure mode and affected path.", + "canonical_pr_reason": null, + "best_issue_reason": "45478 is the strongest stand-alone triage target in the set because it has the clearest repro surface, explicit version boundary, and a narrow model-loading code path.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43576", + "right": "issue:46132", + "accept": false, + "reason": "CLI/env command breakage and AttentionInterface.register behavior are different subsystems and different bugs." + }, + { + "left": "issue:43010", + "right": "issue:43065", + "accept": false, + "reason": "Cache/layer `update()` no_grad decoration and SAM3 pixel decoder dummy conv are unrelated code paths." + }, + { + "left": "issue:45125", + "right": "issue:45478", + "accept": false, + "reason": "Both involve Qwen3.5 MoE, but one is tensor-parallel `_tp_plan` support and the other is a `from_pretrained` loading regression." + }, + { + "left": "issue:30333", + "right": "issue:33453", + "accept": false, + "reason": "MLflow job-state reporting and tokenizer loading are unrelated failures." + }, + { + "left": "issue:44355", + "right": "issue:44749", + "accept": false, + "reason": "Compiled Python file errors and upgrade-related data slowdown are different problems." + }, + { + "left": "issue:42548", + "right": "issue:44351", + "accept": false, + "reason": "Missing `PreTrainedModel` export and missing `HybridCache` export are separate import regressions." + }, + { + "left": "issue:45125", + "right": "issue:45375", + "accept": false, + "reason": "Tensor-parallel plan support and a missing vision config field are not the same bug." + }, + { + "left": "issue:46664", + "right": "issue:46726", + "accept": false, + "reason": "Video `num_frames` semantics and missing vision-backend handling are distinct issues." + }, + { + "left": "issue:43502", + "right": "issue:44961", + "accept": false, + "reason": "`local_files_only=True` network leakage and an unrelated placeholder issue do not match." + }, + { + "left": "issue:43502", + "right": "issue:44222", + "accept": false, + "reason": "Offline-loading behavior and FP8 `save_pretrained` are unrelated." + }, + { + "left": "issue:43010", + "right": "issue:43232", + "accept": false, + "reason": "Both touch generation internals, but the cache `update()` no_grad issue and sync_gpus kwargs handling are different root causes." + }, + { + "left": "issue:30990", + "right": "issue:31795", + "accept": false, + "reason": "Loading hang and documentation confusion about `input_ids`/`past_key_values` are not duplicates." + }, + { + "left": "issue:43827", + "right": "issue:44961", + "accept": false, + "reason": "Pipeline-removal docs breakage and an unrelated placeholder issue are not the same change." + }, + { + "left": "issue:30333", + "right": "issue:36296", + "accept": false, + "reason": "MLflow reporting and tensor-parallel training are unrelated." + }, + { + "left": "issue:43299", + "right": "issue:44556", + "accept": false, + "reason": "Both are version-upgrade load regressions, but on different models and with different symptoms; too broad to merge." + }, + { + "left": "issue:29870", + "right": "issue:31795", + "accept": false, + "reason": "A concrete `generate()` runtime error and a documentation wording issue are different classes of bug." + }, + { + "left": "issue:44355", + "right": "issue:45071", + "accept": false, + "reason": "Compiled-file execution errors and `PretrainedConfig` type-checking are unrelated." + }, + { + "left": "issue:33453", + "right": "issue:34689", + "accept": false, + "reason": "Tokenizer-loading regression and Llama 3.2 Vision model-loading breakage are different model/load paths." + }, + { + "left": "issue:42757", + "right": "issue:43576", + "accept": false, + "reason": "Missing `is_offline_mode` import from `huggingface_hub` and a broken `transformers env` command are unrelated." + }, + { + "left": "issue:43576", + "right": "issue:45071", + "accept": false, + "reason": "CLI breakage and config type-checking regression are different problems." + }, + { + "left": "issue:44315", + "right": "issue:44360", + "accept": false, + "reason": "Liger Kernel application in `model_init` and a DSA indexer discussion are not the same issue." + }, + { + "left": "issue:31795", + "right": "issue:38617", + "accept": false, + "reason": "Docs confusion and missing `layer_type_validation` import are unrelated." + }, + { + "left": "issue:44961", + "right": "issue:45468", + "accept": false, + "reason": "Placeholder text and Gemma-4 audio positional encoding are unrelated." + }, + { + "left": "issue:43576", + "right": "issue:43824", + "accept": false, + "reason": "Broken CLI env command and missing `Qwen2_5_VLForConditionalGeneration` export are different failures." + }, + { + "left": "issue:44485", + "right": "issue:44961", + "accept": false, + "reason": "GLM-5 RoPE implementation and an unrelated placeholder issue do not match." + }, + { + "left": "issue:44351", + "right": "issue:46617", + "accept": false, + "reason": "Missing `HybridCache` export and FileNotFoundError when loading cached trust_remote_code models are different bug classes." + }, + { + "left": "issue:44908", + "right": "issue:44961", + "accept": false, + "reason": "Scheduler kwargs handling and an unrelated placeholder issue are not duplicates." + }, + { + "left": "issue:43576", + "right": "issue:44938", + "accept": false, + "reason": "Broken `transformers env` command and Python 3.14 load failure are different issues." + }, + { + "left": "issue:36296", + "right": "issue:43901", + "accept": false, + "reason": "Tensor-parallel training bug and docs mentioning `return_all_scores` are unrelated." + }, + { + "left": "issue:46143", + "right": "issue:46144", + "accept": false, + "reason": "Both are RoFormer-related, but one is missing `**kwargs` passthrough and the other is attention-interface integration; not the same concrete bug." + }, + { + "left": "issue:43475", + "right": "issue:46617", + "accept": false, + "reason": "SAM3 vision output attribute error and cached trust_remote_code FileNotFoundError are different failures." + }, + { + "left": "issue:44222", + "right": "issue:44230", + "accept": false, + "reason": "FP8 `save_pretrained` bug and FP8 inference support request are not the same change." + }, + { + "left": "issue:41628", + "right": "issue:42548", + "accept": false, + "reason": "Missing `AutoImageProcessor` and missing `PreTrainedModel` imports are separate exports/regressions." + }, + { + "left": "issue:44315", + "right": "issue:46209", + "accept": false, + "reason": "Liger Kernel/model-init application and FP8 inference on B200 are unrelated." + }, + { + "left": "issue:44485", + "right": "issue:46664", + "accept": false, + "reason": "RoPE implementation details and video processor frame semantics are unrelated." + }, + { + "left": "issue:43673", + "right": "issue:44492", + "accept": false, + "reason": "Generation cache missing in v5 and a typo in cache strategies are not the same bug." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 59, + "review_unit_count": 61, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-502", + "nodes": [ + "issue:31795", + "issue:33453", + "issue:41669", + "issue:42371", + "issue:42890", + "issue:43352", + "issue:43408", + "issue:43525", + "issue:43572", + "issue:43576", + "issue:43673", + "issue:43976", + "issue:44038", + "issue:44077", + "issue:44230", + "issue:44246", + "issue:44273", + "issue:44295", + "issue:44297", + "issue:44315", + "issue:44351", + "issue:44360", + "issue:44393", + "issue:44492", + "issue:44556", + "issue:44623", + "issue:44704", + "issue:44829", + "issue:44861", + "issue:44908", + "issue:45003", + "issue:45071", + "issue:45092", + "issue:45125", + "issue:45230", + "issue:45278", + "issue:45341", + "issue:45362", + "issue:45468", + "issue:45478", + "issue:45542", + "issue:45753", + "issue:45850", + "issue:46095", + "issue:46218", + "issue:46328", + "issue:46664" + ], + "soft_pairs": [ + "issue:45362|issue:45850", + "issue:44273|issue:45542", + "issue:44623|issue:45230", + "issue:44360|issue:46664", + "issue:44038|issue:45125", + "issue:44273|issue:45003", + "issue:44393|issue:45071", + "issue:44393|issue:45753", + "issue:43576|issue:44556", + "issue:41669|issue:43673", + "issue:43525|issue:45850", + "issue:44556|issue:45092", + "issue:43572|issue:45850", + "issue:44492|issue:45850", + "issue:45278|issue:45850", + "issue:44230|issue:46095", + "issue:42371|issue:44492", + "issue:41669|issue:46218", + "issue:43408|issue:44315", + "issue:42890|issue:44908", + "issue:44246|issue:44273", + "issue:43576|issue:44908", + "issue:43352|issue:44393", + "issue:43976|issue:44393", + "issue:44908|issue:45071", + "issue:44295|issue:45230", + "issue:44297|issue:45230", + "issue:44704|issue:44908", + "issue:44492|issue:46328", + "issue:41669|issue:44351", + "issue:41669|issue:45341", + "issue:43976|issue:45850", + "issue:44393|issue:45478", + "issue:43976|issue:46218", + "issue:44077|issue:44861", + "issue:31795|issue:33453", + "issue:43576|issue:44273", + "issue:44315|issue:44829", + "issue:44360|issue:45468" + ], + "prepared_review_unit_hash": "a9c656f1fa3c4b3a826725c123107e7692fed4f07ec1cad0c25334ae23f836c8", + "component_budget": { + "node_count": 502, + "item_count": 502, + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 39, + "serialized_chars": 38227, + "estimated_input_tokens": 9557, + "estimated_eval_tokens": 19370 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a loose collection of unrelated Transformers v5 regressions, model-specific incompatibilities, and a few documentation/performance issues. None of the soft pairs looks like the same concrete underlying bug or change.", + "confidence": 0.96, + "canonical_issue_reason": "issue:45850 is the broadest representative of the cluster\u2019s recurring theme of v5 breakage complaints, but it is still only a loose umbrella, not a true duplicate hub.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:45850 is the best global issue candidate because it most directly captures the cluster\u2019s dominant user-visible regression pattern across many models and upgrade paths.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:45362", + "right": "issue:45850", + "accept": false, + "reason": "Different bugs: Qwen3.5 chat crash vs generic v5 model-breakage complaint." + }, + { + "left": "issue:44273", + "right": "issue:45542", + "accept": false, + "reason": "Lazy loading failure and tf backend error are unrelated code paths." + }, + { + "left": "issue:44623", + "right": "issue:45230", + "accept": false, + "reason": "Specific processor save_pretrained file bug vs a generic bug report with no concrete overlap." + }, + { + "left": "issue:44360", + "right": "issue:46664", + "accept": false, + "reason": "DSA indexer ReLU discussion and video num_frames semantics are unrelated." + }, + { + "left": "issue:44038", + "right": "issue:45125", + "accept": false, + "reason": "Same model family, but different failures: unsupported Qwen3-VL-Moe path vs missing tensor-parallel plan." + }, + { + "left": "issue:44273", + "right": "issue:45003", + "accept": false, + "reason": "Both involve import/lazy-loading concerns, but the actual bugs and fixes are different." + }, + { + "left": "issue:44393", + "right": "issue:45071", + "accept": false, + "reason": "Qwen3-VL output bug and PretrainedConfig type-checking regression are unrelated." + }, + { + "left": "issue:44393", + "right": "issue:45753", + "accept": false, + "reason": "Both are Qwen3.5-VL related, but one is bbox output failure and the other is an infinite-loop image-specific bug." + }, + { + "left": "issue:43576", + "right": "issue:44556", + "accept": false, + "reason": "Broken env command and checkpoint reload incompatibility are different upgrade regressions." + }, + { + "left": "issue:41669", + "right": "issue:43673", + "accept": false, + "reason": "Import-* / lazy-loading cleanup vs chunked_prefill cache missing are distinct issues." + }, + { + "left": "issue:43525", + "right": "issue:45850", + "accept": false, + "reason": "Llama4Config pad_token_id attribute bug is model-specific and not the same as broad v5 breakage." + }, + { + "left": "issue:44556", + "right": "issue:45092", + "accept": false, + "reason": "Both are upgrade-related, but one is checkpoint reload failure and the other is remote-code/meta-init incompatibility." + }, + { + "left": "issue:43572", + "right": "issue:45850", + "accept": false, + "reason": "StableLmConfig pad token attribute regression is separate from the generic v5 breakage complaint." + }, + { + "left": "issue:44492", + "right": "issue:45850", + "accept": false, + "reason": "Cache-strategy typo and broad model breakage are not the same underlying bug." + }, + { + "left": "issue:45278", + "right": "issue:45850", + "accept": false, + "reason": "Broad upgrade complaints overlap only thematically; one is import errors, the other generic model breakage." + }, + { + "left": "issue:44230", + "right": "issue:46095", + "accept": false, + "reason": "FP8 support request for Qwen3-VL is unrelated to a DeepSeek weight-conversion question." + }, + { + "left": "issue:42371", + "right": "issue:44492", + "accept": false, + "reason": "TF32 settings API and cache-strategy typo are unrelated." + }, + { + "left": "issue:41669", + "right": "issue:46218", + "accept": false, + "reason": "Import-* lazy-loading issue and missing Qwen3_5 module import are different failures." + }, + { + "left": "issue:43408", + "right": "issue:44315", + "accept": false, + "reason": "Sam3 tracker/model mismatch and Liger Kernel not applied with model_init are unrelated." + }, + { + "left": "issue:42890", + "right": "issue:44908", + "accept": false, + "reason": "Missing test seeds and inverse_sqrt scheduler kwargs are different bugs." + }, + { + "left": "issue:44246", + "right": "issue:44273", + "accept": false, + "reason": "Slow import sometimes and lazy loading broken are related in topic only, not the same defect." + }, + { + "left": "issue:43576", + "right": "issue:44908", + "accept": false, + "reason": "Environment command regression and scheduler kwargs bug are unrelated." + }, + { + "left": "issue:43352", + "right": "issue:44393", + "accept": false, + "reason": "Flash Attention 2 support error for Nemotron and Qwen3-VL bbox hallucination are different model issues." + }, + { + "left": "issue:43976", + "right": "issue:44393", + "accept": false, + "reason": "Python version compatibility regression and Qwen3-VL output bug do not share a concrete code path." + }, + { + "left": "issue:44908", + "right": "issue:45071", + "accept": false, + "reason": "Scheduler kwargs handling and PretrainedConfig type checking are unrelated." + }, + { + "left": "issue:44295", + "right": "issue:45230", + "accept": false, + "reason": "Position_ids buffer access error and a generic bug report are not duplicates." + }, + { + "left": "issue:44297", + "right": "issue:45230", + "accept": false, + "reason": "Tokenizer save_pretrained mismatch is a concrete tokenizer bug, unlike the generic bug report." + }, + { + "left": "issue:44704", + "right": "issue:44908", + "accept": false, + "reason": "AutoProcessor cached_file kwargs forwarding and scheduler kwargs are different subsystems." + }, + { + "left": "issue:44492", + "right": "issue:46328", + "accept": false, + "reason": "Cache-strategy typo and SequenceFeatureExtractor.pad performance issue are unrelated." + }, + { + "left": "issue:41669", + "right": "issue:44351", + "accept": false, + "reason": "Both are import-related, but one is import-* cleanup and the other is HybridCache export/import failure." + }, + { + "left": "issue:41669", + "right": "issue:45341", + "accept": false, + "reason": "Import cleanup regression and a testing_utils bug are not the same issue." + }, + { + "left": "issue:43976", + "right": "issue:45850", + "accept": false, + "reason": "Python-version incompatibility and generic v5 model breakage are only broadly related." + }, + { + "left": "issue:44393", + "right": "issue:45478", + "accept": false, + "reason": "Qwen3-VL bbox error and Qwen3.5 MoE from_pretrained failure are different model-specific bugs." + }, + { + "left": "issue:43976", + "right": "issue:46218", + "accept": false, + "reason": "Python support regression and missing Qwen3_5ForCausalLM import are unrelated." + }, + { + "left": "issue:44077", + "right": "issue:44861", + "accept": false, + "reason": "Patchtsmixer post_init validation and tied-weight-key crash are different code paths." + }, + { + "left": "issue:31795", + "right": "issue:33453", + "accept": false, + "reason": "Documentation confusion and tokenizer-loading regression are not the same underlying bug." + }, + { + "left": "issue:43576", + "right": "issue:44273", + "accept": false, + "reason": "Broken env command and lazy loading malfunction are separate v5 infra issues." + }, + { + "left": "issue:44315", + "right": "issue:44829", + "accept": false, + "reason": "Liger Kernel application problem and flash_attention_3 degenerate training are distinct bugs." + }, + { + "left": "issue:44360", + "right": "issue:45468", + "accept": false, + "reason": "DSA indexer discussion and Gemma-4 audio positional encoding bug are unrelated." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 60, + "review_unit_count": 61, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-502", + "nodes": [ + "issue:28348", + "issue:29870", + "issue:30333", + "issue:31515", + "issue:31795", + "issue:34689", + "issue:36296", + "issue:38617", + "issue:41669", + "issue:42371", + "issue:42913", + "issue:43066", + "issue:43298", + "issue:43352", + "issue:43502", + "issue:43519", + "issue:43525", + "issue:43572", + "issue:43576", + "issue:43618", + "issue:43704", + "issue:43717", + "issue:44230", + "issue:44261", + "issue:44263", + "issue:44273", + "issue:44351", + "issue:44485", + "issue:44492", + "issue:44623", + "issue:44655", + "issue:44861", + "issue:44908", + "issue:44938", + "issue:44998", + "issue:45230", + "issue:45250", + "issue:45290", + "issue:45310", + "issue:45468", + "issue:45850", + "issue:46018", + "issue:46095", + "issue:46144", + "issue:46172", + "issue:46218", + "issue:46655", + "issue:46894" + ], + "soft_pairs": [ + "issue:43519|issue:46655", + "issue:44273|issue:44908", + "issue:45230|issue:45310", + "issue:44655|issue:44861", + "issue:43576|issue:46218", + "issue:43519|issue:44485", + "issue:45468|issue:46894", + "issue:43519|issue:44908", + "issue:41669|issue:43704", + "issue:31795|issue:34689", + "issue:46018|issue:46172", + "issue:43704|issue:44485", + "issue:44485|issue:45468", + "issue:43502|issue:43519", + "issue:42371|issue:43704", + "issue:45850|issue:46218", + "issue:43519|issue:43704", + "issue:44263|issue:45468", + "issue:44261|issue:45468", + "issue:43298|issue:46095", + "issue:44351|issue:46218", + "issue:43352|issue:46218", + "issue:42913|issue:43066", + "issue:28348|issue:45250", + "issue:44908|issue:44998", + "issue:44492|issue:44998", + "issue:43618|issue:44998", + "issue:44998|issue:45850", + "issue:44938|issue:44998", + "issue:44861|issue:45290", + "issue:31515|issue:38617", + "issue:43572|issue:44861", + "issue:43525|issue:44861", + "issue:44230|issue:44623", + "issue:28348|issue:43352", + "issue:29870|issue:31515", + "issue:30333|issue:31515", + "issue:31515|issue:36296", + "issue:31515|issue:31795", + "issue:44230|issue:45310", + "issue:43298|issue:44230", + "issue:43717|issue:46144" + ], + "prepared_review_unit_hash": "f0dfca480ae299ca72a894fa08fc0b7c656ea92ed69bdbacb69a4dd52d27921b", + "component_budget": { + "node_count": 502, + "item_count": 502, + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 42, + "serialized_chars": 39367, + "estimated_input_tokens": 9842, + "estimated_eval_tokens": 19940 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "No soft pair looks like the same underlying bug/change closely enough to merge. Most pairs are only superficially similar by shared model family, a broad subsystem, or a generic regression theme; the concrete symptoms and code paths differ.", + "confidence": 0.92, + "canonical_issue_reason": "issue:43525 is the clearest self-contained bug report in the set: a specific model, a concrete missing attribute, and a narrow actionable failure path.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:43525 is the strongest canonical candidate because it is specific, reproducible, and points to a well-scoped fix surface.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43519", + "right": "issue:46655", + "accept": false, + "reason": "Different models and different processors: timestamp calculation in Qwen3VL vs padding calculation in SmolVLM video processing." + }, + { + "left": "issue:44273", + "right": "issue:44908", + "accept": false, + "reason": "Unrelated subsystems: lazy loading vs learning-rate scheduler kwargs." + }, + { + "left": "issue:45230", + "right": "issue:45310", + "accept": false, + "reason": "One is a generic bug placeholder; the other is a specific Qwen3.5 MoE from_pretrained regression." + }, + { + "left": "issue:44655", + "right": "issue:44861", + "accept": false, + "reason": "Saving pipeline objects and tied-weight-key handling are different code paths with different failures." + }, + { + "left": "issue:43576", + "right": "issue:46218", + "accept": false, + "reason": "CLI/env command breakage is unrelated to an import failure for a specific model class." + }, + { + "left": "issue:43519", + "right": "issue:44485", + "accept": false, + "reason": "Different model-specific bugs: processor timestamp logic vs GLM-5 RoPE implementation." + }, + { + "left": "issue:45468", + "right": "issue:46894", + "accept": false, + "reason": "Different models and symptoms: Gemma-4 audio positional encoding vs CohereASR training-loss shifting." + }, + { + "left": "issue:43519", + "right": "issue:44908", + "accept": false, + "reason": "Processor timestamp logic and scheduler kwargs are unrelated." + }, + { + "left": "issue:41669", + "right": "issue:43704", + "accept": false, + "reason": "Import-star cleanup/perf issue is unrelated to a VRAM leak in multi-threaded dataloading." + }, + { + "left": "issue:31795", + "right": "issue:34689", + "accept": false, + "reason": "Documentation confusion about forward inputs is not the same as a model-loading regression." + }, + { + "left": "issue:46018", + "right": "issue:46172", + "accept": false, + "reason": "Both mention DeepSeek-V4, but the reported bugs are different: gating logic vs a tensor transpose/cast issue." + }, + { + "left": "issue:43704", + "right": "issue:44485", + "accept": false, + "reason": "VRAM leakage in threading and RoPE implementation are distinct bugs." + }, + { + "left": "issue:44485", + "right": "issue:45468", + "accept": false, + "reason": "Different models and different implementation areas; no shared failing code path." + }, + { + "left": "issue:43502", + "right": "issue:43519", + "accept": false, + "reason": "local_files_only network leakage is unrelated to timestamp calculation in a video processor." + }, + { + "left": "issue:42371", + "right": "issue:43704", + "accept": false, + "reason": "TF32 settings guidance and a VRAM leak are unrelated issues." + }, + { + "left": "issue:45850", + "right": "issue:46218", + "accept": false, + "reason": "Broad 'many models break' report is too vague to match a specific Qwen3_5 import failure." + }, + { + "left": "issue:43519", + "right": "issue:43704", + "accept": false, + "reason": "Different failure modes in different model families; no evidence of one underlying bug." + }, + { + "left": "issue:44263", + "right": "issue:45468", + "accept": false, + "reason": "torch.split return handling in GlmMoeDsaIndexer is unrelated to Gemma-4 audio positional encoding." + }, + { + "left": "issue:44261", + "right": "issue:45468", + "accept": false, + "reason": "Different models and different numerical issues; not the same bug." + }, + { + "left": "issue:43298", + "right": "issue:46095", + "accept": false, + "reason": "Continuous batching support for audio models is not a weight-conversion-script request." + }, + { + "left": "issue:44351", + "right": "issue:46218", + "accept": false, + "reason": "Two different import failures with no shared target or code path." + }, + { + "left": "issue:43352", + "right": "issue:46218", + "accept": false, + "reason": "Flash-Attention support for Nemotron and an import error for Qwen3_5 are unrelated." + }, + { + "left": "issue:42913", + "right": "issue:43066", + "accept": false, + "reason": "Both are tokenizer-v5 complaints, but the symptoms are different enough that they do not clearly describe one concrete bug." + }, + { + "left": "issue:28348", + "right": "issue:45250", + "accept": false, + "reason": "GPT2LMHeadModel flash-attention support is more specific than the generic Flash Attention 2.0 topic." + }, + { + "left": "issue:44908", + "right": "issue:44998", + "accept": false, + "reason": "Scheduler kwargs bug and an unrelated generic issue title do not match." + }, + { + "left": "issue:44492", + "right": "issue:44998", + "accept": false, + "reason": "A typo in cache-strategy naming is unrelated to the generic issue." + }, + { + "left": "issue:43618", + "right": "issue:44998", + "accept": false, + "reason": "CLIPOutput attentions assignment regression is unrelated to the generic issue title." + }, + { + "left": "issue:44998", + "right": "issue:45850", + "accept": false, + "reason": "A generic 'too many models break' complaint does not identify the same concrete bug as the generic placeholder issue." + }, + { + "left": "issue:44938", + "right": "issue:44998", + "accept": false, + "reason": "Python 3.14 import/load failure is unrelated to the generic issue." + }, + { + "left": "issue:44861", + "right": "issue:45290", + "accept": false, + "reason": "Tied-weight key handling and chat-template tool-call crashes are different features and failures." + }, + { + "left": "issue:31515", + "right": "issue:38617", + "accept": false, + "reason": "Slow from_pretrained checkpoint loading and an import error for layer_type_validation are different problems." + }, + { + "left": "issue:43572", + "right": "issue:44861", + "accept": false, + "reason": "Missing pad_token_idx in config is not the same as tied-weight key list handling." + }, + { + "left": "issue:43525", + "right": "issue:44861", + "accept": false, + "reason": "Llama4Config missing pad_token_id and tied-weight key handling are distinct code paths." + }, + { + "left": "issue:44230", + "right": "issue:44623", + "accept": false, + "reason": "FP8 inference support request and missing processor save files are unrelated." + }, + { + "left": "issue:28348", + "right": "issue:43352", + "accept": false, + "reason": "Both mention Flash Attention 2.0, but one is GPT2 support and the other is Nemotron unsupported status; not the same concrete bug or change." + }, + { + "left": "issue:29870", + "right": "issue:31515", + "accept": false, + "reason": "BART generate() decoder_attention_mask bug and slow from_pretrained loading are unrelated." + }, + { + "left": "issue:30333", + "right": "issue:31515", + "accept": false, + "reason": "MLFlow reporting on training failure is not the same as checkpoint loading performance." + }, + { + "left": "issue:31515", + "right": "issue:36296", + "accept": false, + "reason": "from_pretrained slowness and tensor-parallel training bugs are different failure modes." + }, + { + "left": "issue:31515", + "right": "issue:31795", + "accept": false, + "reason": "Checkpoint loading performance and forward-API documentation confusion are unrelated." + }, + { + "left": "issue:44230", + "right": "issue:45310", + "accept": false, + "reason": "FP8 inference support request is unrelated to a Qwen3.5 MoE from_pretrained error." + }, + { + "left": "issue:43298", + "right": "issue:44230", + "accept": false, + "reason": "Audio continuous batching support and FP8 inference support are different feature requests." + }, + { + "left": "issue:43717", + "right": "issue:46144", + "accept": false, + "reason": "Mamba init_weights behavior and RoFormer attention interface usage are unrelated implementation issues." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 61, + "review_unit_count": 61, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-502", + "nodes": [ + "issue:37934", + "issue:42936", + "issue:43240", + "issue:43668", + "issue:43813", + "issue:44031", + "issue:46018", + "issue:46144" + ], + "soft_pairs": [ + "issue:42936|issue:46018", + "issue:43668|issue:44031", + "issue:42936|issue:43240", + "issue:43240|issue:46144", + "issue:42936|issue:44031", + "issue:42936|issue:46144", + "issue:42936|issue:43813", + "issue:43240|issue:43668", + "issue:43668|issue:43813", + "issue:37934|issue:43668", + "issue:42936|issue:43668" + ], + "prepared_review_unit_hash": "4cf3715f25058c3c5bf72a5e9b627bdbd4d63ccc254992c82ea27136d4aa4094", + "component_budget": { + "node_count": 502, + "item_count": 502, + "soft_pair_count": 1841, + "serialized_chars": 955860, + "estimated_input_tokens": 238965, + "estimated_eval_tokens": 478186 + }, + "budget": { + "node_count": 8, + "item_count": 8, + "soft_pair_count": 11, + "serialized_chars": 7776, + "estimated_input_tokens": 1944, + "estimated_eval_tokens": 4144 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "These issues cover unrelated bugs across different models/components (norms, loss kwargs, tokenizer warnings, attention interfaces, and a typo). No duplicate cluster is evident, so all soft-edge pairs should be rejected.", + "confidence": 0.93, + "canonical_issue_reason": "issue:43240 is the most concrete and actionable report in the set: it describes a specific loss-kwargs propagation bug with a clear code-path implication.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:43240 is the best representative issue because it is narrowly scoped, implementation-specific, and likely to map to a single fix.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:42936", + "right": "issue:46018", + "accept": false, + "reason": "Different models and different failure modes: Mask2Former ignore_value handling vs DeepSeek-V4 shared expert gating." + }, + { + "left": "issue:43668", + "right": "issue:44031", + "accept": false, + "reason": "A config type-hint mistake is unrelated to a tokenizer regex warning regression." + }, + { + "left": "issue:42936", + "right": "issue:43240", + "accept": false, + "reason": "One is about Mask2Former ignore_value usage; the other is about loss kwargs not being forwarded." + }, + { + "left": "issue:43240", + "right": "issue:46144", + "accept": false, + "reason": "Loss-function kwargs propagation and RoFormer attention interface usage are distinct code-path problems." + }, + { + "left": "issue:42936", + "right": "issue:44031", + "accept": false, + "reason": "Mask2Former ignore_value behavior is unrelated to tokenizer warning behavior." + }, + { + "left": "issue:42936", + "right": "issue:46144", + "accept": false, + "reason": "These concern different subsystems and different bugs; no shared underlying fix is apparent." + }, + { + "left": "issue:42936", + "right": "issue:43813", + "accept": false, + "reason": "Ignore_value handling in a model head is unrelated to a typo in a quantization attribute name." + }, + { + "left": "issue:43240", + "right": "issue:43668", + "accept": false, + "reason": "Forwarding kwargs to a loss function is unrelated to a ModernBERTConfig type annotation issue." + }, + { + "left": "issue:43668", + "right": "issue:43813", + "accept": false, + "reason": "A type-hint error and a string/attribute typo are different defect types with no common code path." + }, + { + "left": "issue:37934", + "right": "issue:43668", + "accept": false, + "reason": "RMS-norm behavior question for Llama4TextL2Norm is unrelated to a ModernBERT config type hint." + }, + { + "left": "issue:42936", + "right": "issue:43668", + "accept": false, + "reason": "Mask2Former loss/ignore_value logic and ModernBERT config typing do not describe the same bug." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 2, + "component_count": 23, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10519-3", + "nodes": [ + "issue:10519", + "issue:11005", + "issue:11307" + ], + "soft_pairs": [ + "issue:10519|issue:11307", + "issue:10519|issue:11005", + "issue:11005|issue:11307" + ], + "prepared_review_unit_hash": "ada96ec7f012a44d1948ab1aff1fb32cf36cee0143dfa255eb191c2367c5de8f", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 2875, + "estimated_input_tokens": 719, + "estimated_eval_tokens": 1694 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 2875, + "estimated_input_tokens": 719, + "estimated_eval_tokens": 1694 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These three issues are not duplicates: one is about token truncation direction, one about ReduceLROnPlateau-like scheduling, and one about Wav2Vec2 word time offsets. The cluster looks like a false-positive similarity grouping rather than a single underlying bug.", + "confidence": 0.98, + "canonical_issue_reason": "No clear canonical issue exists because the topics are unrelated and do not describe the same bug or feature request.", + "canonical_pr_reason": null, + "best_issue_reason": "None of the issues is a good global canonical representative for the cluster; 10519 is the most self-contained, but it still has no relationship to the other two.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:10519", + "right": "issue:11307", + "accept": false, + "reason": "Different domains and requests: tokenizer truncation behavior vs Wav2Vec2 word boundary timing." + }, + { + "left": "issue:10519", + "right": "issue:11005", + "accept": false, + "reason": "Unrelated functionality: truncation option vs learning-rate scheduler behavior." + }, + { + "left": "issue:11005", + "right": "issue:11307", + "accept": false, + "reason": "No shared underlying bug or change; scheduler feature request is unrelated to Wav2Vec2 timestamp extraction." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 3, + "component_count": 23, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-13244-3", + "nodes": [ + "issue:13244", + "issue:15980", + "issue:16618" + ], + "soft_pairs": [ + "issue:13244|issue:16618", + "issue:15980|issue:16618" + ], + "prepared_review_unit_hash": "2fef424e1aa6db6c0457ae8a812a119b6e9c418513283024be98a7e9192bda07", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 2479, + "estimated_input_tokens": 620, + "estimated_eval_tokens": 1496 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 2479, + "estimated_input_tokens": 620, + "estimated_eval_tokens": 1496 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are three unrelated issue reports. The only potentially general model-loading failure is #16618, while #15980 is specifically about an unauthenticated private-model download error and #13244 is about Tapas tokenization behavior, so neither soft pair should be merged.", + "confidence": 0.93, + "canonical_issue_reason": "#16618 is the clearest standalone bug report about model loading failure and is more central than the other two, which describe different subsystems and failure modes.", + "canonical_pr_reason": null, + "best_issue_reason": "#16618 is the best representative issue because it names the concrete user-facing failure, whereas #15980 is an auth/error-message problem and #13244 is a tokenizer parity issue.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:13244", + "right": "issue:16618", + "accept": false, + "reason": "These describe different problems: Tapas tokenization mismatch versus inability to load bert-base-uncased. They are not the same underlying bug." + }, + { + "left": "issue:15980", + "right": "issue:16618", + "accept": false, + "reason": "#15980 is about the error shown when downloading a private model without login, not the same failure as #16618's model-loading issue." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 4, + "component_count": 23, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-26350-2", + "nodes": [ + "issue:26350", + "issue:33260" + ], + "soft_pairs": [ + "issue:26350|issue:33260" + ], + "prepared_review_unit_hash": "8f76a7e6011a335b5c884e5ecc4eb93de67f787cfaa6167a6cae6da475c12cf8", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1590, + "estimated_input_tokens": 398, + "estimated_eval_tokens": 1052 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1590, + "estimated_input_tokens": 398, + "estimated_eval_tokens": 1052 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "The two issues are about different feature requests: Flash Attention 2 support for more architectures vs GGUF support for more architectures. They are not duplicates and should not be clustered together.", + "confidence": 0.98, + "canonical_issue_reason": "No canonical issue: the issues describe distinct enhancement requests for different code paths and formats.", + "canonical_pr_reason": null, + "best_issue_reason": "If a single representative is required, issue 26350 is slightly better due to being older and having more discussion, but it is not a duplicate of issue 33260.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:26350", + "right": "issue:33260", + "accept": false, + "reason": "Different features with different technical goals (Flash Attention 2 vs GGUF support); same broad theme of architecture support is too weak for duplicate/merge triage." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 5, + "component_count": 23, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-27088-9", + "nodes": [ + "issue:27088", + "issue:28936", + "issue:38435", + "issue:43834", + "issue:44365", + "issue:44370", + "issue:46222", + "issue:46226", + "issue:46931" + ], + "soft_pairs": [ + "issue:27088|issue:44370", + "issue:27088|issue:44365", + "issue:27088|issue:43834", + "issue:27088|issue:38435", + "issue:27088|issue:28936", + "issue:28936|issue:43834", + "issue:28936|issue:44365", + "issue:28936|issue:38435", + "issue:28936|issue:44370", + "issue:38435|issue:44370", + "issue:38435|issue:44365", + "issue:38435|issue:43834", + "issue:38435|issue:46222", + "issue:38435|issue:46226", + "issue:43834|issue:44370", + "issue:43834|issue:44365", + "issue:43834|issue:46222", + "issue:43834|issue:46226", + "issue:43834|issue:46931", + "issue:44365|issue:44370", + "issue:44365|issue:46222", + "issue:44365|issue:46226", + "issue:44365|issue:46931", + "issue:44370|issue:46222", + "issue:44370|issue:46226", + "issue:44370|issue:46931", + "issue:46222|issue:46226", + "issue:46222|issue:46931", + "issue:46226|issue:46931" + ], + "prepared_review_unit_hash": "6d6d4edb052383ac3ec27e42ce44ac422e53244189b2489bcec37e39cb5c1046", + "component_budget": { + "node_count": 9, + "item_count": 9, + "soft_pair_count": 29, + "serialized_chars": 15982, + "estimated_input_tokens": 3996, + "estimated_eval_tokens": 8248 + }, + "budget": { + "node_count": 9, + "item_count": 9, + "soft_pair_count": 29, + "serialized_chars": 15982, + "estimated_input_tokens": 3996, + "estimated_eval_tokens": 8248 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are not true duplicates: most are separate docs-translation requests for different languages, while a few are generic template instances. The cluster is driven by template similarity rather than the same underlying issue.", + "confidence": 0.97, + "canonical_issue_reason": "No single issue is a real duplicate canonical because each concrete issue targets a different language translation request; issue 43834 is only the closest generic template match.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 43834 is the best representative only in the narrow sense that it is the generic placeholder form, but it is not a good global canonical because the cluster contains multiple independent language-specific requests.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:27088", + "right": "issue:44370", + "accept": false, + "reason": "Different language translation requests; same template wording but not the same underlying task." + }, + { + "left": "issue:27088", + "right": "issue:44365", + "accept": false, + "reason": "Different language translation requests; they are structurally similar but not duplicates." + }, + { + "left": "issue:27088", + "right": "issue:43834", + "accept": false, + "reason": "One is a concrete Turkish request and the other is a generic placeholder template, so not the same issue." + }, + { + "left": "issue:27088", + "right": "issue:38435", + "accept": false, + "reason": "Different target languages (Turkish vs Romanian), so these are separate requests." + }, + { + "left": "issue:27088", + "right": "issue:28936", + "accept": false, + "reason": "Different target languages (Turkish vs Spanish), so not duplicates." + }, + { + "left": "issue:28936", + "right": "issue:43834", + "accept": false, + "reason": "Generic template vs a concrete Spanish translation request; not the same underlying issue." + }, + { + "left": "issue:28936", + "right": "issue:44365", + "accept": false, + "reason": "Different translation requests despite similar template text." + }, + { + "left": "issue:28936", + "right": "issue:38435", + "accept": false, + "reason": "Different target languages, so separate issues." + }, + { + "left": "issue:28936", + "right": "issue:44370", + "accept": false, + "reason": "Different translation requests; similarity is only in format." + }, + { + "left": "issue:38435", + "right": "issue:44370", + "accept": false, + "reason": "Different instances of the translation template and different actual requests." + }, + { + "left": "issue:38435", + "right": "issue:44365", + "accept": false, + "reason": "Separate language-translation requests, not one underlying bug/change." + }, + { + "left": "issue:38435", + "right": "issue:43834", + "accept": false, + "reason": "Romanian request vs generic placeholder template; not a duplicate." + }, + { + "left": "issue:38435", + "right": "issue:46222", + "accept": false, + "reason": "Different issues; only the wording/template is similar." + }, + { + "left": "issue:38435", + "right": "issue:46226", + "accept": false, + "reason": "Different issues; these are not the same concrete request." + }, + { + "left": "issue:43834", + "right": "issue:44370", + "accept": false, + "reason": "Both are template-like, but they still represent separate translation requests rather than one shared bug/change." + }, + { + "left": "issue:43834", + "right": "issue:44365", + "accept": false, + "reason": "Generic placeholders only; not enough to conclude the same underlying issue." + }, + { + "left": "issue:43834", + "right": "issue:46222", + "accept": false, + "reason": "Different concrete issues even if the template is identical." + }, + { + "left": "issue:43834", + "right": "issue:46226", + "accept": false, + "reason": "Different concrete issues even if the template is identical." + }, + { + "left": "issue:43834", + "right": "issue:46931", + "accept": false, + "reason": "The titles are unrelated beyond matching the template pattern; not a duplicate." + }, + { + "left": "issue:44365", + "right": "issue:44370", + "accept": false, + "reason": "Two separate template-based translation requests; no evidence they are the same request." + }, + { + "left": "issue:44365", + "right": "issue:46222", + "accept": false, + "reason": "Same template pattern but not the same underlying issue." + }, + { + "left": "issue:44365", + "right": "issue:46226", + "accept": false, + "reason": "Same template pattern but not the same underlying issue." + }, + { + "left": "issue:44365", + "right": "issue:46931", + "accept": false, + "reason": "Titles do not indicate the same concrete task or bug." + }, + { + "left": "issue:44370", + "right": "issue:46222", + "accept": false, + "reason": "Same generic translation template, but separate issues." + }, + { + "left": "issue:44370", + "right": "issue:46226", + "accept": false, + "reason": "Same generic translation template, but separate issues." + }, + { + "left": "issue:44370", + "right": "issue:46931", + "accept": false, + "reason": "No clear shared underlying issue beyond superficial similarity." + }, + { + "left": "issue:46222", + "right": "issue:46226", + "accept": false, + "reason": "Even though the text is nearly identical, they appear to be distinct issue entries rather than evidence of one concrete duplicate." + }, + { + "left": "issue:46222", + "right": "issue:46931", + "accept": false, + "reason": "Only superficial textual similarity; not enough to treat as the same issue." + }, + { + "left": "issue:46226", + "right": "issue:46931", + "accept": false, + "reason": "Only superficial textual similarity; not enough to treat as the same issue." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 6, + "component_count": 23, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-36718-19", + "nodes": [ + "issue:36718", + "issue:36817", + "issue:36879", + "issue:37015", + "issue:41115", + "issue:41211", + "issue:42503", + "issue:42738", + "issue:43671", + "issue:43837", + "issue:43909", + "issue:44121", + "issue:44205", + "issue:45027", + "issue:45177", + "issue:45322", + "issue:46133", + "issue:46645", + "issue:46653" + ], + "soft_pairs": [ + "issue:36718|issue:41115", + "issue:36817|issue:41211", + "issue:36817|issue:43671", + "issue:36817|issue:36879", + "issue:36879|issue:41211", + "issue:36879|issue:43909", + "issue:36879|issue:44121", + "issue:36879|issue:43837", + "issue:36879|issue:43671", + "issue:37015|issue:43837", + "issue:37015|issue:44121", + "issue:37015|issue:43909", + "issue:36879|issue:37015", + "issue:37015|issue:42503", + "issue:41115|issue:43837", + "issue:41115|issue:44121", + "issue:41115|issue:42503", + "issue:41115|issue:45027", + "issue:41211|issue:43671", + "issue:42503|issue:43837", + "issue:42503|issue:43671", + "issue:42503|issue:42738", + "issue:42738|issue:45027", + "issue:42738|issue:46653", + "issue:43671|issue:43837", + "issue:43671|issue:45027", + "issue:43671|issue:45177", + "issue:43671|issue:44121", + "issue:43837|issue:43909", + "issue:43837|issue:45322", + "issue:43837|issue:44121", + "issue:43909|issue:44205", + "issue:43909|issue:46645", + "issue:44121|issue:45322", + "issue:43837|issue:45027", + "issue:37015|issue:45027", + "issue:43837|issue:45177", + "issue:45322|issue:46133", + "issue:42503|issue:46653", + "issue:45027|issue:46653" + ], + "prepared_review_unit_hash": "e546ac0256f0a3d3b3d0e3c357719e087b83d4a6fcb13bf75ae2bbc33dc7a0d0", + "component_budget": { + "node_count": 19, + "item_count": 19, + "soft_pair_count": 40, + "serialized_chars": 24672, + "estimated_input_tokens": 6168, + "estimated_eval_tokens": 12592 + }, + "budget": { + "node_count": 19, + "item_count": 19, + "soft_pair_count": 40, + "serialized_chars": 24672, + "estimated_input_tokens": 6168, + "estimated_eval_tokens": 12592 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "This cluster is a loose set of model-add/support requests, not a duplicate set. The titles share a broad pattern ('Add X model/support'), but each pair targets a different model, modality, or feature, so no soft edge looks like the same underlying change. No PRs are present.", + "confidence": 0.78, + "canonical_issue_reason": "issue:36879 is the clearest concrete model-addition request and a good representative of the cluster\u2019s common pattern, even though the cluster itself is not made of true duplicates.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:36879 is the most representative issue overall: specific, well-scoped, and central to the repeated 'add model' theme across the cluster.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:36718", + "right": "issue:41115", + "accept": false, + "reason": "Different artifacts: keypoint matcher vs MiniCPM3 architecture request." + }, + { + "left": "issue:36817", + "right": "issue:41211", + "accept": false, + "reason": "EuroBert config request and DEIMv2 are unrelated model additions." + }, + { + "left": "issue:36817", + "right": "issue:43671", + "accept": false, + "reason": "BERT config change vs Qwen3-TTS support; different model families and code paths." + }, + { + "left": "issue:36817", + "right": "issue:36879", + "accept": false, + "reason": "EuroBert and RF-DETR are separate model requests, not one change." + }, + { + "left": "issue:36879", + "right": "issue:41211", + "accept": false, + "reason": "RF-DETR and DEIMv2 are distinct models with different integration work." + }, + { + "left": "issue:36879", + "right": "issue:43909", + "accept": false, + "reason": "Object detection model vs audio model; no shared underlying bug or feature." + }, + { + "left": "issue:36879", + "right": "issue:44121", + "accept": false, + "reason": "RF-DETR model addition and weight-sparse transformer request are unrelated." + }, + { + "left": "issue:36879", + "right": "issue:43837", + "accept": false, + "reason": "RF-DETR and Qwen3-ASR are different models and different tasks." + }, + { + "left": "issue:36879", + "right": "issue:43671", + "accept": false, + "reason": "Different model families and modalities; not the same implementation." + }, + { + "left": "issue:37015", + "right": "issue:43837", + "accept": false, + "reason": "NeoBERT addition and Qwen3-ASR support are unrelated requests." + }, + { + "left": "issue:37015", + "right": "issue:44121", + "accept": false, + "reason": "NeoBERT and circuit-sparsity transformer changes do not share a concrete code-path fix." + }, + { + "left": "issue:37015", + "right": "issue:43909", + "accept": false, + "reason": "BERT-style model request vs audio model support; different targets." + }, + { + "left": "issue:36879", + "right": "issue:37015", + "accept": false, + "reason": "RF-DETR and NeoBERT are distinct model additions." + }, + { + "left": "issue:37015", + "right": "issue:42503", + "accept": false, + "reason": "NeoBERT and ModernVBERT are separate models; same broad area only." + }, + { + "left": "issue:41115", + "right": "issue:43837", + "accept": false, + "reason": "MiniCPM3 architecture and Qwen3-ASR support are different model implementations." + }, + { + "left": "issue:41115", + "right": "issue:44121", + "accept": false, + "reason": "MiniCPM3 model support vs weight-sparse transformer request are not the same change." + }, + { + "left": "issue:41115", + "right": "issue:42503", + "accept": false, + "reason": "MiniCPM3 and ModernVBERT are different model requests." + }, + { + "left": "issue:41115", + "right": "issue:45027", + "accept": false, + "reason": "MiniCPM3 architecture and Voxtral TTS support are unrelated." + }, + { + "left": "issue:41211", + "right": "issue:43671", + "accept": false, + "reason": "DEIMv2 and Qwen3-TTS are different models and modalities." + }, + { + "left": "issue:42503", + "right": "issue:43837", + "accept": false, + "reason": "ModernVBERT models and Qwen3-ASR support are separate requests." + }, + { + "left": "issue:42503", + "right": "issue:43671", + "accept": false, + "reason": "ModernVBERT and Qwen3-TTS are unrelated model additions." + }, + { + "left": "issue:42503", + "right": "issue:42738", + "accept": false, + "reason": "ModernVBERT models vs a general RoPE-related BERT-like proposal; not the same concrete change." + }, + { + "left": "issue:42738", + "right": "issue:45027", + "accept": false, + "reason": "General RoPE support for BERT-like models and Voxtral TTS support are unrelated." + }, + { + "left": "issue:42738", + "right": "issue:46653", + "accept": false, + "reason": "BERT/RoPE model request vs an unrelated personal-name issue; no overlap." + }, + { + "left": "issue:43671", + "right": "issue:43837", + "accept": false, + "reason": "Qwen3-TTS and Qwen3-ASR are related by name only; distinct support paths." + }, + { + "left": "issue:43671", + "right": "issue:45027", + "accept": false, + "reason": "Both audio-related, but different models and tasks; not one mergeable fix." + }, + { + "left": "issue:43671", + "right": "issue:45177", + "accept": false, + "reason": "TTS support and DeepSeek-OCR2 are different model integrations." + }, + { + "left": "issue:43671", + "right": "issue:44121", + "accept": false, + "reason": "Qwen3-TTS support and weight-sparse transformer request are unrelated." + }, + { + "left": "issue:43837", + "right": "issue:43909", + "accept": false, + "reason": "Qwen3-ASR and LFM2.5 Audio are different audio models with separate work." + }, + { + "left": "issue:43837", + "right": "issue:45322", + "accept": false, + "reason": "ASR support vs EUPE encoder request; different model additions." + }, + { + "left": "issue:43837", + "right": "issue:44121", + "accept": false, + "reason": "Qwen3-ASR support and circuit-sparsity transformer request are not the same feature." + }, + { + "left": "issue:43909", + "right": "issue:44205", + "accept": false, + "reason": "Audio model support and SAM3-LiteText are different integration targets." + }, + { + "left": "issue:43909", + "right": "issue:46645", + "accept": false, + "reason": "LFM2.5 Audio support and Granite SWA+sink option are different changes." + }, + { + "left": "issue:44121", + "right": "issue:45322", + "accept": false, + "reason": "Weight-sparse transformer request and EUPE model request are unrelated." + }, + { + "left": "issue:43837", + "right": "issue:45027", + "accept": false, + "reason": "Qwen3-ASR and Voxtral TTS are separate speech-model support requests." + }, + { + "left": "issue:37015", + "right": "issue:45027", + "accept": false, + "reason": "NeoBERT and Voxtral TTS are different model requests." + }, + { + "left": "issue:43837", + "right": "issue:45177", + "accept": false, + "reason": "Qwen3-ASR support and DeepSeek-OCR2 are distinct model additions." + }, + { + "left": "issue:45322", + "right": "issue:46133", + "accept": false, + "reason": "EUPE and TIPSv2 are different vision-language models; same broad domain only." + }, + { + "left": "issue:42503", + "right": "issue:46653", + "accept": false, + "reason": "ModernVBERT model request and unrelated personal-name issue share no underlying change." + }, + { + "left": "issue:45027", + "right": "issue:46653", + "accept": false, + "reason": "Voxtral TTS support and unrelated personal-name issue are not duplicates." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 7, + "component_count": 23, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-38740-2", + "nodes": [ + "issue:38740", + "issue:46443" + ], + "soft_pairs": [ + "issue:38740|issue:46443" + ], + "prepared_review_unit_hash": "af5bd9264c2ade167cea9d8286bf3cfc54e3cba527c4ffb63c671c77f5a949a5", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1505, + "estimated_input_tokens": 377, + "estimated_eval_tokens": 1010 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1505, + "estimated_input_tokens": 377, + "estimated_eval_tokens": 1010 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These two issues are not duplicates: one is a docs request to add Pruna as an optimization framework, while the other is a technical Vision RoPE refactor. They do not describe the same underlying problem or change.", + "confidence": 0.98, + "canonical_issue_reason": "No canonical issue: the topics are unrelated and there is no shared bug or feature request to collapse onto one issue.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue #46443 is the more substantive and specific technical work item; it is the better representative issue in this pair, though not a duplicate target for #38740.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:38740", + "right": "issue:46443", + "accept": false, + "reason": "Different scopes and intent: documentation/framework listing request vs a Vision RoPE code refactor. No evidence they refer to the same underlying change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 8, + "component_count": 23, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-41084-2", + "nodes": [ + "issue:41084", + "issue:43037" + ], + "soft_pairs": [ + "issue:41084|issue:43037" + ], + "prepared_review_unit_hash": "e2ee96cba9396022e1a3c9926aa435bc1f40fa5bb33042bda070729ba0de732b", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1479, + "estimated_input_tokens": 370, + "estimated_eval_tokens": 996 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1479, + "estimated_input_tokens": 370, + "estimated_eval_tokens": 996 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "The two issues do not look like duplicates: one is about set block decoding, the other about DeepSeek v3.2 support. The soft similarity signal is too weak to justify merging them.", + "confidence": 0.97, + "canonical_issue_reason": "Issue #41084 is the better canonical representative because it is the more general technical topic, while #43037 is a specific model-support request.", + "canonical_pr_reason": null, + "best_issue_reason": "#41084 is the best issue to anchor the cluster since it is broader and more likely to describe the underlying implementation area rather than a single model request.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:41084", + "right": "issue:43037", + "accept": false, + "reason": "Reject: the titles and likely intents are unrelated; 'Set Block Decoding' concerns a decoding mode/implementation detail, while 'DeepSeek v3.2 support' is a model compatibility request. No shared files, targets, or explicit linkage." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 9, + "component_count": 23, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-41306-2", + "nodes": [ + "issue:41306", + "issue:41826" + ], + "soft_pairs": [ + "issue:41306|issue:41826" + ], + "prepared_review_unit_hash": "fc07ad48c73dd5a0e92986297fae7aaf6f7ad52cc2cab0ba11a17939862034fb", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1585, + "estimated_input_tokens": 397, + "estimated_eval_tokens": 1050 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1585, + "estimated_input_tokens": 397, + "estimated_eval_tokens": 1050 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These two issues are both about memory-efficient/tiled model components, but they describe different features in different code paths: one is about exposing a `TiledFusedLogitsLoss` / FLCE flag in `from_pretrained()`, while the other requests integrating `TiledMLP`. They should not be clustered as duplicates.", + "confidence": 0.95, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": "If a representative must be chosen, issue 41306 is slightly more concrete and narrowly scoped, but it is not a true duplicate of 41826.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:41306", + "right": "issue:41826", + "accept": false, + "reason": "Different underlying requests and likely different code paths: FLCE/from_pretrained configuration vs. adding TiledMLP for lower memory use. No explicit linkage or shared filenames/targets." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 10, + "component_count": 23, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-43598-2", + "nodes": [ + "issue:43598", + "issue:43600" + ], + "soft_pairs": [ + "issue:43598|issue:43600" + ], + "prepared_review_unit_hash": "9c9e8f6882ae04cf1beaaf9732aa144760b8007df33d63f82f1f1a0f667ca023", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1583, + "estimated_input_tokens": 396, + "estimated_eval_tokens": 1048 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1583, + "estimated_input_tokens": 396, + "estimated_eval_tokens": 1048 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are two distinct Trainer customization issues: one about `remove_unused_column` behavior, the other about FLOPS estimation with custom trainers. They share the same broad area but not the same underlying bug/change.", + "confidence": 0.91, + "canonical_issue_reason": "Issue #43598 is the more specific and actionable bug report, centered on a concrete Trainer behavior (`remove_unused_column`) and its customizability.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue #43598 is the better representative for the cluster because it is narrowly scoped and directly describes a single Trainer customization problem.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43598", + "right": "issue:43600", + "accept": false, + "reason": "Both mention custom trainers, but they address different code paths and outcomes: column removal customization vs FLOPS estimation. They are not the same underlying bug or change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 11, + "component_count": 23, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-43684-2", + "nodes": [ + "issue:43684", + "issue:44405" + ], + "soft_pairs": [ + "issue:43684|issue:44405" + ], + "prepared_review_unit_hash": "7ad541e71e04a019b7c43d5911ed2e0400342429971c56c5012bc1b227d42a28", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1612, + "estimated_input_tokens": 403, + "estimated_eval_tokens": 1062 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1612, + "estimated_input_tokens": 403, + "estimated_eval_tokens": 1062 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "The two issues are about different model families and different AutoModel APIs: one requests Qwen3-Omni registration for AutoModel/AutoModelForConditionalGeneration, while the other asks for AutoModelForSequenceClassification support for Qwen3.5. They are related only at a broad \u201cadd model support\u201d level, not the same underlying bug or change.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": "Issue #43684 is the broader and more concrete API-registration request, but it is not a duplicate of #44405; the topics and target classes differ.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43684", + "right": "issue:44405", + "accept": false, + "reason": "Different models (Qwen3-Omni vs Qwen3.5) and different auto classes (generation registration vs sequence classification support); not the same underlying change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 12, + "component_count": 23, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-43835-4", + "nodes": [ + "issue:43835", + "issue:43845", + "issue:43846", + "issue:43859" + ], + "soft_pairs": [ + "issue:43835|issue:43845", + "issue:43835|issue:43846", + "issue:43835|issue:43859", + "issue:43845|issue:43846", + "issue:43845|issue:43859", + "issue:43846|issue:43859" + ], + "prepared_review_unit_hash": "e429079caea6c4cc196581fef0609df9cab7b3386b02971e593d613b18d003db", + "component_budget": { + "node_count": 4, + "item_count": 4, + "soft_pair_count": 6, + "serialized_chars": 4262, + "estimated_input_tokens": 1066, + "estimated_eval_tokens": 2388 + }, + "budget": { + "node_count": 4, + "item_count": 4, + "soft_pair_count": 6, + "serialized_chars": 4262, + "estimated_input_tokens": 1066, + "estimated_eval_tokens": 2388 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "All four issues are too sparse to establish real duplicate identity; the three identical-title \"huggingface\" issues are the closest candidates, but there is not enough substantive evidence to confidently merge any pair.", + "confidence": 0.42, + "canonical_issue_reason": "issue:43859 is the best placeholder canonical because it matches the repeated title and has the only discussion activity, but the cluster lacks body/detail evidence.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:43859 is the most suitable cluster representative: it is one of the repeated-title issues and has slightly more activity than the others.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43835", + "right": "issue:43845", + "accept": false, + "reason": "Only weak title similarity; no body details, links, or shared targets to show the same underlying issue." + }, + { + "left": "issue:43835", + "right": "issue:43846", + "accept": false, + "reason": "Only weak title similarity; no substantive evidence of the same bug/problem." + }, + { + "left": "issue:43835", + "right": "issue:43859", + "accept": false, + "reason": "Different title from the repeated huggingface issues, and there is no content evidence tying it to the others." + }, + { + "left": "issue:43845", + "right": "issue:43846", + "accept": false, + "reason": "Identical short title, but the issues are otherwise empty; that is not enough to safely treat them as the same underlying report." + }, + { + "left": "issue:43845", + "right": "issue:43859", + "accept": false, + "reason": "Same generic title, but there is no description or issue-target evidence to confirm a duplicate." + }, + { + "left": "issue:43846", + "right": "issue:43859", + "accept": false, + "reason": "Same generic title, but insufficient evidence to conclude they refer to the same concrete issue." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 13, + "component_count": 23, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-44371-5", + "nodes": [ + "issue:44371", + "issue:44418", + "issue:44572", + "issue:44573", + "issue:44937" + ], + "soft_pairs": [ + "issue:44371|issue:44572", + "issue:44371|issue:44573", + "issue:44371|issue:44418", + "issue:44371|issue:44937", + "issue:44418|issue:44572", + "issue:44418|issue:44573", + "issue:44572|issue:44573", + "issue:44572|issue:44937", + "issue:44573|issue:44937" + ], + "prepared_review_unit_hash": "edd89155e9f2dd29dde7e59e0880390fcb7d12af902bf26da389f92a3bc01cda", + "component_budget": { + "node_count": 5, + "item_count": 5, + "soft_pair_count": 9, + "serialized_chars": 5791, + "estimated_input_tokens": 1448, + "estimated_eval_tokens": 3152 + }, + "budget": { + "node_count": 5, + "item_count": 5, + "soft_pair_count": 9, + "serialized_chars": 5791, + "estimated_input_tokens": 1448, + "estimated_eval_tokens": 3152 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are low-information, likely spam or placeholder issues with no evidence of a shared underlying bug. The similar titles are not enough to treat them as duplicates, so the soft links should all be rejected.", + "confidence": 0.97, + "canonical_issue_reason": "No clear canonical issue exists: the items are unrelated spam/placeholder reports with too little substantive detail to identify a common bug.", + "canonical_pr_reason": null, + "best_issue_reason": "If one must be chosen as a representative, issue 44418 is the least spam-like title, but it still does not represent a real shared issue cluster.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44371", + "right": "issue:44572", + "accept": false, + "reason": "Both are placeholder spam titles, but there is no substantive body, target, or code-path evidence showing the same underlying issue." + }, + { + "left": "issue:44371", + "right": "issue:44573", + "accept": false, + "reason": "Identical spam-style titles are not enough; there is no concrete bug description tying them together." + }, + { + "left": "issue:44371", + "right": "issue:44418", + "accept": false, + "reason": "One is spam and the other is a generic documentation suggestion; no shared underlying issue is shown." + }, + { + "left": "issue:44371", + "right": "issue:44937", + "accept": false, + "reason": "Completely different titles and no supporting evidence of the same problem." + }, + { + "left": "issue:44418", + "right": "issue:44572", + "accept": false, + "reason": "Generic documentation suggestion versus spam placeholder; no concrete common issue." + }, + { + "left": "issue:44418", + "right": "issue:44573", + "accept": false, + "reason": "Generic documentation suggestion versus spam placeholder; no evidence they describe the same bug or change." + }, + { + "left": "issue:44572", + "right": "issue:44573", + "accept": false, + "reason": "Although the titles match, both are bare spam placeholders with no details to verify a shared underlying bug." + }, + { + "left": "issue:44572", + "right": "issue:44937", + "accept": false, + "reason": "Spam placeholder versus unrelated promotional-looking title; no shared issue." + }, + { + "left": "issue:44573", + "right": "issue:44937", + "accept": false, + "reason": "Spam placeholder versus unrelated promotional-looking title; no shared issue." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 14, + "component_count": 23, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-45145-2", + "nodes": [ + "issue:45145", + "issue:45151" + ], + "soft_pairs": [ + "issue:45145|issue:45151" + ], + "prepared_review_unit_hash": "679779e3e7a577e4947fdfe3b6c6a87fe88d9f92348332d8e4fde54ea9ce6e7e", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1646, + "estimated_input_tokens": 412, + "estimated_eval_tokens": 1080 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1646, + "estimated_input_tokens": 412, + "estimated_eval_tokens": 1080 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These two closed issues have identical titles and strong soft-similarity signals, but the available evidence is still limited to similarity rather than a deterministic duplicate link. Issue 45145 remains the better canonical choice because it was created first and already has one inbound reference.", + "confidence": 0.87, + "canonical_issue_reason": "Issue 45145 is the earliest instance and has an inbound reference, so it is the cleaner canonical representative for this likely duplicate report.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 45145 is the best representative because it is first chronologically and slightly better connected, making it the most suitable target for consolidation.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:45145", + "right": "issue:45151", + "accept": false, + "reason": "The titles match and the issues look highly similar, but the evidence shown is only soft similarity and deterministic_accept is false. This is best treated as a likely duplicate rather than a confirmed duplicate pair." + } + ] + }, + "evaluator_result": { + "accept": false, + "feedback": "The canonical-choice reasoning is grounded in the packet (same title, earlier creation time, and one inbound reference vs. none). However, the soft-edge verdict is too strong for the available evidence: the packet only shows identical titles plus soft-similarity signals, and deterministic_accept is false. This should be phrased as a possible/likely duplicate, not a confirmed valid duplicate pair." + }, + "evaluator_used": true, + "retried": true, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 15, + "component_count": 23, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-45295-2", + "nodes": [ + "issue:45295", + "issue:45373" + ], + "soft_pairs": [], + "prepared_review_unit_hash": "f5fd95fada2f530ab0999e343c17c16266cde0c7182bb22509972ca44827cd49", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 0, + "serialized_chars": 1286, + "estimated_input_tokens": 322, + "estimated_eval_tokens": 900 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 0, + "serialized_chars": 1286, + "estimated_input_tokens": 322, + "estimated_eval_tokens": 900 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These two issues describe the same missing Gemma 4 sequence-classification support; the later, more specific open issue is the better cluster representative.", + "confidence": 0.94, + "canonical_issue_reason": "issue #45373 is the clearest canonical tracker: it is open, explicitly names the missing class `Gemma4ForSequenceClassification`, and states the module-level gap relative to Gemma 2/3.", + "canonical_pr_reason": null, + "best_issue_reason": "issue #45373 best captures the concrete request and likely remaining work, while #45295 is a shorter earlier duplicate with less detail and is already closed.", + "best_pr_reason": null, + "soft_edge_verdicts": [] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 16, + "component_count": 23, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-45762-2", + "nodes": [ + "issue:45762", + "issue:45763" + ], + "soft_pairs": [ + "issue:45762|issue:45763" + ], + "prepared_review_unit_hash": "b177fd47360863275875b048f94a02f4a7eef8e4135f8540d58a8a745590cb71", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1477, + "estimated_input_tokens": 370, + "estimated_eval_tokens": 996 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1477, + "estimated_input_tokens": 370, + "estimated_eval_tokens": 996 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These two closed issues do not look like duplicates of the same technical problem; they are just distinct personal-name/date entries with no body or code-related context. There is no meaningful canonical issue or PR in this cluster.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:45762", + "right": "issue:45763", + "accept": false, + "reason": "Titles only share a surname and both are non-technical personal-name/date strings; there is no evidence they describe the same underlying bug or request." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 17, + "component_count": 23, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-45764-2", + "nodes": [ + "issue:45764", + "issue:45765" + ], + "soft_pairs": [ + "issue:45764|issue:45765" + ], + "prepared_review_unit_hash": "148c902d520cd5a41eff0457cd1b90042167eb4a80d340a4d92ae38baada3ab7", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1481, + "estimated_input_tokens": 371, + "estimated_eval_tokens": 998 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1481, + "estimated_input_tokens": 371, + "estimated_eval_tokens": 998 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These two closed issues are title-only stubs with unrelated personal-name/date strings and no bug or feature description. They do not appear to be duplicates or part of the same underlying problem.", + "confidence": 0.98, + "canonical_issue_reason": "No clear canonical issue: neither item describes an actionable problem, and the titles do not indicate the same subject matter beyond both being low-signal stubs.", + "canonical_pr_reason": null, + "best_issue_reason": "If a representative must be chosen, issue 45764 is only marginally preferable because it was created slightly earlier, but it is not a meaningful canonical target.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:45764", + "right": "issue:45765", + "accept": false, + "reason": "Reject: the titles are just different names/dates with no shared bug, code path, or explicit linkage; they do not look like the same underlying issue." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 18, + "component_count": 23, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-45810-2", + "nodes": [ + "issue:45810", + "issue:46295" + ], + "soft_pairs": [ + "issue:45810|issue:46295" + ], + "prepared_review_unit_hash": "aa54513823ba01c4bf26b057029044fc24f3df324bc0923041cf039da12e85e1", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1541, + "estimated_input_tokens": 386, + "estimated_eval_tokens": 1028 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1541, + "estimated_input_tokens": 386, + "estimated_eval_tokens": 1028 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These two issues are unrelated: one requests a new Qwen3_5 token-classification/value-model class, while the other asks for FP8 training support. No duplicate triage relationship is supported.", + "confidence": 0.97, + "canonical_issue_reason": "No clear canonical issue for this cluster; the reports describe different features in different areas of the stack.", + "canonical_pr_reason": null, + "best_issue_reason": "Neither issue is a good cluster representative because they are not duplicates of each other.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:45810", + "right": "issue:46295", + "accept": false, + "reason": "Different underlying requests: model/class support for token classification versus FP8 training support. They do not describe the same bug or change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 19, + "component_count": 23, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-45869-5", + "nodes": [ + "issue:45869", + "issue:45870", + "issue:45871", + "issue:45872", + "issue:45873" + ], + "soft_pairs": [ + "issue:45869|issue:45870", + "issue:45869|issue:45871", + "issue:45870|issue:45871", + "issue:45870|issue:45872", + "issue:45870|issue:45873", + "issue:45871|issue:45872", + "issue:45871|issue:45873", + "issue:45872|issue:45873" + ], + "prepared_review_unit_hash": "26f6de163a91cb4a8c33b7aca5fa45bcaad73bf9990f4529dac448291feebddc", + "component_budget": { + "node_count": 5, + "item_count": 5, + "soft_pair_count": 8, + "serialized_chars": 5727, + "estimated_input_tokens": 1432, + "estimated_eval_tokens": 3120 + }, + "budget": { + "node_count": 5, + "item_count": 5, + "soft_pair_count": 8, + "serialized_chars": 5727, + "estimated_input_tokens": 1432, + "estimated_eval_tokens": 3120 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "This cluster contains two separate exact-duplicate issue pairs: 45870/45871 and 45872/45873. Issue 45869 is only loosely related by financial-data theme, but it describes a different dataset/sample offering rather than the same request or bug.", + "confidence": 0.95, + "canonical_issue_reason": "45870 is the earliest and clearest representative of the exact-duplicate pair 45870/45871, with the same title and apparent content; 45871 is redundant.", + "canonical_pr_reason": null, + "best_issue_reason": "45870 is the best overall issue representative because it is a clean exact-duplicate anchor with the earliest timestamp and a concrete, self-contained topic. The other issues split into a separate duplicate pair or are only thematically related.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:45869", + "right": "issue:45870", + "accept": false, + "reason": "Related financial-data theme, but one is a dataset/sample listing and the other is an example prediction use case; not the same underlying issue." + }, + { + "left": "issue:45869", + "right": "issue:45871", + "accept": false, + "reason": "Same as 45869/45870: broad theme overlap only, not the same request or change." + }, + { + "left": "issue:45870", + "right": "issue:45871", + "accept": true, + "reason": "Exact duplicate titles and highly similar content; same example issue, with 45870 as the earlier copy." + }, + { + "left": "issue:45870", + "right": "issue:45872", + "accept": false, + "reason": "Both mention financial time series/DataForge, but one is an example issue and the other is a scraping-service topic; different underlying subject." + }, + { + "left": "issue:45870", + "right": "issue:45873", + "accept": false, + "reason": "Different concrete topic from 45870; similar financial/DataForge wording is too broad to treat as a duplicate." + }, + { + "left": "issue:45871", + "right": "issue:45872", + "accept": false, + "reason": "Different issue themes: example prediction use case versus scraping-service integration." + }, + { + "left": "issue:45871", + "right": "issue:45873", + "accept": false, + "reason": "Different concrete subject; shared keywords are not enough to merge them." + }, + { + "left": "issue:45872", + "right": "issue:45873", + "accept": true, + "reason": "Exact duplicate titles and matching apparent content; same scraping-service issue, with 45872 as the earlier copy." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet: 45870/45871 share an identical title and body length, and 45872/45873 do as well, while 45869 is only loosely thematically related by title. The canonical choice of the earliest duplicate in each exact pair is conservative, and the soft-edge rejections do not overclaim beyond the available evidence." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 20, + "component_count": 23, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-46439-2", + "nodes": [ + "issue:46439", + "issue:46441" + ], + "soft_pairs": [ + "issue:46439|issue:46441" + ], + "prepared_review_unit_hash": "1e10978e612d0667297c0bd0b45091859ac8f4e6cb4ce33f8043c9b2ab0d3ecd", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1637, + "estimated_input_tokens": 410, + "estimated_eval_tokens": 1076 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1637, + "estimated_input_tokens": 410, + "estimated_eval_tokens": 1076 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both issues point at static-cache failures around `linear_attention`, but they describe different concrete breakages: one is shape corruption during `early_initialization`, the other is a `KeyError` in Qwen3.5 with `cache_implementation=\"static\"`. They are related, but not clearly the same underlying bug.", + "confidence": 0.66, + "canonical_issue_reason": "Issue 46439 is the better canonical issue because it states the likely root cause more directly (`early_initialization` corrupting linear-attention layers) and is broader than a model-specific crash report.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 46439 is the strongest representative for the cluster since it captures the underlying static-cache/linear-attention failure mode rather than a single downstream symptom.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:46439", + "right": "issue:46441", + "accept": false, + "reason": "Same broad area, but the reported failures differ materially: wrong-shape RuntimeError during early initialization vs. a `KeyError: 'linear_attention'` in Qwen3.5. Not enough evidence that they are the same bug or patch target." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 21, + "component_count": 23, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-46551-2", + "nodes": [ + "issue:46551", + "issue:46552" + ], + "soft_pairs": [ + "issue:46551|issue:46552" + ], + "prepared_review_unit_hash": "13ad7501bc622533ac19da1691560dad2d8be28776d0d4634544ffa9ee4e80cd", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1707, + "estimated_input_tokens": 427, + "estimated_eval_tokens": 1110 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1707, + "estimated_input_tokens": 427, + "estimated_eval_tokens": 1110 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These two issues are effectively duplicates: they have the same title, same underlying Wav2Vec2CTCTokenizer v5 bug description, and near-identical evidence. The open issue is the better canonical tracker.", + "confidence": 0.99, + "canonical_issue_reason": "Issue #46552 is open and appears to be the live tracker for the exact same tokenizer bug, so it is the better canonical issue.", + "canonical_pr_reason": null, + "best_issue_reason": "#46552 is the best issue to keep because it is open, current, and describes the same concrete bug as #46551.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:46551", + "right": "issue:46552", + "accept": true, + "reason": "Same title and same reported Wav2Vec2CTCTokenizer v5 behavior; these are the same underlying bug report." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The conclusion is well grounded: the two issues have identical titles, identical body length, and a soft-similarity score of 1.0, so treating them as duplicates and choosing the open issue as canonical is conservative and supported by the packet." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 22, + "component_count": 23, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-46657-2", + "nodes": [ + "issue:46657", + "issue:46692" + ], + "soft_pairs": [ + "issue:46657|issue:46692" + ], + "prepared_review_unit_hash": "28ff32d86f1243ec80bdfb66a679d82433386e7bf3d8558ef54533424b3da81a", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1569, + "estimated_input_tokens": 393, + "estimated_eval_tokens": 1042 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1569, + "estimated_input_tokens": 393, + "estimated_eval_tokens": 1042 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These two issues do not describe the same bug: one is a tensor shape mismatch during object-detection training, the other is a training-time out-of-memory problem in RT-DETR v2 after several epochs. The soft similarity is only at the broad \u201cobject detection training\u201d level, so it should not be merged.", + "confidence": 0.95, + "canonical_issue_reason": "Issue #46657 is the better canonical representative because it describes a concrete, specific failure mode (tensor size mismatch) rather than a broad resource symptom.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue #46657 is the most suitable issue to anchor on because it is more precise and actionable, with a clearer error signature than the OOM report.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:46657", + "right": "issue:46692", + "accept": false, + "reason": "Different underlying problems: one is a shape mismatch error, the other is memory exhaustion during later epochs. They share the training context but not the same code-path bug." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 23, + "component_count": 23, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-46822-2", + "nodes": [ + "issue:46822", + "issue:46823" + ], + "soft_pairs": [ + "issue:46822|issue:46823" + ], + "prepared_review_unit_hash": "15a541f8b122cce2b4d36be096da720e8bdba98fe301b5f6f7d7749d4cac7186", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1672, + "estimated_input_tokens": 418, + "estimated_eval_tokens": 1092 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1672, + "estimated_input_tokens": 418, + "estimated_eval_tokens": 1092 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are two separate model-loading bugs, not duplicates: one is a ZeRO-3/zero.Init partitioning failure causing OOM, the other is a 4-bit device_map/max_memory dispatch error causing CPU placement and ValueError.", + "confidence": 0.97, + "canonical_issue_reason": "issue:46822 is the better canonical representative because it has the richer reproduction context and more discussion, and it is the more specific ZeRO-3/zero.Init bug in this pair.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:46822 is the most representative issue in the cluster, but only because it is the more detailed report; it should not be treated as the same bug as issue:46823.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:46822", + "right": "issue:46823", + "accept": false, + "reason": "Reject: these fail in different code paths and with different mechanisms. 46822 is about ZeRO-3 zero.Init not partitioning a composite submodule and OOMing on load; 46823 is about 4-bit load_in_4bit with device_map='auto' and max_memory still dispatching to CPU and raising ValueError." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 38, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-40546-2", + "nodes": [ + "pull_request:40546", + "pull_request:42781" + ], + "soft_pairs": [ + "pull_request:40546|pull_request:42781" + ], + "prepared_review_unit_hash": "54467223b2ac28cf6ba9bd41777c79f701d7688ba4fbdcde66ed1bf62e9ee252", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 7224, + "estimated_input_tokens": 1806, + "estimated_eval_tokens": 3868 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 7224, + "estimated_input_tokens": 1806, + "estimated_eval_tokens": 3868 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are related VibeVoice PRs, but they do not look like duplicates: one is the main VibeVoice implementation, the other adds VibeVoice Realtime plus a separate acoustic tokenizer and generation changes. Overlap is mostly in shared model/docs scaffolding, not the same underlying change.", + "confidence": 0.91, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 40546 is the main, more established VibeVoice implementation and is better suited as the canonical representative of the feature work; PR 42781 is a narrower realtime extension in draft state.", + "best_issue_reason": null, + "best_pr_reason": "PR 40546 is the best overall PR candidate because it appears to be the primary implementation with substantial review activity and broader feature coverage, whereas PR 42781 is an add-on realtime follow-up.", + "soft_edge_verdicts": [ + { + "left": "pull_request:40546", + "right": "pull_request:42781", + "accept": false, + "reason": "Reject: despite shared VibeVoice filenames, 42781 adds realtime support and a new acoustic tokenizer with additional generation/modeling changes. This is an extension to the same family, not the same concrete bug or one mergeable fix." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 2, + "component_count": 38, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-40755-2", + "nodes": [ + "pull_request:40755", + "pull_request:41224" + ], + "soft_pairs": [ + "pull_request:40755|pull_request:41224" + ], + "prepared_review_unit_hash": "5a18f7cf0d18702711ecf80318d296faae214f3ba27955c5b77c2d471c919a23", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4675, + "estimated_input_tokens": 1169, + "estimated_eval_tokens": 2594 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4675, + "estimated_input_tokens": 1169, + "estimated_eval_tokens": 2594 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These two PRs are unrelated feature additions: one adds TimesFM covariate forecasting, the other adds DINOv3 image-classification support. They do not describe the same bug/change, so the soft duplicate edge should be rejected.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single canonical PR for a duplicate cluster here; if forced to pick a representative, PR 40755 is the more self-contained and substantial change set, while PR 41224 is a separate model feature.", + "best_issue_reason": null, + "best_pr_reason": "PR 40755 is the better cluster representative only by breadth and coherence: it implements one feature across code/docs/tests. PR 41224 is an entirely different model capability and does not fit the same underlying change.", + "soft_edge_verdicts": [ + { + "left": "pull_request:40755", + "right": "pull_request:41224", + "accept": false, + "reason": "Different models and code paths: TimesFM covariate forecasting vs. DINOv3 image-classification support. Shared \"model support\" theme is too broad to indicate the same underlying change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 3, + "component_count": 38, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-41356-2", + "nodes": [ + "pull_request:41356", + "pull_request:44339" + ], + "soft_pairs": [ + "pull_request:41356|pull_request:44339" + ], + "prepared_review_unit_hash": "aead1308c167b0bae6ff11bf9358340baa5e76f816618383418b505a10a9cd01", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 6050, + "estimated_input_tokens": 1513, + "estimated_eval_tokens": 3282 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 6050, + "estimated_input_tokens": 1513, + "estimated_eval_tokens": 3282 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Two DEIMv2 pull requests target the same model family, but they are not the same change: PR 41356 is an early/basic model + image processor add, while PR 44339 is a much broader follow-up that adds loss code, auto-mappings, conversion tooling, modular files, and more integration work.", + "confidence": 0.91, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 44339 is the more complete and representative DEIMv2 implementation; PR 41356 looks like an earlier partial version of the same model bring-up.", + "best_issue_reason": null, + "best_pr_reason": "PR 44339 is the strongest single artifact for the cluster because it subsumes the basic model work and includes the broader integration needed for a full Transformers addition.", + "soft_edge_verdicts": [ + { + "left": "pull_request:41356", + "right": "pull_request:44339", + "accept": false, + "reason": "Same model area, but not the same concrete change: 44339 expands the implementation substantially and adds new code paths beyond the basic PR 41356. They look like related iterations, not mergeable duplicates." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 4, + "component_count": 38, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-41855-15", + "nodes": [ + "pull_request:41855", + "pull_request:42668", + "pull_request:43251", + "pull_request:43254", + "pull_request:43424", + "pull_request:43823", + "pull_request:43995", + "pull_request:44827", + "pull_request:45170", + "pull_request:45213", + "pull_request:45655", + "pull_request:46048", + "pull_request:46267", + "pull_request:46707", + "pull_request:46911" + ], + "soft_pairs": [ + "pull_request:41855|pull_request:43823", + "pull_request:41855|pull_request:43424", + "pull_request:42668|pull_request:43823", + "pull_request:43251|pull_request:43823", + "pull_request:43251|pull_request:44827", + "pull_request:43424|pull_request:46911", + "pull_request:43424|pull_request:46707", + "pull_request:43424|pull_request:44827", + "pull_request:43424|pull_request:43823", + "pull_request:43823|pull_request:45655", + "pull_request:43823|pull_request:43995", + "pull_request:43995|pull_request:45213", + "pull_request:43995|pull_request:44827", + "pull_request:44827|pull_request:46911", + "pull_request:44827|pull_request:46707", + "pull_request:44827|pull_request:45170", + "pull_request:44827|pull_request:46267", + "pull_request:45170|pull_request:46707", + "pull_request:44827|pull_request:45213", + "pull_request:43823|pull_request:45213", + "pull_request:44827|pull_request:45655", + "pull_request:46048|pull_request:46707", + "pull_request:44827|pull_request:46048", + "pull_request:43823|pull_request:46707", + "pull_request:43823|pull_request:46911", + "pull_request:43251|pull_request:43254" + ], + "prepared_review_unit_hash": "abde52afea45308b81feceace91f09d88d8d1da1054a09ee42310decaf762274", + "component_budget": { + "node_count": 15, + "item_count": 15, + "soft_pair_count": 26, + "serialized_chars": 42250, + "estimated_input_tokens": 10563, + "estimated_eval_tokens": 21382 + }, + "budget": { + "node_count": 15, + "item_count": 15, + "soft_pair_count": 26, + "serialized_chars": 42250, + "estimated_input_tokens": 10563, + "estimated_eval_tokens": 21382 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These PRs are overwhelmingly unrelated and should not be clustered as duplicates. The only true overlap is the pair of fixed_cross_entropy changes (43251 and 43254), which both address the same cross-entropy kwargs bug; the rest are distinct features, refactors, or tests in different areas.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single canonical PR exists for the whole set because the items cover unrelated changes. If one must be chosen as the representative duplicate-like change, 43251 is the better canonical PR: it is the broader, issue-linked implementation of the same fixed_cross_entropy kwargs fix that 43254 narrows to.", + "best_issue_reason": null, + "best_pr_reason": "43251 is the strongest standalone PR in this set because it targets a concrete bug, is explicitly linked to an issue, and subsumes the same underlying fix addressed more narrowly by 43254.", + "soft_edge_verdicts": [ + { + "left": "pull_request:41855", + "right": "pull_request:43823", + "accept": false, + "reason": "Unrelated changes: Mistral tokenizer API additions vs adding a new MobileLLM model." + }, + { + "left": "pull_request:41855", + "right": "pull_request:43424", + "accept": false, + "reason": "Different code paths and goals: tokenizer methods vs Executorch dynamic-shape export tests." + }, + { + "left": "pull_request:42668", + "right": "pull_request:43823", + "accept": false, + "reason": "Processor auto-loading robustness is unrelated to MobileLLM model addition." + }, + { + "left": "pull_request:43251", + "right": "pull_request:43823", + "accept": false, + "reason": "Cross-entropy kwargs fix is unrelated to MobileLLM model scaffolding." + }, + { + "left": "pull_request:43251", + "right": "pull_request:44827", + "accept": false, + "reason": "Same broad loss/modeling repo area, but different problems: loss helper kwargs vs Mistral4 test fixes." + }, + { + "left": "pull_request:43424", + "right": "pull_request:46911", + "accept": false, + "reason": "Executorch dynamic export testing is unrelated to Olmo3 RoPE-per-layer behavior." + }, + { + "left": "pull_request:43424", + "right": "pull_request:46707", + "accept": false, + "reason": "Executorch export tests and native FSDP2 migration are distinct changes." + }, + { + "left": "pull_request:43424", + "right": "pull_request:44827", + "accept": false, + "reason": "Different concrete issues: export dynamic shapes vs Mistral4 test fixes." + }, + { + "left": "pull_request:43424", + "right": "pull_request:43823", + "accept": false, + "reason": "No shared bug or change: export testing vs adding MobileLLM." + }, + { + "left": "pull_request:43823", + "right": "pull_request:45655", + "accept": false, + "reason": "New model addition and config-resolution fix are unrelated." + }, + { + "left": "pull_request:43823", + "right": "pull_request:43995", + "accept": false, + "reason": "MobileLLM model files vs Falcon output-collection refactor are different changes." + }, + { + "left": "pull_request:43823", + "right": "pull_request:45213", + "accept": false, + "reason": "Adding MobileLLM is not the same as adding the model-creation skill tooling." + }, + { + "left": "pull_request:43823", + "right": "pull_request:46707", + "accept": false, + "reason": "Model addition vs FSDP2 module migration are unrelated." + }, + { + "left": "pull_request:43823", + "right": "pull_request:46911", + "accept": false, + "reason": "MobileLLM model addition and Olmo3 RoPE changes are unrelated." + }, + { + "left": "pull_request:43995", + "right": "pull_request:44827", + "accept": false, + "reason": "Falcon output-interface refactor is not the same as Mistral4 test fixes." + }, + { + "left": "pull_request:43995", + "right": "pull_request:45213", + "accept": false, + "reason": "Falcon refactor and model-creation skill are entirely different concerns." + }, + { + "left": "pull_request:44827", + "right": "pull_request:45170", + "accept": false, + "reason": "Both touch model code, but one fixes Mistral4 tests while the other corrects a CLIP-family spelling rename; not the same bug." + }, + { + "left": "pull_request:44827", + "right": "pull_request:45213", + "accept": false, + "reason": "Mistral4 tests are unrelated to the model-creation skill PR." + }, + { + "left": "pull_request:44827", + "right": "pull_request:45655", + "accept": false, + "reason": "Different issues: Mistral4 test fixes vs config-resolution ordering." + }, + { + "left": "pull_request:44827", + "right": "pull_request:46048", + "accept": false, + "reason": "Mistral4 tests are unrelated to adding the TDT loss kernel." + }, + { + "left": "pull_request:44827", + "right": "pull_request:46267", + "accept": false, + "reason": "Different subsystems and goals: model tests vs chat parsing updates." + }, + { + "left": "pull_request:44827", + "right": "pull_request:46707", + "accept": false, + "reason": "Mistral4 test fixes and FSDP2 migration are unrelated." + }, + { + "left": "pull_request:44827", + "right": "pull_request:46911", + "accept": false, + "reason": "Mistral4 test fixes do not overlap with Olmo3 RoPE changes." + }, + { + "left": "pull_request:45170", + "right": "pull_request:46707", + "accept": false, + "reason": "Typos in CLIP-like weight renaming are unrelated to native FSDP2 migration." + }, + { + "left": "pull_request:46048", + "right": "pull_request:46707", + "accept": false, + "reason": "TDT loss kernel support and FSDP2 migration are separate feature additions." + }, + { + "left": "pull_request:43251", + "right": "pull_request:43254", + "accept": true, + "reason": "Same underlying bug: both fix fixed_cross_entropy to pass supported kwargs through to nn.functional.cross_entropy, and they could plausibly be merged into one PR." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays conservative. The only supported overlap is 43251/43254, which both touch the same fixed_cross_entropy kwargs bug and look mergeable into one bugfix PR. The rest of the pairwise rejections are consistent with the diffs and titles, and I don\u2019t see any overstatement of evidence." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 5, + "component_count": 38, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-42975-2", + "nodes": [ + "pull_request:42975", + "pull_request:42976" + ], + "soft_pairs": [ + "pull_request:42975|pull_request:42976" + ], + "prepared_review_unit_hash": "90def3c23cf3a673dbce63eee60f32c958723072c9d9e4bb028dd7dacf086a1b", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5934, + "estimated_input_tokens": 1484, + "estimated_eval_tokens": 3224 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5934, + "estimated_input_tokens": 1484, + "estimated_eval_tokens": 3224 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Two workflow-maintenance PRs overlap on GitHub Actions files, but they are different upgrade efforts: one is specifically about Node 24 compatibility for core actions, while the other updates Docker/build-related actions to newer versions. They are related in area, not the same underlying change.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 42975 is the better canonical representative because it is the more broadly scoped and clearly motivated GitHub Actions compatibility upgrade, touching many of the shared workflow files in this cluster.", + "best_issue_reason": null, + "best_pr_reason": "PR 42975 is the best single representative of the cluster: it covers the common workflow-action upgrade theme more comprehensively and has the clearest stated purpose.", + "soft_edge_verdicts": [ + { + "left": "pull_request:42975", + "right": "pull_request:42976", + "accept": false, + "reason": "They both update workflow dependencies, but they target different action families and different concrete goals (Node 24 compatibility vs latest Docker/build action versions). This is not the same underlying change and would not plausibly be merged into one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 6, + "component_count": 38, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-42978-6", + "nodes": [ + "pull_request:42978", + "pull_request:43448", + "pull_request:43451", + "pull_request:45490", + "pull_request:46562", + "pull_request:46658" + ], + "soft_pairs": [ + "pull_request:42978|pull_request:43448", + "pull_request:42978|pull_request:43451", + "pull_request:42978|pull_request:45490", + "pull_request:43448|pull_request:43451", + "pull_request:43448|pull_request:45490", + "pull_request:43451|pull_request:45490", + "pull_request:45490|pull_request:46562", + "pull_request:46562|pull_request:46658" + ], + "prepared_review_unit_hash": "52437740d2b97d939f913d551bbce2952845ce1cfb0c58bdf77bf7b6294b96ac", + "component_budget": { + "node_count": 6, + "item_count": 6, + "soft_pair_count": 8, + "serialized_chars": 21192, + "estimated_input_tokens": 5298, + "estimated_eval_tokens": 10852 + }, + "budget": { + "node_count": 6, + "item_count": 6, + "soft_pair_count": 8, + "serialized_chars": 21192, + "estimated_input_tokens": 5298, + "estimated_eval_tokens": 10852 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are mostly unrelated model-introduction PRs that only overlap on shared boilerplate files. The lone high-similarity pair (Step 3.7 Flash vs a mini-model conversion/debug script) is not a duplicate and should not be merged.", + "confidence": 0.93, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 46562 is the best representative of the cluster: it is the substantive library integration for Step 3.7 Flash, while the other Step 3.7 item is a standalone script/debug artifact rather than the same change.", + "best_issue_reason": null, + "best_pr_reason": "PR 46562 is the strongest single PR in the set because it contains the actual model implementation, docs, auto mappings, and tests; the others are separate model-addition PRs or auxiliary scripts.", + "soft_edge_verdicts": [ + { + "left": "pull_request:42978", + "right": "pull_request:43448", + "accept": false, + "reason": "Different model families (ViT NEPA vs Molmo); shared docs/auto-registration files are boilerplate and do not indicate the same change." + }, + { + "left": "pull_request:42978", + "right": "pull_request:43451", + "accept": false, + "reason": "Different model families (ViT NEPA vs Molmo2) with distinct codepaths and test coverage; overlap is only in shared integration files." + }, + { + "left": "pull_request:42978", + "right": "pull_request:45490", + "accept": false, + "reason": "ViT NEPA and CTSM are unrelated model additions; same broad repo scaffolding, not the same bug or feature." + }, + { + "left": "pull_request:43448", + "right": "pull_request:43451", + "accept": false, + "reason": "Molmo and Molmo2 are separate model integrations with different modules, tests, and conversion logic; naming similarity is not enough." + }, + { + "left": "pull_request:43448", + "right": "pull_request:45490", + "accept": false, + "reason": "Molmo and CTSM target different architectures and codepaths; shared registry/docs files are incidental." + }, + { + "left": "pull_request:43451", + "right": "pull_request:45490", + "accept": false, + "reason": "Molmo2 and CTSM are distinct model implementations, not the same underlying fix or feature." + }, + { + "left": "pull_request:45490", + "right": "pull_request:46562", + "accept": false, + "reason": "CTSM and Step 3.7 Flash are unrelated models; overlap is limited to auto-mapping/boilerplate updates." + }, + { + "left": "pull_request:46562", + "right": "pull_request:46658", + "accept": false, + "reason": "The Step 3.7 Flash PR adds the actual library model, while 46658 is a standalone mini-model/build/debug script for conversion work; they are not the same concrete code-path change and would not be merged as one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 7, + "component_count": 38, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-43291-7", + "nodes": [ + "pull_request:43291", + "pull_request:43665", + "pull_request:44815", + "pull_request:46380", + "pull_request:46539", + "pull_request:46886", + "pull_request:46922" + ], + "soft_pairs": [ + "pull_request:43291|pull_request:43665", + "pull_request:43291|pull_request:44815", + "pull_request:43291|pull_request:46380", + "pull_request:43291|pull_request:46922", + "pull_request:43665|pull_request:46539", + "pull_request:43665|pull_request:44815", + "pull_request:43665|pull_request:46380", + "pull_request:43665|pull_request:46922", + "pull_request:44815|pull_request:46539", + "pull_request:44815|pull_request:46380", + "pull_request:44815|pull_request:46922", + "pull_request:46380|pull_request:46539", + "pull_request:46380|pull_request:46922", + "pull_request:46539|pull_request:46922", + "pull_request:46886|pull_request:46922" + ], + "prepared_review_unit_hash": "3cb60424b34a3f1457eb1f7e196966860dc2960eb886b7669414316de01e2ad8", + "component_budget": { + "node_count": 7, + "item_count": 7, + "soft_pair_count": 15, + "serialized_chars": 19604, + "estimated_input_tokens": 4901, + "estimated_eval_tokens": 10058 + }, + "budget": { + "node_count": 7, + "item_count": 7, + "soft_pair_count": 15, + "serialized_chars": 19604, + "estimated_input_tokens": 4901, + "estimated_eval_tokens": 10058 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Only the two YOLOS pull requests look like the same underlying bug fix; the rest target unrelated models, CI, generation validation, or tests. 46886 is the stronger canonical PR because it has the explicit issue target and the clearer, more focused title.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 46886 is the cleanest representative: it explicitly targets issue 46869, has a precise title about the YOLOS device_map='auto' mismatch, and the diff directly addresses that bug path.", + "best_issue_reason": null, + "best_pr_reason": "PR 46886 is the best overall representative of the cluster because it is the most specific and complete fix for the only true duplicate pair in the set.", + "soft_edge_verdicts": [ + { + "left": "pull_request:43291", + "right": "pull_request:43665", + "accept": false, + "reason": "Different models and different concerns: Whisper tokenizer/test cleanup versus CLIP/ViT test edits." + }, + { + "left": "pull_request:43291", + "right": "pull_request:44815", + "accept": false, + "reason": "Whisper decoding changes are unrelated to FP8 dequantization and model loading logic." + }, + { + "left": "pull_request:43291", + "right": "pull_request:46380", + "accept": false, + "reason": "Whisper tests/tokenization versus generation watermarking validation; different code paths and bugs." + }, + { + "left": "pull_request:43291", + "right": "pull_request:46922", + "accept": false, + "reason": "Whisper tokenizer/test work is unrelated to the YOLOS device-mismatch fix." + }, + { + "left": "pull_request:43665", + "right": "pull_request:46539", + "accept": false, + "reason": "Model test assertions in CLIP/ViT are unrelated to Phimoe CI integration test infrastructure." + }, + { + "left": "pull_request:43665", + "right": "pull_request:44815", + "accept": false, + "reason": "Different subsystems and failure modes: vision-model tests versus FP8 dequant/model-loading changes." + }, + { + "left": "pull_request:43665", + "right": "pull_request:46380", + "accept": false, + "reason": "Both touch tests, but they are for different bugs and different models; not the same change." + }, + { + "left": "pull_request:43665", + "right": "pull_request:46922", + "accept": false, + "reason": "CLIP/ViT test edits do not match the YOLOS runtime device-placement bug." + }, + { + "left": "pull_request:44815", + "right": "pull_request:46539", + "accept": false, + "reason": "FP8 quantization/dequantization fixes are unrelated to CI memory/device-map setup." + }, + { + "left": "pull_request:44815", + "right": "pull_request:46380", + "accept": false, + "reason": "Different core problems: FP8 dequantization versus generation-config validation." + }, + { + "left": "pull_request:44815", + "right": "pull_request:46922", + "accept": false, + "reason": "No shared bug path; FP8 kernel/model-loading work is unrelated to YOLOS device mismatch." + }, + { + "left": "pull_request:46380", + "right": "pull_request:46539", + "accept": false, + "reason": "Generation configuration validation is unrelated to Phimoe integration test CI behavior." + }, + { + "left": "pull_request:46380", + "right": "pull_request:46922", + "accept": false, + "reason": "Different bug domains and code paths: generation config versus YOLOS forward/device placement." + }, + { + "left": "pull_request:46539", + "right": "pull_request:46922", + "accept": false, + "reason": "CI instance-sharing memory capping does not overlap with the YOLOS device_map='auto' model bug." + }, + { + "left": "pull_request:46886", + "right": "pull_request:46922", + "accept": true, + "reason": "Both PRs fix the same YOLOS device-mismatch issue under device_map='auto' in the same file and share the same issue target; they are plausible alternatives for one merged fix." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The analysis is grounded in the packet: the only clearly related pair is PR 46886 and PR 46922, which both target the YOLOS device-map/device-mismatch bug in the same file and are plausibly mergeable into one fix. The rest of the pairwise rejections are conservative and consistent with the differing models, code paths, and issue contexts shown in the packet. The canonical PR choice is also supported by the explicit issue target and the more focused title." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 8, + "component_count": 38, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-43488-3", + "nodes": [ + "pull_request:43488", + "pull_request:43532", + "pull_request:46747" + ], + "soft_pairs": [ + "pull_request:43488|pull_request:46747", + "pull_request:43488|pull_request:43532", + "pull_request:43532|pull_request:46747" + ], + "prepared_review_unit_hash": "b6b254c5cf42c063af6e9a0bf85f6e7d3aa8e2b8f74d64e8f7aa6ed5cea5c6c2", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 6350, + "estimated_input_tokens": 1588, + "estimated_eval_tokens": 3432 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 6350, + "estimated_input_tokens": 1588, + "estimated_eval_tokens": 3432 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "All three PRs are distinct \"don\u2019t merge\" test changes with only superficial overlap in CI/testing areas; they should not be deduplicated.", + "confidence": 0.88, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 46747 is the narrowest and most self-contained change, but the three PRs are not duplicates of one another.", + "best_issue_reason": null, + "best_pr_reason": "46747 is the best representative of the cluster because it is the smallest, clearest workflow-only change.", + "soft_edge_verdicts": [ + { + "left": "pull_request:43488", + "right": "pull_request:46747", + "accept": false, + "reason": "Different changes: 43488 touches packaging/dependency versioning and a test file, while 46747 only switches a workflow trigger for the security gate. Not the same underlying bug or fix." + }, + { + "left": "pull_request:43488", + "right": "pull_request:43532", + "accept": false, + "reason": "Unrelated edits: 43488 is setup/dependency/test scaffolding, while 43532 rewires several CI workflows and helper code. Shared Vit test file is not enough to make them the same change." + }, + { + "left": "pull_request:43532", + "right": "pull_request:46747", + "accept": false, + "reason": "Both are workflow-related, but they fix different code paths: 43532 changes comment/scheduled/slack CI plumbing, while 46747 only changes the security-gate workflow trigger." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 9, + "component_count": 38, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-43612-2", + "nodes": [ + "pull_request:43612", + "pull_request:43613" + ], + "soft_pairs": [ + "pull_request:43612|pull_request:43613" + ], + "prepared_review_unit_hash": "1e7b33a5c2d2c2d4c38d9e46f7caf25093af41176d7f68c0dbd762e58ee133a6", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 6082, + "estimated_input_tokens": 1521, + "estimated_eval_tokens": 3298 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 6082, + "estimated_input_tokens": 1521, + "estimated_eval_tokens": 3298 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are two related but distinct PRs: one adds a Promptable Concept Segmentation pipeline, the other adds a Promptable Visual Segmentation pipeline. They share scaffolding/docs, but target different tasks and model integrations, so they are not duplicates.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": "PR 43613 is the broader, more representative pipeline addition, with more model integrations and tests; however, it is still a separate change from 43612.", + "soft_edge_verdicts": [ + { + "left": "Add Promptable Concept Segmentation pipeline", + "right": "Add Promptable Visual Segmentation pipeline", + "accept": false, + "reason": "Different segmentation tasks and different implementation/test surfaces; shared docs and pipeline boilerplate are not enough to treat them as the same change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 10, + "component_count": 38, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-43651-2", + "nodes": [ + "pull_request:43651", + "pull_request:43663" + ], + "soft_pairs": [ + "pull_request:43651|pull_request:43663" + ], + "prepared_review_unit_hash": "1d0cca5ed35ad72887c3e8bdf665f24ef757a0f4030598386502040ff5e008c6", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4293, + "estimated_input_tokens": 1074, + "estimated_eval_tokens": 2404 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4293, + "estimated_input_tokens": 1074, + "estimated_eval_tokens": 2404 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "The two pull requests both touch `trainer.py` but address different trainer extension points: one changes gradient-accumulation loss scaling, the other refactors signature-column filtering for custom trainers/PEFT models. They are not duplicates of the same underlying change.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": "PR #43651 is the more concrete fix to a specific training-path bug (loss scaling during gradient accumulation), while PR #43663 is a broader refactor/override hook for column filtering.", + "soft_edge_verdicts": [ + { + "left": "pull_request:43651", + "right": "pull_request:43663", + "accept": false, + "reason": "Same file, but different code paths and problems: one fixes loss normalization in `training_step`, the other adds `_get_signature_columns` for input column filtering. They are not the same underlying change and would not be merged into one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 11, + "component_count": 38, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-43656-2", + "nodes": [ + "pull_request:43656", + "pull_request:43842" + ], + "soft_pairs": [ + "pull_request:43656|pull_request:43842" + ], + "prepared_review_unit_hash": "a0bedeeca104af34c1872f05231bb327abaead2218cd1b73e4ebeff2e77eea5d", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 3664, + "estimated_input_tokens": 916, + "estimated_eval_tokens": 2088 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 3664, + "estimated_input_tokens": 916, + "estimated_eval_tokens": 2088 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Two open PRs in the same CLI file both touch TypeAdapter-related typing, but they appear to address different runtime/import issues rather than the same exact fix.", + "confidence": 0.77, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #43842 is the more specific and direct fix: it targets the TypeAdapter NameError when pydantic is not installed, with a minimal TYPE_CHECKING import/quoted annotation change.", + "best_issue_reason": null, + "best_pr_reason": "PR #43842 is the best representative of the cluster because its scope is narrowly aligned to the concrete TypeAdapter failure mode, whereas #43656 is a broader typing/postponed-annotations cleanup.", + "soft_edge_verdicts": [ + { + "left": "pull_request:43656", + "right": "pull_request:43842", + "accept": false, + "reason": "Both modify src/transformers/cli/serve.py and mention TypeAdapter, but they do not look like the same concrete change: #43656 adds postponed annotations and broader typing cleanup, while #43842 specifically fixes optional pydantic/TypeAdapter import handling. The issue targets also differ." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 12, + "component_count": 38, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-43779-2", + "nodes": [ + "pull_request:43779", + "pull_request:43816" + ], + "soft_pairs": [ + "pull_request:43779|pull_request:43816" + ], + "prepared_review_unit_hash": "e7c674adcc031aa8883109a5dfc678b4407fb20b10764302092e384d031ba83a", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4362, + "estimated_input_tokens": 1091, + "estimated_eval_tokens": 2438 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4362, + "estimated_input_tokens": 1091, + "estimated_eval_tokens": 2438 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both pull requests appear to implement the same SwanLab integration change: adding run id/resume support in `integration_utils.py` for issue #43698, with one variant also expanding the docs. They are strong duplicate PR candidates.", + "confidence": 0.94, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #43816 is the cleaner representative: it carries the same SwanLab id/resume support plus the accompanying documentation updates, making it the more complete version of the shared change.", + "best_issue_reason": null, + "best_pr_reason": "PR #43816 is the best global representative because it appears to include the full feature change and docs for SwanLab run id/resume support, while #43779 is the narrower implementation-focused variant.", + "soft_edge_verdicts": [ + { + "left": "pull_request:43779", + "right": "pull_request:43816", + "accept": true, + "reason": "Same underlying SwanLab callback enhancement for id/resume support in the same file and same issue target; the doc differences do not change the core code-path change, so these are duplicate PRs." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet: both PRs target issue #43698, touch the same file, and the previews indicate the same SwanLab id/resume support with #43816 adding docs. The duplicate/mergeable verdict is conservative enough given the shared target and overlapping code change." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 13, + "component_count": 38, + "review_unit_index": 1, + "review_unit_count": 2, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-43996-28", + "nodes": [ + "pull_request:43996", + "pull_request:43997", + "pull_request:44007", + "pull_request:44010", + "pull_request:44013", + "pull_request:44017", + "pull_request:44018", + "pull_request:44019", + "pull_request:44024", + "pull_request:44044", + "pull_request:44056", + "pull_request:44059", + "pull_request:44066", + "pull_request:44068", + "pull_request:44071", + "pull_request:44072", + "pull_request:44073", + "pull_request:44074", + "pull_request:44076", + "pull_request:44085", + "pull_request:44086", + "pull_request:44098", + "pull_request:44101", + "pull_request:44116", + "pull_request:44129", + "pull_request:44154", + "pull_request:44161", + "pull_request:44722" + ], + "soft_pairs": [ + "pull_request:43996|pull_request:44085", + "pull_request:43996|pull_request:44044", + "pull_request:44066|pull_request:44085", + "pull_request:44007|pull_request:44072", + "pull_request:44072|pull_request:44722", + "pull_request:44066|pull_request:44072", + "pull_request:44013|pull_request:44044", + "pull_request:44066|pull_request:44086", + "pull_request:44018|pull_request:44068", + "pull_request:44066|pull_request:44071", + "pull_request:44066|pull_request:44068", + "pull_request:44018|pull_request:44066", + "pull_request:44019|pull_request:44722", + "pull_request:44086|pull_request:44722", + "pull_request:44019|pull_request:44085", + "pull_request:44018|pull_request:44019", + "pull_request:44019|pull_request:44071", + "pull_request:44068|pull_request:44722", + "pull_request:44068|pull_request:44116", + "pull_request:44018|pull_request:44116", + "pull_request:44071|pull_request:44722", + "pull_request:44018|pull_request:44085", + "pull_request:44018|pull_request:44086", + "pull_request:44068|pull_request:44086", + "pull_request:44071|pull_request:44085", + "pull_request:44085|pull_request:44086", + "pull_request:44059|pull_request:44086", + "pull_request:44098|pull_request:44101", + "pull_request:44018|pull_request:44071", + "pull_request:44059|pull_request:44722", + "pull_request:44017|pull_request:44161", + "pull_request:44017|pull_request:44722", + "pull_request:44017|pull_request:44073", + "pull_request:44068|pull_request:44071", + "pull_request:43997|pull_request:44073", + "pull_request:44071|pull_request:44086", + "pull_request:44017|pull_request:44071", + "pull_request:44073|pull_request:44101", + "pull_request:44073|pull_request:44161", + "pull_request:44059|pull_request:44071", + "pull_request:44056|pull_request:44101", + "pull_request:44018|pull_request:44059", + "pull_request:44059|pull_request:44068", + "pull_request:44101|pull_request:44161", + "pull_request:43997|pull_request:44017", + "pull_request:43997|pull_request:44056", + "pull_request:44071|pull_request:44076", + "pull_request:44074|pull_request:44101", + "pull_request:43997|pull_request:44018", + "pull_request:44056|pull_request:44161", + "pull_request:43997|pull_request:44161", + "pull_request:44074|pull_request:44161", + "pull_request:44018|pull_request:44161", + "pull_request:44010|pull_request:44071", + "pull_request:44010|pull_request:44073", + "pull_request:44018|pull_request:44129", + "pull_request:43997|pull_request:44074", + "pull_request:44010|pull_request:44018", + "pull_request:43997|pull_request:44129", + "pull_request:44071|pull_request:44129", + "pull_request:43996|pull_request:44161", + "pull_request:43996|pull_request:44017", + "pull_request:44010|pull_request:44044", + "pull_request:44044|pull_request:44074", + "pull_request:44066|pull_request:44722", + "pull_request:44085|pull_request:44722", + "pull_request:44085|pull_request:44129", + "pull_request:43996|pull_request:44007", + "pull_request:43996|pull_request:44129", + "pull_request:44044|pull_request:44129", + "pull_request:44044|pull_request:44154", + "pull_request:44024|pull_request:44076" + ], + "prepared_review_unit_hash": "35749bac72ee7c14cf4a9ba5997130760252b49bbadf68095381606dcbab60ee", + "component_budget": { + "node_count": 28, + "item_count": 28, + "soft_pair_count": 77, + "serialized_chars": 84511, + "estimated_input_tokens": 21128, + "estimated_eval_tokens": 42512 + }, + "budget": { + "node_count": 28, + "item_count": 28, + "soft_pair_count": 72, + "serialized_chars": 76943, + "estimated_input_tokens": 19236, + "estimated_eval_tokens": 38728 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a broad wave of similar output-tracing refactors across many different model implementations, not one duplicate bug. The only convincing duplicates are within the GPT-Neo and GPT-J families, where multiple PRs touch the same concrete code paths and appear to be overlapping revisions of the same refactor.", + "confidence": 0.87, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 44722 is the strongest representative of the true duplicate subgroup: it covers the GPT-J output-tracing refactor and also extends it to CodeGen, making it the most complete version of that concrete change family.", + "best_issue_reason": null, + "best_pr_reason": "PR 44722 is the best global representative because it is the latest and broadest concrete implementation of the overlapping GPT-J/CodeGen tracing refactor, with explicit issue linkage and the clearest code-path overlap.", + "soft_edge_verdicts": [ + { + "left": "pull_request:43996", + "right": "pull_request:44085", + "accept": false, + "reason": "Different model families and different code paths; same tracking issue is only a shared refactor theme." + }, + { + "left": "pull_request:43996", + "right": "pull_request:44044", + "accept": false, + "reason": "Different models and files; both are output-tracing refactors but not the same concrete bug/change." + }, + { + "left": "pull_request:44066", + "right": "pull_request:44085", + "accept": true, + "reason": "Same GPT-J output-tracing code path in the same file; 44085 looks like a narrower overlap of the broader GPT-J/CodeGen refactor." + }, + { + "left": "pull_request:44007", + "right": "pull_request:44072", + "accept": false, + "reason": "RegNet/ResNet-family vs EfficientNet: same umbrella refactor pattern, but not the same underlying change." + }, + { + "left": "pull_request:44072", + "right": "pull_request:44722", + "accept": false, + "reason": "EfficientNet and GPT-J are unrelated code paths; this is only similarity in refactor style." + }, + { + "left": "pull_request:44066", + "right": "pull_request:44072", + "accept": false, + "reason": "GPT-J/CodeGen and EfficientNet are different implementations; no concrete overlap beyond the generic tracing refactor." + }, + { + "left": "pull_request:44013", + "right": "pull_request:44044", + "accept": false, + "reason": "MobileNetV2 vs DeBERTa are unrelated models; shared issue target is not enough to merge them." + }, + { + "left": "pull_request:44066", + "right": "pull_request:44086", + "accept": false, + "reason": "Different model code paths (GPT-J/CodeGen vs MGP-STR); only the refactor pattern is similar." + }, + { + "left": "pull_request:44018", + "right": "pull_request:44068", + "accept": true, + "reason": "Both are GPT-Neo tracing refactors in the same file/test area and look like overlapping revisions of the same change." + }, + { + "left": "pull_request:44066", + "right": "pull_request:44071", + "accept": false, + "reason": "GPT-J/CodeGen and MPT are unrelated implementations; no same bug/change." + }, + { + "left": "pull_request:44066", + "right": "pull_request:44068", + "accept": false, + "reason": "GPT-J/CodeGen vs GPT-Neo: same theme, different model and different code path." + }, + { + "left": "pull_request:44018", + "right": "pull_request:44066", + "accept": false, + "reason": "GPT-Neo vs GPT-J/CodeGen are different model families; not the same concrete refactor." + }, + { + "left": "pull_request:44019", + "right": "pull_request:44722", + "accept": false, + "reason": "ResNet and GPT-J are unrelated code paths; shared issue label does not imply duplication." + }, + { + "left": "pull_request:44086", + "right": "pull_request:44722", + "accept": false, + "reason": "MGP-STR vs GPT-J are different models and different output paths." + }, + { + "left": "pull_request:44019", + "right": "pull_request:44085", + "accept": false, + "reason": "ResNet and GPT-J are unrelated implementations." + }, + { + "left": "pull_request:44018", + "right": "pull_request:44019", + "accept": false, + "reason": "GPT-Neo and ResNet are different model families; only the refactor theme matches." + }, + { + "left": "pull_request:44019", + "right": "pull_request:44071", + "accept": false, + "reason": "ResNet and MPT are unrelated model code paths." + }, + { + "left": "pull_request:44068", + "right": "pull_request:44722", + "accept": false, + "reason": "GPT-Neo and GPT-J are different model implementations; not the same fix." + }, + { + "left": "pull_request:44068", + "right": "pull_request:44116", + "accept": false, + "reason": "GPT-Neo and Flaubert are unrelated models." + }, + { + "left": "pull_request:44018", + "right": "pull_request:44071", + "accept": false, + "reason": "GPT-Neo and MPT are unrelated code paths." + }, + { + "left": "pull_request:44018", + "right": "pull_request:44085", + "accept": false, + "reason": "GPT-Neo vs GPT-J/CodeGen: same refactor pattern, different model code." + }, + { + "left": "pull_request:44018", + "right": "pull_request:44086", + "accept": false, + "reason": "GPT-Neo and MGP-STR are unrelated implementations." + }, + { + "left": "pull_request:44068", + "right": "pull_request:44086", + "accept": false, + "reason": "GPT-Neo and MGP-STR are unrelated code paths." + }, + { + "left": "pull_request:44071", + "right": "pull_request:44722", + "accept": false, + "reason": "MPT and GPT-J are different model families; no shared concrete bug." + }, + { + "left": "pull_request:44018", + "right": "pull_request:44059", + "accept": false, + "reason": "GPT-Neo and GPT-2 are different models; only the tracing refactor theme overlaps." + }, + { + "left": "pull_request:44059", + "right": "pull_request:44722", + "accept": false, + "reason": "GPT-2 and GPT-J are different code paths." + }, + { + "left": "pull_request:44017", + "right": "pull_request:44161", + "accept": false, + "reason": "Segformer and LongT5 are unrelated models." + }, + { + "left": "pull_request:44017", + "right": "pull_request:44722", + "accept": false, + "reason": "Segformer and GPT-J are unrelated code paths." + }, + { + "left": "pull_request:44017", + "right": "pull_request:44073", + "accept": false, + "reason": "Segformer and VisualBERT are different implementations." + }, + { + "left": "pull_request:44068", + "right": "pull_request:44071", + "accept": false, + "reason": "GPT-Neo and MPT are different model families." + }, + { + "left": "pull_request:43997", + "right": "pull_request:44073", + "accept": false, + "reason": "RegNet and VisualBERT are unrelated models." + }, + { + "left": "pull_request:44073", + "right": "pull_request:44101", + "accept": false, + "reason": "VisualBERT and Flaubert/XLM are different code paths." + }, + { + "left": "pull_request:44074", + "right": "pull_request:44101", + "accept": false, + "reason": "TextNet and Flaubert/XLM are unrelated implementations." + }, + { + "left": "pull_request:44056", + "right": "pull_request:44161", + "accept": false, + "reason": "MPNet and LongT5 are unrelated models." + }, + { + "left": "pull_request:43996", + "right": "pull_request:44017", + "accept": false, + "reason": "Cvt/FNet and Segformer are different model families." + }, + { + "left": "pull_request:44010", + "right": "pull_request:44071", + "accept": false, + "reason": "SqueezeBERT and MPT are unrelated code paths." + }, + { + "left": "pull_request:44010", + "right": "pull_request:44073", + "accept": false, + "reason": "SqueezeBERT and VisualBERT are different models." + }, + { + "left": "pull_request:44010", + "right": "pull_request:44018", + "accept": false, + "reason": "SqueezeBERT and GPT-Neo are unrelated implementations." + }, + { + "left": "pull_request:44010", + "right": "pull_request:44044", + "accept": false, + "reason": "SqueezeBERT and DeBERTa are different model code paths." + }, + { + "left": "pull_request:44044", + "right": "pull_request:44074", + "accept": false, + "reason": "DeBERTa and TextNet are unrelated models." + }, + { + "left": "pull_request:44024", + "right": "pull_request:44076", + "accept": false, + "reason": "FocalNet and ImageGPT are different code paths; same generic refactor style only." + }, + { + "left": "pull_request:44059", + "right": "pull_request:44071", + "accept": false, + "reason": "GPT-2 and MPT are unrelated model families." + }, + { + "left": "pull_request:44059", + "right": "pull_request:44068", + "accept": false, + "reason": "GPT-2 and GPT-Neo are different implementations." + }, + { + "left": "pull_request:44074", + "right": "pull_request:44101", + "accept": false, + "reason": "TextNet and Flaubert/XLM are unrelated models." + }, + { + "left": "pull_request:43997", + "right": "pull_request:44056", + "accept": false, + "reason": "RegNet and MPNet are unrelated code paths." + }, + { + "left": "pull_request:43997", + "right": "pull_request:44074", + "accept": false, + "reason": "RegNet and TextNet are different model families." + }, + { + "left": "pull_request:44071", + "right": "pull_request:44076", + "accept": false, + "reason": "MPT and ImageGPT are unrelated implementations." + }, + { + "left": "pull_request:44074", + "right": "pull_request:44161", + "accept": false, + "reason": "TextNet and LongT5 are different model code paths." + }, + { + "left": "pull_request:44018", + "right": "pull_request:44129", + "accept": false, + "reason": "GPT-Neo and SpeechT5 are unrelated models." + }, + { + "left": "pull_request:44018", + "right": "pull_request:44059", + "accept": false, + "reason": "GPT-Neo and GPT-2 are different model families." + }, + { + "left": "pull_request:44056", + "right": "pull_request:44129", + "accept": false, + "reason": "MPNet and SpeechT5 are unrelated code paths." + }, + { + "left": "pull_request:44066", + "right": "pull_request:44722", + "accept": true, + "reason": "Both touch the same GPT-J tracing refactor; 44722 is the broader follow-up that subsumes the GPT-J part of 44066." + }, + { + "left": "pull_request:44085", + "right": "pull_request:44722", + "accept": true, + "reason": "Same GPT-J file and same output-tracing migration; 44722 is a broader version of the same concrete change." + }, + { + "left": "pull_request:44085", + "right": "pull_request:44129", + "accept": false, + "reason": "GPT-J and SpeechT5 are unrelated models; shared issue target is only thematic." + }, + { + "left": "pull_request:43996", + "right": "pull_request:44007", + "accept": false, + "reason": "Cvt/FNet and RegNet/ResNet-family are unrelated model implementations." + }, + { + "left": "pull_request:43996", + "right": "pull_request:44129", + "accept": false, + "reason": "Cvt/FNet and SpeechT5 are unrelated code paths." + }, + { + "left": "pull_request:44044", + "right": "pull_request:44129", + "accept": false, + "reason": "DeBERTa and SpeechT5 are unrelated models." + }, + { + "left": "pull_request:44044", + "right": "pull_request:44154", + "accept": false, + "reason": "DeBERTa and VITS are different model families." + }, + { + "left": "pull_request:44024", + "right": "pull_request:44076", + "accept": false, + "reason": "FocalNet and ImageGPT are unrelated model code paths." + }, + { + "left": "pull_request:44018", + "right": "pull_request:44161", + "accept": false, + "reason": "GPT-Neo and LongT5 are unrelated implementations." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is well grounded in the packet and stays conservative: it distinguishes the broad refactor wave from the smaller duplicate subgroups, and the accepted soft-edge pairs are supported by shared files/code paths or clear overlap in the same model family. The choice of PR 44722 as the best representative is also justified by its later, broader GPT-J+CodeGen scope and explicit issue linkage." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 13, + "component_count": 38, + "review_unit_index": 2, + "review_unit_count": 2, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-43996-28", + "nodes": [ + "pull_request:43997", + "pull_request:44010", + "pull_request:44056", + "pull_request:44073", + "pull_request:44074", + "pull_request:44129" + ], + "soft_pairs": [ + "pull_request:44073|pull_request:44074", + "pull_request:44056|pull_request:44073", + "pull_request:44056|pull_request:44074", + "pull_request:43997|pull_request:44010", + "pull_request:44010|pull_request:44129" + ], + "prepared_review_unit_hash": "1e3a85c8c78cea756f8f59f5b69370761eb47e4f6e37f76851b63cd583734e3e", + "component_budget": { + "node_count": 28, + "item_count": 28, + "soft_pair_count": 77, + "serialized_chars": 84511, + "estimated_input_tokens": 21128, + "estimated_eval_tokens": 42512 + }, + "budget": { + "node_count": 6, + "item_count": 6, + "soft_pair_count": 5, + "serialized_chars": 13068, + "estimated_input_tokens": 3267, + "estimated_eval_tokens": 6790 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "All six PRs are related by the same general refactor pattern: migrating model-specific output tracing to standardized capture/decorator utilities. However, they touch different model implementations and concrete code paths, so they are not duplicates or mergeable into a single PR.", + "confidence": 0.94, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #44129 is the best representative of the cluster because it has the clearest concrete scope, an explicit issue target, and shows a substantive standardized output-capture refactor in one model implementation.", + "best_issue_reason": null, + "best_pr_reason": "PR #44129 is the strongest standalone representative: it is directly tied to an issue, has a clear refactor goal, and is more fully specified than the others. That said, it is still not a duplicate of the other PRs.", + "soft_edge_verdicts": [ + { + "left": "pull_request:44073", + "right": "pull_request:44074", + "accept": false, + "reason": "Both are output-tracing refactors, but they modify different model files and different code paths (VisualBert vs TextNet). Same theme, not the same underlying change." + }, + { + "left": "pull_request:44056", + "right": "pull_request:44073", + "accept": false, + "reason": "These are separate model refactors in different implementations (MPNet vs VisualBert). The similarity is architectural pattern, not a shared bug or change." + }, + { + "left": "pull_request:44056", + "right": "pull_request:44074", + "accept": false, + "reason": "Different models and different forward/output-capture logic. They are related refactors, but not duplicates and not safely mergeable into one PR." + }, + { + "left": "pull_request:43997", + "right": "pull_request:44010", + "accept": false, + "reason": "RegNet and SqueezeBert each migrate their own output handling; same modernization theme, but no shared code path or concrete duplicate fix." + }, + { + "left": "pull_request:44010", + "right": "pull_request:44129", + "accept": false, + "reason": "SqueezeBert and SpeechT5 are unrelated model implementations. The overlap is only in standardized output tracing, not in the same underlying change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 14, + "component_count": 38, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-43998-13", + "nodes": [ + "pull_request:43998", + "pull_request:43999", + "pull_request:44000", + "pull_request:44001", + "pull_request:44002", + "pull_request:44003", + "pull_request:44004", + "pull_request:44025", + "pull_request:44026", + "pull_request:44027", + "pull_request:44028", + "pull_request:44029", + "pull_request:44030" + ], + "soft_pairs": [ + "pull_request:43998|pull_request:44002", + "pull_request:43998|pull_request:44000", + "pull_request:43998|pull_request:44001", + "pull_request:43998|pull_request:44003", + "pull_request:43998|pull_request:44029", + "pull_request:43999|pull_request:44001", + "pull_request:43999|pull_request:44002", + "pull_request:43999|pull_request:44025", + "pull_request:43999|pull_request:44026", + "pull_request:43999|pull_request:44027", + "pull_request:44000|pull_request:44001", + "pull_request:44000|pull_request:44002", + "pull_request:44000|pull_request:44003", + "pull_request:44000|pull_request:44025", + "pull_request:44001|pull_request:44002", + "pull_request:44001|pull_request:44003", + "pull_request:44001|pull_request:44025", + "pull_request:44002|pull_request:44029", + "pull_request:44002|pull_request:44003", + "pull_request:44003|pull_request:44028", + "pull_request:44003|pull_request:44004", + "pull_request:43998|pull_request:44004", + "pull_request:44001|pull_request:44004", + "pull_request:44002|pull_request:44004", + "pull_request:44000|pull_request:44004", + "pull_request:44025|pull_request:44026", + "pull_request:44025|pull_request:44027", + "pull_request:44025|pull_request:44028", + "pull_request:44025|pull_request:44029", + "pull_request:44025|pull_request:44030", + "pull_request:44026|pull_request:44029", + "pull_request:44026|pull_request:44027", + "pull_request:44026|pull_request:44028", + "pull_request:44026|pull_request:44030", + "pull_request:44027|pull_request:44028", + "pull_request:44027|pull_request:44029", + "pull_request:44027|pull_request:44030", + "pull_request:44028|pull_request:44029", + "pull_request:44028|pull_request:44030", + "pull_request:44029|pull_request:44030" + ], + "prepared_review_unit_hash": "d51c0716fe10b3e4d8639b6423034c198f36517369eff1236a2a965285787597", + "component_budget": { + "node_count": 13, + "item_count": 13, + "soft_pair_count": 40, + "serialized_chars": 40170, + "estimated_input_tokens": 10043, + "estimated_eval_tokens": 20342 + }, + "budget": { + "node_count": 13, + "item_count": 13, + "soft_pair_count": 40, + "serialized_chars": 40170, + "estimated_input_tokens": 10043, + "estimated_eval_tokens": 20342 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These PRs share a generic \"output tracing\" refactor theme, but they touch different model implementations and code paths, so they are related cleanups rather than duplicates.", + "confidence": 0.95, + "canonical_issue_reason": null, + "canonical_pr_reason": "No canonical PR: each change is scoped to a different model family, with distinct forward signatures and implementation details.", + "best_issue_reason": null, + "best_pr_reason": "44025 is the broadest exemplar of the recurring tracing refactor pattern, but it still does not subsume the others or serve as a true duplicate target.", + "soft_edge_verdicts": [ + { + "left": "pull_request:43998", + "right": "pull_request:44002", + "accept": false, + "reason": "Different model families (`timm_backbone` vs `upernet`); same refactor theme, different code paths." + }, + { + "left": "pull_request:43998", + "right": "pull_request:44000", + "accept": false, + "reason": "`timm_backbone` and `vision_text_dual_encoder` are unrelated implementations despite similar tracing cleanup." + }, + { + "left": "pull_request:43998", + "right": "pull_request:44001", + "accept": false, + "reason": "Image backbone refactor vs audio generator refactor; not the same bug/change." + }, + { + "left": "pull_request:43998", + "right": "pull_request:44003", + "accept": false, + "reason": "`timm_backbone` cleanup is separate from the `mamba`/`falcon_mamba` tracing changes." + }, + { + "left": "pull_request:43998", + "right": "pull_request:44029", + "accept": false, + "reason": "Different architectures (`timm_backbone` vs `rwkv`); only the helper pattern is shared." + }, + { + "left": "pull_request:43999", + "right": "pull_request:44001", + "accept": false, + "reason": "MobileNetV1 and UnivNet are different domains and forward paths; not duplicates." + }, + { + "left": "pull_request:43999", + "right": "pull_request:44002", + "accept": false, + "reason": "CNN tracing refactor vs segmentation-model tracing refactor; distinct model code." + }, + { + "left": "pull_request:43999", + "right": "pull_request:44025", + "accept": false, + "reason": "MobileNetV1 and DepthAnything are separate models; shared helper usage is too broad to merge." + }, + { + "left": "pull_request:43999", + "right": "pull_request:44026", + "accept": false, + "reason": "MobileNetV1 refactor is unrelated to the vision-encoder/decoder wrapper change." + }, + { + "left": "pull_request:43999", + "right": "pull_request:44027", + "accept": false, + "reason": "Different model stacks (`mobilenet_v1` vs `speech_encoder_decoder`); not the same underlying change." + }, + { + "left": "pull_request:44000", + "right": "pull_request:44001", + "accept": false, + "reason": "Multimodal dual encoder vs UnivNet audio model; same pattern, different implementation." + }, + { + "left": "pull_request:44000", + "right": "pull_request:44002", + "accept": false, + "reason": "These touch different model families and forward contracts, so they are not duplicates." + }, + { + "left": "pull_request:44000", + "right": "pull_request:44003", + "accept": false, + "reason": "`vision_text_dual_encoder` is unrelated to the `mamba` refactor beyond shared helper cleanup." + }, + { + "left": "pull_request:44000", + "right": "pull_request:44025", + "accept": false, + "reason": "Different vision model families; no shared concrete bug or change." + }, + { + "left": "pull_request:44001", + "right": "pull_request:44002", + "accept": false, + "reason": "UnivNet and UperNet are unrelated architectures; only the refactor style matches." + }, + { + "left": "pull_request:44001", + "right": "pull_request:44003", + "accept": false, + "reason": "Audio generation model vs Mamba variant; not the same code path." + }, + { + "left": "pull_request:44001", + "right": "pull_request:44025", + "accept": false, + "reason": "DepthAnything and UnivNet are separate domains; no duplicate evidence." + }, + { + "left": "pull_request:44002", + "right": "pull_request:44029", + "accept": false, + "reason": "Segmentation model refactor vs RWKV refactor; different subsystems." + }, + { + "left": "pull_request:44002", + "right": "pull_request:44003", + "accept": false, + "reason": "UperNet cleanup does not overlap with the Mamba/Falcon-Mamba change set." + }, + { + "left": "pull_request:44003", + "right": "pull_request:44028", + "accept": false, + "reason": "Falcon/Mamba stateful model refactor and SuperPoint refactor are unrelated." + }, + { + "left": "pull_request:44003", + "right": "pull_request:44004", + "accept": false, + "reason": "Different model code paths (`mamba` vs `codegen`); not one concrete fix." + }, + { + "left": "pull_request:43998", + "right": "pull_request:44004", + "accept": false, + "reason": "Timm backbone cleanup and CodeGen cache/signature refactor are distinct changes." + }, + { + "left": "pull_request:44001", + "right": "pull_request:44004", + "accept": false, + "reason": "UnivNet audio output tracing and CodeGen forward signature changes are not the same issue." + }, + { + "left": "pull_request:44002", + "right": "pull_request:44004", + "accept": false, + "reason": "UperNet segmentation refactor vs CodeGen refactor; different code paths." + }, + { + "left": "pull_request:44000", + "right": "pull_request:44004", + "accept": false, + "reason": "Multimodal encoder/decoder refactor does not match CodeGen's cache/attention API cleanup." + }, + { + "left": "pull_request:44025", + "right": "pull_request:44026", + "accept": false, + "reason": "DepthAnything and VisionEncoderDecoder are different model families; same helper pattern only." + }, + { + "left": "pull_request:44025", + "right": "pull_request:44027", + "accept": false, + "reason": "DepthAnything refactor is unrelated to SpeechEncoderDecoder beyond generic tracing cleanup." + }, + { + "left": "pull_request:44025", + "right": "pull_request:44028", + "accept": false, + "reason": "DepthAnything and SuperPoint are separate vision models with different forward logic." + }, + { + "left": "pull_request:44025", + "right": "pull_request:44029", + "accept": false, + "reason": "DepthAnything and RWKV touch unrelated architectures and code paths." + }, + { + "left": "pull_request:44025", + "right": "pull_request:44030", + "accept": false, + "reason": "DepthAnything refactor is not the same as DPR encoder wrapper cleanup." + }, + { + "left": "pull_request:44026", + "right": "pull_request:44029", + "accept": false, + "reason": "VisionEncoderDecoder wrapper and RWKV model changes are unrelated." + }, + { + "left": "pull_request:44026", + "right": "pull_request:44027", + "accept": false, + "reason": "VisionEncoderDecoder vs SpeechEncoderDecoder are different wrapper implementations; not duplicates." + }, + { + "left": "pull_request:44026", + "right": "pull_request:44028", + "accept": false, + "reason": "Different model families (`vision_encoder_decoder` vs `superpoint`)." + }, + { + "left": "pull_request:44026", + "right": "pull_request:44030", + "accept": false, + "reason": "Encoder-decoder wrapper cleanup and DPR refactor are separate model changes." + }, + { + "left": "pull_request:44027", + "right": "pull_request:44028", + "accept": false, + "reason": "Speech encoder-decoder wrapper and SuperPoint model are unrelated." + }, + { + "left": "pull_request:44027", + "right": "pull_request:44029", + "accept": false, + "reason": "SpeechEncoderDecoder and RWKV differ in architecture and forward path." + }, + { + "left": "pull_request:44027", + "right": "pull_request:44030", + "accept": false, + "reason": "SpeechEncoderDecoder refactor does not match DPR's wrapper change." + }, + { + "left": "pull_request:44028", + "right": "pull_request:44029", + "accept": false, + "reason": "SuperPoint and RWKV are unrelated codebases/architectures." + }, + { + "left": "pull_request:44028", + "right": "pull_request:44030", + "accept": false, + "reason": "SuperPoint vision model refactor is separate from DPR's encoder wrapper cleanup." + }, + { + "left": "pull_request:44029", + "right": "pull_request:44030", + "accept": false, + "reason": "RWKV and DPR changes touch different model families and are not mergeable as one fix." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 15, + "component_count": 38, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-44662-2", + "nodes": [ + "pull_request:44662", + "pull_request:45401" + ], + "soft_pairs": [ + "pull_request:44662|pull_request:45401" + ], + "prepared_review_unit_hash": "b0ca5f39923acbe07b5f023b2800e6d34f3c4362a359c2264496bc737c73dbdf", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 6551, + "estimated_input_tokens": 1638, + "estimated_eval_tokens": 3532 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 6551, + "estimated_input_tokens": 1638, + "estimated_eval_tokens": 3532 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are two separate model-addition PRs (PenguinVL vs Voxtral TTS). They share framework registration/docs boilerplate, but they do not fix the same underlying code-path problem.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single canonical PR: 44662 introduces PenguinVL, while 45401 introduces Voxtral TTS; they target different models and independent implementation paths.", + "best_issue_reason": null, + "best_pr_reason": "45401 is the cleaner anchor for its own change because it is draft, explicitly tied to issue #45027, and has a narrower, well-scoped addition.", + "soft_edge_verdicts": [ + { + "left": "pull_request:44662", + "right": "pull_request:45401", + "accept": false, + "reason": "Different model integrations with distinct configs, processors, and model code; the shared auto-registry/docs edits are generic boilerplate, not evidence of the same bug or change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 16, + "component_count": 38, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-44965-2", + "nodes": [ + "pull_request:44965", + "pull_request:46845" + ], + "soft_pairs": [ + "pull_request:44965|pull_request:46845" + ], + "prepared_review_unit_hash": "3587ac714a751f2d2319d48f0d70c536540f6ec8799d6e46c6e8769c114b8abe", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4219, + "estimated_input_tokens": 1055, + "estimated_eval_tokens": 2366 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4219, + "estimated_input_tokens": 1055, + "estimated_eval_tokens": 2366 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "The two PRs are not duplicates: one changes the PR documentation workflow into a draft-conversion action, while the other adjusts CI branch routing and touches unrelated model code/test plumbing. They do not appear to fix the same underlying problem or implement the same change.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #46845 is the better representative because it has a clearer, concrete purpose (CI branch/routing adjustment) and more substantive context than the placeholder-like PR #44965 titled \"try\".", + "best_issue_reason": null, + "best_pr_reason": "PR #46845 is the strongest standalone PR in this cluster: its intent is more explicit, and the changes are at least internally coherent compared with #44965, which looks exploratory and incomplete.", + "soft_edge_verdicts": [ + { + "left": "pull_request:44965", + "right": "pull_request:46845", + "accept": false, + "reason": "Reject: these PRs target different changes. #44965 rewrites the PR documentation workflow to convert PRs to draft, while #46845 changes CI workflow branch handling and includes an unrelated edit in modeling_utils.py. They do not share the same concrete bug or code path." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 17, + "component_count": 38, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-44979-2", + "nodes": [ + "pull_request:44979", + "pull_request:45363" + ], + "soft_pairs": [ + "pull_request:44979|pull_request:45363" + ], + "prepared_review_unit_hash": "ed94334000f7a11579355824b8ab79d82f0942e062ba91a5ac4d0144091426bf", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4408, + "estimated_input_tokens": 1102, + "estimated_eval_tokens": 2460 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4408, + "estimated_input_tokens": 1102, + "estimated_eval_tokens": 2460 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "The two pull requests are related to module/kernel fusion, but they are not the same underlying change: one introduces a generic Module Fusion API, while the other wires up n-to-1 kernel fusion through KernelConfig and related integration code. They could coexist, not be merged as duplicates.", + "confidence": 0.91, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 44979 is the cleaner canonical representative because it introduces the foundational Module Fusion API itself, which the other PR appears to build on rather than duplicate.", + "best_issue_reason": null, + "best_pr_reason": "PR 44979 is the best cluster representative since it defines the base abstraction; PR 45363 is a narrower follow-on integration/use-case change with example and wiring code.", + "soft_edge_verdicts": [ + { + "left": "pull_request:44979", + "right": "pull_request:45363", + "accept": false, + "reason": "Different scope and code path: 44979 adds a new generic module-fusion API, while 45363 implements n-to-1 kernel fusion via KernelConfig plus integration/example changes. They are related, but not the same fix/change and would not plausibly be merged into one PR as duplicates." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 18, + "component_count": 38, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-45086-2", + "nodes": [ + "pull_request:45086", + "pull_request:45317" + ], + "soft_pairs": [ + "pull_request:45086|pull_request:45317" + ], + "prepared_review_unit_hash": "98fe49e488b0466dc06a0945d1c8819436eb71956388bc98e6474e190b0839f8", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4414, + "estimated_input_tokens": 1104, + "estimated_eval_tokens": 2464 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4414, + "estimated_input_tokens": 1104, + "estimated_eval_tokens": 2464 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both pull requests fix the same Mistral regex AttributeError by switching from `backend_tokenizer.pre_tokenizer` to `tokenizer.pre_tokenizer` in the same code path; 45317 is the more complete change because it adds tests.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 45317 is the clearest canonical PR: same core fix as 45086, but with added test coverage and a more explicit title about the failing `fix_mistral_regex=True` path.", + "best_issue_reason": null, + "best_pr_reason": "PR 45317 is the best representative because it addresses the same bug while also including regression tests, making it the more complete standalone fix.", + "soft_edge_verdicts": [ + { + "left": "pull_request:45086", + "right": "pull_request:45317", + "accept": true, + "reason": "Same underlying AttributeError in `_patch_mistral_regex`, same file and same line-level fix; 45317 appears to be an expanded version with tests rather than a different change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 19, + "component_count": 38, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-45115-2", + "nodes": [ + "pull_request:45115", + "pull_request:45569" + ], + "soft_pairs": [ + "pull_request:45115|pull_request:45569" + ], + "prepared_review_unit_hash": "095a59f3d46d1f32a1e8aeaf5eb7339d1b7c38271459b11a07337fdeb045dee5", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 6066, + "estimated_input_tokens": 1517, + "estimated_eval_tokens": 3290 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 6066, + "estimated_input_tokens": 1517, + "estimated_eval_tokens": 3290 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These PRs are related to Nemotron H, but they address different scopes: #45115 is a conversion-mapping/refactor PR, while #45569 is a broader model split and docs/tests update. They do not look like the same underlying change.", + "confidence": 0.89, + "canonical_issue_reason": null, + "canonical_pr_reason": "pull_request:45569 is the more canonical PR for the actual model-structure change: it introduces NemotronH dense/sparse model files, updates auto/model registration, adds docs, and includes tests. It is the more complete and user-facing implementation.", + "best_issue_reason": null, + "best_pr_reason": "pull_request:45569 is the strongest standalone PR because it covers the full architectural split and integration points. pull_request:45115 is narrower and mostly updates checkpoint conversion mappings for renamed weights.", + "soft_edge_verdicts": [ + { + "left": "pull_request:45115", + "right": "pull_request:45569", + "accept": false, + "reason": "Reject: they share NemotronH files, but one is about conversion key remapping while the other is about introducing dense/sparse model variants plus docs/tests. Different concrete code-path changes, not mergeable as the same fix." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 20, + "component_count": 38, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-45332-2", + "nodes": [ + "pull_request:45332", + "pull_request:45333" + ], + "soft_pairs": [ + "pull_request:45332|pull_request:45333" + ], + "prepared_review_unit_hash": "abcef5461863d14fc1b1e2c7323e994b928b9985e1455696270d622cc97c0d1b", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5953, + "estimated_input_tokens": 1489, + "estimated_eval_tokens": 3234 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5953, + "estimated_input_tokens": 1489, + "estimated_eval_tokens": 3234 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are related but not duplicates: PR 45333 adds per-layer heterogeneous configuration support, while PR 45332 includes that plus broader heterogeneous modeling/runtime changes. They share setup files and some configuration code, but they target different scopes of the feature.", + "confidence": 0.91, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 45332 is the broader, more complete heterogeneous-model feature PR; it includes the config work plus modeling/runtime changes across cache, masking, modeling utils, and model support.", + "best_issue_reason": null, + "best_pr_reason": "PR 45332 is the best representative of the cluster because it covers the full end-to-end heterogeneous support rather than only the configuration subset.", + "soft_edge_verdicts": [ + { + "left": "pull_request:45332", + "right": "pull_request:45333", + "accept": false, + "reason": "They overlap on heterogeneous configuration plumbing, but 45332 also changes modeling/runtime paths and is not just a variant of 45333. The PRs are complementary scope, not the same concrete change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 21, + "component_count": 38, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-45415-2", + "nodes": [ + "pull_request:45415", + "pull_request:45425" + ], + "soft_pairs": [ + "pull_request:45415|pull_request:45425" + ], + "prepared_review_unit_hash": "452aad0baac796558941499008fa92e596b18bfc5b75b7b73061c85e2d781921", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5513, + "estimated_input_tokens": 1379, + "estimated_eval_tokens": 3014 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5513, + "estimated_input_tokens": 1379, + "estimated_eval_tokens": 3014 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are two related typing-focused PRs, but they are not the same underlying change. PR 45415 is a broad repository-wide type-checking pass across many Transformers modules, while PR 45425 adds specific typing helpers/protocols for a narrower modeling_utils/PEFT-related path.", + "confidence": 0.91, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 45415 is the broader umbrella change: it adjusts type-checking config and applies annotations across many files, making it the better canonical representative of the cluster.", + "best_issue_reason": null, + "best_pr_reason": "PR 45415 is the best overall representative because it is the most comprehensive and central typing cleanup in the set; PR 45425 is narrower and more specialized.", + "soft_edge_verdicts": [ + { + "left": "pull_request:45415", + "right": "pull_request:45425", + "accept": false, + "reason": "Reject: both are typing-related, but 45415 is a broad annotation/type-checking sweep while 45425 adds specific protocol/alias definitions for modeling_utils and PEFT. They overlap in a few files, yet they do not appear to fix the same concrete code-path problem or be mergeable as one single PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 22, + "component_count": 38, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-45548-2", + "nodes": [ + "pull_request:45548", + "pull_request:45662" + ], + "soft_pairs": [ + "pull_request:45548|pull_request:45662" + ], + "prepared_review_unit_hash": "e7be306aa6b9347fc0ba5edd7e136b7e1b13f2bb8cd93c2897f39022ec25ef31", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4508, + "estimated_input_tokens": 1127, + "estimated_eval_tokens": 2510 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4508, + "estimated_input_tokens": 1127, + "estimated_eval_tokens": 2510 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are related EP-loading PRs, but they fix different bugs: one preserves distributed config and exposes `has_ep` for Accelerate/ZeRO-3 loading, while the other fixes expert parameters being overwritten in FSDP2/sharded loading. Not duplicates.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #45662 is the more concrete representative of the cluster: it targets a specific expert-parameter overwrite path during distributed loading, with broader code-path changes across loading and MoE utilities.", + "best_issue_reason": null, + "best_pr_reason": "PR #45662 is the strongest standalone PR for this cluster because it addresses a precise runtime failure mode in the loading pipeline and has the clearest implementation impact.", + "soft_edge_verdicts": [ + { + "left": "pull_request:45548", + "right": "pull_request:45662", + "accept": false, + "reason": "Different underlying bugs and fix strategies: #45548 serializes `distributed_config` and adds `has_ep` to make EP/ZeRO-3 loading work; #45662 wraps loaded params post-shard to prevent expert weights being overwritten under FSDP2. Shared EP context and one filename are not enough to treat them as the same change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 23, + "component_count": 38, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-45639-2", + "nodes": [ + "pull_request:45639", + "pull_request:45645" + ], + "soft_pairs": [ + "pull_request:45639|pull_request:45645" + ], + "prepared_review_unit_hash": "ee90f8b30f4b0981c528f8c3190c6cef160e39511180076272234823f6162575", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4671, + "estimated_input_tokens": 1168, + "estimated_eval_tokens": 2592 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4671, + "estimated_input_tokens": 1168, + "estimated_eval_tokens": 2592 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both PRs address the same xdist-safe captured_info logging problem in transformers testing utilities; one is the narrower code fix, the other is the broader end-to-end version that also updates CI log collection.", + "confidence": 0.92, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 45645 is the most complete representative: it fixes the xdist collision in `testing_utils.py` and also updates CI to read all `captured_info*.txt` artifacts, matching the full user-facing impact of the bug.", + "best_issue_reason": null, + "best_pr_reason": "PR 45639 is the cleanest code-path fix and the best narrow representative if you want the smallest change that directly makes patched testing debug logs xdist-safe.", + "soft_edge_verdicts": [ + { + "left": "pull_request:45639", + "right": "pull_request:45645", + "accept": true, + "reason": "Same underlying bug and same core implementation area (`testing_utils.py`); the second PR expands the fix to CI artifact handling, so these could plausibly be merged into one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 24, + "component_count": 38, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-45720-2", + "nodes": [ + "pull_request:45720", + "pull_request:46259" + ], + "soft_pairs": [ + "pull_request:45720|pull_request:46259" + ], + "prepared_review_unit_hash": "e4c22c8aa34b045e0b82e68fc7104be2185bb3e4638de279e8fd6797995bffc0", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 7233, + "estimated_input_tokens": 1809, + "estimated_eval_tokens": 3874 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 7233, + "estimated_input_tokens": 1809, + "estimated_eval_tokens": 3874 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These two PRs are both linter-rule/modeling cleanup changes, but they target different rules and different underlying code paths: PR 45720 adds TRF003 tuple-return handling across several model forwards, while PR 46259 adds TRF018 initialization-rule changes across many model initializers. They are similar in theme but not duplicates.", + "confidence": 0.95, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single canonical PR as a duplicate representative: the changes implement different lint rules (TRF003 vs TRF018) and would not plausibly merge into one patch without conflating unrelated fixes.", + "best_issue_reason": null, + "best_pr_reason": "pull_request:46259 is the broader/more representative linter-rule PR in this cluster, with wider model coverage and a more systematic refactor pattern, though it is not a duplicate of 45720.", + "soft_edge_verdicts": [ + { + "left": "pull_request:45720", + "right": "pull_request:46259", + "accept": false, + "reason": "Different lint rules and different bug/change classes: TRF003 tuple-return handling in forward methods vs TRF018 weight-initialization behavior. Shared files are incidental to the same modeling subsystem, not the same underlying fix." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 25, + "component_count": 38, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-45737-2", + "nodes": [ + "pull_request:45737", + "pull_request:45738" + ], + "soft_pairs": [ + "pull_request:45737|pull_request:45738" + ], + "prepared_review_unit_hash": "cc2ade73a49ec6a8883a7a24c433988645588bcd049e8aad2bce56b1b24daa1c", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4131, + "estimated_input_tokens": 1033, + "estimated_eval_tokens": 2322 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4131, + "estimated_input_tokens": 1033, + "estimated_eval_tokens": 2322 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both PRs address the same MusicgenMelody cache initialization bug in `forward()`: replacing `EncoderDecoderCache(DynamicCache, DynamicCache)` with a single `DynamicCache`. PR #45737 is the more complete duplicate because it bundles the fix with a regression test, while #45738 is the fix-only variant.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #45737 is the best canonical PR because it contains the same code fix as #45738 plus a regression test, making it the most complete representation of the bug fix.", + "best_issue_reason": null, + "best_pr_reason": "PR #45737 is the strongest overall candidate: it fixes the concrete cache bug and adds a targeted regression test, whereas #45738 only changes the implementation.", + "soft_edge_verdicts": [ + { + "left": "pull_request:45737", + "right": "pull_request:45738", + "accept": true, + "reason": "Same underlying code-path bug in `musicgen_melody` cache setup; the diffs match on the core fix and the test-only addition in #45737 could plausibly live in the same PR as #45738." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The assessment is grounded in the packet: both PRs make the same `musicgen_melody` cache initialization change, and #45737 only adds a regression test on top of that fix. The soft-edge accept is conservative enough because the diffs are mergeable into one PR for the same bugfix." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 26, + "component_count": 38, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-45932-3", + "nodes": [ + "pull_request:45932", + "pull_request:46100", + "pull_request:46891" + ], + "soft_pairs": [ + "pull_request:45932|pull_request:46100", + "pull_request:45932|pull_request:46891", + "pull_request:46100|pull_request:46891" + ], + "prepared_review_unit_hash": "97193e916be9f9fed300720cbded4e9d6a528a29418d79c62fd8b35845b55d3d", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 7880, + "estimated_input_tokens": 1970, + "estimated_eval_tokens": 4196 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 7880, + "estimated_input_tokens": 1970, + "estimated_eval_tokens": 4196 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "All three items are open PRs in the same general testing/CI orbit, but they target different code paths and changes: one is a broad temporary model/CI test PR, one is a test migration with new distributed kernel tests, and one is a PEFT CI integration PR. They are not duplicates of the same underlying fix.", + "confidence": 0.91, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single underlying bug/change ties these PRs together. If one representative must be chosen, #46891 is the most self-contained and focused PR, centered on PEFT CI wiring, whereas #45932 and #46100 are broader and touch unrelated test/model areas.", + "best_issue_reason": null, + "best_pr_reason": "#46891 is the best cluster representative because it is the narrowest and most coherent change set. The others are also test/CI-related, but they modify different subsystems and solve different problems.", + "soft_edge_verdicts": [ + { + "left": "pull_request:45932", + "right": "pull_request:46100", + "accept": false, + "reason": "Both are test-oriented, but #45932 is a broad \"testing grafana\" PR touching model code and CI wiring, while #46100 is a test migration adding kernel/distributed test files. They do not appear to fix the same concrete code-path problem and would not plausibly merge as one PR." + }, + { + "left": "pull_request:45932", + "right": "pull_request:46891", + "accept": false, + "reason": "#45932 is a broad temporary testing/CI PR with unrelated model changes; #46891 is specifically about PEFT integration CI and a small peft integration code path. Shared workflow edits are not enough to make them the same change." + }, + { + "left": "pull_request:46100", + "right": "pull_request:46891", + "accept": false, + "reason": "#46100 adds kernel tests and test-migration files, while #46891 changes PEFT integration CI behavior and a PEFT integration helper. They touch different subsystems and address different test/CI concerns." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 27, + "component_count": 38, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-46126-2", + "nodes": [ + "pull_request:46126", + "pull_request:46269" + ], + "soft_pairs": [ + "pull_request:46126|pull_request:46269" + ], + "prepared_review_unit_hash": "93423926f50836b67b67b53d07eb5027ac34f9154d777be0eee97738f07ec042", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5525, + "estimated_input_tokens": 1382, + "estimated_eval_tokens": 3020 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5525, + "estimated_input_tokens": 1382, + "estimated_eval_tokens": 3020 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are not duplicates: PR #46126 is a focused EP/trainer integration change around router/tensor-parallel code, while PR #46269 is a broad distributed-branch base with many unrelated distributed and CI edits. The overlapping filenames look like branch overlap, not the same underlying fix.", + "confidence": 0.91, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #46126 is the better canonical PR because it has a single, concrete scope centered on EP routing and trainer integration; PR #46269 is a wide base branch touching many subsystems and is not a single actionable change.", + "best_issue_reason": null, + "best_pr_reason": "PR #46126 is the best representative of the cluster because it is the most specific and technically coherent change, whereas #46269 is too broad and heterogeneous to serve as a clean duplicate target.", + "soft_edge_verdicts": [ + { + "left": "pull_request:46126", + "right": "pull_request:46269", + "accept": false, + "reason": "Shared distributed files are not enough to treat these as the same change. #46126 targets EP/router and trainer integration; #46269 is a large branch-base PR spanning CI, FSDP, MoE, loading, and other distributed components, so they would not plausibly merge as one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 28, + "component_count": 38, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-46138-10", + "nodes": [ + "pull_request:46138", + "pull_request:46432", + "pull_request:46501", + "pull_request:46502", + "pull_request:46503", + "pull_request:46504", + "pull_request:46554", + "pull_request:46558", + "pull_request:46563", + "pull_request:46716" + ], + "soft_pairs": [ + "pull_request:46138|pull_request:46554", + "pull_request:46138|pull_request:46558", + "pull_request:46138|pull_request:46563", + "pull_request:46138|pull_request:46716", + "pull_request:46138|pull_request:46504", + "pull_request:46432|pull_request:46503", + "pull_request:46432|pull_request:46501", + "pull_request:46432|pull_request:46502", + "pull_request:46432|pull_request:46554", + "pull_request:46432|pull_request:46558", + "pull_request:46501|pull_request:46503", + "pull_request:46501|pull_request:46502", + "pull_request:46501|pull_request:46504", + "pull_request:46501|pull_request:46554", + "pull_request:46502|pull_request:46503", + "pull_request:46502|pull_request:46504", + "pull_request:46502|pull_request:46554", + "pull_request:46503|pull_request:46504", + "pull_request:46503|pull_request:46554", + "pull_request:46504|pull_request:46554", + "pull_request:46504|pull_request:46558", + "pull_request:46554|pull_request:46558", + "pull_request:46554|pull_request:46563", + "pull_request:46554|pull_request:46716", + "pull_request:46558|pull_request:46563", + "pull_request:46558|pull_request:46716", + "pull_request:46563|pull_request:46716", + "pull_request:46504|pull_request:46563", + "pull_request:46504|pull_request:46716" + ], + "prepared_review_unit_hash": "7ac72a0ea3921b83f4a6e1096cfa6cd1225313f9074e36ae49fcb35c9439dbfc", + "component_budget": { + "node_count": 10, + "item_count": 10, + "soft_pair_count": 29, + "serialized_chars": 25919, + "estimated_input_tokens": 6480, + "estimated_eval_tokens": 13216 + }, + "budget": { + "node_count": 10, + "item_count": 10, + "soft_pair_count": 29, + "serialized_chars": 25919, + "estimated_input_tokens": 6480, + "estimated_eval_tokens": 13216 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Most items are independent workflow-hygiene PRs touching different files and code paths. The only tight cluster is pr-ci-security-gate.yml, but those PRs make different hardening edits rather than the same concrete fix, so none of the soft edges should merge.", + "confidence": 0.94, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 46554 is the strongest representative because it makes the core security-gate logic change; the other nearby PRs are smaller, complementary edits (permissions, comments, secrets handling) rather than the same fix.", + "best_issue_reason": null, + "best_pr_reason": "PR 46554 is the best global representative: it changes the substantive execution path in pr-ci-security-gate.yml, while the others are mostly ancillary workflow tweaks or unrelated maintenance changes.", + "soft_edge_verdicts": [ + { + "left": "pull_request:46138", + "right": "pull_request:46554", + "accept": false, + "reason": "Different workflows and different behavior changes; only a broad CI/workflow similarity, not the same bug or fix." + }, + { + "left": "pull_request:46138", + "right": "pull_request:46558", + "accept": false, + "reason": "Unrelated workflow edits: stdin handling in self-comment CI versus security-gate permissions/approval logic." + }, + { + "left": "pull_request:46138", + "right": "pull_request:46563", + "accept": false, + "reason": "Different file and different concern; one is command input handling, the other is a reusable-workflow/security note change." + }, + { + "left": "pull_request:46138", + "right": "pull_request:46716", + "accept": false, + "reason": "Different workflow and different code path; one changes CI message passing, the other changes PR security-gate secrets handling." + }, + { + "left": "pull_request:46138", + "right": "pull_request:46504", + "accept": false, + "reason": "Both are shell-safety workflow tweaks, but they fix different scripts in different workflows and are not the same underlying issue." + }, + { + "left": "pull_request:46432", + "right": "pull_request:46503", + "accept": false, + "reason": "Different workflow files and different trigger logic; no shared concrete bug." + }, + { + "left": "pull_request:46432", + "right": "pull_request:46501", + "accept": false, + "reason": "Different workflow maintenance changes in different jobs; similar namespace only." + }, + { + "left": "pull_request:46432", + "right": "pull_request:46502", + "accept": false, + "reason": "Unrelated workflow edits: SSH runner hardening versus reviewer assignment checkout behavior." + }, + { + "left": "pull_request:46432", + "right": "pull_request:46554", + "accept": false, + "reason": "Both touch CI/workflow infrastructure, but they address different problems: SSH runner hardening versus PR security-gate logic." + }, + { + "left": "pull_request:46432", + "right": "pull_request:46558", + "accept": false, + "reason": "Different workflow and different fix; security-gate permissions/approval gating is not the SSH runner hardening issue." + }, + { + "left": "pull_request:46501", + "right": "pull_request:46503", + "accept": false, + "reason": "Both are workflow updates, but one is anti-slop permissions and the other is benchmark dispatch support; not the same change." + }, + { + "left": "pull_request:46501", + "right": "pull_request:46502", + "accept": false, + "reason": "Different workflows with unrelated behavior changes." + }, + { + "left": "pull_request:46501", + "right": "pull_request:46504", + "accept": false, + "reason": "Different workflow files and different shell/input handling fixes; only superficial similarity." + }, + { + "left": "pull_request:46501", + "right": "pull_request:46554", + "accept": false, + "reason": "Security-gate logic is unrelated to anti-slop permission scoping." + }, + { + "left": "pull_request:46502", + "right": "pull_request:46503", + "accept": false, + "reason": "Different workflow jobs and different fixes; no shared underlying bug." + }, + { + "left": "pull_request:46502", + "right": "pull_request:46504", + "accept": false, + "reason": "Reviewer checkout ref change is unrelated to Docker-image build command quoting." + }, + { + "left": "pull_request:46502", + "right": "pull_request:46554", + "accept": false, + "reason": "Different workflow infrastructure changes with no common code-path problem." + }, + { + "left": "pull_request:46503", + "right": "pull_request:46504", + "accept": false, + "reason": "Benchmark workflow trigger change versus build-image shell quoting; unrelated." + }, + { + "left": "pull_request:46503", + "right": "pull_request:46554", + "accept": false, + "reason": "Different workflows and different functionality; not a duplicate fix." + }, + { + "left": "pull_request:46504", + "right": "pull_request:46554", + "accept": false, + "reason": "Both are CI/workflow edits, but they fix different scripts and different jobs; not one mergeable change." + }, + { + "left": "pull_request:46504", + "right": "pull_request:46558", + "accept": false, + "reason": "Build-script input handling versus security-gate permissions/approval logic are separate concerns." + }, + { + "left": "pull_request:46554", + "right": "pull_request:46558", + "accept": false, + "reason": "Same file, but different changes: 46554 adds checkout/diff-based security checks, while 46558 adjusts permissions and gates approval on success; complementary, not the same fix." + }, + { + "left": "pull_request:46554", + "right": "pull_request:46563", + "accept": false, + "reason": "46563 is just a reusable-workflow/comment clarification, not the same code-path change as 46554's security checks." + }, + { + "left": "pull_request:46554", + "right": "pull_request:46716", + "accept": false, + "reason": "Secrets scoping and security-check logic are related thematically but are different concrete changes." + }, + { + "left": "pull_request:46558", + "right": "pull_request:46563", + "accept": false, + "reason": "Permissions/approval gating is distinct from the reusable-workflow safety comment update." + }, + { + "left": "pull_request:46558", + "right": "pull_request:46716", + "accept": false, + "reason": "Both are security-gate hardening, but one changes permissions/approval flow and the other changes secrets inheritance; not the same fix." + }, + { + "left": "pull_request:46563", + "right": "pull_request:46716", + "accept": false, + "reason": "Comment-only safety note versus actual secrets scoping change; different kinds of edits." + }, + { + "left": "pull_request:46504", + "right": "pull_request:46563", + "accept": false, + "reason": "Unrelated workflow edits; only superficial similarity in being small CI changes." + }, + { + "left": "pull_request:46504", + "right": "pull_request:46716", + "accept": false, + "reason": "Different workflow and different problem domain; build-image command quoting is unrelated to PR CI secrets scoping." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 29, + "component_count": 38, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-46152-2", + "nodes": [ + "pull_request:46152", + "pull_request:46207" + ], + "soft_pairs": [ + "pull_request:46152|pull_request:46207" + ], + "prepared_review_unit_hash": "f95db07950c690999a7689a2013711ae32fbb02d9e4e34cdf1ea69d2a8d89e19", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4275, + "estimated_input_tokens": 1069, + "estimated_eval_tokens": 2394 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4275, + "estimated_input_tokens": 1069, + "estimated_eval_tokens": 2394 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both PRs modernize source inspection to avoid `open()`, but they target different subsystems and distinct code paths, so they are not duplicates.", + "confidence": 0.94, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single canonical PR: 46152 changes `import_utils.py` for traversable resolution, while 46207 changes `modeling_utils.py` for `_can_set_attn_implementation`; they solve different bugs.", + "best_issue_reason": null, + "best_pr_reason": "46207 is the cleaner representative if one must be chosen because it is a focused, test-backed fix for one concrete code path; 46152 addresses a separate helper in a different module.", + "soft_edge_verdicts": [ + { + "left": "pull_request:46152", + "right": "pull_request:46207", + "accept": false, + "reason": "Same broad theme (avoid `open()`), but different modules, different functions, and different behaviors; they are not the same underlying change and would not be merged into one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 30, + "component_count": 38, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-46174-2", + "nodes": [ + "pull_request:46174", + "pull_request:46780" + ], + "soft_pairs": [ + "pull_request:46174|pull_request:46780" + ], + "prepared_review_unit_hash": "a48ce4ca162f555682f79411d7229ac1ccc08d7822dfab654bb881639fccddbc", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 6030, + "estimated_input_tokens": 1508, + "estimated_eval_tokens": 3272 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 6030, + "estimated_input_tokens": 1508, + "estimated_eval_tokens": 3272 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Two PRs target the same reported MPS crash in the sinusoidal position embedding helper, but they implement different fixes and are not safe to treat as duplicates.", + "confidence": 0.89, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 46174 is the broader, more complete fix for the float64-on-MPS crash: it updates the helper across all affected model files and keeps the internal math on CPU float64 before casting back.", + "best_issue_reason": null, + "best_pr_reason": "PR 46174 is the best representative of the cluster because it is the original, more comprehensive fix and has more discussion/review context than the narrower follow-up.", + "soft_edge_verdicts": [ + { + "left": "pull_request:46174", + "right": "pull_request:46780", + "accept": false, + "reason": "They share the same issue target and helper, but the patches are different solutions: 46174 avoids float64 on MPS by doing the arithmetic on CPU first, while 46780 switches the helper to use dtype for internal arithmetic. Those are alternative implementations, not one mergeable duplicate PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 31, + "component_count": 38, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-46281-2", + "nodes": [ + "pull_request:46281", + "pull_request:46890" + ], + "soft_pairs": [ + "pull_request:46281|pull_request:46890" + ], + "prepared_review_unit_hash": "f27a94c190d9a11207de7366a1a4e4b0859090bc0a8e2e598d3be1470bb0146b", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4270, + "estimated_input_tokens": 1068, + "estimated_eval_tokens": 2392 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4270, + "estimated_input_tokens": 1068, + "estimated_eval_tokens": 2392 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both PRs harden path handling against traversal, but they fix different inputs in different code paths: one guards sharded checkpoint filenames from the index `weight_map`, the other guards `transformers_weights` in model config resolution. They are related security hardening, not duplicates.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": "46281 is the cleaner representative of the cluster theme: it fixes a concrete checkpoint-shard loading path and the containment check is explicit and well-scoped. Still, it is not a duplicate of 46890.", + "best_issue_reason": null, + "best_pr_reason": "46281 is the better standalone PR because the bug is narrowly defined, the affected path is clear, and the fix is self-contained. 46890 addresses a separate config-driven file-resolution path.", + "soft_edge_verdicts": [ + { + "left": "pull_request:46281", + "right": "pull_request:46890", + "accept": false, + "reason": "Different underlying bugs and different code paths: `get_checkpoint_shard_files` uses index `weight_map` entries, while `_get_resolved_checkpoint_files` uses `transformers_weights`. Same vulnerability class, but not the same fix or mergeable into one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 32, + "component_count": 38, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-46310-2", + "nodes": [ + "pull_request:46310", + "pull_request:46546" + ], + "soft_pairs": [ + "pull_request:46310|pull_request:46546" + ], + "prepared_review_unit_hash": "7aaab9e0aeda2c627210bb0d5640ca3a07432b6a9d3922f7580e39a595676187", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4849, + "estimated_input_tokens": 1213, + "estimated_eval_tokens": 2682 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4849, + "estimated_input_tokens": 1213, + "estimated_eval_tokens": 2682 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both PRs fix the same off-by-one bug in Moonshine streaming\u2019s right-side sliding window mask (`-dist <= right_window_size` vs `<`). PR 46546 is the more complete representation because it includes the Moonshine fix plus tests and an additional analogous Gemma4 fix.", + "confidence": 0.92, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 46546 is the better canonical PR because it contains the Moonshine streaming boundary fix, adds test coverage, and is the more complete patch set.", + "best_issue_reason": null, + "best_pr_reason": "PR 46546 best represents the cluster since it covers the same Moonshine streaming bug as 46310 and includes validation plus the related Gemma4 adjustment.", + "soft_edge_verdicts": [ + { + "left": "pull_request:46310", + "right": "pull_request:46546", + "accept": true, + "reason": "Same underlying Moonshine streaming right-window boundary bug; the diff is the same off-by-one correction in the same mask code path, and 46546 is essentially a superset of 46310 for that fix." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 33, + "component_count": 38, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-46376-6", + "nodes": [ + "pull_request:46376", + "pull_request:46417", + "pull_request:46425", + "pull_request:46509", + "pull_request:46850", + "pull_request:46917" + ], + "soft_pairs": [ + "pull_request:46376|pull_request:46417", + "pull_request:46376|pull_request:46425", + "pull_request:46425|pull_request:46509", + "pull_request:46425|pull_request:46917", + "pull_request:46425|pull_request:46850", + "pull_request:46850|pull_request:46917" + ], + "prepared_review_unit_hash": "613d36cd047c28497d745197401c5023aff1e5e68ff677a57055047cf3f5090c", + "component_budget": { + "node_count": 6, + "item_count": 6, + "soft_pair_count": 6, + "serialized_chars": 16838, + "estimated_input_tokens": 4210, + "estimated_eval_tokens": 8676 + }, + "budget": { + "node_count": 6, + "item_count": 6, + "soft_pair_count": 6, + "serialized_chars": 16838, + "estimated_input_tokens": 4210, + "estimated_eval_tokens": 8676 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These PRs are not duplicates of one another. They span unrelated changes: multimodal continuous batching, HunYuan VL model support, RADIO model support, Nemotron Omni support, Gemma masking fixes, and deprecated API removal. The only similarities are broad feature-addition patterns and shared framework files, which are not enough to merge PRs.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": "PR 46425 is the most self-contained model-support addition and the closest thing to a cluster representative, but it is not a true duplicate canonical target because the set is heterogeneous.", + "soft_edge_verdicts": [ + { + "left": "pull_request:46376", + "right": "pull_request:46417", + "accept": false, + "reason": "Different features and code paths: continuous batching multimodal support vs adding a new HunYuan VL model package." + }, + { + "left": "pull_request:46376", + "right": "pull_request:46425", + "accept": false, + "reason": "Both are large feature additions, but one changes continuous batching internals while the other adds a specific vision-language model; no shared bug or mergeable code-path." + }, + { + "left": "pull_request:46425", + "right": "pull_request:46509", + "accept": false, + "reason": "Distinct model integrations for different architectures (RADIO vs Nemotron Omni); same broad subsystem only, not the same change." + }, + { + "left": "pull_request:46425", + "right": "pull_request:46917", + "accept": false, + "reason": "RADIO model support and deprecated API cleanup are unrelated; no common underlying bug or implementation target." + }, + { + "left": "pull_request:46425", + "right": "pull_request:46850", + "accept": false, + "reason": "Model addition versus a Gemma attention-masking bug fix; completely different concrete behavior and files." + }, + { + "left": "pull_request:46850", + "right": "pull_request:46917", + "accept": false, + "reason": "Gemma sliding-window masking fix and deprecated training/image-backend cleanup are unrelated changes." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 34, + "component_count": 38, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-46574-2", + "nodes": [ + "pull_request:46574", + "pull_request:46580" + ], + "soft_pairs": [ + "pull_request:46574|pull_request:46580" + ], + "prepared_review_unit_hash": "fd347c25ad1076d0ee8a6814c2a5defa5c91f0eda8a2578129a2047a972aced6", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4408, + "estimated_input_tokens": 1102, + "estimated_eval_tokens": 2460 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4408, + "estimated_input_tokens": 1102, + "estimated_eval_tokens": 2460 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These two PRs fix different tokenizer-related problems in different models: CodeLlama decode-space stripping vs. DeepSeek-Coder/DeepSeek-LLM auto-tokenizer misclassification. They are not duplicates and should not be merged.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single canonical PR: the changes address separate bugs in different code paths and model families.", + "best_issue_reason": null, + "best_pr_reason": "Neither PR is a global canonical choice because they solve unrelated issues; each is only best for its own specific bug.", + "soft_edge_verdicts": [ + { + "left": "pull_request:46574", + "right": "pull_request:46580", + "accept": false, + "reason": "Different failure modes and code paths: CodeLlama decoder setup vs. AutoTokenizer checkpoint-prefix override for DeepSeek models. Not the same underlying change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 35, + "component_count": 38, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-46603-2", + "nodes": [ + "pull_request:46603", + "pull_request:46604" + ], + "soft_pairs": [ + "pull_request:46603|pull_request:46604" + ], + "prepared_review_unit_hash": "cc33ca8c203482762107b8b95bdb103ffae57774d83082ea5263be55e87d3f90", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5496, + "estimated_input_tokens": 1374, + "estimated_eval_tokens": 3004 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5496, + "estimated_input_tokens": 1374, + "estimated_eval_tokens": 3004 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Two related but distinct Mistral PRs: one moves MistralConverter into a dedicated integrations package, the other adds tekken tokenizer detection/conversion/save utilities. They overlap on files and subsystem but do not appear to be the same fix or mergeable into one PR as-is.", + "confidence": 0.94, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #46604 is the better representative of the cluster because it introduces the broader tokenizer functionality (detection, conversion, and save utilities) rather than only relocating existing converter code.", + "best_issue_reason": null, + "best_pr_reason": "PR #46604 is the strongest standalone change set: it covers the concrete new Mistral tokenizer behavior, while #46603 is mostly a refactor/package move.", + "soft_edge_verdicts": [ + { + "left": "pull_request:46603", + "right": "pull_request:46604", + "accept": false, + "reason": "Reject: #46603 is a code move/refactor for MistralConverter, while #46604 adds tekken tokenizer detection, conversion, and save utilities. They share Mistral files but address different underlying changes and would not plausibly be merged as one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 36, + "component_count": 38, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-46741-4", + "nodes": [ + "pull_request:46741", + "pull_request:46773", + "pull_request:46781", + "pull_request:46937" + ], + "soft_pairs": [ + "pull_request:46741|pull_request:46773", + "pull_request:46741|pull_request:46781", + "pull_request:46773|pull_request:46937" + ], + "prepared_review_unit_hash": "540b0506cea828d90cce328a1ba3976f6e7719d81844620af3dbddf39f8cdabb", + "component_budget": { + "node_count": 4, + "item_count": 4, + "soft_pair_count": 3, + "serialized_chars": 11149, + "estimated_input_tokens": 2788, + "estimated_eval_tokens": 5832 + }, + "budget": { + "node_count": 4, + "item_count": 4, + "soft_pair_count": 3, + "serialized_chars": 11149, + "estimated_input_tokens": 2788, + "estimated_eval_tokens": 5832 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Four open PRs cluster around hybrid/linear-attention model behavior, but they address different bugs: chunked-prefill/speculative decoding cache handling, padding-mask application for batch size 1, QuantizedCache construction for hybrid models, and assisted-generation statefulness. No duplicate pair is strong enough to merge.", + "confidence": 0.93, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #46741 is the strongest representative: it fixes a concrete cache/state handling bug in chunked-prefill/speculative decoding and applies it across multiple affected hybrid models with targeted code changes and tests.", + "best_issue_reason": null, + "best_pr_reason": "PR #46741 is the best standalone change in the cluster because it targets a specific runtime failure mode in a well-defined code path and has the broadest, most coherent fix surface across affected models.", + "soft_edge_verdicts": [ + { + "left": "pull_request:46741", + "right": "pull_request:46773", + "accept": false, + "reason": "Both touch Bamba and other linear-attention models, but #46741 fixes precomputed-state handling for chunked prefill/speculative decoding, while #46773 fixes padding-mask application when batch size is 1. Different bug, different code path." + }, + { + "left": "pull_request:46741", + "right": "pull_request:46781", + "accept": false, + "reason": "#46741 changes model forward/cache-state logic in specific architectures; #46781 changes QuantizedCache layer construction in cache_utils for hybrid models. Related subsystem, but not the same underlying defect." + }, + { + "left": "pull_request:46773", + "right": "pull_request:46937", + "accept": false, + "reason": "#46773 addresses padding-mask skipping; #46937 marks LFM2/LFM2-MoE as stateful to reject assisted generation. These are unrelated behavior fixes." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 37, + "component_count": 38, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-46895-3", + "nodes": [ + "pull_request:46895", + "pull_request:46898", + "pull_request:46916" + ], + "soft_pairs": [ + "pull_request:46895|pull_request:46898", + "pull_request:46895|pull_request:46916", + "pull_request:46898|pull_request:46916" + ], + "prepared_review_unit_hash": "aedf8d5c5e1ff1a317506bb871751cfc19777863aba87605ad51c3cde63b6f03", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 7716, + "estimated_input_tokens": 1929, + "estimated_eval_tokens": 4114 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 7716, + "estimated_input_tokens": 1929, + "estimated_eval_tokens": 4114 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "All three PRs address the same broad training-loss double-shift pattern, but they are model-specific patches and not true duplicates. PR 46916 is the best representative because it spans both CohereASR and Florence2 and subsumes the other two fixes.", + "confidence": 0.83, + "canonical_issue_reason": null, + "canonical_pr_reason": "46916 is the broadest and most representative patch: it covers both CohereASR and Florence2 training-loss double-shift fixes, while the others each cover only one model.", + "best_issue_reason": null, + "best_pr_reason": "46916 best captures the cluster theme and includes both code paths, making it the strongest anchor PR.", + "soft_edge_verdicts": [ + { + "left": "pull_request:46895", + "right": "pull_request:46898", + "accept": false, + "reason": "Same symptom, but different model code paths (CohereASR vs Florence2); these are parallel model-specific fixes, not one duplicate change." + }, + { + "left": "pull_request:46895", + "right": "pull_request:46916", + "accept": false, + "reason": "46916 is a broader superset patch that also changes Florence2; it overlaps on CohereASR but is not the same single concrete PR change." + }, + { + "left": "pull_request:46898", + "right": "pull_request:46916", + "accept": false, + "reason": "46916 includes extra CohereASR work in addition to Florence2, so it is not just a duplicate of the Florence2-only fix." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 38, + "component_count": 38, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-46907-5", + "nodes": [ + "pull_request:46907", + "pull_request:46908", + "pull_request:46926", + "pull_request:46927", + "pull_request:46928" + ], + "soft_pairs": [ + "pull_request:46907|pull_request:46927", + "pull_request:46907|pull_request:46926", + "pull_request:46907|pull_request:46908", + "pull_request:46907|pull_request:46928", + "pull_request:46908|pull_request:46926", + "pull_request:46908|pull_request:46928", + "pull_request:46908|pull_request:46927", + "pull_request:46926|pull_request:46927", + "pull_request:46926|pull_request:46928", + "pull_request:46927|pull_request:46928" + ], + "prepared_review_unit_hash": "e0a57655ac9c6e71cdce62aa886fdff05dd64e5a1f8f002fa8f66586c9c1180e", + "component_budget": { + "node_count": 5, + "item_count": 5, + "soft_pair_count": 10, + "serialized_chars": 11920, + "estimated_input_tokens": 2980, + "estimated_eval_tokens": 6216 + }, + "budget": { + "node_count": 5, + "item_count": 5, + "soft_pair_count": 10, + "serialized_chars": 11920, + "estimated_input_tokens": 2980, + "estimated_eval_tokens": 6216 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These PRs are not duplicates despite similar triage labels and soft similarity: each changes a different model or subsystem and fixes a different failure mode. The cluster is a set of unrelated model-specific test/load fixes, not one underlying bug.", + "confidence": 0.94, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": "No single PR is a good cluster representative: the changes target different code paths (EdgeTAM forward kwargs, DeepSeek/MiniMax load-time offloading, GLM46V input_id dtype handling, and BLIP-2 tokenizer registration).", + "soft_edge_verdicts": [ + { + "left": "pull_request:46907", + "right": "pull_request:46927", + "accept": false, + "reason": "Different models and fixes: EdgeTAM strips an invalid processor kwarg before backbone forward, while GLM46V adds input_id casting in its forward method. Not the same bug or mergeable into one PR." + }, + { + "left": "pull_request:46907", + "right": "pull_request:46926", + "accept": false, + "reason": "Different model families and failure modes: EdgeTAM integration-test import/config issue vs MiniMax-M3-VL load_error workaround via offload_folder." + }, + { + "left": "pull_request:46907", + "right": "pull_request:46908", + "accept": false, + "reason": "EdgeTAM code-path fix vs DeepSeek-V3.2 test harness load_error fix; unrelated models and unrelated runtime problems." + }, + { + "left": "pull_request:46907", + "right": "pull_request:46928", + "accept": false, + "reason": "EdgeTAM forward-kwargs fix is unrelated to BLIP-2 tokenizer auto-registration; different subsystem and different bug." + }, + { + "left": "pull_request:46908", + "right": "pull_request:46926", + "accept": false, + "reason": "Both are load_error-related test fixes, but for different models and different loaders/offload behavior. Not the same underlying issue." + }, + { + "left": "pull_request:46908", + "right": "pull_request:46928", + "accept": false, + "reason": "DeepSeek load-time offloading workaround is unrelated to BLIP-2 tokenizer mapping." + }, + { + "left": "pull_request:46908", + "right": "pull_request:46927", + "accept": false, + "reason": "DeepSeek load_error workaround vs GLM46V dtype conversion in forward; different code paths and failure modes." + }, + { + "left": "pull_request:46926", + "right": "pull_request:46927", + "accept": false, + "reason": "MiniMax-M3-VL offload_folder test fix is unrelated to GLM46V forward input_id casting." + }, + { + "left": "pull_request:46926", + "right": "pull_request:46928", + "accept": false, + "reason": "MiniMax load_error workaround vs BLIP-2 tokenizer registry update; no shared underlying bug." + }, + { + "left": "pull_request:46927", + "right": "pull_request:46928", + "accept": false, + "reason": "GLM46V input_id long-casting fix is unrelated to BLIP-2 auto tokenizer support." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + } + ] +} diff --git a/snapshots/20260628T000017Z/analysis-runs/hybrid-model-20260628t000017z/manifest.json b/snapshots/20260628T000017Z/analysis-runs/hybrid-model-20260628t000017z/manifest.json new file mode 100644 index 0000000000000000000000000000000000000000..2415c34483d5cd93876ac89a07b90504bec7fb8e --- /dev/null +++ b/snapshots/20260628T000017Z/analysis-runs/hybrid-model-20260628t000017z/manifest.json @@ -0,0 +1,14 @@ +{ + "analysis_id": "hybrid-model-20260628t000017z", + "artifacts": { + "hybrid": "snapshots/20260628T000017Z/analysis-runs/hybrid-model-20260628t000017z/analysis-report-hybrid.json", + "hybrid_reviews": "snapshots/20260628T000017Z/analysis-runs/hybrid-model-20260628t000017z/analysis-report-hybrid.llm-reviews.json" + }, + "channel": "canonical", + "model": null, + "published_at": "2026-06-28T00:10:09Z", + "repo": "huggingface/transformers", + "schema_version": 1, + "snapshot_id": "20260628T000017Z", + "variant": "hybrid" +} diff --git a/snapshots/20260628T000017Z/manifest.json b/snapshots/20260628T000017Z/manifest.json index 6c565737090ef34e1c5e28ba48950d7f6e4b8073..56ea56dadb27529640b4dc70bad57829bd95f2a2 100644 --- a/snapshots/20260628T000017Z/manifest.json +++ b/snapshots/20260628T000017Z/manifest.json @@ -32,6 +32,24 @@ "timeline_events": 351 }, "extracted_at": "2026-06-28T00:00:17Z", + "published_analysis": { + "canonical_analysis_id": "hybrid-model-20260628t000017z", + "runs": { + "hybrid-model-20260628t000017z": { + "analysis_id": "hybrid-model-20260628t000017z", + "artifacts": { + "hybrid": "snapshots/20260628T000017Z/analysis-runs/hybrid-model-20260628t000017z/analysis-report-hybrid.json", + "hybrid_reviews": "snapshots/20260628T000017Z/analysis-runs/hybrid-model-20260628t000017z/analysis-report-hybrid.llm-reviews.json" + }, + "channel": "canonical", + "manifest_path": "snapshots/20260628T000017Z/analysis-runs/hybrid-model-20260628t000017z/manifest.json", + "model": null, + "published_at": "2026-06-28T00:10:09Z", + "variant": "hybrid" + } + }, + "schema_version": 1 + }, "repo": "huggingface/transformers", "snapshot_id": "20260628T000017Z", "watermark": {