diff --git a/analysis/current/analysis-report-hybrid.json b/analysis/current/analysis-report-hybrid.json index e9f94722d73d3654f3ccebc539eb64dcf7870df3..077f43bc950eec8e8d92d54cd29fb1c8c89dd3e7 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": "20260627T000016Z", - "generated_at": "2026-06-27T00:09:44Z", + "snapshot_id": "20260627T060016Z", + "generated_at": "2026-06-27T06:09:28Z", "evidence_quality": "full", "llm_enrichment": true, "meta_bugs": [ @@ -1050,21 +1050,117 @@ ] }, { - "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": 0.55, + "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-46895-4", + "summary": "Cluster of 1 issues and 3 PRs centered on issue #46897.", "status": "open", "confidence": 0.8, "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": [ @@ -1084,17 +1180,6 @@ "shared_filenames": [], "shared_file_areas": [] }, - { - "left_pr_number": 46895, - "right_pr_number": 46903, - "code_similarity": 0.249, - "size_similarity": 0.874, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.495, - "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 - ] - ] - } - ] } ] }, @@ -1432,6 +1396,60 @@ } ] }, + { + "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.", @@ -1753,60 +1771,6 @@ } ] }, - { - "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": 0.732, - "size_similarity": 0.857, - "file_overlap": 1.0, - "area_overlap": 0.5, - "patch_similarity": 0.571, - "shared_filenames": [ - "src/transformers/models/yolos/modeling_yolos.py" - ], - "shared_file_areas": [ - { - "filename": "src/transformers/models/yolos/modeling_yolos.py", - "left_ranges": [ - [ - 434, - 440 - ] - ], - "right_ranges": [ - [ - 434, - 440 - ], - [ - 446, - 452 - ] - ] - } - ] - } - ] - }, { "cluster_id": "cluster-45561-3", "summary": "Cluster of 1 issues and 2 PRs centered on issue #45561.", @@ -2082,6 +2046,35 @@ } ] }, + { + "cluster_id": "cluster-44002-2", + "summary": "Cluster of 2 related pull requests linked by soft_similarity.", + "status": "open", + "confidence": 0.5, + "canonical_issue_number": null, + "canonical_pr_number": 44002, + "issue_numbers": [], + "pr_numbers": [ + 44002, + 44029 + ], + "evidence_types": [ + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 44002, + "right_pr_number": 44029, + "code_similarity": 0.135, + "size_similarity": 0.43, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.329, + "shared_filenames": [], + "shared_file_areas": [] + } + ] + }, { "cluster_id": "cluster-44018-2", "summary": "Cluster of 2 related pull requests linked by soft_similarity.", @@ -2358,17 +2351,7 @@ ] } ], - "duplicate_issues": [ - { - "cluster_id": "cluster-46894-7", - "canonical_issue_number": 46897, - "duplicate_issue_numbers": [ - 46894, - 46901 - ], - "reason": "Issues in cluster-46894-7 are treated as duplicates because they share closing_reference, shared_issue_target, soft_similarity evidence." - } - ], + "duplicate_issues": [], "duplicate_prs": [ { "cluster_id": "cluster-41211-3", @@ -2414,6 +2397,15 @@ "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-44002-2", + "canonical_pr_number": 44002, + "duplicate_pr_numbers": [ + 44029 + ], + "target_issue_number": null, + "reason": "PRs in cluster-44002-2 are treated as duplicates because they share soft_similarity evidence." + }, { "cluster_id": "cluster-44018-2", "canonical_pr_number": 44068, @@ -2478,22 +2470,21 @@ "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-46895-4", + "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-46895-4 are treated as duplicates because they converge on issue #46897 with closing_reference, shared_issue_target, soft_similarity evidence." } ], "best_issue": { - "cluster_id": "cluster-46894-7", + "cluster_id": "cluster-46895-4", "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 4 artifacts, and carries 0 discussion comments plus 9 inbound references.", + "score": 149.0 }, "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 492c8396705fffb3dfa973d2e4900ddef9b183de..ce9680039f87f860885d47014df122c63b564335 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": "20260627T000016Z", - "generated_at": "2026-06-27T00:09:44Z", + "snapshot_id": "20260627T060016Z", + "generated_at": "2026-06-27T06:09:28Z", "model": "gpt-5.4-mini?service_tier=flex", "reviews": [ { @@ -132,418 +132,370 @@ "issue:45237|issue:46015", "issue:43039|issue:44805" ], - "prepared_review_unit_hash": "f27c9dab842b01c1c54ea093d0737424a4600e462160f23a60088362a2e659ea", + "prepared_review_unit_hash": "f0d510929ef519664e5f4f6f3903a8a98b16db6eda79a20cf63f2b02096456ee", "component_budget": { "node_count": 500, "item_count": 500, "soft_pair_count": 1839, - "serialized_chars": 953913, - "estimated_input_tokens": 238479, - "estimated_eval_tokens": 477214 + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 63, - "serialized_chars": 46569, - "estimated_input_tokens": 11643, - "estimated_eval_tokens": 23542 + "serialized_chars": 46573, + "estimated_input_tokens": 11644, + "estimated_eval_tokens": 23544 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Most pairs are false positives across unrelated models/features. The only strong duplicate clusters are DeepSpeed ZeRO-3 zero-sized-tensor index errors, Qwen3.5 MoE rotary-embedding offload/materialization, and a torch 2.4.1 compatibility regression.", - "confidence": 0.83, - "canonical_issue_reason": "issue:43638 is the best anchor: it clearly names the ZeRO-3 `index 0 is out of bounds for dimension 0 with size 0` failure on a Bert model, and it lines up with the closely related Gemma3 ZeRO-3 report.", + "summary": "This cluster is mostly a mix of unrelated bug reports. Only two soft pairs look like true duplicates: the ZeRO-3 BertModel index-out-of-bounds reports, and the torch 2.4.1 compatibility reports.", + "confidence": 0.93, + "canonical_issue_reason": "Issue 43596 is the clearest canonical for the Bert/ZeRO-3 duplicate pair: it describes the concrete index-out-of-bounds failure on a standard BertModel, and 43638 is just the non-pretrained variant of the same bug.", "canonical_pr_reason": null, - "best_issue_reason": "issue:43638 is the most suitable issue to keep as the representative bug report: concrete error, clear repro context, and it matches the same underlying ZeRO-3 failure seen in the other Bert/Gemma3 report.", + "best_issue_reason": "Issue 43596 is the best single issue in the set to anchor triage because it has the strongest true duplicate match and a concrete, reusable failure signature.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:44322", "right": "issue:45464", "accept": false, - "reason": "Different model and failure mode: config attribute error vs streaming chat/completions issue." + "reason": "Different bugs: Qwen3.5 config attribute missing vs streaming chat/completions failure." }, { "left": "issue:45237", "right": "issue:45685", "accept": false, - "reason": "Different subsystems and errors: AMD GPU model failure vs MPS histogram op limitation." + "reason": "Unrelated failures: AMD GPU support for GPT-OSS vs an MPS histogram op error." }, { "left": "issue:43716", "right": "issue:45237", "accept": false, - "reason": "Different bug classes: dtype mismatch in preprocessing vs AMD GPU/device-map failure." + "reason": "Different model/runtime issues: Mistral-3 image dtype mismatch vs GPT-OSS on AMD GPUs." }, { "left": "issue:43828", "right": "issue:45237", "accept": false, - "reason": "Both are runtime mismatches, but on different models/code paths and not the same concrete bug." + "reason": "Different bugs: autocast dtype mismatch in Phi-tiny-MoE vs GPT-OSS AMD GPU failure." }, { "left": "issue:45237", "right": "issue:46468", "accept": false, - "reason": "Different models and failures; Nemotron example breakage is unrelated to GPT-OSS on AMD GPUs." + "reason": "Different models and symptoms; no shared concrete code path is evident." }, { "left": "issue:45237", "right": "issue:45507", "accept": false, - "reason": "Different failures: GPU compatibility vs invalid method call in GraniteMoEHybrid." + "reason": "Different issues: GPT-OSS AMD GPU support vs GraniteMoEHybrid invalid method call." }, { "left": "issue:30277", "right": "issue:35545", "accept": false, - "reason": "Unrelated errors: DeepSpeed collective mismatch vs ONNX export error." + "reason": "Different failures: DeepSpeed ZeRO-3 collective mismatch vs ONNX export error." }, { "left": "issue:35545", "right": "issue:42915", "accept": false, - "reason": "Different models and problems: ONNX export vs FineGrainedFP8 failure." + "reason": "Different problems: ONNX export of ModernBERT vs Qwen3Moe FP8 failure." }, { "left": "issue:35545", "right": "issue:41720", "accept": false, - "reason": "Different failures: export-time ONNX issue vs runtime device-map assert on Qwen3." + "reason": "Different code paths: export/tracing bug vs Qwen3 auto device-map cuda assert." }, { "left": "issue:45901", "right": "issue:46015", "accept": false, - "reason": "Different tasks: table QA crash vs outdated RAG docs/examples." + "reason": "Different topics: table QA crash vs outdated RAG docs." }, { "left": "issue:44509", "right": "issue:44512", "accept": false, - "reason": "Both are v5 docs cleanup, but they reference different removed docs content and are not the same report." + "reason": "Both are v5 docs cleanup, but they remove different obsolete references in different documentation areas." }, { "left": "issue:35545", "right": "issue:36010", "accept": false, - "reason": "Different code paths: ModernBERT ONNX export vs missing GenerationMixin import." + "reason": "Different bugs: ONNX export error vs GenerationMixin import regression." }, { "left": "issue:44918", "right": "issue:46165", "accept": false, - "reason": "Different training issues: Qwen3.5 embedding unpacking vs resume-with-remote-code failure." + "reason": "Different training failures: Qwen3.5 embedding unpacking vs resume-training with trust_remote_code." }, { "left": "issue:46165", "right": "issue:46869", "accept": false, - "reason": "Different models and symptoms: resume/trust-remote-code issue vs YOLO multi-GPU device-map failure." + "reason": "Unrelated model/runtime problems; no common underlying bug is clear." }, { "left": "issue:44805", "right": "issue:45823", "accept": false, - "reason": "Both mention mismatches, but one is a mask-shape error and the other is a device mismatch in Gemma4 PLE." + "reason": "Different symptoms and subsystems: mask shape/indexing mismatch vs Gemma4 device-map PLE mismatch." }, { "left": "issue:43638", "right": "issue:44805", "accept": false, - "reason": "Different tensor-shape failures and different model contexts." + "reason": "Both mention tensor/index errors, but the failures are different and not the same code path." }, { "left": "issue:43638", "right": "issue:46165", "accept": false, - "reason": "Different problems: ZeRO-3 init index error vs resume/trust-remote-code issue." + "reason": "Different issues: Bert ZeRO-3 index error vs resume-training with local model and remote code." }, { "left": "issue:43596", "right": "issue:44805", "accept": false, - "reason": "Different error signatures and model paths; only share a generic IndexError wording." + "reason": "Different failures: Bert ZeRO-3 loading bug vs mask/tensor indexing mismatch." }, { "left": "issue:44805", "right": "issue:46294", "accept": false, - "reason": "Different modalities and causes: mask/tensor shape mismatch vs video token/feature mismatch." + "reason": "Different modalities and errors: mask shape mismatch vs video token/feature mismatch." }, { "left": "issue:44805", "right": "issue:45161", "accept": false, - "reason": "Different systems: generic mask-shape error vs GPT-OSS tensor-parallel issue." + "reason": "Different problems: tensor mask indexing error vs GPT-OSS MoE tensor-parallel issue." }, { "left": "issue:46165", "right": "issue:46468", "accept": false, - "reason": "Different models and failure modes; not the same underlying bug." + "reason": "Different bugs with different models and call paths." }, { "left": "issue:44560", "right": "issue:46165", "accept": false, - "reason": "Unrelated: video embedding StopIteration vs model resume/remote-code issue." + "reason": "Different issues: Qwen3-vl video StopIteration vs training resume with remote code." }, { "left": "issue:34634", "right": "issue:35545", "accept": false, - "reason": "Different areas entirely: BarkProcessor preset handling vs ModernBERT export." + "reason": "Different components: BarkProcessor voice_preset vs ModernBERT ONNX export." }, { "left": "issue:30277", "right": "issue:43638", "accept": false, - "reason": "Both involve DeepSpeed, but the errors and models differ; not the same bug." + "reason": "Both involve DeepSpeed, but one is a collective mismatch and the other is a Bert ZeRO-3 index error." }, { "left": "issue:43596", "right": "issue:43638", "accept": true, - "reason": "Same ZeRO-3 `index 0 is out of bounds for dimension 0 with size 0` failure during model init/loading; model wording differs but the bug appears identical." + "reason": "Same underlying bug: both are DeepSpeed ZeRO-3 BertModel loads failing with index 0 out of bounds; only the pretrained status differs." }, { "left": "issue:43039", "right": "issue:45161", "accept": false, - "reason": "Different features and failure points: Liger Kernel cross-entropy call vs tensor-parallel MoE issue." + "reason": "Different failures: Liger Kernel cross-entropy still being called vs GPT-OSS MoE tensor-parallel behavior." }, { "left": "issue:43975", "right": "issue:44457", "accept": false, - "reason": "Different bugs: detokenization regression vs LoRA merge/save/load mismatch." + "reason": "Different bugs: DeepSeek-Coder detokenization vs LoRA merge/save/reload output mismatch." }, { "left": "issue:44292", "right": "issue:44457", "accept": false, - "reason": "Different models and errors: NVFP4 runtime issue vs LoRA weight merge inconsistency." + "reason": "Different problems: Qwen-3-8B-NVFP4 runtime error vs LoRA merge/reload inconsistency." }, { "left": "issue:44841", "right": "issue:46661", "accept": false, - "reason": "Different bug classes: processor failure vs remote-code model_class config_class AttributeError." - }, - { - "left": "issue:43716", - "right": "issue:46869", - "accept": false, - "reason": "Different models and issues: dtype mismatch in preprocessing vs YOLO multi-GPU device-map failure." - }, - { - "left": "issue:24643", - "right": "issue:30277", - "accept": false, - "reason": "Both mention DeepSpeed, but the concrete runtime errors and scenarios differ." - }, - { - "left": "issue:34928", - "right": "issue:44805", - "accept": false, - "reason": "Different failure modes: activation-checkpointing/FSDP recompute mismatch vs mask-shape IndexError." + "reason": "Different issues: processor failure for Voxtral vs remote-code config_class AttributeError." }, { "left": "issue:45859", "right": "issue:45902", - "accept": true, - "reason": "Same Qwen3.5 MoE rotary-embedding/offload path; both point to broken `Qwen3_5MoeTextRotaryEmbedding` handling around materialization/offload." + "accept": false, + "reason": "Same class name, but different bugs: CPU-offload forward compatibility vs uninitialized inv_freq after meta materialization." }, { "left": "issue:45685", "right": "issue:46869", "accept": false, - "reason": "Different hardware/backend problems: MPS histogram op vs YOLO multi-GPU device-map failure." + "reason": "Different hardware/runtime problems: MPS Int histogram op vs YOLO multi-GPU device-map auto failure." }, { "left": "issue:43638", "right": "issue:45161", "accept": false, - "reason": "Different subsystems and symptoms: ZeRO-3 Bert index error vs GPT-OSS tensor-parallel issue." + "reason": "Different bugs: Bert ZeRO-3 index error vs GPT-OSS MoE tensor-parallel issue." }, { "left": "issue:34928", "right": "issue:36331", "accept": false, - "reason": "Different training bugs: FSDP/activation checkpointing recompute mismatch vs CustomTrainer argument error." + "reason": "Different training failures: activation checkpointing/FSDP tensor recomputation vs CustomTrainer compute_loss signature mismatch." }, { "left": "issue:43596", "right": "issue:45161", "accept": false, - "reason": "Different failure classes: ZeRO-3 index error vs tensor-parallel MoE issue." + "reason": "Different bugs and models; no shared concrete failure mode." }, { "left": "issue:43716", "right": "issue:45507", "accept": false, - "reason": "Different models and bugs: image dtype mismatch vs GraniteMoEHybrid invalid method." + "reason": "Different model-specific issues: Mistral-3 dtype mismatch vs GraniteMoEHybrid invalid method." }, { "left": "issue:44292", "right": "issue:45237", "accept": false, - "reason": "Different model/runtime failures: Qwen NVFP4 issue vs GPT-OSS AMD GPU issue." + "reason": "Different model/runtime failures; not the same underlying bug." }, { "left": "issue:45874", "right": "issue:46294", "accept": false, - "reason": "Both are Gemma/Qwen multimodal issues, but one is inputs_embeds reverse lookup and the other is video token/template mismatch." + "reason": "Different features: reverse embedding lookup on inputs_embeds vs video chat-template feature/token mismatch." }, { "left": "issue:39290", "right": "issue:45081", "accept": false, - "reason": "Different models and failure points: Gemma3 config attribute vs Mistral tokenizer regex patch crash." + "reason": "Different breakages: Gemma3 vLLM config attribute error vs mistral regex patch tokenizer backend error." }, { "left": "issue:45800", "right": "issue:46515", "accept": true, - "reason": "Same torch 2.4.1 compatibility regression; the second report is a shorter version of the same incompatibility." + "reason": "Same compatibility issue: both report torch 2.4.1 incompatibility, with no meaningful distinction in the failure itself." }, { "left": "issue:45823", "right": "issue:45874", "accept": false, - "reason": "Different Gemma4 issues: device mismatch in forward vs expensive reverse embedding lookup." + "reason": "Different Gemma4 issues: PLE device mismatch vs expensive reverse embedding lookup." }, { "left": "issue:43425", "right": "issue:44292", "accept": false, - "reason": "Generic incompatibility vs a specific Qwen NVFP4 runtime failure; not the same bug." + "reason": "Different problems: torch 2.10 compatibility vs Qwen-3-8B-NVFP4 runtime error." }, { "left": "issue:34928", "right": "issue:41720", "accept": false, - "reason": "Different failures: FSDP/activation checkpointing vs Qwen3 auto device mapping assert." + "reason": "Different issues: FSDP/activation-checkpointing size mismatch vs Qwen3 auto device-map cuda assert." }, { "left": "issue:43828", - "right": "issue:45161", - "accept": false, - "reason": "Different model paths and causes; only both are device/dtype-related at a high level." - }, - { - "left": "issue:43541", - "right": "issue:43716", + "right": "issue:45685", "accept": false, - "reason": "Different models and errors: Mixtral torch dynamo tracing failure vs dtype mismatch in Mistral preprocessing." + "reason": "Different bugs: autocast dtype mismatch vs MPS histogram op not implemented for Int." }, { "left": "issue:43828", "right": "issue:44292", "accept": false, - "reason": "Different models and runtime errors; not the same underlying issue." + "reason": "Different failures and models; no evidence they share a root cause." }, { "left": "issue:43716", "right": "issue:46468", "accept": false, - "reason": "Different model families and bugs; no evidence of shared root cause." + "reason": "Different model families and symptoms." }, { "left": "issue:44292", "right": "issue:44918", "accept": false, - "reason": "Different model/runtime paths: Qwen NVFP4 failure vs Qwen3.5 embeddings with TRL SFT trainer." + "reason": "Different code paths: Qwen NVFP4 runtime error vs TRL SFT trainer input embedding unpacking." }, { "left": "issue:45507", "right": "issue:45685", "accept": false, - "reason": "Unrelated failures: invalid method in GraniteMoEHybrid vs unimplemented MPS histogram op." + "reason": "Different bugs: GraniteMoEHybrid invalid method call vs MPS histogram op unsupported for Int." }, { "left": "issue:43716", "right": "issue:44292", "accept": false, - "reason": "Different models and symptoms; both are runtime errors but not the same bug." + "reason": "Different model/runtime failures; not the same bug." }, { "left": "issue:44918", "right": "issue:45161", "accept": false, - "reason": "Different code paths: TRL SFT embedding unpacking vs tensor-parallel GPT-OSS issue." + "reason": "Different issues: TRL SFT embedding unpacking vs GPT-OSS MoE tensor-parallel behavior." }, { "left": "issue:43716", "right": "issue:45685", "accept": false, - "reason": "Different models and backends; not the same bug." + "reason": "Different subsystems and failure modes." }, { "left": "issue:43039", "right": "issue:44918", "accept": false, - "reason": "Different areas: Liger Kernel cross-entropy vs TRL SFT embedding handling." + "reason": "Different issues: Liger Kernel cross-entropy invocation vs Qwen3.5 TRL embedding unpacking." }, { "left": "issue:41762", "right": "issue:43638", - "accept": true, - "reason": "Same DeepSpeed ZeRO-3 `index 0 is out of bounds for dimension 0 with size 0` failure during model loading/init; model names differ, but the underlying bug looks identical." + "accept": false, + "reason": "Both are ZeRO-3 index errors, but they affect different models and loading paths; too broad to treat as the same bug." }, { "left": "issue:43827", "right": "issue:44512", "accept": false, - "reason": "Both are v5 docs updates, but they reference different removed commands/tasks and should not be merged as the same issue." + "reason": "Different docs removals: pipeline task references vs transformers run command references." }, { "left": "issue:43828", "right": "issue:45685", "accept": false, - "reason": "Different backends and failures: autocast dtype mismatch vs MPS integer op limitation." - }, - { - "left": "issue:43039", - "right": "issue:43828", - "accept": false, - "reason": "Different model paths and different mismatch types; not the same bug." - }, - { - "left": "issue:30277", - "right": "issue:36331", - "accept": false, - "reason": "Unrelated: DeepSpeed collective mismatch vs CustomTrainer API breakage." - }, - { - "left": "issue:34928", - "right": "issue:36010", - "accept": false, - "reason": "Different subsystems and errors; no shared root cause." - }, - { - "left": "issue:45237", - "right": "issue:46015", - "accept": false, - "reason": "Completely different topics: AMD GPU model failure vs RAG example/documentation issue." + "reason": "Different bugs: autocast dtype mismatch vs MPS histogram op unsupported for Int." }, { "left": "issue:43039", "right": "issue:44805", "accept": false, - "reason": "Different failures and code paths; only a generic runtime-shape flavor in common." + "reason": "Different failures: Liger Kernel cross_entropy call vs mask shape/indexing mismatch." } ] }, "evaluator_result": { "accept": true, - "feedback": "The summary is grounded in the packet, and the soft-edge decisions stay conservative. The accepted clusters are supported by closely matching titles and error signatures (the ZeRO-3 index-out-of-bounds reports, the Qwen3.5 MoE rotary/offload reports, and the torch 2.4.1 incompatibility reports), while the rejected pairs generally differ by model, backend, or failure mode. No pull-request mergeability issue applies here." + "feedback": "The summary is grounded in the packet and stays conservative. The two accepted duplicate pairs are well supported by the issue titles and descriptions, and the rejected soft pairs are phrased cautiously without overclaiming shared root cause. The canonical Bert ZeRO-3 issue choice is also defensible." }, "evaluator_used": true, "retried": false, @@ -657,292 +609,274 @@ "issue:43572|issue:45070", "issue:45800|issue:45823" ], - "prepared_review_unit_hash": "259d051695b415c3d8c1bf70d8344bb29ff8e8248eb3bdbfc9b9e2c32d8aaf58", + "prepared_review_unit_hash": "d55e3f913db7038362b35e55a5d54edd731dafe62cbcf19ba693a4368b81a49a", "component_budget": { "node_count": 500, "item_count": 500, "soft_pair_count": 1839, - "serialized_chars": 953913, - "estimated_input_tokens": 238479, - "estimated_eval_tokens": 477214 + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 42, - "serialized_chars": 40047, - "estimated_input_tokens": 10012, - "estimated_eval_tokens": 20280 + "serialized_chars": 40040, + "estimated_input_tokens": 10010, + "estimated_eval_tokens": 20276 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster is mostly a grab bag of unrelated model/runtime regressions, but there are two plausible duplicate pockets: ZeRO-3 loading index errors and AutoTokenizer v5 regression reports. No PRs are present.", - "confidence": 0.78, - "canonical_issue_reason": "issue:45920 is the best canonical issue because it describes the broadest instance of the tokenizer regression across multiple models, subsuming the narrower Granite-only report in issue:45812.", + "summary": "This cluster is mostly a grab-bag of unrelated Transformers bugs, with one clear duplicate pair around the v5 AutoTokenizer token-ID regression.", + "confidence": 0.76, + "canonical_issue_reason": "issue:45920 is the broadest and most representative report of the shared tokenizer regression, and it subsumes the Granite-only repro in issue:45812.", "canonical_pr_reason": null, - "best_issue_reason": "issue:45920 is the most suitable global issue target: broad scope, clear symptom, and strongest chance of collecting related reports about the same tokenizer bug.", + "best_issue_reason": "issue:45920 best represents the actual duplicate-worthy topic in the cluster: a v5 AutoTokenizer regression affecting multiple model families.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43866", "right": "issue:46402", "accept": false, - "reason": "Both are model-load failures, but they affect different architectures and failure modes; not the same concrete bug." + "reason": "Different models and failures: Ovis2 checkpoint corruption vs Sam3Model text-encoder weights not loaded." }, { "left": "issue:45800", "right": "issue:46304", "accept": false, - "reason": "Different incompatibility reports (torch/version matrix vs USE_DISTRIBUTED=0); too broad to merge." + "reason": "Both mention compatibility, but one is a generic torch/transformers version mismatch while the other is a USE_DISTRIBUTED=0 breakage." }, { "left": "issue:43296", "right": "issue:46304", "accept": false, - "reason": "Different runtime stacks and symptoms; one is vLLM/PaddleOCR-VL loading, the other is PyTorch distributed support." + "reason": "PaddleOCR-VL loading in vLLM and USE_DISTRIBUTED=0 are unrelated code paths and failure modes." }, { "left": "issue:41762", "right": "issue:43596", - "accept": true, - "reason": "Same ZeRO-3 loading path and same out-of-bounds-zero-size indexing failure; model-specific titles but likely one underlying bug." + "accept": false, + "reason": "Both hit a zero3-related IndexError, but the repros are on different model-loading paths and there is not enough evidence they are the same bug." }, { "left": "issue:39290", "right": "issue:41720", "accept": false, - "reason": "Both mention model/runtime failures, but the concrete problems are unrelated." + "reason": "Gemma3TextConfig missing sliding_window_pattern and Qwen3 auto device-map cuda asserts are different regressions." }, { "left": "issue:43901", "right": "issue:44512", "accept": false, - "reason": "Docs-only issues about different stale references; not the same change." + "reason": "Separate documentation issues in different areas, not the same underlying bug or change." }, { "left": "issue:44589", "right": "issue:46304", "accept": false, - "reason": "Unrelated runtime errors with different root causes." + "reason": "Float8 storage lookup failure is unrelated to the distributed-support regression in transformers v5.9.0." }, { "left": "issue:45237", "right": "issue:45901", "accept": false, - "reason": "Different subsystems and symptoms; no shared code-path problem." + "reason": "Different subsystems and symptoms: AMD GPU support for GPT-OSS vs table-question-answering crash." }, { "left": "issue:43975", "right": "issue:46015", "accept": false, - "reason": "One is a tokenizer/detokenization bug, the other is outdated RAG docs." + "reason": "A DeepSeek detokenization regression and outdated RAG docs are unrelated." }, { "left": "issue:43039", "right": "issue:44560", "accept": false, - "reason": "Different feature paths and failures; no shared concrete bug." + "reason": "Liger Kernel cross-entropy dispatch and Qwen3.5-vl video StopIteration are different bugs." }, { "left": "issue:45823", "right": "issue:46294", "accept": false, - "reason": "Different model families and different failure mechanisms." + "reason": "Different models and mismatches: Gemma4 PLE device placement vs Qwen3.5 video/chat-template token mismatch." }, { "left": "issue:44841", "right": "issue:45823", "accept": false, - "reason": "Processor failure and device-map mismatch are unrelated." + "reason": "Voxtral processor failure and Gemma4 device mismatch are separate model-specific problems." }, { "left": "issue:43866", "right": "issue:46772", "accept": false, - "reason": "Different model-loading/compatibility problems without a shared code fix." + "reason": "Ovis2 checkpoint corruption is unrelated to DiffusionGemma NVFP4 compatibility." }, { "left": "issue:44863", "right": "issue:46015", "accept": false, - "reason": "Checkpoint-loading bug vs docs issue; unrelated." + "reason": "NemotronH checkpoint loading and RAG docs being outdated are unrelated." }, { "left": "issue:45020", "right": "issue:46304", "accept": false, - "reason": "Both involve regressions in recent versions, but the specific breakages differ." + "reason": "Remote-code model breakage and USE_DISTRIBUTED=0 support regression are not the same underlying issue." }, { "left": "issue:45812", "right": "issue:45920", "accept": true, - "reason": "Same underlying AutoTokenizer v5 regression; 45920 is the broader multi-model report and 45812 looks like a narrower instance." + "reason": "Both report the same v5 AutoTokenizer wrong-token-ID regression; 45920 is the broader multi-model report and 45812 is a narrower Granite repro." }, { "left": "issue:43866", "right": "issue:44863", "accept": false, - "reason": "Both are load failures, but for different models and likely different causes." + "reason": "Different model-loading failures with no shared concrete code path." }, { "left": "issue:43872", "right": "issue:46304", "accept": false, - "reason": "bitsandbytes parameter incompatibility is unrelated to distributed support changes." + "reason": "bitsandbytes Int8Params API incompatibility is unrelated to the distributed-build regression." }, { "left": "issue:44292", "right": "issue:46015", "accept": false, - "reason": "Different issues: model execution on NVFP4 vs documentation cleanup." + "reason": "Qwen-3-8B-NVFP4 runtime failure and outdated RAG examples are unrelated." }, { "left": "issue:39290", "right": "issue:43531", "accept": false, - "reason": "Both mention sliding_window-related behavior, but the concrete bugs and contexts differ." + "reason": "Both mention sliding_window, but one is a Gemma3 config attribute issue and the other is a Qwen3-MoE model-specific sliding-window bug." }, { "left": "issue:41093", "right": "issue:43638", "accept": false, - "reason": "Both are index errors, but the tensor shapes, trigger conditions, and model paths are different." + "reason": "The mask-shape mismatch and the zero3 BertModel index error are different failure signatures and likely different bugs." }, { "left": "issue:43792", "right": "issue:45901", "accept": false, - "reason": "Unrelated task/runtime failures." + "reason": "Whisper-large-v2 runtime failure and table-question-answering crash are unrelated." }, { "left": "issue:43854", "right": "issue:43866", "accept": false, - "reason": "Different model-loading failures with no clear shared root cause." + "reason": "GLM-4.7-Flash loading in tests and Ovis2 checkpoint corruption are distinct issues." }, { "left": "issue:43425", "right": "issue:45070", "accept": false, - "reason": "General compatibility complaint vs pydantic/config regression; not the same bug." + "reason": "Both are version-related, but they affect different compatibility surfaces and are not the same regression." }, { "left": "issue:43975", "right": "issue:45901", "accept": false, - "reason": "Tokenizer regression vs task crash; unrelated." + "reason": "Detokenization regression and table QA crash are unrelated." }, { "left": "issue:43792", "right": "issue:45070", "accept": false, - "reason": "Different model/runtime problems." + "reason": "Whisper runtime failure is unrelated to the PretrainedConfig pydantic breakage." }, { "left": "issue:43596", "right": "issue:45800", "accept": false, - "reason": "ZeRO-3 loading index error and torch-version incompatibility are unrelated." + "reason": "Zero3 BertModel IndexError and torch/transformers incompatibility are not the same bug." }, { "left": "issue:45362", "right": "issue:45464", "accept": false, - "reason": "Different Qwen3.5 failures with different triggers and code paths." - }, - { - "left": "issue:43541", - "right": "issue:46869", - "accept": false, - "reason": "Different model families and different execution paths." - }, - { - "left": "issue:44292", - "right": "issue:45901", - "accept": false, - "reason": "Different features and symptoms; no shared bug." - }, - { - "left": "issue:39290", - "right": "issue:45823", - "accept": false, - "reason": "Different models and different failure modes." + "reason": "Both touch Qwen3.5 chat flows, but one is a chat-template crash and the other is a streaming API failure; too different to merge as duplicates." }, { "left": "issue:45464", "right": "issue:45685", "accept": false, - "reason": "Streaming API failure vs MPS histogram op limitation; unrelated." + "reason": "Chat streaming failure and an MPS histogram op issue are unrelated." }, { "left": "issue:36010", "right": "issue:39290", "accept": false, - "reason": "ImportError vs config-attribute error; not the same underlying issue." + "reason": "GenerationMixin import error and Gemma3 sliding_window attribute error are unrelated." }, { "left": "issue:43541", "right": "issue:45685", "accept": false, - "reason": "torch dynamo/grouped_mm tracing error and MPS histogram error are unrelated." + "reason": "Mixtral torch-dynamo grouped_mm tracing and the MPS histogram issue are separate bugs." }, { "left": "issue:43541", "right": "issue:43828", "accept": false, - "reason": "Different numerical/runtime failures on different model paths." + "reason": "Both involve model execution, but one is a grouped_mm tracing failure and the other is an autocast dtype mismatch." }, { "left": "issue:44841", "right": "issue:45084", "accept": false, - "reason": "Processor bug and template-compilation error are unrelated." + "reason": "Voxtral processor failure and a compile error about non-template nodes are unrelated." }, { "left": "issue:43404", "right": "issue:43866", "accept": false, - "reason": "Weight-tying bug and corrupted checkpoint/load failure are different problems." + "reason": "Untied lm_head weights in Mistral3 and Ovis2 checkpoint corruption are different issues." }, { "left": "issue:45507", "right": "issue:46869", "accept": false, - "reason": "Invalid method call in one model vs multi-GPU device-map failure in another." + "reason": "GraniteMoEHybrid invalid method and YOLO multi-GPU auto device-map failure are unrelated." }, { "left": "issue:41093", "right": "issue:41762", "accept": false, - "reason": "Both are index errors, but they arise from different shapes and different failure points." + "reason": "Different error shapes and symptoms: mask-size mismatch vs zero3 index-out-of-bounds during Gemma3 loading." }, { "left": "issue:45399", "right": "issue:46661", "accept": false, - "reason": "Different compatibility and loading issues." + "reason": "Flash-attn fallback gating and remote-code model_class/config_class handling are unrelated." }, { "left": "issue:43572", "right": "issue:45070", "accept": false, - "reason": "Missing config field vs pydantic model field breakage; not the same bug." + "reason": "StableLm pad_token_idx regression and a PretrainedConfig pydantic breakage are different compatibility bugs." }, { "left": "issue:45800", "right": "issue:45823", "accept": false, - "reason": "Torch/distributed incompatibility and Gemma4 device placement are unrelated." + "reason": "Torch/transformers version incompatibility and Gemma4 device-mapping mismatch are not the same concrete bug." } ] }, "evaluator_result": { "accept": true, - "feedback": "The summary is grounded in the packet and stays conservative. The two accepted duplicate pockets are supported by the issue titles/body patterns (ZeRO-3 index errors; AutoTokenizer v5 regression), and the other pair judgments are cautious. No PRs are present, so the null PR fields are appropriate." + "feedback": "The summary is well grounded in the packet: issue:45920 and issue:45812 are the only clearly duplicate-worthy pair, and the rest of the cluster appears heterogeneous. The soft-edge verdicts are conservative overall and do not overclaim evidence. No PR-pair mergeability concerns apply here." }, "evaluator_used": true, "retried": false, @@ -1045,226 +979,226 @@ "issue:45183|issue:46463", "issue:46489|issue:46491" ], - "prepared_review_unit_hash": "934302821de3b2b9708d602d2a7af23eadd463cbcf50dc8d003421d634922bdc", + "prepared_review_unit_hash": "6a538354b3f80a6e75e4adc1734b0412bce501176282a4e361eb87891925c4fb", "component_budget": { "node_count": 500, "item_count": 500, "soft_pair_count": 1839, - "serialized_chars": 953913, - "estimated_input_tokens": 238479, - "estimated_eval_tokens": 477214 + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 31, - "serialized_chars": 36884, - "estimated_input_tokens": 9221, - "estimated_eval_tokens": 18698 + "serialized_chars": 36887, + "estimated_input_tokens": 9222, + "estimated_eval_tokens": 18700 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly heterogeneous, with only a few soft pairs looking like true duplicates: the Qwen3.5 video regression pair, the trust_remote_code custom-module cache collision pair, and the shared input-validation/error-message pair. The rest appear to be different bugs, models, or subsystems despite surface similarity.", - "confidence": 0.63, - "canonical_issue_reason": "Issue 45183 is the best umbrella issue among the likely duplicates because it describes a shared validation path and spans multiple models, while the other items are mostly narrower model-specific manifestations or unrelated bugs.", + "summary": "This cluster is mostly heterogeneous; most soft edges are false positives. The only plausible duplicate groups are around trust_remote_code custom-module loading, Qwen3.5 video-input token accounting, and input validation/error-message wording.", + "confidence": 0.67, + "canonical_issue_reason": "No single issue cleanly represents the whole cluster, but issue 45698 is the strongest anchor among the near-duplicates because it captures the broad custom-module misloading symptom that also covers the leaf-directory collision case.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 45183 is the strongest global issue anchor in this set: broad enough to subsume the T5-specific follow-up, concrete enough to be actionable, and closer to a shared code-path bug than the other candidates.", + "best_issue_reason": "Issue 45698 is the most actionable and generally worded issue among the candidates, and it plausibly subsumes the closely related 45632 report.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43296", "right": "issue:43366", "accept": false, - "reason": "Different problem areas: PaddleOCR-VL loading in vLLM vs GGUF gpt-oss architecture support." + "reason": "Different model/setup problems: PaddleOCR-VL load failure vs GGUF gpt-oss architecture support." }, { "left": "issue:44387", "right": "issue:45005", "accept": false, - "reason": "Both mention training/model correctness, but one is int4 memory/OOM and the other is tied-weights behavior on translation models." + "reason": "One is int4 quantization memory/OOM behavior; the other is tied-weights handling for translation models." }, { "left": "issue:44805", "right": "issue:44918", "accept": false, - "reason": "Both are shape-related errors, but the model and failure path differ; not the same concrete bug." + "reason": "Both are runtime errors, but the shapes and code paths differ: mask indexing vs TRL input embedding unpacking." }, { "left": "issue:43541", "right": "issue:45161", "accept": false, - "reason": "Both involve MoE models, but one is a torch-dynamo grouped_mm tracing error and the other is a tensor-parallel issue." + "reason": "Both mention MoE/Mixtral/GPT-OSS, but one is torch dynamo grouped_mm tracing and the other is tensor-parallel support." }, { "left": "issue:43828", "right": "issue:45464", "accept": false, - "reason": "Different modalities and code paths: autocast dtype mismatch vs chat/completions streaming failure." + "reason": "Different multimodal/dtype failures with different models and failure modes." }, { "left": "issue:43366", "right": "issue:45464", "accept": false, - "reason": "Unrelated topics: GGUF architecture support vs Qwen3.5 streaming inference." + "reason": "Unrelated: GGUF gpt-oss architecture support vs Qwen3.5 streaming inference failure." }, { "left": "issue:44291", "right": "issue:45070", "accept": false, - "reason": "init_empty_weights/_is_hf_initialized loading issue is unrelated to the PretrainedConfig pydantic break." + "reason": "Both are v5 regressions, but one is init_empty_weights API breakage and the other is pydantic config-field behavior." }, { "left": "issue:43531", "right": "issue:45070", "accept": false, - "reason": "Qwen3-MoE sliding-window handling and pydantic config-field break are different bugs." + "reason": "Qwen3-MoE sliding_window behavior is unrelated to PretrainedConfig/pydantic issues." }, { "left": "issue:41093", "right": "issue:43596", "accept": false, - "reason": "Both raise IndexError, but one is a mask/tensor length mismatch and the other is a ZeRO-3 init empty-tensor indexing bug." + "reason": "Similar error type only; the mask-size mismatch and zero3/BertModel index error are different bugs." }, { "left": "issue:45020", "right": "issue:45800", "accept": false, - "reason": "One is remote_code breakage in recent transformers; the other is a torch/transformers version incompatibility." + "reason": "Remote-code breakage and torch/transformers version incompatibility are not the same concrete issue." }, { "left": "issue:43716", "right": "issue:46294", "accept": false, - "reason": "Both are multimodal mismatches, but image-preprocessor dtype mismatch and video-token/feature mismatch are different code paths." + "reason": "Both involve multimodal inputs, but one is image-preprocessor dtype mismatch and the other is video-template token accounting." }, { "left": "issue:43054", "right": "issue:43994", "accept": false, - "reason": "Same model family, but one is a quality/regression report and the other is a bad AutoModel/pipeline output bug; not clearly the same root cause." + "reason": "Both are about Siglip2 quality, but one is text-embedding degradation and the other is incorrect overall outputs/loading behavior." }, { "left": "issue:44479", "right": "issue:46294", "accept": true, - "reason": "46294 looks like a concrete Qwen3.5 video-input repro of the broader v5.3.0 video regression described in 44479." + "reason": "Both describe the same Qwen3.5 video-input regression where chat-template/video-token accounting mismatches the provided video features." }, { "left": "issue:46468", "right": "issue:46869", "accept": false, - "reason": "Official Nemotron example breakage and YOLO multi-GPU device-map failure are unrelated." + "reason": "Different domains: Nemotron example breakage vs YOLO multi-GPU device-map auto failure." }, { "left": "issue:45759", "right": "issue:45902", "accept": false, - "reason": "Composite config unwrapping and uninitialized rotary embedding memory are different model-init bugs." + "reason": "Composite-config unwrapping and uninitialized rotary-embedding memory are separate bugs." }, { "left": "issue:43596", "right": "issue:46165", "accept": false, - "reason": "ZeRO-3/BertModel indexing failure is unrelated to locally downloaded remote-code resume behavior." + "reason": "Zero3/BertModel index error and trust_remote_code resume issues are unrelated." }, { "left": "issue:46640", "right": "issue:46661", "accept": false, - "reason": "Flash Attention 2 integration and remote-code model_class/config_class loading are separate issues." + "reason": "Flash Attention 2 usage and remote-code config_class attribute errors are different code paths." }, { "left": "issue:43716", "right": "issue:44918", "accept": false, - "reason": "Different failures: multimodal dtype mismatch vs SFT trainer unpacking error." + "reason": "Both are multimodal-related, but image dtype mismatch is not the same as the TRL embedding-unpacking failure." }, { "left": "issue:39186", "right": "issue:44918", "accept": false, - "reason": "FSDP weight-dimension RuntimeError and Qwen3.5 input-embedding unpacking are not the same bug." + "reason": "FSDP 2-D weight enforcement is unrelated to the Qwen3.5/TRl tensor-shape issue." }, { "left": "issue:40990", "right": "issue:44960", "accept": false, - "reason": "High perplexity on gpt-oss and GLM5 model-loading/behavior are unrelated." + "reason": "High perplexity on gpt-oss and GLM5 are unrelated model-quality reports." }, { "left": "issue:43645", "right": "issue:45399", "accept": false, - "reason": "Custom-model Jupyter initialization and flash-attn2 fallback blocking are different issues." + "reason": "Custom-model notebook initialization and flash-attn2 fallback logic are unrelated." }, { "left": "issue:45632", "right": "issue:45698", "accept": true, - "reason": "Both point to trust_remote_code custom-module cache resolution returning the wrong module; the second looks like another manifestation of the same cache-key/path-collision bug." + "reason": "Both point to the same trust_remote_code custom-module loading/caching bug; 45632 is the specific leaf-directory collision variant of 45698's wrong-module symptom." }, { "left": "issue:43404", "right": "issue:45005", "accept": false, - "reason": "Both involve tied weights, but they affect different model families and code paths." + "reason": "Tied lm_head weights in Mistral3 and translation-model tied-weight issues are different model families and failures." }, { "left": "issue:42915", "right": "issue:45399", "accept": false, - "reason": "FineGrainedFP8Config failure on Qwen3Moe is unrelated to flash-attn2 fallback checks." + "reason": "FineGrainedFP8Config failure and flash-attn2 fallback gating are unrelated." }, { "left": "issue:43873", "right": "issue:45005", "accept": false, - "reason": "Quantization offloading/OOM behavior is not the same as tied-weights regression." + "reason": "Quantization/offloading behavior is not the same bug as tied-weight handling." }, { "left": "issue:44530", "right": "issue:45005", "accept": false, - "reason": "PagedAttentionCache linear_attention crash and tied-weights regression are different subsystems and symptoms." + "reason": "PagedAttentionCache linear_attention crash and tied weights are unrelated." }, { "left": "issue:44360", "right": "issue:44512", "accept": false, - "reason": "DSA indexer/ReLU discussion is unrelated to a docs typo about the removed run command." + "reason": "A DSA indexer note and a docs typo are not the same underlying change." }, { "left": "issue:44492", "right": "issue:44512", "accept": false, - "reason": "Cache-strategies typo and removed run-command docs issue are separate documentation bugs." + "reason": "Cache-strategy typo and removed run-command docs are unrelated documentation issues." }, { "left": "issue:44863", "right": "issue:45005", "accept": false, - "reason": "NemotronH checkpoint loading failure is unrelated to translation-model tied weights." + "reason": "NemotronH checkpoint loading and translation-model tied weights are different bugs." }, { "left": "issue:45183", "right": "issue:46463", "accept": true, - "reason": "Both describe the shared input-validation path producing wrong or misleading decoder/input_ids/inputs_embeds errors; 46463 is a T5-specific instance of the broader 45183 bug." + "reason": "Both are the same validation-family bug: incorrect XOR/input validation between input_ids and inputs_embeds leading to misleading error messages, with 46463 as a T5-specific manifestation." }, { "left": "issue:46489", "right": "issue:46491", "accept": false, - "reason": "Different tokenizer regressions on different model families; no clear shared concrete bug." + "reason": "Both are tokenizer regressions, but they affect different models and different round-trip behaviors." } ] }, "evaluator_result": { "accept": true, - "feedback": "The summary is grounded in the packet and stays conservative. The three accepted soft pairs are plausibly duplicate/near-duplicate manifestations based on the issue titles, and the rejected pairs are described cautiously without overclaiming. No obvious overstatement beyond the evidence in the packet." + "feedback": "The summary is conservative and stays within what the packet supports. The accepted soft edges are plausibly related by the titles, and the rejected pairs are not over-merged. The only potentially close call is 45632/45698, but the reasoning is still cautious enough to pass." }, "evaluator_used": true, "retried": false, @@ -1367,14 +1301,14 @@ "issue:43881|issue:43883", "issue:45685|issue:45823" ], - "prepared_review_unit_hash": "d7238da07d9010c42ab0df1887da34a87f3254ab3fec343a82eab203993be88d", + "prepared_review_unit_hash": "9e1943f5cfb7c064917e63f085e3a01ff3815c954071dc7854a976de8481c9a4", "component_budget": { "node_count": 500, "item_count": 500, "soft_pair_count": 1839, - "serialized_chars": 953913, - "estimated_input_tokens": 238479, - "estimated_eval_tokens": 477214 + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 }, "budget": { "node_count": 48, @@ -1389,204 +1323,204 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Most items are unrelated, but there are a few real duplicate subclusters: the strongest is the Transformers 5.x non-persistent-buffer corruption regression, plus a smaller static-cache seq-length contract issue and a likely SigLIP2 implementation discrepancy pair. No PRs are present.", - "confidence": 0.86, - "canonical_issue_reason": "issue:43950 is the clearest umbrella report for the non-persistent-buffer regression: it names the concrete failure (`register_buffer(persistent=False)`) and frames it as a Transformers 5.x regression, which subsumes the narrower duplicate reports in this set.", + "summary": "The cluster mostly contains unrelated issue reports, but a few soft pairs do look like the same underlying bug: non-persistent buffer corruption in v5, StaticCache.get_seq_length returning a tensor instead of int, tokenizer regression in v5+, and the SigLIP2 implementation discrepancy. The rest are different models or code paths.", + "confidence": 0.78, + "canonical_issue_reason": "issue:43950 is the clearest root-cause regression report: it names the transformers 5.x non-persistent-buffer corruption directly and cleanly subsumes narrower manifestations like issue:44534 and likely issue:45902.", "canonical_pr_reason": null, - "best_issue_reason": "issue:43950 is the strongest canonical issue for the cluster because it is specific, broadly applicable, and has the cleanest root-cause framing among the duplicate candidates.", + "best_issue_reason": "issue:43950 is the best canonical anchor because it is broad, technically specific, and framed as the regression itself rather than a single model-specific symptom.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43975", "right": "issue:44292", "accept": false, - "reason": "Different problems: tokenizer detokenization regression vs an NVFP4 loading/runtime error." + "reason": "Different failure modes and models: DeepSeek-Coder detokenization vs Qwen NVFP4 loading/runtime error." }, { "left": "issue:44918", "right": "issue:45507", "accept": false, - "reason": "Different model/training paths and different failure modes; no shared concrete bug." + "reason": "Different models and code paths; one is Qwen3.5 input-embedding unpacking with TRL, the other GraniteMoEHybrid method invocation." }, { "left": "issue:43950", "right": "issue:44534", "accept": true, - "reason": "Same underlying bug: Transformers 5.x corrupting non-persistent buffers / filling them with junk." + "reason": "Same underlying bug: transformers v5 non-persistent buffer handling/corruption, with 44534 phrased as the same regression." }, { "left": "issue:39692", "right": "issue:43054", "accept": false, - "reason": "Both involve SigLIP2, but one is a docs/example mismatch/quantization issue and the other is a quality regression; not the same concrete bug." + "reason": "One is a docs/example mismatch and quantization failure; the other is a SigLIP2 quality/performance complaint." }, { "left": "issue:43644", "right": "issue:43950", - "accept": true, - "reason": "Same regression: `from_pretrained()` mishandling non-persistent buffers in Transformers 5.x." + "accept": false, + "reason": "Unrelated regressions: missing config fields from trust_remote_code vs non-persistent buffer corruption." }, { "left": "issue:36331", "right": "issue:41762", "accept": false, - "reason": "Unrelated issues: custom `compute_loss` signature vs DeepSpeed ZeRO-3 loading failure." + "reason": "Different subsystems: Trainer loss signature change vs DeepSpeed ZeRO-3 model loading IndexError." }, { "left": "issue:43872", "right": "issue:44589", "accept": false, - "reason": "Different incompatibilities with different root causes; no evidence they are the same bug." + "reason": "Both are quantization/backend issues, but they describe different concrete incompatibilities and code paths." }, { "left": "issue:43828", "right": "issue:44928", "accept": false, - "reason": "Both touch Qwen/Phi attention behavior, but the concrete failures differ (dtype mismatch vs 3D position_ids/SDPA fallback/Nan explosion)." + "reason": "Different models and symptoms; dtype mismatch in Phi-tiny-MoE is not the same bug as Qwen3.5 RLHF gradient explosion." }, { "left": "issue:45987", "right": "issue:46858", "accept": true, - "reason": "Same API contract issue around `cache.get_seq_length()` returning the wrong type/device-shaped value, with 46858 being a backend-specific manifestation." + "reason": "Same bug family: get_seq_length returning a tensor/device tensor instead of an int, causing downstream static-cache arithmetic failures." }, { "left": "issue:45920", "right": "issue:46489", - "accept": false, - "reason": "Both are tokenizer regressions, but 46489 is a model-specific gap in a prior fix while 45920 is a broader multi-model token-ID regression; not clearly the same bug." + "accept": true, + "reason": "Same tokenizer regression in transformers v5+: both report incorrect token IDs/output for affected model families." }, { "left": "issue:43054", "right": "issue:45701", "accept": false, - "reason": "Different scope: SigLIP2 embedding quality vs a general tokenization-version change." + "reason": "SigLIP2 embedding quality complaint vs a broad tokenizer-change report; not the same concrete issue." }, { "left": "issue:38175", "right": "issue:43054", "accept": false, - "reason": "Related model family, but zero probabilities and degraded text embeddings are not the same concrete failure." + "reason": "Both mention SigLIP2, but one is zero probabilities and the other is relative embedding degradation; no clear shared root cause." }, { "left": "issue:43950", "right": "issue:45902", "accept": true, - "reason": "Same non-persistent-buffer corruption class; 45902 is a specific manifestation after `meta -> to_empty(cuda)` materialization." + "reason": "Same non-persistent-buffer regression, with 45902 describing a specific manifestation after meta\u2192to_empty(cuda) materialization." }, { "left": "issue:41553", "right": "issue:43883", "accept": false, - "reason": "Different model-loading failures with unrelated causes." + "reason": "Bad AutoTokenizer error message for Voxtral is unrelated to Molmo's missing tied-weights attribute." }, { "left": "issue:43054", "right": "issue:43493", "accept": true, - "reason": "Likely same SigLIP2 implementation discrepancy: 43493 states the HF-vs-JAX mismatch directly, and 43054 reports the quality impact of that mismatch." + "reason": "Likely the same SigLIP2 implementation discrepancy: HF vs JAX mismatch plausibly explains the degraded text embeddings." }, { "left": "issue:44910", "right": "issue:45859", "accept": false, - "reason": "Different bugs: Flash Attention packed-sequence misinterpretation vs CPU-offload incompatibility in a rotary embedding." + "reason": "Both involve Qwen3.5/Qwen3_5Moe, but one is FlashAttention position_id handling and the other is CPU-offload incompatibility in rotary embedding." }, { "left": "issue:42947", "right": "issue:44387", "accept": false, - "reason": "Unrelated: LoRA gradient checkpointing behavior vs int4 quantization memory growth/OOM." + "reason": "Different performance problems: LoRA gradient checkpointing vs int4 quantization memory bloat/OOM." }, { "left": "issue:45663", "right": "issue:45902", "accept": false, - "reason": "Different models and different mechanisms; no shared code-path bug is evident." + "reason": "Different model paths and failures: Gemma4 FSDP2 attention KeyError vs non-persistent buffer materialization." }, { "left": "issue:41553", "right": "issue:43881", "accept": false, - "reason": "Different loading errors and no clear common root cause." + "reason": "Different model-loading failures; no shared underlying bug is evident." }, { "left": "issue:44403", "right": "issue:44451", "accept": false, - "reason": "One is about noisy logging during load, the other is a real model-loading failure." + "reason": "Generic load-time noise vs a specific model loading failure for ScandiBERT." }, { "left": "issue:43975", "right": "issue:45520", "accept": false, - "reason": "Tokenizer detokenization bug vs import error in flash-attn feature detection; unrelated." + "reason": "Tokenizer detokenization bug vs Python 3.13 flash_attn import mapping issue." }, { "left": "issue:44403", "right": "issue:44488", "accept": false, - "reason": "Different severity and symptoms; noisy loading vs failing to load a specific model." + "reason": "Different artifacts and failures; both are loading-related but not the same bug." }, { "left": "issue:24643", "right": "issue:28530", "accept": false, - "reason": "Different training-time issues: DeepSpeed weight shape error vs early-stopping metric lookup." + "reason": "Training with DeepSpeed weight-shape error vs early-stopping metric selection; unrelated." }, { "left": "issue:42947", "right": "issue:43856", "accept": false, - "reason": "Both mention training efficiency, but they describe different models and different problems." + "reason": "Different memory-related training issues with different causes and model setups." }, { "left": "issue:43643", "right": "issue:45020", "accept": false, - "reason": "Both involve `remote_code`, but one is missing fields in `AutoConfig` results and the other is a broad model-loading breakage; not the same concrete bug." + "reason": "Related theme of remote_code, but one is missing config fields and the other is a broader compatibility regression; not the same concrete bug." }, { "left": "issue:41762", "right": "issue:42915", "accept": false, - "reason": "Different models and backend/config failures; no duplicate evidence." + "reason": "Gemma3 ZeRO-3 loading IndexError vs Qwen3MoE FineGrainedFP8 failure; different code paths." }, { "left": "issue:44991", "right": "issue:45399", "accept": false, - "reason": "Tokenizer loading regression vs flash-attn fallback gating issue; unrelated." + "reason": "Tokenizer loading failure and flash-attn fallback blocking are unrelated." }, { "left": "issue:43824", "right": "issue:43872", "accept": false, - "reason": "ImportError for a missing model class vs bitsandbytes `Int8Params` keyword incompatibility." + "reason": "Importing a model class vs bitsandbytes Int8Params constructor incompatibility." }, { "left": "issue:43782", "right": "issue:44960", "accept": false, - "reason": "Different model families and unrelated failures." + "reason": "Qwen3VL weight_only load error vs a vague GLM5 issue; no shared root cause." }, { "left": "issue:43881", "right": "issue:43883", "accept": false, - "reason": "Different concrete failures: model loading vs missing attribute on a different model class." + "reason": "Both are model-loading issues, but they affect different models and failure mechanisms." }, { "left": "issue:45685", "right": "issue:45823", "accept": false, - "reason": "Both are device/backend related, but the failures are different and not the same code-path bug." + "reason": "MPS histogram/int limitation vs Gemma4 PLE device mismatch are different backend bugs." } ] }, "evaluator_result": { "accept": true, - "feedback": "The summary is well grounded in the packet, and the soft-edge judgments stay conservative overall. The identified duplicate subclusters are plausible from the issue titles alone, especially the non-persistent-buffer regression cluster and the seq_length contract pair. No PR mergeability concerns apply because there are no PRs in the packet." + "feedback": "Accept. The canonical anchor choice and the soft-pair judgments are conservative and grounded in the packet: the accepted pairs are all clear same-bug families by title/body framing (non-persistent buffers, StaticCache.get_seq_length, tokenizer regression, SigLIP2 discrepancy), and the rejected pairs are appropriately kept separate by model/code-path differences. I don\u2019t see an obvious overstatement." }, "evaluator_used": true, "retried": false, @@ -1675,8 +1609,8 @@ "issue:43792|issue:44220", "issue:43257|issue:45305", "issue:41553|issue:44991", - "issue:42947|issue:45923", "issue:36331|issue:41093", + "issue:42947|issue:45923", "issue:43854|issue:44863", "issue:43940|issue:45313", "issue:41553|issue:44488", @@ -1687,217 +1621,223 @@ "issue:44530|issue:44863", "issue:44863|issue:46402" ], - "prepared_review_unit_hash": "667c52ecf593181218a08f2525f8cbd8d98077e20be3fbe24015f82ee3b00c61", + "prepared_review_unit_hash": "100af846bdef61f1ee147fce57474a9894ecc6623c94204688c10dec265418af", "component_budget": { "node_count": 500, "item_count": 500, "soft_pair_count": 1839, - "serialized_chars": 953913, - "estimated_input_tokens": 238479, - "estimated_eval_tokens": 477214 + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 29, - "serialized_chars": 37083, - "estimated_input_tokens": 9271, - "estimated_eval_tokens": 18798 + "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 set is mostly heterogeneous transformer/model-loading and docs issues. I see three true duplicate-style pairs: the v5 docs cleanup, the training-loss double-shift regression, and the identical mask/shape-mismatch runtime error. The rest are related by subsystem only, not the same underlying bug.", - "confidence": 0.77, - "canonical_issue_reason": "Issue 46897 is the strongest canonical anchor among the true-duplicate candidates: it has the clearest duplicate-style title, the highest inbound reference count in that mini-cluster, and cleanly captures the underlying training-loss double-shift regression.", + "summary": "The packet is mostly heterogeneous. The docs regression pair and the mask-shape IndexError pair look like true duplicates; the Whisper/fbank pair is only a weak thematic match and should not be accepted on the evidence shown.", + "confidence": 0.81, + "canonical_issue_reason": "No single issue represents the whole packet well. The clearest duplicate subclusters are 43827/44509 for the stale v5 docs references and 41093/44805 for the same mask-shape IndexError; 43827 is the cleanest docs representative.", "canonical_pr_reason": null, - "best_issue_reason": "46897 is the best overall issue representative for this cluster because it is explicit, high-signal, and tightly matches another report without extra model-specific noise.", + "best_issue_reason": "43827 is the strongest overall issue representative because it cleanly states the v5 docs regression and has a clear duplicate in 44509.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:45081", "right": "issue:45399", "accept": false, - "reason": "Same general tokenizer/flash-attn area, but different failure modes: regex patch crash vs fallback gating checks." + "reason": "Tokenizer regex patching and flash-attn2 fallback checks are different bugs with different code paths." }, { "left": "issue:45020", "right": "issue:45823", "accept": false, - "reason": "Both involve model loading, but one is remote_code breakage and the other is a device mismatch in Gemma4 PLE." + "reason": "Remote-code loading breakage and a device-map forward mismatch are not the same concrete failure." }, { "left": "issue:44661", "right": "issue:45020", "accept": false, - "reason": "Different bugs: one is add-new-model-like failing on TOKENIZER_MAPPING_NAMES, the other is remote_code regressions." + "reason": "add-new-model-like/tokenizer mapping logic is unrelated to the broader remote_code regression." }, { "left": "issue:42491", "right": "issue:44368", "accept": false, - "reason": "Both mention Qwen/LoRA, but one is a hf4.x vs hf5.x load failure and the other is a warning about tie_word_embeddings." + "reason": "LoRA compatibility across versions is not the same bug as a missing gradient-checkpointing flag warning." }, { "left": "issue:43828", "right": "issue:45823", "accept": false, - "reason": "Different concrete problems: dtype mismatch under autocast vs device placement mismatch during forward." + "reason": "Dtype mismatch under autocast and device mismatch with auto device_map are different failure modes." }, { "left": "issue:45020", "right": "issue:45356", "accept": false, - "reason": "These are distinct tokenizer/model-loading regressions affecting different models and code paths." + "reason": "Generic remote_code breakage and a tokenizer codec regression do not share one concrete bug path." }, { "left": "issue:45715", "right": "issue:46315", "accept": false, - "reason": "Related tokenizer performance regressions, but different APIs and symptoms: convert_ids_to_tokens vs convert_tokens_to_ids." + "reason": "These are both tokenizer performance issues, but they affect opposite conversion directions and likely different code paths." }, { "left": "issue:43827", "right": "issue:44509", "accept": true, - "reason": "Both report the same docs cleanup needed after v5 pipeline removals; the titles describe the same underlying documentation issue." + "reason": "Same stale v5 docs regression: both report documentation still referencing removed pipeline tasks." }, { "left": "issue:45488", "right": "issue:46489", "accept": false, - "reason": "Both concern tokenizer behavior in DeepSeek-family models, but the concrete bugs are different and likely need different fixes." + "reason": "Both involve tokenizers, but the affected models and bugs are different and not shown to be the same underlying issue." }, { "left": "issue:46894", "right": "issue:46897", - "accept": true, - "reason": "Same training-loss double-shift regression, only the model name differs; these look like the same underlying bug report." + "accept": false, + "reason": "The titles suggest a similar training-loss pattern, but different models and insufficient evidence prevent treating them as the same bug." }, { "left": "issue:43872", "right": "issue:46869", "accept": false, - "reason": "Different failures: bitsandbytes _is_hf_initialized incompatibility vs YOLO multi-GPU device_map auto behavior." + "reason": "bitsandbytes constructor incompatibility is unrelated to the YOLO multi-GPU device_map failure." }, { "left": "issue:43404", "right": "issue:43950", "accept": false, - "reason": "Different code paths: a tied lm_head bug in Mistral3 versus silent corruption of non-persistent buffers in from_pretrained()." + "reason": "lm_head tying in a specific multimodal model is unrelated to non-persistent buffer corruption in from_pretrained()." }, { "left": "issue:45759", "right": "issue:46661", "accept": false, - "reason": "Both are config/model-loading issues, but one is about unwrapping text_config and the other about remote-code config_class lookup." + "reason": "Different model-construction failures: text_config unwrapping versus remote-code config_class access." }, { "left": "issue:43872", "right": "issue:44291", "accept": false, - "reason": "Same unexpected _is_hf_initialized keyword theme, but one is bitsandbytes Int8Params and the other is init_empty_weights loading." + "reason": "Both mention _is_hf_initialized, but the concrete failures differ: bitsandbytes Int8Params versus init_empty_weights." }, { "left": "issue:45020", "right": "issue:46661", "accept": false, - "reason": "Related remote-code/model-loading surface, but the specific bugs differ enough that they should not be merged." + "reason": "Broad remote-code breakage and missing config_class on model_class are distinct loading failures." }, { "left": "issue:43792", "right": "issue:44220", "accept": false, - "reason": "Likely related audio/Whisper area, but the titles are too different to confirm the same underlying bug." + "reason": "Both mention Whisper/fbank, but the evidence here does not establish the same concrete buggy code path, so this is only uncertain." }, { "left": "issue:43257", "right": "issue:45305", "accept": false, - "reason": "Different DeepSpeed/Qwen regressions: missing converted MoE weights vs gradient averaging with GAS and loss kwargs." + "reason": "Both touch DeepSpeed/Qwen training, but one is missing MoE weight conversion and the other is gradient averaging with GAS." }, { "left": "issue:41553", "right": "issue:44991", "accept": false, - "reason": "Different tokenizer failures for different models; one is a bad error message, the other is an actual load failure." + "reason": "Bad tokenizer error messaging for Voxtral is unrelated to the est-roberta tokenizer load failure." }, { - "left": "issue:42947", - "right": "issue:45923", + "left": "issue:36331", + "right": "issue:41093", "accept": false, - "reason": "Both mention LoRA/checkpointing, but one is a gradient checkpointing issue and the other is a missing supports_gradient_checkpointing flag." + "reason": "Unexpected compute_loss kwarg and mask/tensor shape mismatch are different trainer/runtime bugs." }, { - "left": "issue:36331", - "right": "issue:41093", + "left": "issue:42947", + "right": "issue:45923", "accept": false, - "reason": "Different training/runtime errors: unexpected compute_loss kwarg vs a mask/tensor shape mismatch." + "reason": "Gradient checkpointing being ineffective with PEFT LoRA is not the same as a missing supports_gradient_checkpointing flag." }, { "left": "issue:43854", "right": "issue:44863", "accept": false, - "reason": "Different model-loading bugs for different architectures; no clear shared underlying change." + "reason": "Different model-loading problems for different model families; no shared underlying bug is evident." }, { "left": "issue:43940", "right": "issue:45313", "accept": false, - "reason": "Both are ZeRO-3 loading failures on Qwen variants, but the model-specific symptoms suggest separate bugs." + "reason": "Both are ZeRO-3 loading failures on Qwen variants, but the symptoms and affected components differ enough to keep them separate." }, { "left": "issue:41553", "right": "issue:44488", "accept": false, - "reason": "Different tokenizer-loading issues for different models and error surfaces." + "reason": "Different tokenizer-loading failures: a Voxtral error message issue versus a model loading failure for cjvt/sleng-bert." }, { "left": "issue:44928", "right": "issue:45588", "accept": false, - "reason": "Different attention/cache failures: RLHF gradient explosion vs flash_attention.py crashing on s_aux=None." + "reason": "RLHF NaN/SDPA fallback issue is unrelated to flash_attention.py crashing on sink-less models." }, { "left": "issue:41093", "right": "issue:44805", "accept": true, - "reason": "These report the same mask/tensor shape-mismatch exception signature; the numerical sizes differ but the failure looks identical." + "reason": "Same mask-shape IndexError; 44805 is a later, more detailed report of the same bug." }, { "left": "issue:44479", "right": "issue:44560", "accept": false, - "reason": "Both involve video-related regressions, but one is a broad v5.3.0 video-input regression and the other is a StopIteration in Qwen3-vl-embedding." + "reason": "Both are video regressions in Qwen-family models, but the affected models and failure signatures differ." }, { "left": "issue:24643", "right": "issue:30064", "accept": false, - "reason": "Unrelated failures: DeepSpeed training weight shape error versus image processor handling of void segmentation maps." + "reason": "DeepSpeed training weight-shape error is unrelated to void segmentation map processing." }, { "left": "issue:44530", "right": "issue:44863", "accept": false, - "reason": "Different model-specific issues: PagedAttentionCache linear_attention crash vs NemotronH checkpoint loading." + "reason": "PagedAttentionCache linear_attention crash and NemotronH checkpoint loading are different model/backend failures." }, { - "left": "issue:44863", - "right": "issue:46402", + "left": "issue:43872", + "right": "issue:44291", + "accept": false, + "reason": "Reiterating the shared keyword is not enough; the constructor/type mismatch contexts differ and should stay separate." + }, + { + "left": "issue:45020", + "right": "issue:46661", "accept": false, - "reason": "Different model-loading bugs affecting different architectures and checkpoints." + "reason": "Remote-code breakage and model_class.config_class access errors are distinct problems." } ] }, "evaluator_result": { - "accept": true, - "feedback": "Grounded and conservative overall. The three accepted soft pairs are well-supported by the packet: the v5 docs cleanup pair, the training-loss double-shift pair, and the mask/shape-mismatch pair. The canonical issue choice is also justified by the inbound reference counts. The rejected pairs stay appropriately cautious about same-subsystem but different-bug cases." + "accept": false, + "feedback": "Mostly grounded, but one soft-edge acceptance is too strong for the packet: 43792/44220 only share a Whisper/fbank theme from titles, and the evidence shown here does not establish the same concrete bug path. The docs pair (43827/44509) and mask-shape pair (41093/44805) look fine, but the Whisper pair should be downgraded to uncertain rather than accepted." }, "evaluator_used": true, - "retried": false, + "retried": true, "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null @@ -2001,14 +1941,14 @@ "issue:38175|issue:39692", "issue:43644|issue:44534" ], - "prepared_review_unit_hash": "0028b39f69d4b5c8769717ad6d6a2dd1fb4dbd12bff69b049e3c177f5dc848cb", + "prepared_review_unit_hash": "6e2d7f93d566bf7e242bc3d9f595168bf62ba865e2a93038586c65a87e34cf98", "component_budget": { "node_count": 500, "item_count": 500, "soft_pair_count": 1839, - "serialized_chars": 953913, - "estimated_input_tokens": 238479, - "estimated_eval_tokens": 477214 + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 }, "budget": { "node_count": 48, @@ -2023,228 +1963,228 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster is mostly a set of unrelated issue reports that only look similar by shared Transformers v5 / model-family keywords. A few true duplicate pairs stand out around non-persistent buffer corruption, Gemma4 4-bit multimodal dtype casting, custom model initialization, and SigLIP2 inference regressions.", - "confidence": 0.74, - "canonical_issue_reason": "Issue 43644 is the cleanest canonical report: it describes a broad Transformers v5 regression, is self-contained, and has an obvious duplicate in 44534 with nearly identical symptoms.", + "summary": "This cluster is mostly heterogeneous transformer bug reports, with only a few true duplicate pairs: the v5 non-persistent-buffer regression, the custom-model initialization regression, a SigLIP2 output bug pair, and a Gemma4 4-bit dtype/casting pair. Most other soft links are broad-similarity false positives across different models or code paths.", + "confidence": 0.86, + "canonical_issue_reason": "43644 is the clearest canonical anchor because it is a tightly scoped v5 regression with a direct duplicate in 44534 and cleanly represents one real duplicate subgroup in an otherwise mixed cluster.", "canonical_pr_reason": null, - "best_issue_reason": "43644 is the best issue candidate because it is the broadest, most actionable regression report in the set and directly anchors one of the clearest duplicate pairs.", + "best_issue_reason": "43644 is the best issue representative: concrete, reproducible, and backed by an exact duplicate (44534), making it the strongest anchor for this cluster.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43366", "right": "issue:43828", "accept": false, - "reason": "Different bugs: GGUF/gpt-oss support vs a Float8 storage lookup error in unrelated loading code." + "reason": "Different bugs and different model families: GGUF architecture support vs Phi autocast dtype mismatch." }, { "left": "issue:43596", "right": "issue:46515", "accept": false, - "reason": "Only shares a vague compatibility theme; one is a Bert/DeepSpeed zero3 IndexError, the other is a generic torch version incompatibility." + "reason": "Zero3/BertModel index error is unrelated to a general torch version incompatibility report." }, { "left": "issue:45325", "right": "issue:45902", "accept": false, - "reason": "Both involve rotary/position-id logic, but they target different model families and failure modes." + "reason": "Both involve model internals, but one is a Qwen2.5-VL rope scaling bug and the other is Qwen3.5-MoE uninitialized memory after meta materialization." }, { "left": "issue:43366", "right": "issue:43531", "accept": false, - "reason": "Unrelated issues: GGUF gpt-oss support vs a sliding_window problem in Qwen3-MoE." + "reason": "GGUF gpt-oss support and Qwen3-MoE sliding_window are unrelated issues." }, { "left": "issue:43638", "right": "issue:46515", "accept": false, - "reason": "The Bert zero3 training IndexError is not the same concrete bug as a broad torch compatibility report." + "reason": "Same broad area of loading/runtime failures, but the concrete bugs are different: Bert/zero3 index error vs torch 2.4.1 incompatibility." }, { "left": "issue:45663", "right": "issue:45823", "accept": false, - "reason": "Both are Gemma4-related, but one is an FSDP2 shared_kv_states bug and the other is a device_map auto offload mismatch." + "reason": "Both are Gemma4, but one is an FSDP2 shared_kv_states training bug and the other is a device_map forward mismatch." }, { "left": "issue:44155", "right": "issue:45584", "accept": false, - "reason": "Different models and code paths: AudioFlamingo3 batched inference leak vs Whisper empty-transcription generation failure." + "reason": "Both are audio-related, but one is batched inference leakage in AudioFlamingo3 and the other is Whisper empty-transcription generation failure." }, { "left": "issue:43366", "right": "issue:44589", "accept": false, - "reason": "No concrete overlap: GGUF architecture support vs missing Float8 storage handling." + "reason": "GGUF architecture support is unrelated to the Float8 storage-object loading error." }, { "left": "issue:43824", "right": "issue:43957", "accept": false, - "reason": "An import error for a specific class is unrelated to meta-device model loading failures." + "reason": "Importing Qwen2_5_VLForConditionalGeneration and meta-device loading failures are different code paths and different symptoms." }, { "left": "issue:43950", "right": "issue:45834", "accept": false, - "reason": "Non-persistent buffer corruption and a Kosmos2.5 long-input index error are different bugs." + "reason": "Non-persistent buffer corruption and Kosmos2.5 OCR index errors are unrelated." }, { "left": "issue:42915", "right": "issue:43957", "accept": false, - "reason": "Qwen3-MoE FP8 config failure is not the same issue as meta-device loading regressions." + "reason": "Qwen3-MoE FineGrainedFP8Config failure is unrelated to generic meta-device loading breakage." }, { "left": "issue:44910", "right": "issue:46858", "accept": false, - "reason": "Both touch attention/cache mechanics, but one is a flash-attention position_id crash and the other is a static-cache slice-arithmetic bug." + "reason": "Both involve attention/cache mechanics, but the concrete failures are different: packed 3D position_ids on Qwen3.5 vs device tensor seq_length arithmetic on non-CUDA backends." }, { "left": "issue:43716", "right": "issue:44560", "accept": false, - "reason": "Different subsystems and symptoms: multimodal dtype mismatch vs Qwen3-vl video StopIteration." + "reason": "Mistral-3 image dtype mismatch and Qwen3-vl video StopIteration are distinct multimodal bugs." }, { "left": "issue:43645", "right": "issue:43646", "accept": true, - "reason": "These are effectively the same custom-model initialization regression; one is the notebook-specific variant of the other." + "reason": "These are the same underlying Transformers v5 custom model initialization regression; the notebook mention is just a reproduction context." }, { "left": "issue:44560", "right": "issue:45685", "accept": false, - "reason": "Qwen3-vl video embedding failure and an MPS histogram error in MoE are unrelated." + "reason": "Qwen3-vl video StopIteration is unrelated to the MPS histogram implementation error for MoE." }, { "left": "issue:43550", "right": "issue:44910", "accept": false, - "reason": "Different concrete failures: torch.compile/SDPA on Bamba vs flash-attention with 3D position_ids on Qwen3.5." + "reason": "Different failure modes: torch.compile with SDPA on Bamba vs FlashAttention illegal memory access on Qwen3.5." }, { "left": "issue:42617", "right": "issue:43366", "accept": false, - "reason": "No shared code path: 3d_parallel.py runtime failure vs GGUF gpt-oss support." + "reason": "3d_parallel.py runtime failure and GGUF gpt-oss support are unrelated." }, { "left": "issue:44683", "right": "issue:46515", "accept": false, - "reason": "A torch>=2.9 flex_attention compilation failure is not the same as a torch 2.4.1 incompatibility report." + "reason": "Compiled flex_attention on torch >=2.9 and torch 2.4.1 incompatibility are different issues." }, { "left": "issue:44560", "right": "issue:44918", "accept": false, - "reason": "Same broad model family only; the reported failures are different (video StopIteration vs SFT embedding unpacking)." + "reason": "Different Qwen3.x multimodal/training failures with no shared concrete code path." }, { "left": "issue:36010", "right": "issue:42915", "accept": false, - "reason": "Importing GenerationMixin and Qwen3-MoE FP8 config handling are unrelated." + "reason": "GenerationMixin import error and Qwen3-MoE FP8 failure are unrelated." }, { "left": "issue:45812", "right": "issue:46489", "accept": false, - "reason": "Both are tokenizer regressions, but they affect different model families and appear to be different concrete bugs." + "reason": "Both are tokenizer regressions, but they affect different model families and different tokenizer code paths." }, { "left": "issue:45823", "right": "issue:46515", "accept": false, - "reason": "A Gemma4 device-mapping mismatch is not the same bug as a generic torch compatibility issue." + "reason": "Gemma4 device_map mismatch and generic torch incompatibility are not the same bug." }, { "left": "issue:44661", "right": "issue:45823", "accept": false, - "reason": "Tokenizer-mapping registration failure and Gemma4 PLE device mismatch are unrelated." + "reason": "Tokenizer-mapping failure and Gemma4 PLE device mismatch are unrelated." }, { "left": "issue:38175", "right": "issue:43994", "accept": true, - "reason": "Same SigLIP2 model regression: both describe nonsensical/zero-output inference on google/siglip2-base-patch16-224." + "reason": "Same SigLIP2 model, same symptom class of bad/incorrect outputs; these look like the same underlying model/processor bug." }, { "left": "issue:44560", "right": "issue:44805", "accept": false, - "reason": "Different error shapes and code paths; both are index-type failures only at a very high level." + "reason": "Qwen3-vl video StopIteration and a generic mask/tensor shape IndexError are different failures." }, { "left": "issue:45672", "right": "issue:46899", "accept": true, - "reason": "Same underlying Gemma4 multimodal 4-bit quantization dtype-cast problem, affecting audio/vision embedders." + "reason": "Both describe Gemma4 4-bit inference breakage caused by unsafe dtype handling in multimodal/audio-vision embedders." }, { "left": "issue:44877", "right": "issue:45030", "accept": false, - "reason": "Both mention config validation, but they are different models and not the same concrete regression." + "reason": "Config validation regressions on different model families are too broad to merge as one duplicate." }, { "left": "issue:43632", "right": "issue:45834", "accept": false, - "reason": "Transformers v5 initialization flag breakage and a Kosmos2.5 OCR index error are unrelated." + "reason": "_is_hf_initialized flag breakage is unrelated to Kosmos2.5 OCR indexing." }, { "left": "issue:43638", "right": "issue:44661", "accept": false, - "reason": "DeepSpeed zero3 Bert IndexError and add-new-model-like tokenizer mapping failure are distinct problems." + "reason": "Zero3/Bert index error and add-new-model-like/tokenizer mapping failure are unrelated." }, { "left": "issue:43596", "right": "issue:44661", "accept": false, - "reason": "A zero-sized Bert tensor IndexError is unrelated to tokenizer mapping registration logic." + "reason": "Same as above: Bert zero3 initialization bug vs tokenizer mapping bug." }, { "left": "issue:42491", "right": "issue:43720", "accept": false, - "reason": "Qwen3-MoE LoRA/hf5.x compatibility and BitNet packed-weight unpacking are different failures." + "reason": "LoRA hf4.x/hf5.x compatibility and BitNet packed-weight loading are different problems." }, { "left": "issue:44661", "right": "issue:44805", "accept": false, - "reason": "Tokenizer mapping registration and mask/tensor shape mismatch are unrelated." + "reason": "Tokenizer mapping failure and generic mask-shape IndexError are unrelated." }, { "left": "issue:43317", "right": "issue:44756", "accept": false, - "reason": "Both concern loading behavior, but the first is gpu+cpu offload dequantization and the second is mmap OOM on Strix Halo." + "reason": "device_map auto dequantized-model offload failure and Strix Halo mmap OOM are different loading/resource issues." }, { "left": "issue:38175", "right": "issue:39692", "accept": false, - "reason": "The docs example issue includes multiple errors and a processor mismatch, but it is not clearly the same bug as the SigLIP2 zero-probability regression." + "reason": "Both involve SigLIP2, but 39692 is a docs/example-specific model/processor mismatch plus quantization failure, not the same concrete runtime bug as 38175." }, { "left": "issue:43644", "right": "issue:44534", "accept": true, - "reason": "Near-exact duplicate: both report Transformers v5 filling non-persistent buffers with junk." + "reason": "Exact same non-persistent-buffer regression in Transformers v5; 44534 is a later duplicate of 43644." } ] }, "evaluator_result": { "accept": true, - "feedback": "The summary is grounded in the packet, and the duplicate calls are generally conservative. The canonical choice of 43644 vs 44534 is well supported by the titles, and the rejected soft-edge pairs mostly stay within cautious bounds given only title-level evidence." + "feedback": "The cluster summary is grounded in the packet and stays conservative: it treats most links as false positives, and the accepted soft edges are supported by closely aligned titles (especially the near-identical v5 buffer-regression pair and the custom-model init pair). The canonical 43644 anchor is well supported by its exact duplicate 44534." }, "evaluator_used": true, "retried": false, @@ -2345,20 +2285,20 @@ "issue:43425|issue:43957", "issue:41720|issue:44155" ], - "prepared_review_unit_hash": "587e33a21a3f78845232d752b4ba9671641d54464ccbc54d6f26ff8e1267bd09", + "prepared_review_unit_hash": "2c1e5e71bea78e2ac069db7d107676d421e06d1a52e1391d59f9800899bbf0e3", "component_budget": { "node_count": 500, "item_count": 500, "soft_pair_count": 1839, - "serialized_chars": 953913, - "estimated_input_tokens": 238479, - "estimated_eval_tokens": 477214 + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 30, - "serialized_chars": 36098, + "serialized_chars": 36099, "estimated_input_tokens": 9025, "estimated_eval_tokens": 18306 }, @@ -2367,198 +2307,198 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Mostly heterogeneous cluster of unrelated Transformers regressions. Only three soft edges look like true duplicates: the Scandinavian-BERT loading failures, the resize_token_embeddings propagation bug, and the training-loss double-shift bug.", - "confidence": 0.8, - "canonical_issue_reason": "issue:45276 is the best representative among the confirmed duplicate-like items: it states the resize_token_embeddings regression at the right level of abstraction and is broader than the narrower 45335 variant.", + "summary": "This cluster is mostly a loose collection of Transformers 5.x/model-specific regressions rather than one unified bug. I only see one strong duplicate pair: the `resize_token_embeddings` / output-embedding regression across Gemma4 and T5Gemma; the rest are too model-, API-, or failure-mode-specific to merge safely.", + "confidence": 0.58, + "canonical_issue_reason": "`issue:45276` is the clearest representative of the duplicate `resize_token_embeddings` regression: it names the concrete API, the affected embedding surfaces, and the expected propagation failure.", "canonical_pr_reason": null, - "best_issue_reason": "issue:45276 is the most useful cluster anchor because it is a clear, actionable bug report with a broader formulation that subsumes its duplicate partner.", + "best_issue_reason": "`issue:45276` is the strongest issue to anchor the cluster because it is specific, actionable, and broad enough to explain the matched T5Gemma report as the same root problem.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:42491", "right": "issue:45276", "accept": false, - "reason": "Different bugs: Qwen3 MoE LoRA compatibility across HF versions vs Gemma4 resize_token_embeddings propagation." + "reason": "Different bugs: LoRA/qwen3_moe compatibility versus a `resize_token_embeddings` embedding-propagation regression." }, { "left": "issue:41720", "right": "issue:42915", "accept": false, - "reason": "Different failure modes and models: A800 cudaErrorAssert under auto device mapping vs FineGrainedFP8Config on Qwen3Moe." + "reason": "Both mention Qwen3-related failures, but one is auto device mapping/cuda assert and the other is fine-grained FP8 config." }, { "left": "issue:42915", "right": "issue:44756", "accept": false, - "reason": "Unrelated: Qwen3Moe FP8 config issue vs Strix Halo mmap/OOM behavior." + "reason": "Distinct failures: Qwen3Moe FP8 config versus mmap/OOM on Strix Halo." }, { "left": "issue:44451", "right": "issue:44488", - "accept": true, - "reason": "Same underlying loading regression for closely related Scandinavian BERT models; both report the latest/current version cannot load the model." + "accept": false, + "reason": "Both are model-loading complaints, but they target different checkpoints and no shared root cause is evident." }, { "left": "issue:44960", "right": "issue:44977", "accept": false, - "reason": "Different models and symptoms: GLM5 vs Qwen3.5 flash-attention generation failure." + "reason": "Unrelated model/runtime failures: GLM5 versus Qwen3.5 flash-attention generation." }, { "left": "issue:44863", "right": "issue:46846", "accept": false, - "reason": "Different code paths: NemotronH checkpoint loading vs Qwen3.5 TP plan omissions causing OOM/shape errors." + "reason": "Different code paths: NemotronH checkpoint loading versus Qwen3.5 tensor-parallel planning." }, { "left": "issue:46459", "right": "issue:46512", "accept": false, - "reason": "Both mention dtype, but one is composite-checkpoint loading and the other is from_config stale config state; not the same bug." + "reason": "Both involve load-time state/dtype handling, but one is composite-checkpoint dtype propagation and the other is `from_config` stale config metadata." }, { "left": "issue:45313", "right": "issue:45357", "accept": false, - "reason": "Different regressions: ZeRO-3 loading of language_model weights vs save_pretrained writing incorrect visual encoder keys." + "reason": "ZeRO-3 weight loading and save_pretrained visual-key regression are separate bugs." }, { "left": "issue:44488", "right": "issue:44991", "accept": false, - "reason": "Both are tokenizer/model-loading regressions, but they affect different models and likely different root causes." + "reason": "Both are tokenizer-load regressions, but they affect different models and likely different failure points." }, { "left": "issue:45507", "right": "issue:46468", "accept": false, - "reason": "Unrelated: GraniteMoEHybrid invalid method call vs broken Nemotron example." + "reason": "Different model implementations and symptoms: invalid method call in GraniteMoEHybrid versus a broken Nemotron example." }, { "left": "issue:43644", "right": "issue:43645", "accept": false, - "reason": "Both are Transformers 5.0.0 regressions, but one is non-persistent buffer initialization and the other is Jupyter custom-model init." + "reason": "Transformers 5.0 regressions, but one is buffer initialization and the other is Jupyter custom-model construction." }, { "left": "issue:42491", "right": "issue:45292", "accept": false, - "reason": "Different bugs: Qwen3 MoE LoRA compatibility vs resize_token_embeddings not updating output embeddings." + "reason": "LoRA/qwen3_moe incompatibility is not the same bug as `resize_token_embeddings` not updating output embeddings." }, { "left": "issue:39692", "right": "issue:43994", "accept": false, - "reason": "Both involve SigLIP2, but one is a broken doc example while the other is incorrect inference outputs; not the same issue." + "reason": "Different SigLIP/SigLIP2 model-loading and inference problems; not the same concrete defect." }, { "left": "issue:43257", "right": "issue:43854", "accept": false, - "reason": "Different models and loading paths: Qwen3 MoE accelerate+deepspeed conversion vs GLM-4.7-Flash unit-test loading." + "reason": "Qwen3 MoE deepspeed conversion and GLM-4.7-Flash unit-test loading are unrelated." }, { "left": "issue:43856", "right": "issue:43873", "accept": false, - "reason": "Both are memory-related, but one is Qwen3 MoE training efficiency and the other is quantization/offloading behavior." + "reason": "Both are quantization/memory complaints, but one is training-memory usage and the other is offloading behavior." }, { "left": "issue:43856", "right": "issue:44387", "accept": false, - "reason": "Related only at a high level; the specific quantization/OOM regressions are different." + "reason": "Same broad area, but the specific problem differs: training memory usage versus int4 quantization reserved-memory OOM." }, { "left": "issue:43854", "right": "issue:43931", "accept": false, - "reason": "Different bugs: GLM-4.7-Flash test loading vs Qwen3-VL weight-shape mismatch." + "reason": "Different models and failure modes: GLM load issue versus Qwen3-VL weight-shape mismatch." }, { "left": "issue:43708", "right": "issue:44075", "accept": false, - "reason": "Completely different training/resume logic vs SGD argument handling." + "reason": "Trainer resume/max-steps logic and SGD argument handling are unrelated." }, { "left": "issue:42491", "right": "issue:45335", "accept": false, - "reason": "Different bugs: Qwen3 MoE compatibility across versions vs t5gemma resize_token_embeddings propagation." + "reason": "LoRA/qwen3_moe incompatibility is not the same concrete issue as T5Gemma embedding resizing." }, { "left": "issue:44960", "right": "issue:46050", "accept": false, - "reason": "Different models and failure modes: GLM5 vs sinq save/load breakage." + "reason": "GLM5 loading/generation versus quantized-model save/load corruption are different bugs." }, { "left": "issue:46897", "right": "issue:46901", - "accept": true, - "reason": "Same training-loss double-shift bug pattern; only the affected model differs." + "accept": false, + "reason": "They share a \"double-shift\" theme, but the affected models and exact training paths differ, so this is too weak to merge." }, { "left": "issue:45488", "right": "issue:45812", "accept": false, - "reason": "Both are tokenizer regressions in v5, but they appear to be different tokenizer conversion bugs affecting different model families." + "reason": "Both are tokenizer regressions, but one is a pre-tokenizer override bug and the other is wrong token IDs for Granite models." }, { "left": "issue:45276", "right": "issue:45335", "accept": true, - "reason": "Same resize_token_embeddings propagation bug; 45276 is the broader Gemma4 manifestation and 45335 is the narrower t5gemma case." + "reason": "Both report the same `resize_token_embeddings` propagation bug: model-specific output/decoder embeddings are not updated after resizing." }, { "left": "issue:45362", "right": "issue:45406", "accept": false, - "reason": "Different crashes and models: Qwen3.5 chat issue vs Gemma4Processor missing _tokenizer in serve." + "reason": "Qwen3.5 chat crash and Gemma4Processor serve crash are unrelated failures." }, { "left": "issue:44164", "right": "issue:44291", "accept": false, - "reason": "Both concern model initialization/loading internals, but the reported failures are different and not the same code-path bug." + "reason": "Save/from_pretrained extra_state handling and init_empty_weights argument compatibility are different loading paths." }, { "left": "issue:44910", "right": "issue:45588", "accept": false, - "reason": "Both involve flash-attention, but one is a Qwen3.5 packed-sequence position_ids crash and the other is an s_aux=None AttributeError." + "reason": "Both are flash-attention crashes, but one is illegal memory access from 3D position IDs and the other is an AttributeError on `s_aux=None`." }, { "left": "issue:44402", "right": "issue:44488", "accept": false, - "reason": "Different issues: out-of-vocab special tokens in one model vs loading failure for a different BERT model." + "reason": "Tokenizer vocab-size mismatch and model-loading failure are not the same bug." }, { "left": "issue:39692", "right": "issue:44960", "accept": false, - "reason": "Unrelated: SigLIP2 documentation/example errors vs GLM5 model issue." + "reason": "Completely different model families and symptoms: SigLIP2 example errors versus GLM5." }, { "left": "issue:43425", "right": "issue:43957", "accept": false, - "reason": "Different incompatibilities: Torch 2.10 support vs meta-device model loading regression." + "reason": "Torch version incompatibility and meta-device model-loading regression are separate issues." }, { "left": "issue:41720", "right": "issue:44155", "accept": false, - "reason": "Unrelated: Qwen3 device-mapping crash vs AudioFlamingo3 batched inference token/embedding leak." + "reason": "Qwen3 device-mapping CUDA assert and AudioFlamingo3 batched inference leakage are unrelated." } ] }, "evaluator_result": { "accept": true, - "feedback": "The summary is grounded in the packet and stays mostly conservative. The three accepted soft edges are reasonable cluster-level duplicates/near-duplicates based on the issue titles and the pair similarity, and the rejected edges are not overstated. The canonical issue choice for the resize_token_embeddings pair is also defensible from the broader Gemma4 wording." + "feedback": "The summary is grounded in the packet and stays conservative: it correctly treats most pairs as model-specific or failure-mode-specific, and the only accepted duplicate pair (issue:45276 with issue:45335) is well supported by the titles. No overstatement of evidence stood out." }, "evaluator_used": true, "retried": false, @@ -2661,226 +2601,226 @@ "issue:44589|issue:44756", "issue:43425|issue:43828" ], - "prepared_review_unit_hash": "21563b6c705511fcc945a0d4f62d1d6d319cf94d42249cd9376a9ade62f7dca6", + "prepared_review_unit_hash": "4b15f3806416f168c260cbeeed709a0526076d87446ba77e255c614a01a60181", "component_budget": { "node_count": 500, "item_count": 500, "soft_pair_count": 1839, - "serialized_chars": 953913, - "estimated_input_tokens": 238479, - "estimated_eval_tokens": 477214 + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 31, - "serialized_chars": 37129, - "estimated_input_tokens": 9283, - "estimated_eval_tokens": 18822 + "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 cluster of unrelated Transformers issues. Only the resize_token_embeddings reports around 45276/45292 look like the same underlying bug; the rest are distinct model- or subsystem-specific failures.", - "confidence": 0.86, - "canonical_issue_reason": "Choose issue 45292 as the canonical issue: it is the broader, cleaner report of the resize_token_embeddings/output_embeddings problem that also appears in the Gemma4-specific issue 45276.", + "summary": "This cluster is mostly noisy and heterogeneous. The only clear duplicate family is the resize_token_embeddings regression across Gemma/t5gemma variants; most other pairs are merely superficially similar and should stay separate.", + "confidence": 0.71, + "canonical_issue_reason": "issue:45292 is the best anchor for the only real duplicate family here: it states the generic resize_token_embeddings failure, while issue:45276 and issue:45335 are narrower model-specific variants of the same regression.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 45292 is the best representative because it states the core bug generically and matches the only clear duplicate-like family in the set.", + "best_issue_reason": "issue:45292 is the most representative and reusable issue title in the cluster: it describes the shared bug at the right abstraction level without overfitting to one model name.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:45663", "right": "issue:45759", "accept": false, - "reason": "Different model families and different failures: Gemma4 FSDP2 shared_kv_states vs Qwen multimodal text_config unwrapping." + "reason": "Different subsystems and failure modes: Gemma4 FSDP2 shared_kv_states rebuilding vs Qwen multimodal config unwrapping." }, { "left": "issue:44079", "right": "issue:46829", "accept": false, - "reason": "Both involve output objects, but one is a generic ModelOutput key bug and the other is a VideoPrism classifier field assignment bug." + "reason": "Both involve output containers, but one is a ModelOutput key assignment bug and the other is a VideoPrism hidden_states field mix-up." }, { "left": "issue:43404", "right": "issue:43454", "accept": false, - "reason": "Both are lm_head tying issues, but they affect different models/code paths and are not the same concrete bug." + "reason": "Both mention lm_head tying, but they are different model implementations and not clearly the same code-path bug." }, { "left": "issue:44291", "right": "issue:44756", "accept": false, - "reason": "Unrelated problems: init_empty_weights argument handling vs mmap/OOM on Strix Halo." + "reason": "init_empty_weights argument handling and mmap/OOM on Strix Halo are unrelated issues." }, { "left": "issue:43873", "right": "issue:44387", "accept": false, - "reason": "Both mention quantization and memory, but one is offloading behavior and the other is int4 reserved-memory OOM." + "reason": "Both touch quantization memory behavior, but one is offloading and the other is int4 reserved-memory/OOM; too broad to merge." }, { "left": "issue:42947", "right": "issue:45704", "accept": false, - "reason": "Gradient checkpointing/PEFT LoRA is unrelated to the T5 apex.FusedRMSNorm memory leak." + "reason": "Gradient checkpointing/LoRA effectiveness and T5 apex RMSNorm memory leak are unrelated." }, { "left": "issue:44292", "right": "issue:44912", "accept": false, - "reason": "Different quantization/load failures on different models; not the same code-path problem." + "reason": "Different quantization formats and symptoms; one is NVFP4 loading, the other MXFP4 fallback to bf16." }, { "left": "issue:36010", "right": "issue:39404", "accept": false, - "reason": "ImportError for GenerationMixin and Whisper return_language pipeline regression are unrelated." + "reason": "ImportError for GenerationMixin and Whisper pipeline return_language are unrelated." }, { "left": "issue:45663", "right": "issue:45859", "accept": false, - "reason": "Different model-specific offload/training bugs; no shared concrete code path." + "reason": "FSDP2 shared_kv_states KeyError and CPU-offload incompatibility in Qwen rotary embedding are different bugs." }, { "left": "issue:44387", "right": "issue:44756", "accept": false, - "reason": "Both are memory-related, but one is int4 quantization OOM and the other is a Strix Halo mmap workaround." + "reason": "Both mention OOM, but the causes are different: int4 quantization memory growth vs disabling mmap on Strix Halo." }, { "left": "issue:42947", "right": "issue:43381", "accept": false, - "reason": "Gradient checkpointing with LoRA is not the same as gradient checkpointing failing in eval mode." + "reason": "LoRA gradient checkpointing ineffectiveness is not the same as checkpointing being disallowed in eval mode." }, { "left": "issue:44155", "right": "issue:45823", "accept": false, - "reason": "AudioFlamingo3 batched inference token leak and Gemma4 device_map mismatch are different bugs." + "reason": "AudioFlamingo3 batch leakage and Gemma4 device mismatch are unrelated model-specific forward bugs." }, { "left": "issue:43957", "right": "issue:44756", "accept": false, - "reason": "Meta-device model loading breakage is unrelated to disabling mmap for OOM." + "reason": "meta-device loading regressions and Strix Halo mmap OOM are not the same issue." }, { "left": "issue:43335", "right": "issue:45663", "accept": false, - "reason": "Sparse-layer config creation and Gemma4 FSDP2 shared_kv_states are unrelated." + "reason": "SwitchTransformers sparse-layer construction and Gemma4 FSDP2 state rebuilding are unrelated." }, { "left": "issue:45276", "right": "issue:45292", "accept": true, - "reason": "Same resize_token_embeddings bug family: 45276 is a Gemma4-specific manifestation of the broader output_embeddings issue in 45292." + "reason": "Same underlying resize_token_embeddings regression; 45292 is the generic version and 45276 is a model-specific manifestation." }, { "left": "issue:44863", "right": "issue:46468", "accept": false, - "reason": "NemotronH checkpoint loading and the official Nemotron example breakage are not the same concrete issue." + "reason": "NemotronH checkpoint loading and the official Nemotron example being broken are related only at the model family level, not clearly the same bug." }, { "left": "issue:44479", "right": "issue:44877", "accept": false, - "reason": "Video input regression and strict config loading failure are different model/config bugs." + "reason": "Video input regression for Qwen VL models and strict config blocking granite_speech loading are unrelated." }, { "left": "issue:43381", "right": "issue:44928", "accept": false, - "reason": "Eval-mode gradient checkpointing and RLHF NaN explosion from 3D position_ids are different failures." + "reason": "Checkpointing-in-eval and RLHF gradient explosion from 3D position_ids are different problems." }, { "left": "issue:44164", "right": "issue:44756", "accept": false, - "reason": "save/from_pretrained extra_state handling is unrelated to mmap-induced OOM." + "reason": "save/from_pretrained extra_state handling and Strix Halo mmap OOM are unrelated." }, { "left": "issue:45684", "right": "issue:45698", "accept": false, - "reason": "Read-only permission propagation from save_pretrained is a different bug from loading the wrong custom module." + "reason": "Both involve custom-model save/load, but one is read-only file permissions during save_pretrained and the other is loading the wrong custom module." }, { "left": "issue:45588", "right": "issue:45663", "accept": false, - "reason": "flash_attention s_aux=None crash and Gemma4 FSDP2 shared_kv_states are unrelated." + "reason": "Flash attention s_aux=None crash and Gemma4 shared_kv_states rebuilding are unrelated." }, { "left": "issue:43854", "right": "issue:45216", "accept": false, - "reason": "Unit-test model loading for GLM-4.7-Flash is not the same as Qwen3.5 save_pretrained checkpoint corruption." + "reason": "GLM-4.7-Flash unit-test loading and Qwen3.5 checkpoint save_pretrained regression are different model-specific issues." }, { "left": "issue:42796", "right": "issue:46326", "accept": false, - "reason": "Tokenizer BPE-marker handling and continue_final_message on mistral_common are different tokenizer bugs." + "reason": "Both are tokenizer-related, but BPE marker handling and continue_final_message behavior are distinct bugs." }, { "left": "issue:45292", "right": "issue:45335", - "accept": false, - "reason": "Both involve resize_token_embeddings, but they affect different model-specific targets (output_embeddings vs decoder.embed_tokens) and are not clearly the same code path." + "accept": true, + "reason": "Same resize_token_embeddings bug family; both report embeddings not being updated after resize in different model wrappers." }, { "left": "issue:46894", "right": "issue:46901", "accept": false, - "reason": "Both describe double-shift training losses, but they are separate model implementations and not clearly the same fixable code-path." + "reason": "Both describe double-shift loss errors, but they are in different model implementations and not clearly one shared code-path bug." }, { "left": "issue:43828", "right": "issue:43957", "accept": false, - "reason": "Autocast dtype mismatch and meta-device loading breakage are unrelated." + "reason": "Autocast dtype mismatch and meta-device loading failures are unrelated." }, { "left": "issue:44402", "right": "issue:44451", "accept": false, - "reason": "Different tokenizer/vocab loading issues for different models; not the same underlying bug." + "reason": "Tokenizer vocab-size mismatch and ScandiBERT loading failure are not the same bug." }, { "left": "issue:43646", "right": "issue:43708", "accept": false, - "reason": "Custom model initialization in Jupyter and Trainer resume max_steps miscalculation are unrelated." + "reason": "Custom model initialization in notebooks and Trainer resume max_steps miscalculation are unrelated." }, { "left": "issue:43645", "right": "issue:44164", "accept": false, - "reason": "Both touch save/load flows, but one is custom model init in notebooks and the other is extra_state serialization." + "reason": "Custom model init in Jupyter and extra_state save/load failures are different defects." }, { "left": "issue:44589", "right": "issue:44756", "accept": false, - "reason": "Float8 storage lookup failure is unrelated to Strix Halo mmap/OOM." + "reason": "Float8 storage lookup error and mmap-related OOM are unrelated." }, { "left": "issue:43425", "right": "issue:43828", "accept": false, - "reason": "Torch 2.10 incompatibility is not the same as Phi-tiny-MoE autocast dtype mismatch." + "reason": "Torch 2.10 compatibility and Phi-tiny-MoE autocast dtype mismatch are not the same issue." } ] }, "evaluator_result": { "accept": true, - "feedback": "The cluster summary is well grounded in the packet. The only clearly related pair the analyst elevates, 45276/45292, is supported by the very close title/body overlap, while the other soft-edge denials stay conservative and model-specific. No evidence of overclaiming or mergeability issues." + "feedback": "Grounded and conservative overall. The resize_token_embeddings family is the only clearly supported duplicate cluster in the packet, and the other pairwise rejections stay appropriately cautious rather than over-claiming equivalence." }, "evaluator_used": true, "retried": false, @@ -2985,232 +2925,232 @@ "issue:44488|issue:45081", "issue:43611|issue:43644" ], - "prepared_review_unit_hash": "f75d2b96ffeb1c1e645d5bcd69dd314be1c39768bbe665d37d5a8e818a4874ce", + "prepared_review_unit_hash": "dc9f7882e8112fc4bf12345cf5e8b69d51a6dc7ee971808871054e820c56df2b", "component_budget": { "node_count": 500, "item_count": 500, "soft_pair_count": 1839, - "serialized_chars": 953913, - "estimated_input_tokens": 238479, - "estimated_eval_tokens": 477214 + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 33, - "serialized_chars": 37740, - "estimated_input_tokens": 9435, - "estimated_eval_tokens": 19126 + "serialized_chars": 37732, + "estimated_input_tokens": 9433, + "estimated_eval_tokens": 19122 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is a loose mix of mostly unrelated Transformers v5 regressions and docs issues. No soft pair looks like the same concrete bug/change under a strict duplicate policy, so I would keep all pairings separate. The broadest representative issue is the generic model-loading regression around base_model_prefix.", - "confidence": 0.64, - "canonical_issue_reason": "Issue 43611 is the broadest, most general loading-regression report in the set, so it is the best umbrella representative.", + "summary": "A loose v5-regression cluster dominated by tokenizer/model-loading failures, with a few unrelated export/training/memory bugs mixed in. I would keep it mostly split; I don\u2019t see any soft pair that is clearly the same underlying bug.", + "confidence": 0.72, + "canonical_issue_reason": "issue:43611 is the broadest and most central v5 loading-regression report in the set; it best serves as an umbrella anchor, even though the cluster itself is heterogeneous.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 43611 is the best cluster representative because it targets a core v5 model-loading path and is broader than the more model-specific reports.", + "best_issue_reason": "issue:43611 is the best representative because it describes a generic Transformers 5 loading breakage rather than a single model-specific symptom.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:46315", "right": "issue:46491", "accept": false, - "reason": "One is tokenizer-conversion slowdown; the other is a CodeLlama round-trip regression. Different bugs and code paths." + "reason": "Both are tokenizer regressions, but one is a slowdown in convert_tokens_to_ids and the other is a CodeLlama encode/decode space bug; different code paths and fixes." }, { "left": "issue:45812", "right": "issue:46315", "accept": false, - "reason": "Granite token IDs regression is not the same as slow convert_tokens_to_ids for slow tokenizers." + "reason": "Granite token ID corruption and slow slow-tokenizer conversion are distinct bugs; same subsystem, not the same underlying defect." }, { "left": "issue:39404", "right": "issue:45701", "accept": false, - "reason": "Whisper pipeline return_language breakage is separate from general tokenization-change reports." + "reason": "Whisper pipeline return_language regression is unrelated to a general tokenization behavior change report." }, { "left": "issue:39404", "right": "issue:46665", "accept": false, - "reason": "Pipeline return_language and from_pretrained-in-classes are unrelated failure modes." + "reason": "Pipeline return_language failure and from_pretrained-in-classes inconsistency are different API behaviors and different fixes." }, { "left": "issue:44530", "right": "issue:45084", "accept": false, - "reason": "Qwen3.5 paged-attention crash and template-node compilation error are different problems." + "reason": "PagedAttentionCache invalid group type and template-node compilation failure are separate internals with no clear shared root cause." }, { "left": "issue:43611", "right": "issue:43645", "accept": false, - "reason": "Base_model_prefix loading regression is not the same as custom-model initialization in Jupyter." + "reason": "base_model_prefix loading breakage and Jupyter custom-model initialization issues are both v5 regressions, but not the same code-path problem." }, { "left": "issue:44683", "right": "issue:45823", "accept": false, - "reason": "Compiled flex_attention on torch>=2.9 and Gemma4 device mismatch are distinct issues." + "reason": "Compiled flex_attention on torch>=2.9 and Gemma4 device mismatch are unrelated compatibility bugs." }, { "left": "issue:43425", "right": "issue:46869", "accept": false, - "reason": "Torch version incompatibility is not the same as YOLO multi-GPU device-map failure." + "reason": "Torch 2.10 incompatibility is not the same as multi-GPU device_map auto failure on YOLO models." }, { "left": "issue:43425", "right": "issue:43716", "accept": false, - "reason": "Torch compatibility and Mistral-3 dtype mismatch are unrelated." + "reason": "Torch version incompatibility and Mistral-3 image-preprocessor dtype mismatch are different issues." }, { "left": "issue:43425", "right": "issue:45823", "accept": false, - "reason": "Torch 2.10 incompatibility does not match Gemma4 forward device mismatch." + "reason": "General torch incompatibility does not match the Gemma4 PLE device-mapping bug." }, { "left": "issue:44589", "right": "issue:45084", "accept": false, - "reason": "Float8 storage lookup failure is different from the non-template-node compilation error." + "reason": "Missing Float8 storage and non-template-node compilation errors are unrelated failures." }, { "left": "issue:44479", "right": "issue:44805", "accept": false, - "reason": "Video-input regression for Qwen VL models is not the same as a mask-shape IndexError." + "reason": "Qwen video-input regression and a mask-shape IndexError are not the same underlying bug." }, { "left": "issue:43906", "right": "issue:44530", "accept": false, - "reason": "A reproduction of another issue and a PagedAttentionCache linear_attention crash are different bugs." + "reason": "An isolated reproduction of a prior issue and a Qwen3.5 PagedAttentionCache crash are different problems." }, { "left": "issue:43901", "right": "issue:44509", "accept": false, - "reason": "Both are docs-related, but they cover different API removals and wording changes." + "reason": "These are both docs regressions, but they concern different removed/changed pipeline APIs." }, { "left": "issue:44671", "right": "issue:44991", "accept": false, - "reason": "CamemBERT MLM prediction regression and est-roberta tokenizer loading failure are unrelated." + "reason": "CamemBERT masked-LM prediction errors and est-roberta tokenizer loading failures are unrelated." }, { "left": "issue:43994", "right": "issue:44079", "accept": false, - "reason": "Siglip2 output weirdness and ModelOutput key assignment are different code paths." + "reason": "SigLIP2 nonsensical outputs and ModelOutput key assignment with None are different bugs." }, { "left": "issue:44265", "right": "issue:44898", "accept": false, - "reason": "torch.export with torch_compilable_check and Perceiver interpolation failure are separate issues." + "reason": "torch.export failure with torch_compilable_check and Perceiver non-default resolution failure are distinct export/model issues." }, { "left": "issue:36032", "right": "issue:43452", "accept": false, - "reason": "T5Tokenizer special-tokens conflict and gguf_file loading breakage are not the same bug." + "reason": "T5 tokenizer method-name conflict and gguf_file load failures are different tokenizer/loading bugs." }, { "left": "issue:39404", "right": "issue:44912", "accept": false, - "reason": "Whisper pipeline behavior and MXFP4 quantization fallback are unrelated." + "reason": "Whisper pipeline regression and MXFP4 quantization fallback are unrelated." }, { "left": "issue:46050", "right": "issue:46326", "accept": false, - "reason": "Quantized save/load breakage and mistral_common continue_final_message are different failures." - }, - { - "left": "issue:44164", - "right": "issue:45759", - "accept": false, - "reason": "extra_state serialization and composite-config unwrapping are separate model-loading problems." + "reason": "sinq save/load breakage and continue_final_message on mistral_common are separate features and code paths." }, { "left": "issue:43906", "right": "issue:46326", "accept": false, - "reason": "These concern unrelated tokenizer/pipeline behaviors." + "reason": "The reproduction of an older issue is not the same as the mistral_common continue_final_message bug." }, { "left": "issue:44991", "right": "issue:45081", "accept": false, - "reason": "Different tokenizer-loading bugs with different models and failure points." + "reason": "Tokenizer load failure for est-roberta and a Mistral regex patch backend_tokenizer error are different tokenizer issues." }, { "left": "issue:43299", "right": "issue:43931", "accept": false, - "reason": "Qwen3VL MoE loading breakage and Qwen3-VL shape mismatch are not the same concrete defect." + "reason": "Both involve Qwen3-VL loading, but one is a dev0 MoE loading regression and the other is a specific weight-shape mismatch; too little evidence of one shared bug." }, { "left": "issue:45823", "right": "issue:46899", "accept": false, - "reason": "Gemma4 device mismatch and Gemma4 4-bit dtype-cast corruption are distinct multimodal issues." + "reason": "Gemma4 device-mismatch during forward and Gemma4 4-bit multimodal embedder dtype-cast breakage are separate runtime bugs." }, { "left": "issue:45479", "right": "issue:46894", "accept": false, - "reason": "Sequence-classification zero-loss bug and CohereASR double-shift training loss bug are different." + "reason": "Single-label classification zero-loss behavior is not the same as a training-loss double-shift issue." }, { "left": "issue:43856", "right": "issue:44756", "accept": false, - "reason": "Qwen3 MoE memory usage and Strix Halo mmap OOM are unrelated performance/memory reports." + "reason": "Qwen3 MoE memory usage and Strix Halo mmap OOM are different resource issues." }, { "left": "issue:43596", "right": "issue:44683", "accept": false, - "reason": "Deepspeed ZeRO3/BertModel IndexError and flex_attention torch-version failure are different." + "reason": "Deepspeed ZeRO3/BertModel index error and torch>=2.9 compiled flex_attention failure are unrelated." }, { "left": "issue:45479", "right": "issue:46897", "accept": false, - "reason": "Different models and different training-loss bugs." + "reason": "Degenerate classification loss and Florence2 double-shift training loss are not the same defect." }, { "left": "issue:44164", "right": "issue:46821", "accept": false, - "reason": "extra_state handling and TimesFM window_size AttributeError are unrelated." + "reason": "extra_state save/load handling and TimesFM window_size attribute access are different from-pretrained/config issues." }, { "left": "issue:44488", "right": "issue:44987", "accept": false, - "reason": "Two separate model-loading failures for different repos and symptoms." + "reason": "Loading cjvt/sleng-bert and loading physical-intelligence/fast on transformers>=5.1.0 are model-specific failures with no clear shared root cause." + }, + { + "left": "issue:44488", + "right": "issue:45081", + "accept": false, + "reason": "Generic model loading failure and Mistral regex patch backend_tokenizer crash are different problems." }, { "left": "issue:44488", "right": "issue:45081", "accept": false, - "reason": "Different tokenizer loaders and distinct exception causes." + "reason": "Same pair as above; the symptoms and code paths do not align." }, { "left": "issue:43611", "right": "issue:43644", "accept": false, - "reason": "Base_model_prefix loading regression and non-persistent buffer junk are unrelated." + "reason": "Model-loading with base_model_prefix and non-persistent buffer initialization are separate v5 regressions." } ] }, @@ -3315,221 +3255,224 @@ "issue:43883|issue:44991", "issue:43646|issue:44534" ], - "prepared_review_unit_hash": "0b3d265b15b486d78b485c2758e58e19c3d673705d32e97a66d9e046368b6b22", + "prepared_review_unit_hash": "af9882f1d3323fb9213d212c18ff5b929da7aee986ea066b9d82de2d3a47dbf4", "component_budget": { "node_count": 500, "item_count": 500, "soft_pair_count": 1839, - "serialized_chars": 953913, - "estimated_input_tokens": 238479, - "estimated_eval_tokens": 477214 + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 30, - "serialized_chars": 36701, - "estimated_input_tokens": 9176, - "estimated_eval_tokens": 18608 + "serialized_chars": 36693, + "estimated_input_tokens": 9174, + "estimated_eval_tokens": 18604 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Mostly a heterogeneous bundle of unrelated Transformers regressions (tokenizer loading, model init, device/offload, audio, specific model heads). No soft edge looks like a true duplicate of the same underlying bug/change.", - "confidence": 0.89, - "canonical_issue_reason": "issue:43646 is the broadest, most representative regression title in the set (\u201cTransformers 5.0.0 breaks custom model initialization\u201d) and best matches the cluster\u2019s recurring v5 initialization/loading theme.", + "summary": "Most pairs are near-matches only by subsystem or release theme, not the same concrete bug. The strongest duplicates are the audio feature-extraction reports and the GGUF loading reports; the rest should stay separate.", + "confidence": 0.71, + "canonical_issue_reason": "Issue 43452 is the broadest, most concrete loading regression here: it describes a generic GGUF/from_pretrained break that can subsume specific model reports more cleanly than the narrower, model-specific issues.", "canonical_pr_reason": null, - "best_issue_reason": "issue:43646 is the best single issue to anchor the cluster because it is generic enough to cover several nearby v5 initialization regressions, while the other issues are more model-specific or narrower.", + "best_issue_reason": "43452 is the best canonical issue candidate because it states the failure mode at the right abstraction level (GGUF loading via from_pretrained) while still naming the impacted APIs clearly.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:44462", "right": "issue:46489", "accept": false, - "reason": "Both are tokenizer-loading complaints, but one is a repo-level tokenizer.json loading issue and the other is a DeepSeek-Coder-specific output regression tied to a separate fix gap." + "reason": "Both are tokenizer-loading regressions, but 46489 is a model-specific DeepSeek-Coder v1 bug tied to a gap in PR #44801, not clearly the same root cause as \"ignores tokenizer.json\"." }, { "left": "issue:42796", "right": "issue:43404", "accept": false, - "reason": "Different subsystems and failures: tokenizer BPE marker handling vs. Mistral3 lm_head weight tying." + "reason": "Tokenizer BPE-marker handling and Mistral3 lm_head tying are unrelated code paths and symptoms." }, { "left": "issue:46285", "right": "issue:46402", "accept": false, - "reason": "Both concern model loading in v5, but they are different models and different breakages (CLIP structure vs. Sam3 text encoder weights)." + "reason": "One is a CLIP model structure inconsistency, the other is missing weight loading in Sam3Model; different model families and bugs." }, { "left": "issue:36032", "right": "issue:42915", "accept": false, - "reason": "Unrelated: T5 tokenizer method conflict vs. Qwen3Moe FP8 configuration failure." + "reason": "T5 tokenizer special-token name conflict and Qwen3Moe FP8 failure are different failures with no shared concrete code path." }, { "left": "issue:32090", "right": "issue:39404", "accept": false, - "reason": "Different code paths and symptoms: Trainer GPU broadcast NoneType error vs. Whisper pipeline return_language regression." + "reason": "Trainer GPU broadcast NoneType error and Whisper return_language pipeline regression are unrelated." }, { "left": "issue:43716", "right": "issue:45084", "accept": false, - "reason": "Different bugs: image preprocessor/model dtype mismatch vs. template compilation error." + "reason": "Image preprocessor dtype mismatch is not the same bug as template compilation failing on non-template nodes." }, { "left": "issue:44737", "right": "issue:45859", "accept": false, - "reason": "Both are device-related, but one is XLNet CPU tensor creation and the other is Qwen rotary embedding CPU offload incompatibility." + "reason": "Both involve device placement, but XLNet CPU arange and Qwen3_Moe CPU offload are distinct implementation bugs." }, { "left": "issue:44661", "right": "issue:44683", "accept": false, - "reason": "No shared underlying bug: tokenizer mapping/model addition logic vs. compiled flex_attention on torch >=2.9." + "reason": "Tokenizer mapping/model-addition logic and compiled flex_attention on torch 2.9 are unrelated." }, { "left": "issue:43643", "right": "issue:44530", "accept": false, - "reason": "Different failures: missing fields from AutoConfig with trust_remote_code vs. PagedAttentionCache linear_attention crash on Qwen3.5." + "reason": "Missing fields from trust_remote_code config return and PagedAttentionCache linear_attention failure are different issues." }, { "left": "issue:44451", "right": "issue:44987", "accept": false, - "reason": "Both are loading regressions, but they affect different models and likely different causes; not mergeable as one fix." + "reason": "Both are model-loading failures, but the affected models and likely failure points differ; too broad to merge." }, { "left": "issue:40990", "right": "issue:43421", "accept": false, - "reason": "Perplexity regression on gpt-oss-20b is unrelated to runtime post-processor update behavior in TokenizersBackend." + "reason": "High perplexity on a model benchmark and runtime post-processor updates after special-token edits are not the same bug." }, { "left": "issue:42915", "right": "issue:44530", "accept": false, - "reason": "Different model families and distinct runtime/configuration errors." + "reason": "FP8 config failure for Qwen3Moe and PagedAttentionCache group-type crash are separate model/runtime problems." }, { "left": "issue:43541", "right": "issue:45084", "accept": false, - "reason": "Torch Dynamo tracing/grouped_mm float32 failure is unrelated to the template-node compilation error." + "reason": "Torch dynamo grouped_mm tracing failure and template compilation errors are different code paths." }, { "left": "issue:43644", "right": "issue:43646", "accept": false, - "reason": "Both are v5 custom-model/init regressions, but the reported failures differ (non-persistent buffer junk vs. custom model initialization breakage) and do not look like the same concrete bug." + "reason": "Both are v5 regressions around custom model initialization, but one is specifically about non-persistent buffers being junk-filled while the other is broader; not enough to treat as the same bug." }, { "left": "issue:43582", "right": "issue:43632", "accept": false, - "reason": "Different problems: Apple Silicon allocator warmup TypeError vs. _is_hf_initialized flag regression." + "reason": "Apple Silicon warmup TypeError and _is_hf_initialized flag breakage are unrelated regressions." }, { "left": "issue:34567", "right": "issue:40444", "accept": false, - "reason": "TrainerState token counting bug is unrelated to Qwen2.5-VL iterable multi-image finetuning failure." + "reason": "TrainerState token counting and multi-image iterable dataset finetuning failures are not the same issue." }, { "left": "issue:42175", "right": "issue:46178", "accept": false, - "reason": "Different backend/package issues: missing tensorflow in pip extra vs. unsupported tensorflow_text backend mapping." + "reason": "Backend packaging omission for pip extras and an unsupported tensorflow_text backend mapping error are distinct problems." }, { "left": "issue:43883", "right": "issue:44488", "accept": false, - "reason": "Both mention model/tokenizer loading, but they are different failure modes and different models; not the same bug." + "reason": "Missing tied-weights metadata on Molmo and a specific tokenizer load failure for cjvt/sleng-bert are unrelated." }, { "left": "issue:40444", "right": "issue:46326", "accept": false, - "reason": "Different multimodal training bugs: iterable multi-image finetuning vs. mistral_common continue_final_message behavior." + "reason": "Multi-image Qwen2.5-VL finetuning and Mistral-common continue_final_message behavior are different features and failures." }, { "left": "issue:44206", "right": "issue:44220", - "accept": false, - "reason": "Related audio stack, but one is unsupported center arg in feature extractor and the other is a separate _torch_extract_fbank_features issue." + "accept": true, + "reason": "These look like the same audio feature-extraction bug: both point at _torch_extract_fbank_features() / feature extractor crashes, with the second being the helper-function manifestation of the first." }, { "left": "issue:44811", "right": "issue:45584", "accept": false, - "reason": "Both touch Whisper, but one is batch_decode skip_special_tokens handling and the other is empty-transcription failure after align_special_tokens." + "reason": "Both are Whisper-related, but one is batch_decode skip_special_tokens handling and the other is empty-transcription failure after align_special_tokens; different behavior." }, { "left": "issue:43452", "right": "issue:43482", - "accept": false, - "reason": "Both involve gguf loading, but one is a generic from_pretrained breakage and the other is a Qwen2.5-specific Transformers v5 incompatibility." + "accept": true, + "reason": "Both are GGUF loading regressions in transformers v5: 43452 describes the generic from_pretrained break, and 43482 is a specific Qwen2.5-GGUF manifestation of that same loading path failure." }, { "left": "issue:36032", "right": "issue:43645", "accept": false, - "reason": "Tokenizer class conflict vs. custom model initialization in Jupyter notebooks; separate issues." + "reason": "T5 tokenizer method conflict and custom model init failures in notebooks are different regressions." }, { "left": "issue:44987", "right": "issue:44991", "accept": false, - "reason": "Both are version-gated loading failures, but for different repositories/models and likely different root causes." + "reason": "Both are version-dependent tokenizer/model load failures, but they affect different models and do not clearly share the same root cause." }, { "left": "issue:43317", "right": "issue:43482", "accept": false, - "reason": "Dequantized model device_map offload failure is unrelated to Qwen2.5-GGUF loading incompatibility." + "reason": "GPU/CPU offload dequantized-model loading and GGUF loading on v5 are different failure modes." }, { "left": "issue:45084", "right": "issue:45507", "accept": false, - "reason": "Template compilation error and GraniteMoEHybrid invalid method call are distinct bugs." + "reason": "Template compilation errors and GraniteMoEHybrid invalid method calls are unrelated." }, { "left": "issue:43262", "right": "issue:43377", "accept": false, - "reason": "Audio chat-template sampling-rate default vs. MIMI encoder padding-mask batching bug are unrelated." + "reason": "Sampling-rate defaulting in audio chat templates and MIMI batched-vs-single padding mask mismatch are different bugs." }, { "left": "issue:43746", "right": "issue:46326", "accept": false, - "reason": "PEFT local checkpoint loading failure is unrelated to mistral_common continue_final_message behavior." + "reason": "PEFT adapter loading for GraniteSpeech and mistral_common continue_final_message behavior are unrelated." }, { "left": "issue:43883", "right": "issue:44991", "accept": false, - "reason": "Different tokenizer/model loading regressions affecting different models and code paths." + "reason": "Molmo tied-weights metadata and tokenizer loading for EMBEDDIA/est-roberta are not the same problem." }, { "left": "issue:43646", "right": "issue:44534", "accept": false, - "reason": "Both are v5 regressions, but the concrete failures differ: custom model init vs. non-persistent buffer initialization." + "reason": "Both are Transformers v5 regressions, but one is custom model initialization and the other is non-persistent buffers being filled with junk; related theme, not the same concrete bug." } ] }, - "evaluator_result": null, - "evaluator_used": false, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays conservative overall. The analyst mostly rejects weak subsystem-level similarities, and the two accepted pairs are plausibly the same regressions based on the issue titles alone: the audio feature-extraction pair shares the same helper/path, and the GGUF pair is a generic vs. specific manifestation of the same loading failure. No PR-pair mergeability issue applies here." + }, + "evaluator_used": true, "retried": false, - "accepted_nontrivial_soft_edge": false, + "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null }, @@ -3604,8 +3547,8 @@ "issue:43377|issue:43819", "issue:43404|issue:44466", "issue:44464|issue:45987", - "issue:44987|issue:45081", "issue:44008|issue:46899", + "issue:44987|issue:45081", "issue:43994|issue:44610", "issue:44451|issue:45081", "issue:28530|issue:34567", @@ -3627,14 +3570,14 @@ "issue:44220|issue:44683", "issue:43611|issue:44534" ], - "prepared_review_unit_hash": "7b723b9edf0b2f2655b6990843987f39cf87c8797c9bdab99f5c127351925a35", + "prepared_review_unit_hash": "d37f949bf763e90852afa16d45be39485f2ccb3984715c31c86b2efba4f29a4d", "component_budget": { "node_count": 500, "item_count": 500, "soft_pair_count": 1839, - "serialized_chars": 953913, - "estimated_input_tokens": 238479, - "estimated_eval_tokens": 477214 + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 }, "budget": { "node_count": 48, @@ -3649,198 +3592,192 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "These are mostly unrelated transformer/model bugs across different subsystems and model families. I only see one soft duplicate pair: the two Qwen3.5 `save_pretrained` regressions. No PRs are present.", - "confidence": 0.68, - "canonical_issue_reason": "issue:45216 is the best canonical issue: it states the broader Qwen3.5 `save_pretrained` regression, and issue:45357 reads like a later, narrower report of the same broken serialization path.", + "summary": "This cluster is mostly a loose collection of Transformers regressions; only two pairs look like true duplicates. No PRs are present.", + "confidence": 0.44, + "canonical_issue_reason": "issue:45216 is the clearest duplicate anchor because issue:45357 looks like the same Qwen3.5 save_pretrained / visual-encoder-key regression continuing into a later release.", "canonical_pr_reason": null, - "best_issue_reason": "issue:45216 is the strongest representative because it names the regression source (`5.4.0`) and the general failure mode, while issue:45357 appears to be a follow-up on the same bug.", + "best_issue_reason": "issue:43611 is the broadest and most central issue in the set, covering a core Transformers v5 model-loading regression theme rather than a single model-specific symptom.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:45084", "right": "issue:46869", "accept": false, - "reason": "Different failures and code paths: compile/template-node error vs YOLO multi-GPU device-map crash." + "reason": "Different subsystems and failure modes: template compilation vs YOLO multi-GPU device mapping." }, { "left": "issue:43792", "right": "issue:46869", "accept": false, - "reason": "Whisper runtime failure and YOLO device-map failure are unrelated." + "reason": "Unrelated models and bugs; Whisper loading/can't run is not the same as YOLO device-map failures." }, { "left": "issue:44403", "right": "issue:45081", "accept": false, - "reason": "Generic loading noise vs Mistral regex patch crash; not the same bug." + "reason": "Both touch loading behavior, but one is generic loading noise and the other is a tokenizer regex crash; not the same bug." }, { "left": "issue:43645", "right": "issue:44534", "accept": false, - "reason": "Custom model init in notebooks vs non-persistent buffer junk; different regressions." + "reason": "Both are v5 regressions, but one is custom-model init in notebooks and the other is non-persistent buffer corruption." }, { "left": "issue:44849", "right": "issue:45325", "accept": false, - "reason": "Different Qwen model bugs: hidden-states output vs still-image rope/position-id scaling." + "reason": "Same family, but different concrete symptoms: hidden-states output bug vs RoPE/position-id scaling bug." }, { "left": "issue:43377", "right": "issue:43819", "accept": false, - "reason": "Separate audio-model bugs with different symptoms and code paths." + "reason": "Both are audio-model issues, but they involve different models and different code paths." }, { "left": "issue:43404", "right": "issue:44466", "accept": false, - "reason": "Tied-weights bug in Mistral3 vs serialization inconsistency of `lm_head.weight`; related theme, different issue." + "reason": "Both mention tied lm_head weights, but one is missing tying and the other is serialization varying by device; not the same concrete defect." }, { "left": "issue:44464", "right": "issue:45987", "accept": false, - "reason": "Compiled chunked generation mismatch vs cache sequence-length return type; not the same defect." + "reason": "Compiled generation inconsistency and cache sequence-length return type are different issues." }, { - "left": "issue:44987", - "right": "issue:45081", + "left": "issue:44008", + "right": "issue:46899", "accept": false, - "reason": "Both involve tokenizers, but the failing models and root causes differ." + "reason": "Different multimodal models and distinct bugs: variable shadowing vs unsafe dtype casting." }, { - "left": "issue:44008", - "right": "issue:46899", + "left": "issue:44987", + "right": "issue:45081", "accept": false, - "reason": "Gemma 3n name-collision bug vs Gemma4 dtype-cast bug; distinct multimodal failures." + "reason": "Different tokenizer/loading failures; no clear shared code-path or defect." }, { "left": "issue:43994", "right": "issue:44610", "accept": false, - "reason": "Different preprocessing/model mismatch bugs for different vision models." + "reason": "Different model families and symptoms: bad SigLIP outputs vs OmDet-Turbo image-size mismatch." }, { "left": "issue:44451", "right": "issue:45081", "accept": false, - "reason": "ScandiBERT loading failure is not the same as the Mistral regex patch crash." + "reason": "Unrelated loading failure vs Mistral tokenizer regex crash." }, { "left": "issue:28530", "right": "issue:34567", "accept": false, - "reason": "Early-stopping metric lookup and `num_input_tokens_seen` updating are unrelated Trainer issues." + "reason": "Trainer early-stopping config issue and token-count tracking bug are unrelated." }, { "left": "issue:42915", "right": "issue:44291", "accept": false, - "reason": "FineGrainedFP8Config failure and `_is_hf_initialized` init-empty-weights TypeError are different bugs." + "reason": "Different problems: quantization/config handling vs init_empty_weights argument compatibility." }, { "left": "issue:36032", "right": "issue:43742", "accept": false, - "reason": "Tokenizer method-name conflict vs model load KeyError; no shared underlying bug." + "reason": "Tokenizer method-name collision and MobileLLM key error are not the same bug." }, { "left": "issue:44062", "right": "issue:44987", "accept": false, - "reason": "`AddedToken` argument collision and a separate tokenizer-loading failure are not duplicates." + "reason": "Both are tokenizer-related, but one is an AddedToken argument conflict and the other is a model/tokenizer loading failure." }, { "left": "issue:44265", "right": "issue:44610", "accept": false, - "reason": "Torch export check failure vs wrong image processor output size; different code paths." + "reason": "torch.export/compilable-check failure is unrelated to an image processor size mismatch." }, { "left": "issue:42915", "right": "issue:44589", "accept": false, - "reason": "Both mention float8, but one is a config-level model failure and the other is a storage deserialization issue." + "reason": "Both involve version compatibility, but the concrete failures differ completely." }, { "left": "issue:43089", "right": "issue:45987", "accept": false, - "reason": "Generation performance overhead and cache length type issues are unrelated." + "reason": "Generation performance overhead is not the same as cache length contract behavior." }, { "left": "issue:43322", "right": "issue:43540", "accept": false, - "reason": "Different multimodal model load/processing failures." + "reason": "Both are multimodal loading/runtime issues, but they affect different models and different failure points." }, { "left": "issue:44479", "right": "issue:45325", - "accept": false, - "reason": "Same broad Qwen VL family, but one is a video regression and the other is a still-image position-id bug." + "accept": true, + "reason": "These look like the same Qwen VL RoPE/temporal-position regression in v5.3.0, with one report focusing on video inputs and the other on still-image position_ids." }, { "left": "issue:43632", "right": "issue:46821", "accept": false, - "reason": "_is_hf_initialized regression and TimesFM `window_size` AttributeError are unrelated." + "reason": "Different regressions: HF initialization flag breakage vs TimesFM window_size attribute error." }, { "left": "issue:36032", "right": "issue:41720", "accept": false, - "reason": "Tokenizer conflict vs Qwen3 auto device-map CUDA assert; no shared bug." + "reason": "Tokenizer loading conflict and Qwen3 auto device-map CUDA assert are unrelated." }, { "left": "issue:45216", "right": "issue:45357", "accept": true, - "reason": "Both report the same Qwen3.5 `save_pretrained` serialization bug; the later issue narrows the symptom to visual encoder keys." + "reason": "Both report the same Qwen3.5 save_pretrained / checkpoint serialization regression, with 45357 sounding like a later manifestation of the same bug." }, { "left": "issue:44991", "right": "issue:46315", "accept": false, - "reason": "Tokenizer loading failure vs performance regression in `convert_tokens_to_ids`; different problems." + "reason": "Tokenizer loading failure and tokenizer conversion slowdown are different issues." }, { "left": "issue:43531", "right": "issue:44877", "accept": false, - "reason": "Qwen3-MoE sliding-window bug and granite_speech config strictness are unrelated." - }, - { - "left": "issue:43425", - "right": "issue:44403", - "accept": false, - "reason": "Torch version incompatibility and loading noise are not the same underlying defect." + "reason": "Different model/config problems with no shared concrete bug." }, { "left": "issue:43296", "right": "issue:43531", "accept": false, - "reason": "PaddleOCR-VL/vLLM load failure and Qwen3-MoE sliding-window issue are distinct." + "reason": "vLLM PaddleOCR-VL load failure and Qwen3-MoE sliding_window behavior are unrelated." }, { "left": "issue:44220", "right": "issue:44683", "accept": false, - "reason": "Audio feature extraction issue and compiled flex_attention failure are unrelated." + "reason": "Audio feature extraction and compiled flex_attention failures are unrelated." }, { "left": "issue:43611", "right": "issue:44534", "accept": false, - "reason": "Base-model-prefix loading regression and non-persistent buffer junk are different bugs." + "reason": "Both are v5 regressions, but one is base_model_prefix loading and the other is junk-filled non-persistent buffers; not the same underlying defect." } ] }, "evaluator_result": { "accept": true, - "feedback": "The summary is grounded in the packet and stays conservative. The only accepted soft pair, issue:45216 vs issue:45357, is reasonably supported by the titles as the same Qwen3.5 save_pretrained serialization regression with a later narrower symptom. The rest of the soft-edge rejections are cautious and match the distinct issue titles. No PRs are present." + "feedback": "The summary is grounded in the packet: there are no PRs, and the claim that only two issue pairs look duplicate is conservative. The accepted soft pairs are supported by closely related titles, and the rejected pairs avoid overclaiming sameness. The best-issue and canonical-issue rationales are subjective but still consistent with the available evidence." }, "evaluator_used": true, "retried": false, @@ -3940,208 +3877,208 @@ "issue:43296|issue:46821", "issue:43377|issue:45491" ], - "prepared_review_unit_hash": "c8de3799585f0558eb0308e8fd8c5afff2ef38dcfedcabe31a8c36403fe392d3", + "prepared_review_unit_hash": "99911d182514ad6d720d6d268e78aaf4eed313eca8c95de75c7da0bc5e36b3bc", "component_budget": { "node_count": 500, "item_count": 500, "soft_pair_count": 1839, - "serialized_chars": 953913, - "estimated_input_tokens": 238479, - "estimated_eval_tokens": 477214 + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 29, - "serialized_chars": 35910, - "estimated_input_tokens": 8978, - "estimated_eval_tokens": 18212 + "serialized_chars": 35904, + "estimated_input_tokens": 8976, + "estimated_eval_tokens": 18208 }, "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 across different models and subsystems; none of the soft pairs look like true duplicates.", - "confidence": 0.72, - "canonical_issue_reason": "issue:45812 is the strongest representative: it is a concrete, high-impact v4\u2192v5 regression affecting all Granite models, with clear scope and inbound references.", + "summary": "This cluster is very heterogeneous: most items are separate model-specific regressions, tokenizer quirks, packaging issues, or docs bugs. I do not see any true duplicate PR/issue pair among the soft edges; at best they share broad themes like v5 regressions or tokenizer behavior.", + "confidence": 0.69, + "canonical_issue_reason": "Issue 44534 is the broadest core regression in the set and the best representative of the cluster\u2019s overall v5-era breakage theme. It is still not a duplicate of the others, just the most canonical umbrella-style report.", "canonical_pr_reason": null, - "best_issue_reason": "issue:45812 is the best issue to anchor the cluster because it is specific, broadly impactful, and well-described compared with the more niche reports.", + "best_issue_reason": "44534 has the widest scope and is closest to an underlying platform-level regression, whereas most other issues are narrower model- or subsystem-specific reports.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43716", "right": "issue:46899", "accept": false, - "reason": "Both are multimodal dtype problems, but they hit different model families and different code paths (image preprocessor mismatch vs 4-bit embedders cast)." + "reason": "Both are multimodal dtype problems, but they affect different models and code paths; not the same bug." }, { "left": "issue:40444", "right": "issue:45325", "accept": false, - "reason": "Same Qwen2.5-VL family, but one is iterable-dataset multi-image finetuning failure and the other is a rope-index temporal position bug; different defects." + "reason": "Both are Qwen2.5-VL, but one is a finetuning/data-pipeline failure and the other is a temporal position-id scaling bug." }, { "left": "issue:45812", "right": "issue:46491", "accept": false, - "reason": "Both are tokenizer regressions, but Granite token IDs and CodeLlama leading-space round-trip are distinct behaviors with different fixes." + "reason": "Both are tokenizer regressions, but Granite token IDs and CodeLlama leading-space round-tripping are different behaviors." }, { "left": "issue:44403", "right": "issue:44991", "accept": false, - "reason": "General noisy load logs versus a specific tokenizer-loading failure for a particular model; not the same underlying bug." + "reason": "Loading-related, but one is log noise and the other is a concrete tokenizer loading failure for a specific model." }, { "left": "issue:43381", "right": "issue:45923", "accept": false, - "reason": "These concern different gradient-checkpointing problems: eval-mode usage versus a missing capability flag on a trust_remote_code model." + "reason": "Both mention gradient checkpointing, but one is an eval-mode restriction and the other is a missing config flag on a trust_remote_code variant." }, { "left": "issue:43632", "right": "issue:44756", "accept": false, - "reason": "One is a Transformers v5 initialization flag regression; the other is an Apple Silicon mmap/OOM workaround issue." + "reason": "Completely different problems: an internal initialization flag regression vs mmap-related OOM avoidance." }, { "left": "issue:42796", "right": "issue:43746", "accept": false, - "reason": "A tokenizer BPE-marker bug and a PEFT checkpoint-loading bug are unrelated model-loading failures." + "reason": "Tokenizer BPE marker handling is unrelated to PEFT/local checkpoint loading." }, { "left": "issue:43792", "right": "issue:44589", "accept": false, - "reason": "Whisper runtime failure and missing Float8 storage are different low-level issues with no shared code path." + "reason": "No shared code path: Whisper runtime failure versus a float8 storage deserialization/type error." }, { "left": "issue:33666", "right": "issue:40444", "accept": false, - "reason": "Both involve Qwen VL training, but multi-GPU training and iterable-dataset multi-image failures are different problems." + "reason": "Both are Qwen vision-training issues, but they describe different failures and likely different fixes." }, { "left": "issue:43404", "right": "issue:46899", "accept": false, - "reason": "Missing lm_head tying and unsafe dtype casting in multimodal embedders are separate bugs in different parts of the model." + "reason": "Different multimodal bugs: weight tying in Mistral3 vs unsafe dtype casting in Gemma4 embedders." }, { "left": "issue:40990", "right": "issue:44811", "accept": false, - "reason": "Perplexity regression on a language model versus Whisper batch_decode special-token handling are unrelated." + "reason": "Unrelated: perplexity evaluation issue vs processor.batch_decode special-token handling." }, { "left": "issue:44291", "right": "issue:44589", "accept": false, - "reason": "`init_empty_weights` initialization and Float8 storage lookup are different failure modes with different fixes." + "reason": "Both are v5 compatibility failures, but one is init_empty_weights argument handling and the other is float8 storage lookup." }, { "left": "issue:42796", "right": "issue:46710", "accept": false, - "reason": "Tokenizer marker handling for Ministral is not the same as the DeepSeek accuracy regression after upgrade." + "reason": "Tokenizer marker handling and accuracy regression are different bug classes with no shared concrete code path." }, { "left": "issue:43482", "right": "issue:43525", "accept": false, - "reason": "A GGUF load failure and a missing `pad_token_id` attribute are distinct load-time issues." + "reason": "Both are model-loading problems, but they affect different models and different missing/unsupported config fields." }, { "left": "issue:35707", "right": "issue:40444", "accept": false, - "reason": "Progressive generation with embeddings/past key values is unrelated to Qwen2.5-VL multi-image finetuning." + "reason": "Progressive generation with inputs_embeds/past_key_values is unrelated to Qwen2.5-VL finetuning with multiple images." }, { "left": "issue:44079", "right": "issue:45663", "accept": false, - "reason": "ModelOutput key assignment and Gemma4 shared_kv_states rebuilding are unrelated implementation bugs." + "reason": "ModelOutput key assignment and Gemma4 FSDP2 shared_kv_states are unrelated." }, { "left": "issue:41553", "right": "issue:43582", "accept": false, - "reason": "Bad tokenizer error messaging and an Apple Silicon TypeError in allocator warmup are unrelated." + "reason": "Bad tokenizer error message and AppleSilicon TypeError are unrelated." }, { "left": "issue:43482", "right": "issue:45538", "accept": false, - "reason": "Both are tokenizer-related, but one is a v5 GGUF loading regression and the other is CLIPTokenizer's max-length constant." + "reason": "GGUF loading failure and CLIPTokenizer max length are different tokenizer/model-loading issues." }, { "left": "issue:43421", "right": "issue:44568", "accept": false, - "reason": "Runtime post-processor updates for special tokens are not the same as a model-specific BOS/EOS insertion regression." + "reason": "Both involve special-token behavior, but one is runtime post-processor updates and the other is add_special_tokens not adding BOS/EOS for a specific tokenizer." }, { "left": "issue:43632", "right": "issue:44534", "accept": false, - "reason": "Different Transformers v5 regressions: one breaks `_is_hf_initialized`, the other corrupts non-persistent buffers." + "reason": "Different regression types: initialization flag breakage versus non-persistent buffer corruption." }, { "left": "issue:43531", "right": "issue:44291", "accept": false, - "reason": "A sliding_window bug in Qwen3-MoE does not match an `_is_hf_initialized` argument mismatch during init." + "reason": "Qwen3-MoE sliding-window behavior is unrelated to init_empty_weights/_is_hf_initialized." }, { "left": "issue:42175", "right": "issue:45542", "accept": false, - "reason": "Both touch TensorFlow packaging, but one is extras installation/backend inclusion and the other is a tensorboard-without-TF runtime backend error." + "reason": "Both touch TensorFlow packaging, but one is missing TF in the torch extra and the other is tensorboard-installed-without-TF backend handling." }, { "left": "issue:44462", "right": "issue:45812", "accept": false, - "reason": "Ignoring repository tokenizer.json and wrong Granite token IDs are related to tokenizers broadly but not the same bug." + "reason": "Both are tokenizer regressions, but one is ignoring tokenizer.json and the other is wrong token IDs for Granite models." }, { "left": "issue:41720", "right": "issue:45081", "accept": false, - "reason": "A device-mapping CUDA assert and a Mistral regex patch crash are different subsystems and failure points." + "reason": "Auto device mapping cuda assert and Mistral regex patch crash are unrelated." }, { "left": "issue:43317", "right": "issue:43856", "accept": false, - "reason": "Dequantized model offload loading and MoE training memory usage are distinct performance/loading issues." + "reason": "Device-map/offload load failure and Qwen3 MoE memory inefficiency are different problems." }, { "left": "issue:44079", "right": "issue:44534", "accept": false, - "reason": "ModelOutput key handling and buffer-filling regressions are unrelated core behaviors." + "reason": "ModelOutput key handling is unrelated to junk-filled non-persistent buffers." }, { "left": "issue:44509", "right": "issue:46909", "accept": false, - "reason": "Both are documentation issues, but they concern different docs and different examples, not one underlying change." + "reason": "Both are docs issues, but they are different documents and different defects." }, { "left": "issue:43296", "right": "issue:46821", "accept": false, - "reason": "PaddleOCR-VL loading and TimesFM window_size AttributeError are unrelated model-specific failures." + "reason": "PaddleOCR-VL/vLLM load failure and TimesFM window_size AttributeError are unrelated." }, { "left": "issue:43377", "right": "issue:45491", "accept": false, - "reason": "MIMI padding-mask batching bug and Gemma3 NaN embeddings from sliding-window padding are different bugs with different causes." + "reason": "Both mention batching/padding, but they are different models and different failure modes (encoder mismatch vs NaN embeddings)." } ] }, @@ -4245,20 +4182,20 @@ "issue:44186|issue:44265", "issue:41720|issue:45823" ], - "prepared_review_unit_hash": "7d9bc7b8ec12df5e8fa62810759984ac55fcd841c77ccc216071a0e00242fe0c", + "prepared_review_unit_hash": "a0fb84dccc83f43d5e18157864a399b6692f95b229f3fad85e0e9deb2c2e7562", "component_budget": { "node_count": 500, "item_count": 500, "soft_pair_count": 1839, - "serialized_chars": 953913, - "estimated_input_tokens": 238479, - "estimated_eval_tokens": 477214 + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 30, - "serialized_chars": 36228, + "serialized_chars": 36225, "estimated_input_tokens": 9057, "estimated_eval_tokens": 18370 }, @@ -4267,198 +4204,198 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The set is mostly heterogeneous and does not form one coherent duplicate cluster. Two soft duplicate subgroups stand out: SigLIP2 correctness regressions and Qwen2.5-VL position-id regressions. Everything else is too weakly related to merge as the same bug.", - "confidence": 0.79, - "canonical_issue_reason": "There is no single cluster-wide canonical issue, but if one issue must anchor triage, 45325 is the best representative because it is the most concrete code-path-specific report (Qwen2.5-VL `get_rope_index` / position-id regression) and cleanly captures one of the only true duplicate-like subgroups.", + "summary": "This cluster is mostly a collection of unrelated Transformers issues across different models/subsystems. Only the two SigLIP2 reports look like a plausible duplicate pair.", + "confidence": 0.74, + "canonical_issue_reason": "issue:43493 is the better canonical target for the SigLIP2 duplicate pair because it states the underlying HF-vs-JAX implementation discrepancy, which likely explains the downstream bad outputs reported in issue:43994.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 45325 is the strongest overall representative for this set: it names a specific failure mode, a likely fix location, and sits in the clearest duplicate pair in the cluster.", + "best_issue_reason": "issue:43873 is the strongest overall standalone issue in the cluster: it has the broadest scope, the most discussion/inbound references, and a concrete reproducible failure mode around quantized offloading.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43611", "right": "issue:43646", "accept": false, - "reason": "Both are Transformers 5 loading regressions, but the failure modes differ: `base_model_prefix` handling vs custom model initialization." + "reason": "Different failure modes: loading with `base_model_prefix` vs. custom model initialization under Transformers 5." }, { "left": "issue:43818", "right": "issue:46402", "accept": false, - "reason": "Different model families and different bugs: Video-LLaVA temporal-attention/weight-sharing issue vs Sam3 text-encoder loading." + "reason": "Both involve model internals, but one is a Video-LLaVA tower/temporal-attention bug and the other is Sam3 text-encoder weight loading." }, { "left": "issue:44829", "right": "issue:45479", "accept": false, - "reason": "Both mention degenerate training, but one is tied to `flash_attention_3` while the other is a `num_labels=1` classification-loss behavior across models." + "reason": "Both describe degenerate training, but one is tied to `flash_attention_3` and the other is a `num_labels=1` single-label classification loss issue." }, { "left": "issue:43644", "right": "issue:44079", "accept": false, - "reason": "Unrelated defects: junk-filled non-persistent buffers vs `ModelOutput` key assignment when a field was previously `None`." + "reason": "Non-persistent buffer serialization junk and `ModelOutput` key assignment are different code paths." }, { "left": "issue:42175", "right": "issue:43296", "accept": false, - "reason": "These are unrelated backend/package-loading problems; no shared code path or same bug is evident." + "reason": "Packaging/backend dependency issue versus a model-loading failure for PaddleOCR-VL." }, { "left": "issue:43818", "right": "issue:45357", "accept": false, - "reason": "One is missing temporal attention / weight sharing in Video-LLaVA; the other is incorrect visual-encoder keys during `save_pretrained`." + "reason": "Video tower architecture bug vs. incorrect `save_pretrained` visual-encoder key handling." }, { "left": "issue:44265", "right": "issue:45072", "accept": false, - "reason": "Different issues: `torch.export` incompatibility with `torch_compilable_check` vs bfloat16 dtype mismatches in inference." + "reason": "`torch.export`/`torch_compilable_check` failure is unrelated to a bfloat16 dtype mismatch during inference." }, { "left": "issue:43493", "right": "issue:43994", "accept": true, - "reason": "Same underlying SigLIP2 correctness problem: both report wrong/nonsensical outputs for the same model family, one via JAX parity mismatch and one via bad inference results." + "reason": "Same SigLIP2 output-correctness bug family: one reports HF-vs-JAX discrepancy, the other reports nonsensical outputs via AutoModel/pipeline, both pointing to the same model implementation path." }, { "left": "issue:40444", "right": "issue:45381", "accept": false, - "reason": "Different Qwen-family multimodal failures; one is iterable-dataset multi-image finetuning, the other is video input position IDs." + "reason": "Multiple-images finetuning with IterableDataset is a different bug from Qwen2.5-VL video position-id handling." }, { "left": "issue:45325", "right": "issue:45834", "accept": false, - "reason": "Both mention vision/OCR-ish inputs, but one is a Qwen2.5-VL position-id regression and the other is a Kosmos2.5 long-input index error." + "reason": "Qwen2.5-VL rope/position-id scaling bug and Kosmos2.5 OCR index error are unrelated." }, { "left": "issue:45325", "right": "issue:45381", - "accept": true, - "reason": "Both are Qwen2.5-VL position-id regressions in 5.3.0 and likely stem from the same rope/vision position computation bug across image/video paths." + "accept": false, + "reason": "Both are Qwen2.5-VL multimodal regressions, but one is still-image temporal position scaling and the other is video `vision_position_ids`; not the same concrete bug." }, { "left": "issue:43454", "right": "issue:44466", "accept": false, - "reason": "Related to tied weights / serialization, but one is a failed tie causing garbage generation while the other is device-dependent serialization inconsistency." + "reason": "One is a tied-weight generation bug in AyaVision, the other is device-dependent serialization inconsistency for `lm_head.weight`." }, { "left": "issue:42915", "right": "issue:43482", "accept": false, - "reason": "Different model-loading failures on different models and triggers; no evidence they share the same root cause." + "reason": "FineGrainedFP8Config failure and GGUF loading failure are separate model/version issues." }, { "left": "issue:44291", "right": "issue:44877", "accept": false, - "reason": "One is an `init_empty_weights` constructor-argument regression; the other is a strict config validation failure for GraniteSpeech." + "reason": "Unexpected `_is_hf_initialized` argument and strict config rejection are different loading regressions." }, { "left": "issue:43701", "right": "issue:45325", "accept": false, - "reason": "`resume_from_checkpoint` key mismatch is unrelated to the Qwen2.5-VL position-id bug." + "reason": "Checkpoint key mismatch is unrelated to Qwen2.5-VL rope-index scaling." }, { "left": "issue:43873", "right": "issue:45701", "accept": false, - "reason": "Quantization/offloading behavior and tokenizer-version changes are not the same bug." + "reason": "Quantized offloading behavior and version-dependent tokenization changes do not share the same code path." }, { "left": "issue:42796", "right": "issue:45381", "accept": false, - "reason": "Tokenizer BPE-marker handling and Qwen2.5-VL video position IDs are unrelated." + "reason": "Tokenizer BPE-marker handling is not the same as Qwen2.5-VL video position-id regression." }, { "left": "issue:43606", "right": "issue:44265", "accept": false, - "reason": "Different code paths: Bark CPU-offload device mismatch vs `torch.export` failure on `torch_compilable_check`." + "reason": "CPU offload device mismatch and `torch.export` failure on `torch_compilable_check` are distinct problems." }, { "left": "issue:40990", "right": "issue:45584", "accept": false, - "reason": "Perplexity on WikiText-2 and Whisper empty-transcription generation are unrelated symptoms and models." + "reason": "Perplexity regression on GPT-OSS and Whisper empty-transcription failure are unrelated model behaviors." }, { "left": "issue:43262", "right": "issue:43746", "accept": false, - "reason": "Audio processor chat-template sampling-rate defaulting is unrelated to GraniteSpeech PEFT checkpoint loading." + "reason": "Audio chat-template sampling-rate defaulting and PEFT checkpoint loading are different subsystems." }, { "left": "issue:44479", "right": "issue:45823", "accept": false, - "reason": "Different regressions in different models: Qwen video input handling vs Gemma4 PLE device mismatch." + "reason": "Qwen video-input regression and Gemma4 PLE device mismatch are unrelated model-specific bugs." }, { "left": "issue:43881", "right": "issue:44488", "accept": false, - "reason": "Two separate model-loading failures with no clear shared root cause." + "reason": "GLM-4V loading failure and sleng-bert loading failure are separate model issues." }, { "left": "issue:35707", "right": "issue:41720", "accept": false, - "reason": "Progressive generation with embeds/past key values is unrelated to Qwen3 device-map cuda asserts." + "reason": "Progressive generation with `inputs_embeds/past_key_values` and Qwen3 auto device-mapping CUDA asserts are different failures." }, { "left": "issue:43828", "right": "issue:46821", "accept": false, - "reason": "Autocast dtype mismatch and `window_size` AttributeError are different failure classes." + "reason": "Autocast dtype mismatch and TimesFM `window_size` AttributeError are unrelated." }, { "left": "issue:34634", "right": "issue:35707", "accept": false, - "reason": "Bark voice preset handling and progressive generation with `inputs_embeds`/`past_key_values` are unrelated." + "reason": "BarkProcessor voice preset behavior and generation with `inputs_embeds/past_key_values` are unrelated." }, { "left": "issue:39692", "right": "issue:40444", "accept": false, - "reason": "Both are multimodal, but one is SigLIP2 documentation/model mismatch and the other is Qwen2.5-VL multi-image finetuning." + "reason": "SigLIP2 doc-example mistakes and Qwen2.5-VL multi-image finetuning failure are different issues." }, { "left": "issue:43454", "right": "issue:43550", "accept": false, - "reason": "AyaVision weight tying and Bamba-9B-v2 torch.compile/SDPA are unrelated bugs." + "reason": "AyaVision tied-weight loading/generation bug and Bamba SDPA compile bug do not match." }, { "left": "issue:43606", "right": "issue:43819", "accept": false, - "reason": "Bark CPU-offload device mismatch and DAC `from_latents` vs forward mismatch are different model behaviors and code paths." + "reason": "CPU offload device mismatch is unrelated to DAC `from_latents` vs forward-pass mismatch." }, { "left": "issue:44186", "right": "issue:44265", "accept": false, - "reason": "LayoutLMv2 tokenizer padding/NER crash and `torch.export` + `torch_compilable_check` failure are unrelated." + "reason": "Tokenizer crashes on NER/padding and `torch.export` failures are different code paths." }, { "left": "issue:41720", "right": "issue:45823", "accept": false, - "reason": "Different model families and error modes: Qwen3 auto device mapping assert vs Gemma4 PLE forward device mismatch." + "reason": "Qwen3 auto device-mapping CUDA assert and Gemma4 PLE forward mismatch are unrelated." } ] }, "evaluator_result": { "accept": true, - "feedback": "The summary is grounded in the packet and stays conservative about most pairs. The two accepted soft edges are plausibly mergeable as same-bug duplicates, and the rejected pairs are explained with model/code-path distinctions that match the titles. The overall conclusion that the cluster is mostly heterogeneous is also supported." + "feedback": "The summary is well grounded in the packet. The choice of issue:43873 as the strongest standalone issue is supported by its higher discussion and inbound-reference counts, and the SigLIP2 pairing of issue:43493 with issue:43994 is a conservative, evidence-based duplicate call. The soft-edge verdicts are appropriately cautious and do not overstate similarity." }, "evaluator_used": true, "retried": false, @@ -4496,6 +4433,7 @@ "issue:43450", "issue:43454", "issue:43526", + "issue:43550", "issue:43572", "issue:43720", "issue:43749", @@ -4508,7 +4446,6 @@ "issue:44534", "issue:44568", "issue:44610", - "issue:44756", "issue:44877", "issue:44898", "issue:44977", @@ -4553,228 +4490,228 @@ "issue:36331|issue:42915", "issue:44466|issue:45357", "issue:43450|issue:45701", - "issue:44877|issue:44991", "issue:43065|issue:46402", + "issue:44877|issue:44991", "issue:40990|issue:45701", "issue:44186|issue:45072", "issue:26413|issue:28530", "issue:41720|issue:42617", - "issue:43873|issue:44756" + "issue:43550|issue:43720" ], - "prepared_review_unit_hash": "1b11a363a806510fac781d823e6bff0d2e7d278f1e6808350d58ba2b0ab51197", + "prepared_review_unit_hash": "d909e08fba73060cf519b69c3cc4f531d24287d4bf57e3ca343cc453cfcd59e0", "component_budget": { "node_count": 500, "item_count": 500, "soft_pair_count": 1839, - "serialized_chars": 953913, - "estimated_input_tokens": 238479, - "estimated_eval_tokens": 477214 + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 31, - "serialized_chars": 36990, - "estimated_input_tokens": 9248, - "estimated_eval_tokens": 18752 + "serialized_chars": 37008, + "estimated_input_tokens": 9252, + "estimated_eval_tokens": 18760 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster is heterogeneous: tokenizer regressions, model-loading/offloading bugs, checkpointing issues, and processor/shape problems. None of the soft pairs appear to be the same underlying bug or change.", - "confidence": 0.92, - "canonical_issue_reason": "issue:43873 is the broadest active report in the set, covering offloading/quantization behavior and having multiple inbound references, so it is the closest umbrella representative for this mixed cluster.", + "summary": "This cluster is a heterogeneous mix of Transformers bug reports across tokenization, model loading/offload, quantization, config, and training behavior. The pairwise similarities are mostly superficial, so none of the soft pairs look like true duplicates.", + "confidence": 0.93, + "canonical_issue_reason": "No single canonical issue stands out: the cluster spans unrelated bugs and subsystems, and the apparent overlaps are mostly broad theme similarity rather than the same underlying defect.", "canonical_pr_reason": null, - "best_issue_reason": "issue:43873 is the best single representative because it is open, broadly worded, and central to the offloading/quantization theme; the other issues are more specific one-off regressions.", + "best_issue_reason": "No global best issue is suitable as a representative duplicate target because the items are too diverse to collapse into one concrete bug report.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43377", "right": "issue:44610", "accept": false, - "reason": "Different bugs: MIMI batching/padding-mask inconsistency vs OmDet-Turbo processor output-size mismatch." + "reason": "Both involve batched input handling, but one is a MIMI padding-mask discrepancy and the other is an OmDet image size mismatch; different code paths and fixes." }, { "left": "issue:46305", "right": "issue:46315", "accept": false, - "reason": "Different failures: sliding-window off-by-one in masking vs slower token ID conversion for slow tokenizers." + "reason": "Sliding-window off-by-one in streaming masks and slow `convert_tokens_to_ids` are unrelated bugs." }, { "left": "issue:43257", "right": "issue:45923", "accept": false, - "reason": "Different code paths: accelerate/deepspeed weight conversion failure vs missing gradient-checkpointing flag on a trust_remote_code model." + "reason": "Qwen3 MoE weight conversion/loading and a missing gradient-checkpointing flag are different problems." }, { "left": "issue:44568", "right": "issue:45920", "accept": false, - "reason": "Both involve tokenization regressions, but one is missing BOS/EOS addition for a specific tokenizer while the other is wrong token IDs across several models." + "reason": "Both are tokenizer regressions, but one is missing BOS/EOS insertion for a specific tokenizer while the other is wrong token IDs across many models; not the same bug." }, { "left": "issue:42796", "right": "issue:46209", "accept": false, - "reason": "Different problems: BPE marker handling in a tokenizer vs FP8 inference failure on B200." + "reason": "Tokenizer BPE marker handling and FP8 inference failure on B200 are unrelated." }, { "left": "issue:43454", "right": "issue:45072", "accept": false, - "reason": "Different issues: tied lm_head weights causing bad generation vs dtype mismatch in bfloat16 inference." + "reason": "AyaVision weight tying and dtype mismatches in inference are separate model/runtime issues." }, { "left": "issue:45823", "right": "issue:46821", "accept": false, - "reason": "Different bugs in different subsystems: device_map auto device mismatch vs TimesFM window_size AttributeError." + "reason": "Device-map forward mismatch for Gemma4 PLE and a TimesFM `window_size` AttributeError are different failures." }, { "left": "issue:43377", "right": "issue:44898", "accept": false, - "reason": "Different image-model bugs: batched-vs-single output inconsistency vs non-default-resolution failure despite interpolate_pos_encoding." + "reason": "Both are batching-related, but one is missing padding-mask support and the other is a Perceiver resolution/interpolation bug; not the same defect." }, { "left": "issue:43454", "right": "issue:43720", "accept": false, - "reason": "Different failures: weight tying/garbage text generation vs packed-weight unpacking during accelerate loading." + "reason": "AyaVision tie-weights failure and BitNet packed-weight unpacking during accelerate loading are unrelated." }, { "left": "issue:40990", "right": "issue:42947", "accept": false, - "reason": "Different domains: perplexity anomaly on GPT-OSS vs ineffective gradient checkpointing with PEFT LoRA." + "reason": "High perplexity on a specific model and ineffective gradient checkpointing with LoRA are different training/inference issues." }, { "left": "issue:42617", "right": "issue:42915", "accept": false, - "reason": "Different concrete failures: 3d_parallel.py runtime issue vs FineGrainedFP8Config failure on Qwen3Moe." + "reason": "3d_parallel.py execution failure and Qwen3Moe fine-grained FP8 failure are unrelated." }, { "left": "issue:43749", "right": "issue:44534", "accept": false, - "reason": "Different bugs: FSDP CPU RAM efficient loading vs non-persistent buffers being filled with junk." + "reason": "Both touch loading/state handling, but FSDP CPU RAM efficient loading and non-persistent buffer junk are different concrete bugs." }, { "left": "issue:43317", "right": "issue:43873", "accept": false, - "reason": "Related offloading/quantization theme, but not the same concrete bug: dequantized model loading on gpu+cpu offload vs generic offloading not working with quantization." + "reason": "Both mention offload/quantization, but one is dequantized-model loading with device_map auto and the other is a broader quantization offloading issue; too different to merge." }, { "left": "issue:43572", "right": "issue:45759", "accept": false, - "reason": "Different issues: missing pad_token_idx in StableLmConfig vs from_config not unwrapping text_config for composite Qwen multimodal configs." + "reason": "Missing `pad_token_idx` in config and failure to unwrap `text_config` are different config-loading bugs." }, { "left": "issue:43526", "right": "issue:43883", "accept": false, - "reason": "Different model bugs: BeitImageProcessorFast reduce_labels behavior vs MolmoForCausalLM missing all_tied_weights_keys." + "reason": "BEiT label reduction and missing `all_tied_weights_keys` are unrelated." }, { "left": "issue:43322", "right": "issue:46050", "accept": false, - "reason": "Different failures: Llava Next loading segfault vs quantized model save/load breakage after sinq." + "reason": "LLaVA Next segmentation fault and sinq save/load breakage are unrelated loading bugs." }, { "left": "issue:41720", "right": "issue:43322", "accept": false, - "reason": "Different failure modes and models: Qwen3 auto device-map cuda assert vs Llava Next loading segfault." + "reason": "Qwen3 auto device-mapping CUDA assert and LLaVA Next segfault are different model/runtime failures." }, { "left": "issue:43881", "right": "issue:44991", "accept": false, - "reason": "Different tokenizer/loading problems for different models; not the same underlying bug." + "reason": "glm-4v-9b loading failure and tokenizer loading regression are unrelated." }, { "left": "issue:36331", "right": "issue:42617", "accept": false, - "reason": "Different issues: CustomTrainer compute_loss signature breakage vs 3d_parallel.py execution failure." + "reason": "`compute_loss` signature breakage and `3d_parallel.py` execution failure are unrelated trainer/script issues." }, { "left": "issue:44387", "right": "issue:45704", "accept": false, - "reason": "Different memory/performance bugs: int4 quantization OOM vs apex FusedRMSNorm memory leak." + "reason": "Both are memory-related, but one is int4 quantization memory growth and the other is an apex RMSNorm leak; different causes and fixes." }, { "left": "issue:41720", "right": "issue:44977", "accept": false, - "reason": "Different Qwen3 failures: auto device-map cuda assert vs flash-attention generation corruption." + "reason": "Qwen3 auto device-map CUDA assert and Qwen3.5 flash-attention generation issues are not the same bug." }, { "left": "issue:36331", "right": "issue:42915", "accept": false, - "reason": "Different problems: trainer compute_loss API mismatch vs Qwen3Moe FP8 config failure." + "reason": "`compute_loss` keyword mismatch and Qwen3Moe FP8 failure are unrelated." }, { "left": "issue:44466", "right": "issue:45357", "accept": false, - "reason": "Both are serialization regressions, but one is lm_head tied-weight serialization and the other is incorrect visual encoder keys in Qwen3.5 save_pretrained." + "reason": "Both are serialization regressions, but one concerns tied `lm_head.weight` behavior and the other incorrect visual encoder keys; different artifacts." }, { "left": "issue:43450", "right": "issue:45701", "accept": false, - "reason": "Different classes of bugs: batched video processor shape issue vs version-dependent tokenization behavior." + "reason": "Batched video shape and version-dependent tokenization are unrelated." }, { - "left": "issue:44877", - "right": "issue:44991", + "left": "issue:43065", + "right": "issue:46402", "accept": false, - "reason": "Different failures: strict config rejecting granite_speech vs tokenizer loading regression for EMBEDDIA/est-roberta." + "reason": "Sam3PixelDecoder dummy Conv2d and Sam3 text encoder weights not loading are different subcomponent issues." }, { - "left": "issue:43065", - "right": "issue:46402", + "left": "issue:44877", + "right": "issue:44991", "accept": false, - "reason": "Same model family, but different issues: dummy Conv2d in Sam3PixelDecoder vs text encoder weights not loaded in Sam3Model." + "reason": "Strict config blocking `granite_speech` and tokenizer loading regression are unrelated." }, { "left": "issue:40990", "right": "issue:45701", "accept": false, - "reason": "Different bugs: high perplexity on gpt-oss vs tokenization changes across transformers versions." + "reason": "Extremely high perplexity and tokenization changes across versions are not the same issue." }, { "left": "issue:44186", "right": "issue:45072", "accept": false, - "reason": "Different code paths: LayoutLMv2Tokenizer NER/padding crash vs bfloat16 dtype mismatch in model inference." + "reason": "LayoutLMv2 tokenizer crashes and dtype mismatches in inference are unrelated." }, { "left": "issue:26413", "right": "issue:28530", "accept": false, - "reason": "Different training-control issues: resume-from-checkpoint epoch iterator failure vs early-stopping metric selection/disablement." + "reason": "Checkpoint resume epoch-iterator failure and early-stopping metric lookup are different trainer behaviors." }, { "left": "issue:41720", "right": "issue:42617", "accept": false, - "reason": "Different runtime failures: Qwen3 auto device-map cuda assert vs 3d_parallel.py not running." + "reason": "CUDA assert on Qwen3 device mapping and running 3d_parallel.py are unrelated." }, { - "left": "issue:43873", - "right": "issue:44756", + "left": "issue:43550", + "right": "issue:43720", "accept": false, - "reason": "Both concern memory/offloading, but one is generic quantized offloading behavior while the other is mmap disabling on Strix Halo to avoid OOM; not the same bug." + "reason": "Bamba torch.compile SDPA failure and BitNet packed-weight unpacking are different code paths." } ] }, @@ -4808,7 +4745,6 @@ "issue:43295", "issue:43316", "issue:43317", - "issue:43322", "issue:43404", "issue:43425", "issue:43526", @@ -4819,7 +4755,7 @@ "issue:43606", "issue:43646", "issue:43708", - "issue:43720", + "issue:43873", "issue:43881", "issue:44008", "issue:44186", @@ -4830,6 +4766,7 @@ "issue:44671", "issue:44734", "issue:44743", + "issue:44756", "issue:44898", "issue:44913", "issue:45161", @@ -4848,12 +4785,12 @@ "issue:46869" ], "soft_pairs": [ - "issue:43550|issue:43720", + "issue:43873|issue:44756", "issue:46612|issue:46858", "issue:43577|issue:43708", "issue:43295|issue:45701", - "issue:45685|issue:46821", "issue:43550|issue:43606", + "issue:45685|issue:46821", "issue:44186|issue:44610", "issue:43646|issue:46665", "issue:43404|issue:45357", @@ -4875,211 +4812,198 @@ "issue:44186|issue:44361", "issue:43526|issue:43881", "issue:43316|issue:44913", - "issue:43425|issue:46821", - "issue:42915|issue:43322" + "issue:43425|issue:46821" ], - "prepared_review_unit_hash": "de282f0be3130130b33e622fcfa2ba8b84ee0bab97aa6b8b3044dfd702be8d10", + "prepared_review_unit_hash": "ccf1e3ba3f196e5fc3e8a476e854487221cd15c0c472a4446629d5c194e4533e", "component_budget": { "node_count": 500, "item_count": 500, "soft_pair_count": 1839, - "serialized_chars": 953913, - "estimated_input_tokens": 238479, - "estimated_eval_tokens": 477214 + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 }, "budget": { "node_count": 48, "item_count": 48, - "soft_pair_count": 29, - "serialized_chars": 36490, - "estimated_input_tokens": 9123, - "estimated_eval_tokens": 18502 + "soft_pair_count": 28, + "serialized_chars": 36122, + "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: it mixes unrelated training, loading, tokenizer, multimodal, quantization, and generation/cache bugs. None of the soft edges look like true duplicates.", - "confidence": 0.93, - "canonical_issue_reason": "No single canonical issue is safe here; the items span distinct failures and feature regressions rather than one underlying defect.", + "summary": "This cluster is heterogeneous: it mixes unrelated Transformers regressions across loading, tokenization, generation, quantization, offloading, and serve code paths. None of the soft pairs look like the same concrete bug/change, so all candidate edges are rejected. The broad custom-model initialization regression is the best representative issue in the set.", + "confidence": 0.74, + "canonical_issue_reason": "Issue 43646 is the broadest, most central regression report here: a Transformers 5.0 breakage around custom model initialization, which best fits the cluster\u2019s overall version/loading-regression theme.", "canonical_pr_reason": null, - "best_issue_reason": "If forced to pick a representative, issue 46858 is the most concrete and broadly technical runtime bug in a shared generation/cache path, but it should not absorb the rest of the cluster.", + "best_issue_reason": "43646 is the best single representative because it is concrete, high-level, and closest to several custom-loading/version-regression reports, even though the cluster is not a tight duplicate set.", "best_pr_reason": null, "soft_edge_verdicts": [ { - "left": "issue:43550", - "right": "issue:43720", + "left": "issue:43873", + "right": "issue:44756", "accept": false, - "reason": "Both involve advanced execution modes, but one is a Bamba torch.compile+SDPA failure and the other is packed-weight unpacking during accelerate loading. Different code paths and fixes." + "reason": "Different bugs: quantization/offload behavior versus mmap/OOM on Strix Halo; no shared code path." }, { "left": "issue:46612", "right": "issue:46858", "accept": false, - "reason": "Both touch generation cache handling, but 46612 is missing beam-search cache reorder for several models, while 46858 is a seq_length device-tensor arithmetic bug in static-cache prefill." + "reason": "Both involve cache/generation, but one is beam-search reorder skipping in specific models and the other is static-cache slice arithmetic on non-CUDA backends." }, { "left": "issue:43577", "right": "issue:43708", "accept": false, - "reason": "BLIP2 dtype propagation on load and Trainer max_steps recomputation on checkpoint resume are unrelated bugs." + "reason": "Unrelated: dtype propagation in Blip2 loading versus Trainer max_steps calculation on checkpoint resume." }, { "left": "issue:43295", "right": "issue:45701", "accept": false, - "reason": "Both are regression reports around tokenization, but one is processor/tokenizer API breakage and the other is version-dependent tokenization drift. Not the same defect." + "reason": "Both are version-related regressions, but one is processor/tokenizer API breakage and the other is tokenization output changes; different underlying issues." }, { - "left": "issue:45685", - "right": "issue:46821", + "left": "issue:43550", + "right": "issue:43606", "accept": false, - "reason": "These are different model-specific AttributeErrors: MPS histogram for Int in a MoE path versus TimesFM window_size handling." + "reason": "Different models and failures: torch.compile with SDPA on Bamba versus CPU-offload device mismatch on bark-small." }, { - "left": "issue:43550", - "right": "issue:43606", + "left": "issue:45685", + "right": "issue:46821", "accept": false, - "reason": "Both are runtime failures, but one is Bamba torch.compile with SDPA and the other is a CPU-offload device mismatch for Bark. Different underlying bugs." + "reason": "MoE MPS histogram support error is unrelated to TimesFM window_size AttributeError." }, { "left": "issue:44186", "right": "issue:44610", "accept": false, - "reason": "Tokenizer crash on NER/padding versus processor image-size mismatch. Same broad multimodal area, but not the same code-path failure." + "reason": "Different model-specific preprocessing bugs: LayoutLMv2 tokenizer padding/NER crash versus OmDet-Turbo image-size mismatch." }, { "left": "issue:43646", "right": "issue:46665", "accept": false, - "reason": "Both concern model loading behavior, but one is Transformers 5.0 custom model initialization and the other is from_pretrained inside Python classes. Distinct issues." + "reason": "Related broadly to loading/custom code, but one is a Transformers 5.0 initialization regression and the other is from_pretrained behavior inside Python classes." }, { "left": "issue:43404", "right": "issue:45357", "accept": false, - "reason": "Weight tying in Mistral3 conditional generation and incorrect saved visual-encoder keys in Qwen3.5 are different bugs." + "reason": "Different models and failure modes: untied lm_head in Mistral3 versus wrong visual encoder keys saved for Qwen3.5." }, { "left": "issue:44734", "right": "issue:46858", "accept": false, - "reason": "Both are cache/continuation related, but 44734 is wrong indexing in serve KV-cache continuation while 46858 is prefill slice arithmetic on a device tensor." + "reason": "Both touch cache/generation, but one crashes in serve KV-cache continuation and the other is a static-cache device-tensor arithmetic bug." }, { "left": "issue:44610", "right": "issue:44898", "accept": false, - "reason": "Both involve vision input sizing, but one is an OmDet processor/model size mismatch and the other is a Perceiver interpolation failure at non-default resolution. Not duplicates." + "reason": "Both concern input sizing, but they affect different models and mechanisms; not the same bug." }, { "left": "issue:42915", "right": "issue:43317", "accept": false, - "reason": "FineGrainedFP8Config failure and dequantized-model loading with gpu+cpu offload are different quantization/offload bugs." + "reason": "Both involve quantization/offload, but one is FineGrainedFP8 on Qwen3MoE and the other is dequantized loading with device_map auto." }, { "left": "issue:36010", "right": "issue:42617", "accept": false, - "reason": "ImportError for GenerationMixin versus failure running 3d_parallel.py are unrelated." + "reason": "Completely unrelated: import error for GenerationMixin versus inability to run a 3d_parallel script." }, { "left": "issue:44423", "right": "issue:45406", "accept": false, - "reason": "Both crash in serve with multimodal models, but one is a str.to AttributeError in continuous batching and the other is a missing _tokenizer on Gemma4Processor." + "reason": "Both are serve multimodal crashes, but the exceptions and root causes differ: string `.to()` misuse versus missing `_tokenizer` on Gemma4Processor." }, { "left": "issue:44008", "right": "issue:45823", "accept": false, - "reason": "Gemma3n variable-name collision in forward and Gemma4 PLE device mismatch under auto device map are different problems." - }, - { - "left": "issue:46821", - "right": "issue:46869", - "accept": false, - "reason": "TimesFM window_size AttributeError and YOLO multi-GPU device-map failure are unrelated." + "reason": "Different Gemma model families and distinct forward-time failures: variable-name collision versus device mismatch with device_map auto." }, { "left": "issue:45632", "right": "issue:45684", "accept": false, - "reason": "Cache-path collision for trust_remote_code and read-only permission propagation during save_pretrained are separate filesystem/export issues." + "reason": "Different filesystem/save_pretrained problems: trust_remote_code cache collisions versus read-only permission propagation." }, { "left": "issue:44464", "right": "issue:46858", "accept": false, - "reason": "Chunked generation inconsistency with compiled forward is not the same as static-cache prefill slice arithmetic failing on a device tensor." + "reason": "Both involve compiled generation, but one is inconsistent outputs with chunked generation and the other is a cache-length tensor arithmetic failure." }, { "left": "issue:41720", "right": "issue:45161", "accept": false, - "reason": "Qwen3 auto device mapping cudaErrorAssert and GPT-OSS MoE tensor-parallel failure are different parallelism/device issues." + "reason": "Different model-parallelism issues: cudaErrorAssert on Qwen3 auto device mapping versus TP not working on GPT-OSS MoE." }, { "left": "issue:28530", "right": "issue:33666", "accept": false, - "reason": "Early-stopping metric mismatch and Qwen2-VL multi-GPU training are unrelated." + "reason": "Unrelated: early stopping metric resolution versus Qwen2-VL multi-GPU training." }, { "left": "issue:44671", "right": "issue:44743", "accept": false, - "reason": "CamemBERT masked-LM prediction regression and Qwen3.5 recurrent-state reset under cache are different model bugs." + "reason": "Different bugs in different models: CamemBERT MLM predictions versus Qwen3.5 recurrent state reset with cache." }, { "left": "issue:43531", "right": "issue:43572", "accept": false, - "reason": "Sliding-window behavior in Qwen3-MoE and missing pad_token_idx in StableLmConfig are different configuration bugs." + "reason": "Unrelated model-specific issues: sliding window behavior in Qwen3-MoE versus missing pad_token_idx in StableLmConfig." }, { "left": "issue:42915", "right": "issue:46291", "accept": false, - "reason": "MoE FP8 failure and kernels version/revision import requirement are unrelated." + "reason": "Different problems: Qwen3MoE FP8 failure versus pipeline import failing due to kernels version/revision requirements." }, { "left": "issue:36331", "right": "issue:41720", "accept": false, - "reason": "Trainer compute_loss signature breakage and Qwen3 auto device-mapping assert are unrelated." + "reason": "Trainer compute_loss signature mismatch is unrelated to Qwen3 device-map CUDA assert." }, { "left": "issue:44186", "right": "issue:44361", "accept": false, - "reason": "Two different tokenizer AttributeErrors on different models/tasks; no shared root cause is evident." + "reason": "Both are tokenizer/task crashes, but they affect different tokenizers and different error conditions." }, { "left": "issue:43526", "right": "issue:43881", "accept": false, - "reason": "BeitImageProcessorFast label reduction bug and glm-4v-9b loading failure are unrelated." + "reason": "Different issues: BeitImageProcessorFast reduce_labels bug versus glm-4v-9b loading failure." }, { "left": "issue:43316", "right": "issue:44913", "accept": false, - "reason": "Gemma3TextConfig API discrepancy and GPTNeoXConfig rotary_pct reload regression are separate config issues." + "reason": "Both are config-related, but one is an API discrepancy in Gemma3TextConfig and the other is a GPTNeoX rotary_pct reload bug." }, { "left": "issue:43425", "right": "issue:46821", "accept": false, - "reason": "Torch 2.10 compatibility and TimesFM window_size handling are unrelated." - }, - { - "left": "issue:42915", - "right": "issue:43322", - "accept": false, - "reason": "FineGrainedFP8Config failure and Llava Next segmentation fault are different bugs." + "reason": "No meaningful overlap: Torch 2.10 incompatibility versus TimesFM window_size AttributeError." } ] }, @@ -5113,6 +5037,8 @@ "issue:36032", "issue:41720", "issue:42490", + "issue:42915", + "issue:43322", "issue:43329", "issue:43388", "issue:43531", @@ -5135,9 +5061,7 @@ "issue:44589", "issue:44625", "issue:44898", - "issue:44987", "issue:44991", - "issue:45042", "issue:45059", "issue:45072", "issue:45357", @@ -5153,6 +5077,7 @@ "issue:46858" ], "soft_pairs": [ + "issue:42915|issue:43322", "issue:28530|issue:30064", "issue:33357|issue:35707", "issue:29159|issue:36032", @@ -5174,224 +5099,223 @@ "issue:43531|issue:46821", "issue:34634|issue:41720", "issue:43540|issue:46858", - "issue:44451|issue:44991", "issue:45759|issue:45859", + "issue:44451|issue:44991", "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:44898|issue:45072" ], - "prepared_review_unit_hash": "f4b8d1ebec0f3d4eeea2effd00aa2369b841d87b3174b7e431564e1cc534abab", + "prepared_review_unit_hash": "cf41b9722903f31118a16702984d6cc895a3ce6d69d24f5eb13a3b03f7816897", "component_budget": { "node_count": 500, "item_count": 500, "soft_pair_count": 1839, - "serialized_chars": 953913, - "estimated_input_tokens": 238479, - "estimated_eval_tokens": 477214 + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 30, - "serialized_chars": 36799, - "estimated_input_tokens": 9200, - "estimated_eval_tokens": 18656 + "serialized_chars": 36754, + "estimated_input_tokens": 9189, + "estimated_eval_tokens": 18634 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The provided soft edges are mostly topical similarities, not true duplicates. I would reject every candidate pair as a duplicate; none look like the same underlying bug/change. No PRs were provided.", - "confidence": 0.94, - "canonical_issue_reason": "No single canonical duplicate cluster is evident from the provided edges; if a representative issue must be named, issue:43644 is the most concrete standalone regression report in this set.", + "summary": "This cluster is mostly a loose semantic grouping of unrelated Transformers issues. The only clearly duplicate-like theme is the v5 non-persistent-buffer regression, but none of the provided soft-edge pairs are strong enough to merge as true duplicates/changes.", + "confidence": 0.33, + "canonical_issue_reason": "Issue 43644 is the clearest concrete regression report in the cluster and is the most canonical anchor for the v5 non-persistent-buffer bug theme; the nearby 44534 report looks like a rephrase of the same problem.", "canonical_pr_reason": null, - "best_issue_reason": "Issue:43644 is the clearest, narrowly-scoped bug report among the items; most others are broader, more speculative, or clearly unrelated.", + "best_issue_reason": "43644 is the best representative because it describes a specific, reproducible framework regression with a clear failure mode, while most other issues here are model-specific or unrelated.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:42915", + "right": "issue:43322", + "accept": false, + "reason": "Different models and failures: FP8 config loading vs Llava Next segfault." + }, { "left": "issue:28530", "right": "issue:30064", "accept": false, - "reason": "Different problems: early-stopping metric selection vs image processor handling of void segmentation maps." + "reason": "Early stopping/metric selection is unrelated to void segmentation map processing." }, { "left": "issue:33357", "right": "issue:35707", "accept": false, - "reason": "Unrelated symptoms and code paths: MacOS CLIP bus error vs generation with inputs_embeds/past_key_values." + "reason": "MacOS CLIP bus error is a different code path from generation with inputs_embeds/past_key_values." }, { "left": "issue:29159", "right": "issue:36032", "accept": false, - "reason": "Both tokenizer-related, but one is slow/fast inconsistency while the other is a T5Tokenizer load-time special-token conflict." + "reason": "Both involve tokenizers, but one is slow/fast inconsistency and the other is a T5 tokenizer class-loading conflict." }, { "left": "issue:43540", "right": "issue:43782", "accept": false, - "reason": "Both involve Qwen multimodal loading, but one is a video-input processing ValueError and the other is a weight_only from_pretrained error." + "reason": "Different Qwen3 multimodal bugs: video input handling vs weight_only loading." }, { "left": "issue:43550", "right": "issue:43653", "accept": false, - "reason": "Completely different subsystems: torch.compile/SDPA on Bamba vs BigBirdTokenizer special-token registration." + "reason": "Unrelated subsystems: torch.compile/SDPA vs BigBird tokenizer special-token registration." }, { "left": "issue:44387", "right": "issue:45701", "accept": false, - "reason": "Different regressions: int4 quantization memory growth/OOM vs tokenization changes across versions." + "reason": "Quantization OOM and tokenization changes are different issues with no shared bug path." }, { "left": "issue:43873", "right": "issue:45704", "accept": false, - "reason": "Both mention memory/offloading, but one is quantization offload behavior and the other is a T5 RMSNorm memory leak." + "reason": "Offloading/quantization behavior is unrelated to T5's RMSNorm memory leak." }, { "left": "issue:30064", "right": "issue:34634", "accept": false, - "reason": "Different models and failures: segmentation map processing vs BarkProcessor voice_preset handling." + "reason": "Image segmentation-map processing and BarkProcessor voice_preset are unrelated." }, { "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": "A typo in cache strategies is not the same as stale docs mentioning removed pipeline tasks." }, { "left": "issue:43606", "right": "issue:43720", "accept": false, - "reason": "Both are loading/offload-related CI bugs, but the concrete failures differ: device mismatch vs unpacking packed weights." + "reason": "Different failure modes and models: CPU offload device mismatch vs packed-weight unpacking during accelerate loading." }, { "left": "issue:44534", "right": "issue:46665", "accept": false, - "reason": "Both touch loading, but one is a non-persistent-buffer corruption regression and the other is inconsistent from_pretrained behavior in Python classes." + "reason": "Non-persistent buffer corruption is unrelated to from_pretrained behavior inside Python classes." }, { "left": "issue:43531", "right": "issue:45759", "accept": false, - "reason": "Different Qwen 3.5 issues: sliding_window behavior vs unwrapping text_config for composite configs." + "reason": "Qwen3-MoE sliding_window bug and Qwen multimodal config unwrapping are different problems." }, { "left": "issue:41720", "right": "issue:43746", "accept": false, - "reason": "Unrelated: auto device mapping cudaErrorAssert vs PEFT adapters failing to load local checkpoints." + "reason": "Auto device-mapping CUDA assert is unrelated to GraniteSpeech PEFT local checkpoint loading." }, { "left": "issue:42490", "right": "issue:43329", "accept": false, - "reason": "Path normalization import warning is unrelated to the missing video-path helpers in multimodal token counting." + "reason": "Trailing-slash import warning and undefined video-branch variables are different bugs." }, { "left": "issue:43388", "right": "issue:45059", "accept": false, - "reason": "Different areas: metric gathering last-batch tuple handling vs SAM3 text/bounding-box behavior." + "reason": "Metric gathering tuple-label truncation is unrelated to SAM3 text/box behavior." }, { "left": "issue:45406", "right": "issue:45464", "accept": false, - "reason": "Serving crash from missing processor state is not the same as streaming chat/completions failure on Qwen3.5-0.8B." + "reason": "Gemma4Processor serve crash and Qwen3.5 streaming API failure are distinct issues." }, { "left": "issue:43606", "right": "issue:43653", "accept": false, - "reason": "Device offload mismatch and tokenizer special-token registration are unrelated bugs." + "reason": "CPU offload device mismatch is not the same as tokenizer special-token registration." }, { "left": "issue:34567", "right": "issue:35707", "accept": false, - "reason": "TrainerState token counting bug vs progressive generation with inputs_embeds/past_key_values." + "reason": "TrainerState token counting and progressive generation are unrelated." }, { "left": "issue:43531", "right": "issue:46821", "accept": false, - "reason": "Both mention window_size, but the affected models and failure modes differ." + "reason": "Different models and bugs: Qwen3-MoE sliding_window vs TimesFM window_size attribute." }, { "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 auto device mapping are unrelated." }, { "left": "issue:43540", "right": "issue:46858", "accept": false, - "reason": "Video-input processing in Qwen3OmniMoe is unrelated to static-cache seq-length arithmetic on non-CUDA backends." + "reason": "Video-input processing and static-cache seq-length tensor arithmetic are different code paths." }, { - "left": "issue:44451", - "right": "issue:44991", + "left": "issue:45759", + "right": "issue:45859", "accept": false, - "reason": "Different loading failures for different models; no shared underlying bug is evident." + "reason": "Both are Qwen3.5-related, but one is config unwrapping and the other is CPU-offload rotary embedding compatibility." }, { - "left": "issue:45759", - "right": "issue:45859", + "left": "issue:44451", + "right": "issue:44991", "accept": false, - "reason": "Composite-config unwrapping and CPU-offload rotary embedding compatibility are separate issues." + "reason": "Model-specific tokenizer load failures are not enough to treat these as the same bug." }, { "left": "issue:45357", "right": "issue:46402", "accept": false, - "reason": "Both concern multimodal weight loading/saving, but the concrete bugs are different: save_pretrained key corruption vs text encoder weights not loading." + "reason": "Saving incorrect visual encoder keys and missing text-encoder weights on load are different directions and symptoms." }, { "left": "issue:43531", "right": "issue:44589", "accept": false, - "reason": "Qwen3-MoE sliding_window and missing Float8 storage are unrelated." + "reason": "Sliding-window behavior and missing Float8 storage are unrelated." }, { "left": "issue:44077", "right": "issue:44625", "accept": false, - "reason": "Model post_init policy and Qwen3.5 num_labels propagation are different configuration bugs." + "reason": "patchtsmixer post_init policy and Qwen3.5 num_labels propagation are unrelated." }, { "left": "issue:43644", "right": "issue:46665", "accept": false, - "reason": "One is a buffer-initialization regression; the other is a from_pretrained behavioral inconsistency." + "reason": "Junk-filled non-persistent buffers are unrelated to from_pretrained behavior in user-defined classes." }, { "left": "issue:44062", "right": "issue:44451", "accept": false, - "reason": "AddedToken constructor argument conflict does not match the model-loading failure for ScandiBERT." + "reason": "AddedToken keyword conflicts and ScandiBERT loading failure are not the same concrete bug." }, { "left": "issue:44898", "right": "issue:45072", "accept": false, - "reason": "Perceiver positional interpolation and dtype mismatch in bfloat16 inference are distinct failures." - }, - { - "left": "issue:44987", - "right": "issue:45042", - "accept": false, - "reason": "Physical-intelligence/fast loading failure and PIL image processors requiring torchvision are different regressions." + "reason": "Perceiver interpolation failure and bfloat16 dtype mismatch are unrelated." } ] }, @@ -5420,7 +5344,6 @@ "issue:39692", "issue:41720", "issue:42886", - "issue:43295", "issue:43381", "issue:43421", "issue:43493", @@ -5443,7 +5366,9 @@ "issue:44843", "issue:44918", "issue:44936", + "issue:44987", "issue:44995", + "issue:45042", "issue:45081", "issue:45356", "issue:45357", @@ -5451,7 +5376,6 @@ "issue:45479", "issue:45704", "issue:45759", - "issue:45834", "issue:45907", "issue:45923", "issue:45987", @@ -5465,6 +5389,7 @@ "issue:46858" ], "soft_pairs": [ + "issue:44987|issue:45042", "issue:44625|issue:45479", "issue:43381|issue:43844", "issue:44368|issue:45357", @@ -5494,229 +5419,228 @@ "issue:43720|issue:43819", "issue:45987|issue:46829", "issue:46829|issue:46832", - "issue:43653|issue:43720", - "issue:43295|issue:45834" + "issue:43653|issue:43720" ], - "prepared_review_unit_hash": "ac51e530f0fc7fc81e69b520ef539abc492ea35cceebc7118fb2f12de5924f40", + "prepared_review_unit_hash": "49330532c7e9d13990193e435fe55c4141b47fe9c45321075041108012557ee3", "component_budget": { "node_count": 500, "item_count": 500, "soft_pair_count": 1839, - "serialized_chars": 953913, - "estimated_input_tokens": 238479, - "estimated_eval_tokens": 477214 + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 31, - "serialized_chars": 37564, - "estimated_input_tokens": 9391, - "estimated_eval_tokens": 19038 + "serialized_chars": 37549, + "estimated_input_tokens": 9388, + "estimated_eval_tokens": 19032 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly heterogeneous. Only a couple of pairs look like true duplicates; the rest are same-subsystem but different bugs. No PRs are present.", - "confidence": 0.83, - "canonical_issue_reason": "issue:44843 is the clearest root-cause report for the offline tokenizer regression because it points to the exact failing path (_patch_mistral_regex -> model_info) and explains the HF_HUB_OFFLINE breakage; issue:42886 is the broader symptom report.", + "summary": "This cluster is mostly a set of unrelated issues that were linked by loose lexical similarity. The only clear duplicate pair is the two Qwen2_5_VLProcessor batched padding=False crash reports; everything else should stay separate.", + "confidence": 0.96, + "canonical_issue_reason": "issue:44545 is the cleanest representative of the one true duplicate pair: it states the batched padding=False crash clearly and matches issue:44514 on the same underlying bug.", "canonical_pr_reason": null, - "best_issue_reason": "issue:44843 is the best overall issue anchor in this set: it is specific, actionable, and directly captures the underlying failure mode rather than only the user-visible symptom.", + "best_issue_reason": "issue:44545 is the strongest canonical issue candidate because it is the clearer, more concise report of the duplicated Qwen2_5_VLProcessor.apply_chat_template failure.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:44987", + "right": "issue:45042", + "accept": false, + "reason": "Different failures in different subsystems: loading a specific model vs PIL backend image processors requiring torchvision." + }, { "left": "issue:44625", "right": "issue:45479", "accept": false, - "reason": "Both are classification/config bugs, but one is num_labels propagation and the other is degenerate zero-loss behavior across models; different code paths." + "reason": "One is Qwen3.5 config propagation into text_config; the other is a generic sequence-classification loss bug across models." }, { "left": "issue:43381", "right": "issue:43844", "accept": false, - "reason": "Gradient checkpointing in eval mode is unrelated to the DeepSpeed/ZeRO-3 gradient growth regression." + "reason": "Eval-mode gradient checkpointing restriction is unrelated to a ZeRO-3 gradient explosion regression." }, { "left": "issue:44368", "right": "issue:45357", "accept": false, - "reason": "One is a LoRA warning about tie_word_embeddings; the other is incorrect visual encoder keys on save_pretrained. Different bugs." + "reason": "A warning about tie_word_embeddings in LoRA fine-tuning is not the same bug as incorrect visual encoder keys being saved." }, { "left": "issue:39692", "right": "issue:44610", "accept": false, - "reason": "SigLIP2 doc/example errors are unrelated to OmDet-Turbo's processor/model image-size mismatch." + "reason": "Both mention vision models, but one is a SigLIP2 docs/model mismatch and the other is an OmDet-Turbo image-size mismatch." }, { "left": "issue:46655", "right": "issue:46693", "accept": false, - "reason": "Video processor padding is unrelated to the FA2 long-context performance regression." + "reason": "Video padding calculation in a processor is unrelated to a FlashAttention2 long-context performance regression." }, { "left": "issue:43421", "right": "issue:43927", "accept": false, - "reason": "Special-token post-processor updates at runtime are distinct from DiaConfig losing custom token IDs on save/load." + "reason": "Runtime special-token post-processing and save/load of custom token IDs are different tokenizer bugs." }, { "left": "issue:41720", "right": "issue:45704", "accept": false, - "reason": "Qwen3 device-mapping cudaErrorAssert is unrelated to the T5 apex.FusedRMSNorm memory leak." + "reason": "Qwen3 device-mapping CUDA asserts and T5 apex RMSNorm memory leak are unrelated." }, { "left": "issue:43746", "right": "issue:46050", "accept": false, - "reason": "Both involve loading/saving, but one is PEFT local checkpoint loading and the other is quantized sinq save/load corruption; not the same bug." + "reason": "PEFT/local checkpoint loading failure is a different code path from quantized save_pretrained/from_pretrained corruption." }, { "left": "issue:43381", "right": "issue:45704", "accept": false, - "reason": "Gradient checkpointing eval-mode behavior does not match a T5 RMSNorm memory leak." + "reason": "Eval-mode checkpointing behavior is unrelated to the T5 apex memory leak." }, { "left": "issue:42886", "right": "issue:44843", - "accept": true, - "reason": "Same offline-mode failure: 42886 reports the cache-load symptom, while 44843 identifies the exact _patch_mistral_regex/model_info cause." + "accept": false, + "reason": "Both involve offline mode, but one is cache loading and the other is an unconditional model_info() call in _patch_mistral_regex." }, { "left": "issue:45704", "right": "issue:45923", "accept": false, - "reason": "T5 apex memory leak and missing gradient-checkpointing flag on a trust_remote_code model are unrelated." + "reason": "T5 apex memory leak and missing supports_gradient_checkpointing on a trust_remote_code model are separate issues." }, { "left": "issue:44918", "right": "issue:44936", "accept": false, - "reason": "TRL SFT input-embedding unpacking and trainer.evaluate() after train() are different training-loop issues." + "reason": "Qwen3.5 embedding unpacking in TRL SFT is not the same as trainer.evaluate() failing after trainer.train()." }, { "left": "issue:44514", "right": "issue:44545", "accept": true, - "reason": "Exact same crash: Qwen2_5_VLProcessor.apply_chat_template fails on batched input with padding=False." + "reason": "Same bug report twice: Qwen2_5_VLProcessor.apply_chat_template crashes on batched input when padding=False." }, { "left": "issue:45759", "right": "issue:45923", "accept": false, - "reason": "AutoModelForCausalLM config unwrapping for Qwen multimodal configs is unrelated to the missing gradient-checkpointing flag bug." + "reason": "Unwrapping text_config in composite multimodal configs is unrelated to the gradient-checkpointing flag on trust_remote_code variants." }, { "left": "issue:43746", "right": "issue:43873", "accept": false, - "reason": "Local PEFT checkpoint loading and quantization/offloading behavior are different failure modes." + "reason": "Local checkpoint loading for GraniteSpeech adapters is unrelated to quantization offloading behavior." }, { "left": "issue:43421", "right": "issue:43653", "accept": false, - "reason": "Runtime special-token updates and BigBirdTokenizer mask-token registration are related thematically but not the same underlying bug." + "reason": "Post-processor updates after special-token mutation are different from BigBirdTokenizer not registering mask as special." }, { "left": "issue:43994", "right": "issue:44671", "accept": false, - "reason": "Different model families and different failures: SigLIP2 pipeline output vs CamemBERT masked-LM predictions in v5." + "reason": "SigLIP2 pipeline/model output issues and CamemBERT masked LM regressions are unrelated." }, { "left": "issue:39692", - "right": "issue:43873", + "right": "issue:43493", "accept": false, - "reason": "SigLIP2 example/model mismatch is unrelated to quantization offloading behavior." + "reason": "Both are SigLIP2-related, but one is a broken documentation example and the other is an implementation discrepancy against JAX." }, { "left": "issue:46032", "right": "issue:46463", "accept": false, - "reason": "Mamba2 cache/use_cache correctness is unrelated to a T5 decoder-input error message." + "reason": "A silent use_cache correctness bug is not the same as an incorrect error message in T5 decoder input validation." }, { "left": "issue:43421", "right": "issue:44186", "accept": false, - "reason": "Special-token post-processing updates are unrelated to LayoutLMv2Tokenizer crashes on NER/batched padding." + "reason": "Tokenizer post-processor updates and LayoutLMv2 NER/batching crashes are different bugs." }, { "left": "issue:43782", "right": "issue:45381", "accept": false, - "reason": "Qwen3VL weight_only loading and Qwen2.5-VL video vision_position_ids are distinct multimodal bugs." + "reason": "Qwen3VL weight_only loading error and Qwen2.5-VL video position-id bug are separate multimodal issues." }, { "left": "issue:45081", "right": "issue:45356", "accept": false, - "reason": "Both touch Mistral/Kimi tokenizer code, but one is an AttributeError in fix_mistral_regex while the other is a codec-handling regression with a warning issue." + "reason": "Both touch tokenizer regex handling, but one is a crash in _patch_mistral_regex and the other is a Kimi-K2.5 codec/regression report." }, { "left": "issue:44995", "right": "issue:46858", "accept": false, - "reason": "Stale indexer cache on second forward pass is different from static-cache seq-length returning a device tensor." + "reason": "A stale indexer cache on second forward pass is unrelated to static-cache seq_length returning a device tensor." }, { "left": "issue:30064", "right": "issue:36010", "accept": false, - "reason": "Image processor void segmentation maps and GenerationMixin ImportError are unrelated." + "reason": "Void segmentation map processing and a GenerationMixin import error have no meaningful overlap." }, { "left": "issue:45356", "right": "issue:45907", "accept": false, - "reason": "Kimi tokenizer regression is unrelated to list_repo_files not forwarding revision in CI." + "reason": "Tokenizer codec/regex regression and list_repo_files revision forwarding are unrelated." }, { "left": "issue:39692", "right": "issue:43873", "accept": false, - "reason": "SigLIP2 documentation/model mismatch is unrelated to quantization offloading issues." + "reason": "SigLIP2 example/model issues and quantization offloading behavior are different bugs." }, { "left": "issue:43720", "right": "issue:43819", "accept": false, - "reason": "BitNet packed-weight loading and DAC.from_latents missing STE are different model-specific bugs." + "reason": "BitNet packed-weight loading and DAC from_latents/STE mismatch are unrelated code paths." }, { "left": "issue:45987", "right": "issue:46829", "accept": false, - "reason": "StaticCache.get_seq_length() returning a tensor is unrelated to VideoPrism hidden_states plumbing." + "reason": "StaticCache seq_length contract bug and VideoPrism output-field mismatch are unrelated." }, { "left": "issue:46829", "right": "issue:46832", "accept": false, - "reason": "VideoPrism output field mix-up is unrelated to RT-DETR forward failing when feature levels exceed backbone outputs." + "reason": "VideoPrism hidden_states misassignment is different from RT-DETR feature-level forwarding errors." }, { "left": "issue:43653", "right": "issue:43720", "accept": false, - "reason": "BigBirdTokenizer special-token registration and BitNet accelerate loading are unrelated." - }, - { - "left": "issue:43295", - "right": "issue:45834", - "accept": false, - "reason": "Processor/tokenizer regression in v4.57.5 is not the same as Kosmos2.5 long-OCR index errors." + "reason": "BigBirdTokenizer special-token registration and BitNet accelerate-loading issues are unrelated." } ] }, "evaluator_result": { "accept": true, - "feedback": "The summary is grounded in the packet, stays conservative on the soft edges, and does not overclaim duplicates or PR mergeability. The accepted duplicate pairs are well supported by the issue titles/descriptions, and the rejected pairs are phrased cautiously." + "feedback": "Grounded and conservative overall. The only clearly duplicate pair in the packet is issue:44514/issue:44545, which have near-identical titles and matching descriptions. The rest of the soft-edge rejections are appropriately cautious and tied to materially different failure modes or code paths." }, "evaluator_used": true, "retried": false, @@ -5743,6 +5667,7 @@ "issue:36331", "issue:39692", "issue:41720", + "issue:43295", "issue:43335", "issue:43531", "issue:43540", @@ -5786,6 +5711,7 @@ "issue:46858" ], "soft_pairs": [ + "issue:43295|issue:45834", "issue:43550|issue:44361", "issue:43550|issue:43927", "issue:44062|issue:44488", @@ -5815,202 +5741,208 @@ "issue:45183|issue:45479", "issue:45584|issue:45834" ], - "prepared_review_unit_hash": "49bb01aa12b15f671cb383ce5e5c7809a474171bf5236473194fbf2886af714a", + "prepared_review_unit_hash": "59d160e69839dc23ae5a4ec954f7ca06ad2280edd4875bf5099e422609e66620", "component_budget": { "node_count": 500, "item_count": 500, "soft_pair_count": 1839, - "serialized_chars": 953913, - "estimated_input_tokens": 238479, - "estimated_eval_tokens": 477214 + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 }, "budget": { - "node_count": 47, - "item_count": 47, - "soft_pair_count": 28, - "serialized_chars": 35678, - "estimated_input_tokens": 8920, - "estimated_eval_tokens": 18096 + "node_count": 48, + "item_count": 48, + "soft_pair_count": 29, + "serialized_chars": 36588, + "estimated_input_tokens": 9147, + "estimated_eval_tokens": 18550 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "No true duplicate cluster here: the soft pairs mostly share broad subsystem keywords but describe different models, code paths, or failure modes. I would reject all proposed merges.", - "confidence": 0.94, - "canonical_issue_reason": "There is no single canonical duplicate candidate in this set because the issues are largely unrelated. If one issue must be chosen as the representative standalone report, issue 44843 is the most concrete and actionable.", + "summary": "The cluster is heterogeneous: most items are distinct model-, tokenizer-, config-, or performance-specific issues rather than duplicates. None of the soft edge pairs looks like the same underlying bug/change. If one issue must serve as the anchor, #43611 is the broadest regression in the set.", + "confidence": 0.86, + "canonical_issue_reason": "Issue #43611 is the most broadly applicable regression here: it affects model loading across Transformers v5 via `base_model_prefix`, making it a better cluster anchor than the more model-specific reports.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 44843 is a strong standalone report: it names a specific helper, has a clear trigger, and appears reproducible with a narrow fix surface.", + "best_issue_reason": "#43611 is the best representative issue because it describes a cross-cutting runtime regression with wide user impact, whereas the others are narrower or isolated to specific models/code paths.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:43295", + "right": "issue:45834", + "accept": false, + "reason": "Different bugs: one is a processor/tokenizer regression, the other is a Kosmos2.5 long-OCR index error." + }, { "left": "issue:43550", "right": "issue:44361", "accept": false, - "reason": "Different model and failure mode: torch.compile/SDPA vs tokenizer AttributeError." + "reason": "Different subsystems and failure modes: Bamba torch.compile/SDPA vs MLukeTokenizer task handling." }, { "left": "issue:43550", "right": "issue:43927", "accept": false, - "reason": "Both are model bugs, but the symptoms and code paths are unrelated." + "reason": "Unrelated issues: Bamba compile failure vs DiaConfig losing custom token IDs on save/load." }, { "left": "issue:44062", "right": "issue:44488", "accept": false, - "reason": "Both concern tokenizers, but one is an AddedToken construction bug and the other is a model loading issue." + "reason": "Both are tokenizer-related, but one is an AddedToken constructor error and the other is a model-loading failure for a specific tokenizer." }, { "left": "issue:43596", "right": "issue:44220", "accept": false, - "reason": "Completely different subsystems: DeepSpeed/BertModel init vs fbank feature extraction." + "reason": "Different code paths: deepspeed ZeRO-3/BertModel init vs `_torch_extract_fbank_features()`." }, { "left": "issue:45381", "right": "issue:46693", "accept": false, - "reason": "Different regression categories: video position IDs vs long-context performance." + "reason": "Qwen2.5-VL video position IDs bug is not the same as the FA2 long-context performance regression." }, { "left": "issue:43540", "right": "issue:46294", "accept": false, - "reason": "Both mention video inputs, but the affected models and root causes differ." + "reason": "Both involve Qwen video flows, but they are different models and different mismatches/errors in the pipeline." }, { "left": "issue:44008", "right": "issue:46858", "accept": false, - "reason": "Different code paths: Gemma3n forward variable collision vs cache sequence-length arithmetic." + "reason": "Gemma3n forward variable collision is unrelated to the static-cache slice arithmetic bug." }, { "left": "issue:43531", "right": "issue:43643", "accept": false, - "reason": "Unrelated issues: sliding-window handling in Qwen3-MoE vs AutoConfig missing fields." + "reason": "Sliding-window behavior in Qwen3-MoE is unrelated to `trust_remote_code=True` returning incomplete configs." }, { "left": "issue:43611", "right": "issue:46665", "accept": false, - "reason": "Both involve loading models, but one is a base_model_prefix regression and the other is from_pretrained behavior inside classes." + "reason": "Both concern loading behavior, but the specific regression is different: base_model_prefix loading vs from_pretrained inside Python classes." }, { "left": "issue:43653", "right": "issue:43819", "accept": false, - "reason": "Tokenizer special-token registration and DAC from_latents/STE mismatch are unrelated bugs." + "reason": "Tokenizer special-token registration bug is unrelated to DAC.from_latents not matching forward without STE." }, { "left": "issue:44220", "right": "issue:44661", "accept": false, - "reason": "Different problems: audio feature extraction vs add-new-model-like / tokenizer mapping logic." + "reason": "Different features entirely: fbank extraction vs add-new-model-like failing on TOKENIZER_MAPPING_NAMES." }, { "left": "issue:44186", "right": "issue:44898", "accept": false, - "reason": "Different model families and failures: tokenizer padding/truncation crash vs Perceiver image classification rescaling issue." + "reason": "LayoutLMv2 tokenizer padding/NER crash is not the same bug as Perceiver image classification at non-default resolution." }, { "left": "issue:43618", "right": "issue:43697", "accept": false, - "reason": "Different symptoms and code paths: CLIPOutput attentions field vs RTDetrV2 output divergence." + "reason": "CLIPOutput attentions assignment and RTDetrV2 output drift are separate model bugs." }, { "left": "issue:41720", "right": "issue:44991", "accept": false, - "reason": "Different areas: Qwen3 device-mapping crash vs tokenizer load failure for a specific model." + "reason": "Qwen3 auto device-map CUDA assert and est-roberta tokenizer loading failure are unrelated." }, { "left": "issue:30064", "right": "issue:36331", "accept": false, - "reason": "No overlap in underlying bug: image processor segmentation maps vs CustomTrainer loss signature." + "reason": "Image processor segmentation-map failure and `compute_loss` signature mismatch are completely different issues." }, { "left": "issue:43643", "right": "issue:44877", "accept": false, - "reason": "Both are config/loading related, but one is missing fields with trust_remote_code and the other is strict config validation for granite_speech." + "reason": "Both are config-loading problems, but one is missing fields from `trust_remote_code`, the other is strict config rejection for granite_speech." }, { "left": "issue:43550", "right": "issue:43819", "accept": false, - "reason": "Different model/runtime problems: Bamba torch.compile/SDPA vs DAC latent decoding mismatch." + "reason": "Torch.compile SDPA failure in Bamba is unrelated to DAC latent decoding mismatch." }, { "left": "issue:43335", "right": "issue:43756", "accept": false, - "reason": "Different models and defects: SwitchTransformers sparse-layer creation vs Smollm3 RoPE layer handling." + "reason": "SwitchTransformers sparse-layer creation and Smollm3 RoPE-layer dropping are different model-configuration bugs." }, { "left": "issue:44843", "right": "issue:45081", "accept": false, - "reason": "Same helper area, but distinct bugs and triggers: offline model_info call vs backend_tokenizer attribute access." + "reason": "Same helper name, but different root causes: offline `model_info()` call vs missing `backend_tokenizer` attribute." }, { "left": "issue:44821", "right": "issue:45084", "accept": false, - "reason": "Unrelated failures: AutoImageProcessor loading vs template compilation error." + "reason": "AutoImageProcessor URL loading and template-node compilation are unrelated failures." }, { "left": "issue:44186", "right": "issue:44442", "accept": false, - "reason": "Tokenizer crashes in LayoutLMv2 vs AutoTokenizer failing to load a different tokenizer are separate issues." + "reason": "LayoutLMv2 tokenizer crashes and AutoTokenizer failing on FastSpeech2ConformerTokenizer are distinct tokenizer bugs." }, { "left": "issue:39692", "right": "issue:45834", "accept": false, - "reason": "Both are multimodal docs/model problems, but the bugs are unrelated and model-specific." + "reason": "Documentation/example errors for SigLIP2 are unrelated to Kosmos2.5 OCR long-input indexing." }, { "left": "issue:44448", "right": "issue:44671", "accept": false, - "reason": "Both are v5 behavior regressions, but they affect different models and likely different underlying causes." + "reason": "Different models and causes: Pegasus v4/v5 output drift vs CamemBERT masked-LM prediction regression." }, { "left": "issue:43540", "right": "issue:44008", "accept": false, - "reason": "Different model families and error sources: video processing ValueError vs Gemma3n attribute collision." + "reason": "Qwen3OmniMoe video processing error and Gemma3n audio mask variable collision are unrelated multimodal bugs." }, { "left": "issue:43644", "right": "issue:43749", "accept": false, - "reason": "Different issue types: non-persistent buffer initialization vs FSDP CPU RAM efficient loading." + "reason": "Non-persistent buffer initialization junk and FSDP CPU RAM efficient loading are different v5 regressions." }, { "left": "issue:34634", "right": "issue:36010", "accept": false, - "reason": "Both mention processor/import problems, but Bark voice preset and GenerationMixin import errors are unrelated." + "reason": "BarkProcessor voice preset handling and GenerationMixin import failure do not describe the same defect." }, { "left": "issue:45183", "right": "issue:45479", "accept": false, - "reason": "Different validation semantics across different model families; not the same bug." + "reason": "Input validation error messages across models are unrelated to the degenerate zero-loss classification bug." }, { "left": "issue:45584", "right": "issue:45834", "accept": false, - "reason": "Different modalities and failure modes: Whisper empty transcription vs Kosmos2.5 long OCR input." + "reason": "Whisper empty-transcription generation failure is not the same as Kosmos2.5 long OCR input handling." } ] }, @@ -6096,8 +6028,8 @@ "issue:44991|issue:46291", "issue:43421|issue:44811", "issue:44561|issue:45020", - "issue:43950|issue:45987", "issue:43716|issue:44008", + "issue:43950|issue:45987", "issue:43931|issue:46846", "issue:46294|issue:46463", "issue:43825|issue:43827", @@ -6112,203 +6044,206 @@ "issue:44671|issue:45701", "issue:42175|issue:45081" ], - "prepared_review_unit_hash": "373853cbbb44fbb8cce55d8e88565cb8c7c9ec7bd3271dd4ed918e157403532a", + "prepared_review_unit_hash": "667680724238723077c7d844a2931b7d5e3aaf8ae47df7cb973d6057fd5353c7", "component_budget": { "node_count": 500, "item_count": 500, "soft_pair_count": 1839, - "serialized_chars": 953913, - "estimated_input_tokens": 238479, - "estimated_eval_tokens": 477214 + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 27, - "serialized_chars": 36216, - "estimated_input_tokens": 9054, - "estimated_eval_tokens": 18364 + "serialized_chars": 36220, + "estimated_input_tokens": 9055, + "estimated_eval_tokens": 18366 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is heterogeneous and mostly contains distinct Transformer v5-era regressions across tokenization, loading, multimodal inputs, and dtype/device handling. The soft pairs are thematically similar in places, but none look like the same concrete bug/change.", - "confidence": 0.41, - "canonical_issue_reason": "issue:45020 is the broadest umbrella in the set, covering recent v5/remote_code breakages and serving as the least narrow representative of the cluster.", + "summary": "This cluster is mostly a loose mix of unrelated Transformers bugs, with one clear duplicate pair around tokenizer added-token spacing/slow-vs-fast inconsistency. I would only merge the tokenizer pair; the rest are different models, code paths, or failure modes.", + "confidence": 0.84, + "canonical_issue_reason": "issue:28218 is the most concrete description of the shared tokenizer regression (extra space after an added token), and issue:29159 reads like a broader manifestation of the same problem.", "canonical_pr_reason": null, - "best_issue_reason": "issue:45020 is the best single issue representative because it is the most general regression report and sits closest to the overall theme of the cluster.", + "best_issue_reason": "issue:28218 is the best issue anchor because it is specific, actionable, and captures the actual symptom more directly than the broader tokenizer-consistency report.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43606", "right": "issue:45072", "accept": false, - "reason": "Both are inference-time mismatch bugs, but one is CPU offload/device placement on a specific model and the other is bfloat16 dtype mismatch on different wrappers; not the same code-path defect." + "reason": "Both mention model/backend incompatibility, but one is a CPU-offload device mismatch and the other is a bfloat16 dtype mismatch on different models." }, { "left": "issue:43322", "right": "issue:45685", "accept": false, - "reason": "Completely different failures: Llava Next loading segfault vs an MPS histogram op limitation in MoE code." + "reason": "Different failures: Llava Next loading segfault versus an MPS histogram kernel not implemented for Int." }, { "left": "issue:43550", "right": "issue:45072", "accept": false, - "reason": "One is a torch.compile/SDPA failure for Bamba, the other is dtype mismatch inference bugs; similar area, different underlying issue." + "reason": "torch.compile+SDPA compile failure is a different code-path problem from dtype mismatch inference errors." }, { "left": "issue:43388", "right": "issue:46032", "accept": false, - "reason": "Different bugs: last-batch label truncation in gather_for_metrics vs incorrect use_cache behavior in Mamba2Mixer." + "reason": "Metric gathering drops labels in the last batch; the other is incorrect cache behavior for seq_len>1. Not the same bug." }, { "left": "issue:44479", "right": "issue:45834", "accept": false, - "reason": "Both are multimodal/video-ish errors, but the specific models, failure modes, and code paths differ substantially." + "reason": "Both are multimodal input shape issues, but they affect different models and different preprocessing paths." }, { "left": "issue:43825", "right": "issue:43994", "accept": false, - "reason": "One is an error-message/documentation issue around pipeline removals; the other is a model output correctness problem." + "reason": "One is an error-message/translation-support issue; the other is a bad model output regression with AutoModel/pipeline." }, { "left": "issue:43782", "right": "issue:44977", "accept": false, - "reason": "Qwen3VL loading with weight_only=True and Qwen3.5 flash-attention generation are unrelated regression types." + "reason": "Weight-only loading failure and flash-attention generation failure are distinct problems." }, { "left": "issue:43582", "right": "issue:46512", "accept": false, - "reason": "Apple Silicon TypeError in warmup code vs stale config dtype after from_config; not the same bug." + "reason": "Apple Silicon warmup TypeError is unrelated to stale config dtype after from_config(dtype=...)." }, { "left": "issue:43577", "right": "issue:43818", "accept": false, - "reason": "Blip2 dtype propagation vs Video-LLaVA missing temporal attention / shared weights are different model-loading/model-architecture problems." + "reason": "BLIP2 dtype propagation and Video-LLaVA temporal-attention/weight-sharing are different model-specific bugs." }, { "left": "issue:44991", "right": "issue:46291", "accept": false, - "reason": "Tokenizer loading failure for a specific model vs import failure due to kernels package version/revision requirements; different dependency issues." + "reason": "Tokenizer loading failure and kernels dependency/version import failure are unrelated." }, { "left": "issue:43421", "right": "issue:44811", "accept": false, - "reason": "Special-token post-processor update at runtime and Whisper batch_decode skip_special_tokens are related to tokenization, but not the same code-path bug." + "reason": "Post-processor not updating after special-token changes is different from batch_decode ignoring skip_special_tokens." }, { "left": "issue:44561", "right": "issue:45020", "accept": false, - "reason": "44561 is one specific regression caused by removing is_torch_fx_available; 45020 is a broad umbrella for remote_code breakages." + "reason": "Both involve recent-version remote_code breakage, but the evidence is too broad to treat them as the same concrete issue." }, { - "left": "issue:43950", - "right": "issue:45987", + "left": "issue:43716", + "right": "issue:44008", "accept": false, - "reason": "Non-persistent buffer corruption in from_pretrained is unrelated to StaticCache.get_seq_length returning the wrong type/shape." + "reason": "Image preprocessor/model dtype mismatch is unrelated to a variable-name collision in Gemma3nModel.forward()." }, { - "left": "issue:43716", - "right": "issue:44008", + "left": "issue:43950", + "right": "issue:45987", "accept": false, - "reason": "Different multimodal bugs: dtype mismatch in Mistral-3 preprocessing vs variable-name collision causing an AttributeError in Gemma3n." + "reason": "Silent corruption of non-persistent buffers is a different regression from StaticCache.get_seq_length returning the wrong type." }, { "left": "issue:43931", "right": "issue:46846", "accept": false, - "reason": "Both mention shape issues in Qwen VL models, but one is a specific load-time weight mismatch and the other is TP-plan missing entries causing OOM/shape errors." + "reason": "Qwen3-VL weight-shape loading errors and Qwen3.5 TP-plan OOM/shape issues are not the same bug." }, { "left": "issue:46294", "right": "issue:46463", "accept": false, - "reason": "Qwen3.5 video feature/token mismatch is unrelated to the T5 decoder-input validation error-message issue." + "reason": "Qwen3.5 video chat-template mismatch and T5 decoder-input validation messaging are unrelated." }, { "left": "issue:43825", "right": "issue:43827", "accept": false, - "reason": "Both are about v5 pipeline removals, but one is an error message and the other is docs cleanup; not the same defect." + "reason": "One concerns a runtime pipeline error message; the other is documentation still referencing pipeline() after removals." }, { "left": "issue:28218", "right": "issue:29159", - "accept": false, - "reason": "The titles are tokenizer-related, but one is specifically added-token spacing while the other is a broader slow/fast inconsistency report." + "accept": true, + "reason": "These look like the same tokenizer regression: extra spacing after added tokens fits the broader slow/fast tokenizer inconsistency report." }, { "left": "issue:44805", "right": "issue:44936", "accept": false, - "reason": "An indexing shape mismatch is not the same bug as trainer.evaluate failing after trainer.train()." + "reason": "Mask/index shape mismatch at inference is unrelated to trainer.evaluate() failing after trainer.train()." }, { "left": "issue:44610", "right": "issue:45072", "accept": false, - "reason": "Processor image-size mismatch and dtype mismatch in inference are different issues." + "reason": "Processor image-size mismatch and dtype mismatch inference failures are different issues." }, { "left": "issue:43540", "right": "issue:45381", "accept": false, - "reason": "Both are video-input regressions, but they affect different models and likely different preprocessing/position-id logic." + "reason": "Both are video-input bugs, but they affect different Qwen model families and different failure modes." }, { "left": "issue:43720", "right": "issue:45072", "accept": false, - "reason": "Packed-weight handling during accelerate loading is unrelated to bfloat16 dtype mismatch inference bugs." + "reason": "Packed-weight loading during accelerate is unrelated to dtype mismatches in inference." }, { "left": "issue:43749", "right": "issue:44291", "accept": false, - "reason": "FSDP CPU RAM-efficient loading broken and init_empty_weights rejecting _is_hf_initialized are both loading-related, but not the same underlying failure." + "reason": "FSDP CPU RAM-efficient loading is a different loading path from init_empty_weights rejecting an unexpected argument." }, { "left": "issue:43577", "right": "issue:43582", "accept": false, - "reason": "Blip2 dtype propagation and Apple Silicon caching_allocator_warmup TypeError are unrelated." + "reason": "These are unrelated dtype/model-loading and Apple Silicon allocator problems." }, { "left": "issue:43645", "right": "issue:46665", "accept": false, - "reason": "Both concern model instantiation patterns, but one is a Jupyter/custom-model regression and the other is inconsistent from_pretrained behavior inside classes." + "reason": "Jupyter notebook custom-model initialization and from_pretrained-inside-classes are separate contexts with different failure causes." }, { "left": "issue:44671", "right": "issue:45701", "accept": false, - "reason": "CamemBERT masked-LM correctness and version-dependent tokenization changes are not the same bug." + "reason": "Masked LM prediction regression and tokenizer behavior changes are too different to merge as one bug." }, { "left": "issue:42175", "right": "issue:45081", "accept": false, - "reason": "Packaging/back-end dependency omission is unrelated to the Mistral tokenizer regex patch crash." + "reason": "Package/backend install issues are unrelated to a Mistral tokenizer regex patch crash." } ] }, - "evaluator_result": null, - "evaluator_used": false, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays mostly conservative on the soft pairs. The only accepted pair, issue:28218 with issue:29159, is a plausible tokenizer duplicate based on the titles and cluster context. The other verdicts avoid overreach and correctly treat different models, code paths, or failure modes as separate." + }, + "evaluator_used": true, "retried": false, - "accepted_nontrivial_soft_edge": false, + "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null }, @@ -6404,202 +6339,202 @@ "issue:42490|issue:43257", "issue:43720|issue:44610" ], - "prepared_review_unit_hash": "529847686d1ac2bc65569b09de977b24582adabc0317aa1b1d64130d208c6b1b", + "prepared_review_unit_hash": "4b884a015de29ab89f0312a8a215e01ccef9cb2ab4407b0f7154239fde6a38f1", "component_budget": { "node_count": 500, "item_count": 500, "soft_pair_count": 1839, - "serialized_chars": 953913, - "estimated_input_tokens": 238479, - "estimated_eval_tokens": 477214 + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 28, - "serialized_chars": 36526, - "estimated_input_tokens": 9132, - "estimated_eval_tokens": 18520 + "serialized_chars": 36524, + "estimated_input_tokens": 9131, + "estimated_eval_tokens": 18518 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "These soft pairs are mostly superficial similarities across unrelated bugs (loaders, processors, trainers, and model-specific regressions). I do not see any pair that clearly reflects the same underlying bug or change, so no soft edges should be accepted.", - "confidence": 0.92, - "canonical_issue_reason": "No single issue cleanly serves as a canonical duplicate target; the set spans unrelated subsystems and failure modes.", + "summary": "All suggested soft matches are false positives: the paired issues mostly share only broad themes like loading, trainers, or image processors, but they describe different models, different code paths, or different failure modes. None of these pairs look like the same underlying bug/change.", + "confidence": 0.95, + "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "No issue is a strong global representative for deduping this cluster, since the apparent similarities are broad and mostly incidental.", + "best_issue_reason": null, "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:34634", "right": "issue:36331", "accept": false, - "reason": "Different areas and failures: BarkProcessor voice preset vs CustomTrainer compute_loss signature change." + "reason": "Unrelated subsystems: BarkProcessor voice presets vs a Trainer compute_loss signature regression." }, { "left": "issue:45003", "right": "issue:46285", "accept": false, - "reason": "One is unsafe sys.modules access in modeling_utils; the other is a CLIP model structure inconsistency. Related theme, but not the same bug." + "reason": "Both touch modeling code, but one is unsafe sys.modules access and the other is CLIP text-model structure mismatch; different bugs." }, { "left": "issue:46697", "right": "issue:46728", "accept": false, - "reason": "Both are image-processor shape/padding issues, but for different models and different logic paths." + "reason": "Both are image-processor counting/padding issues, but for different models and different mechanisms." }, { "left": "issue:43653", "right": "issue:44912", "accept": false, - "reason": "BigBirdTokenizer special-token handling and MXFP4 quantization fallback are unrelated." + "reason": "Tokenizer special-token registration vs MXFP4 quantization loading; no shared bug." }, { "left": "issue:43262", "right": "issue:43819", "accept": false, - "reason": "Audio processor sampling-rate default vs DAC latent reconstruction mismatch are different bugs." + "reason": "Audio-related, but one is chat-template sampling-rate defaulting and the other is DAC latent reconstruction/STE behavior." }, { "left": "issue:43782", "right": "issue:45325", "accept": false, - "reason": "Different Qwen-VL regressions affecting different loading/rope-index code paths." + "reason": "Both involve Qwen VL models, but one is a from_pretrained weight_only load error and the other is a rope index scaling bug." }, { "left": "issue:45081", "right": "issue:45406", "accept": false, - "reason": "Mistral regex patch crash and Gemma4Processor missing _tokenizer are separate tokenizer/processor issues." + "reason": "Both are processor/tokenizer loading problems, but they fail on different missing attributes and different model families." }, { "left": "issue:45372", "right": "issue:46326", "accept": false, - "reason": "Different mistral_common-related failures: missing ReasoningEffort import vs continue_final_message behavior." + "reason": "One is an import dependency break in mistral_common; the other is continue_final_message behavior for a different tokenizer path." }, { "left": "issue:43582", "right": "issue:44857", "accept": false, - "reason": "Apple Silicon allocator TypeError and CUDA float16 loss crash are unrelated." + "reason": "A platform-specific allocator TypeError and a CUDA AMP loss crash are not the same code-path problem." }, { "left": "issue:44589", "right": "issue:45042", "accept": false, - "reason": "Float8 storage lookup failure is not the same as PIL backend requiring torchvision." + "reason": "Float8 storage lookup failure vs PIL backend torchvision dependency regression; unrelated." }, { "left": "issue:43550", "right": "issue:44912", "accept": false, - "reason": "torch.compile SDPA failure in Bamba is unrelated to quantized loading fallback in git-oss-20b." + "reason": "Different model/runtime bugs: torch.compile with SDPA vs MXFP4 loading fallback." }, { "left": "issue:43742", "right": "issue:45020", "accept": false, - "reason": "KeyError loading a specific model is not the same as the broad remote_code regression." + "reason": "A specific MobileLLM load key error is not the same as the broad remote_code regression." }, { "left": "issue:43329", "right": "issue:45600", "accept": false, - "reason": "Multimodal token counting video bug and auto_mappings CI breakage are unrelated." + "reason": "Video multimodal token counting bug vs auto_mappings referencing removed configs; different affected code." }, { "left": "issue:44060", "right": "issue:44368", "accept": false, - "reason": "Both mention tie_word_embeddings/warnings in Qwen variants, but the reported behavior and affected code paths are not clearly the same bug." + "reason": "Both mention tied-weights warnings, but they concern different Qwen variants and different causes." }, { "left": "issue:45084", "right": "issue:45538", "accept": false, - "reason": "Template compilation error and CLIP tokenizer max_length bug are unrelated." + "reason": "Template compilation error and CLIPTokenizer max_length issue are unrelated." }, { "left": "issue:43452", "right": "issue:44991", "accept": false, - "reason": "gguf_file loading breakage and a specific tokenizer load failure are different issues." + "reason": "Both are from_pretrained loading failures, but one is gguf_file handling and the other is a tokenizer regression for a specific model." }, { "left": "issue:43526", "right": "issue:43697", "accept": false, - "reason": "BEiT label reduction bug and RTDetrV2 output drift are separate image-model regressions." + "reason": "One is a BeitImageProcessorFast label-reduction bug; the other is a model output mismatch in RTDetrV2." }, { "left": "issue:43257", "right": "issue:43329", "accept": false, - "reason": "Qwen3 MoE loading/weight conversion and multimodal token counting are unrelated." + "reason": "Qwen3 MoE loading with accelerate/deepspeed is unrelated to multimodal token counting in video branches." }, { "left": "issue:43526", "right": "issue:44554", "accept": false, - "reason": "Different models and failure modes: label reduction vs MPS attention correctness." + "reason": "Different domains and failure modes: image-label reduction vs MPS attention correctness." }, { "left": "issue:43906", "right": "issue:46209", "accept": false, - "reason": "A reproduction of an older issue is not the same as an FP8 inference failure 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": "ImportError for GenerationMixin and compute_loss signature mismatch are unrelated regressions." + "reason": "Both are API breakages, but import path breakage and Trainer signature change are distinct regressions." }, { "left": "issue:43819", "right": "issue:43927", "accept": false, - "reason": "DAC latent reconstruction mismatch and DiaConfig save/load token-ID loss are different bugs." + "reason": "DAC latent decoding mismatch vs DiaConfig token-ID persistence; different components." }, { "left": "issue:36010", "right": "issue:43645", "accept": false, - "reason": "GenerationMixin import failure and Jupyter custom-model initialization breakage are different code paths." + "reason": "GenerationMixin import failure is unrelated to custom-model initialization in notebooks." }, { "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 bug and PIL/torchvision dependency regression are unrelated." }, { "left": "issue:43638", "right": "issue:44936", "accept": false, - "reason": "DeepSpeed ZeRO3 index error and trainer.evaluate() after train() failure are not the same bug." + "reason": "Training/evaluation failures, but one is a deepspeed zero3 index error and the other is evaluate-after-train behavior; no clear same root cause." }, { "left": "issue:42175", "right": "issue:46291", "accept": false, - "reason": "Packaging/backends issue for torch install and kernels import-version requirement are unrelated." + "reason": "Different packaging/import failures: missing TensorFlow backend in install extras vs kernels version/revision requirement." }, { "left": "issue:42490", "right": "issue:43257", "accept": false, - "reason": "Trailing-slash local path warning and Qwen3 MoE weight conversion are unrelated." + "reason": "Local-path trailing-slash import warning and Qwen3 MoE accelerate/deepspeed loading are separate loading bugs." }, { "left": "issue:43720", "right": "issue:44610", "accept": false, - "reason": "Packed-weight loading bug and OmDet-Turbo processor size mismatch are different problems." + "reason": "Packed-weight unpacking during accelerate loading and OmDet-Turbo processor size mismatch are unrelated." } ] }, @@ -6689,8 +6624,8 @@ "issue:42915|issue:43742", "issue:43749|issue:44756", "issue:34567|issue:40990", - "issue:44291|issue:45042", "issue:43650|issue:43906", + "issue:44291|issue:45042", "issue:42915|issue:44821", "issue:29159|issue:34567", "issue:43606|issue:45859", @@ -6701,196 +6636,196 @@ "issue:43525|issue:46291", "issue:44568|issue:45356" ], - "prepared_review_unit_hash": "f298fb9b1a898533cfa97f412117adcb3c6cb85f05ef8aae128209c35b655b0d", + "prepared_review_unit_hash": "e2e421971c3dbc471809f2450382084f156e860c7eae227c3ab9466f61a5b4b4", "component_budget": { "node_count": 500, "item_count": 500, "soft_pair_count": 1839, - "serialized_chars": 953913, - "estimated_input_tokens": 238479, - "estimated_eval_tokens": 477214 + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 27, - "serialized_chars": 36011, - "estimated_input_tokens": 9003, - "estimated_eval_tokens": 18262 + "serialized_chars": 36014, + "estimated_input_tokens": 9004, + "estimated_eval_tokens": 18264 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This set is mostly heterogeneous. The soft pairs share broad subsystems at best, but they do not look like the same underlying bug/change, so all soft edges should be rejected.", - "confidence": 0.95, - "canonical_issue_reason": null, + "summary": "This cluster is a grab bag of mostly unrelated closed issues across loading, tokenization, multimodal/video, quantization, cache, and CPU-offload regressions. None of the soft pairs look like true duplicates; they share only broad subsystem keywords, not the same concrete bug path.", + "confidence": 0.78, + "canonical_issue_reason": "issue:43749 is the strongest anchor: it has the most discussion/inbound references and describes a concrete, high-impact loading regression with a clearly reproducible scope.", "canonical_pr_reason": null, - "best_issue_reason": "If an issue must be used as a loose anchor, issue:43329 is the most concrete code-path bug report (specific undefined variables in the video multimodal helper), but it is not a true representative of the whole set.", + "best_issue_reason": "issue:43749 is the best representative issue for the cluster because it is well-scoped, impactful, and more central than the other candidates; it is also easier to use as a triage anchor than the narrower model-specific reports.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:39692", "right": "issue:44912", "accept": false, - "reason": "Different problems: SigLIP2 doc-example errors vs GPT-oss MXFP4 quantization fallback/OOM behavior." + "reason": "Different models and failures: SigLIP2 doc-example errors vs gpt-oss-20b MXFP4 quantization fallback/OOM behavior." }, { "left": "issue:43746", "right": "issue:44811", "accept": false, - "reason": "Unrelated models and failure modes: PEFT/local-checkpoint loading vs Whisper batch_decode skipping special tokens." + "reason": "Different code paths: PEFT local-checkpoint loading for GraniteSpeech vs Whisper processor batch_decode skip_special_tokens handling." }, { "left": "issue:43854", "right": "issue:44977", "accept": false, - "reason": "Different model-loading/regression reports; one is a unit-test load issue, the other is a flash-attention generation bug." + "reason": "Different model families and bugs: GLM-4.7-Flash unit-test load failure vs Qwen3.5 flash-attention generation regression." }, { "left": "issue:43643", "right": "issue:45759", "accept": false, - "reason": "Both touch config loading, but one is about trust_remote_code dropping fields and the other about unwrapping text_config for composite multimodal configs." + "reason": "Both involve config loading, but one is missing fields with trust_remote_code and the other is text_config unwrapping for composite multimodal configs; not the same bug." }, { "left": "issue:38175", "right": "issue:44610", "accept": false, - "reason": "Different symptoms and code paths: zero probabilities in SigLIP2 vs processor/model image-size mismatch in OmDet-Turbo." + "reason": "Different models and symptoms: siglip2 zero probabilities vs OmDet-Turbo processor/model input-size mismatch." }, { "left": "issue:43257", "right": "issue:43696", "accept": false, - "reason": "Qwen MoE weight conversion/loading bug vs a separate CUDA OOM report for GPT-oss-20b." + "reason": "Qwen3 MOE weight conversion under accelerate+deepspeed is unrelated to GPT-oss-20b CUDA OOM." }, { "left": "issue:43262", "right": "issue:44811", "accept": false, - "reason": "Audio chat-template sampling-rate default vs Whisper decode token-skipping; same broad area only, not same bug." + "reason": "Audio template sampling-rate defaulting and Whisper decode token filtering are separate processor behaviors." }, { "left": "issue:44387", "right": "issue:44617", "accept": false, - "reason": "Both mention OOM, but one is a Transformers 5.x int4 memory regression and the other is a model-specific video OOM report." + "reason": "Quantization memory regression in Transformers 5.x is not the same as Sam3Video model-specific OOM." }, { "left": "issue:43257", "right": "issue:43931", "accept": false, - "reason": "Both involve Qwen, but one is a MoE loading/accelerate-deepspeed issue and the other is a weight-shape mismatch in Qwen3-VL." + "reason": "Different Qwen-related failures: MOE weight conversion vs Qwen3-VL shape mismatch on load." }, { "left": "issue:45459", "right": "issue:45803", "accept": false, - "reason": "Both are about exception handling hiding errors, but they affect different tokenizer/feature code paths and are not the same bug." + "reason": "Both mention hidden errors, but they are different exception-handling bugs in different components and code paths." }, { "left": "issue:43329", "right": "issue:43450", "accept": false, - "reason": "Video-related, but one is undefined variables in multimodal token counting and the other is incorrect batched output shape from video processors." + "reason": "Undefined helper/variable usage in multimodal token counting is not the same as batched video processor shape regression." }, { "left": "issue:46421", "right": "issue:46424", "accept": false, - "reason": "Related cache subsystem, but one is recompilation from lazy init and the other is max_cache_length being honored only for Quantized Cache." + "reason": "Both are cache-related, but one is lazy-init recompilation and the other is max_cache_length applicability." }, { "left": "issue:43329", "right": "issue:44479", "accept": false, - "reason": "Shared video/VLM area, but 44479 is a broad regression across multiple models while 43329 is a specific undefined-variable bug in a helper." + "reason": "The first is an undefined-name bug in token counting; the second is a broader video-input regression across Qwen VL variants." }, { "left": "issue:42915", "right": "issue:43742", "accept": false, - "reason": "Different models and failures: FineGrainedFP8Config handling vs a key error when loading MobileLLM-125M." + "reason": "Different failures: FineGrainedFP8Config handling for Qwen3Moe vs MobileLLM-125M load key error." }, { "left": "issue:43749", "right": "issue:44756", "accept": false, - "reason": "Both mention memory/loading concerns, but one is FSDP CPU RAM efficient loading and the other is mmap on Strix Halo causing OOM." + "reason": "Both are loading/memory issues, but one is FSDP CPU RAM efficient loading and the other is mmap behavior on Strix Halo." }, { "left": "issue:34567", "right": "issue:40990", "accept": false, - "reason": "Unrelated training-state bookkeeping bug vs unexpectedly high perplexity on GPT-oss-20b." + "reason": "TrainerState token counters not updating is unrelated to GPT-oss-20b perplexity on WikiText-2." }, { - "left": "issue:44291", - "right": "issue:45042", + "left": "issue:43650", + "right": "issue:43906", "accept": false, - "reason": "Different regressions: init_empty_weights/_is_hf_initialized argument handling vs PIL image processors requiring torchvision." + "reason": "No shared bug: placeholder issue text versus an isolated reproduction of a different upstream issue." }, { - "left": "issue:43650", - "right": "issue:43906", + "left": "issue:44291", + "right": "issue:45042", "accept": false, - "reason": "Not the same issue; one is a vague data request and the other is an isolated reproduction of a different tracked bug." + "reason": "Model init_empty_weights argument handling is unrelated to PIL image processor torchvision dependency regression." }, { "left": "issue:42915", "right": "issue:44821", "accept": false, - "reason": "Different subsystems: FP8 model loading failure vs loading AutoImageProcessor from URL." + "reason": "GPT-oss-20b loading/quantization issues do not match AutoImageProcessor URL-loading failure." }, { "left": "issue:29159", "right": "issue:34567", "accept": false, - "reason": "Tokenizer inconsistency vs TrainerState token-count tracking; no concrete overlap." + "reason": "Tokenizer slow/fast inconsistencies are unrelated to TrainerState token-count tracking." }, { "left": "issue:43606", "right": "issue:45859", "accept": false, - "reason": "Both involve CPU offload, but they are model-specific failures on different code paths and should not be merged." + "reason": "Both mention CPU offload, but they affect different models and likely different offload paths." }, { "left": "issue:43927", "right": "issue:44912", "accept": false, - "reason": "DiaConfig save/load token-ID loss is unrelated to GPT-oss MXFP4 quantization fallback." + "reason": "DiaConfig save/load token-ID loss is unrelated to GPT-oss-20b MXFP4 quantization fallback." }, { "left": "issue:43540", "right": "issue:46701", "accept": false, - "reason": "Different multimodal pipelines and symptoms: Qwen3OmniMoe video input ValueError vs DiffusionGemma image-text-to-text pipeline failure." + "reason": "Different modalities and failures: Qwen3OmniMoe video-input ValueError vs DiffusionGemma pipeline output issue." }, { "left": "issue:28282", "right": "issue:29870", "accept": false, - "reason": "Completely different problems: missing PyTorch dependency ImportError vs Bart generate() UnboundLocalError." + "reason": "PyTorch-missing ImportError and Bart generate() decoder_attention_mask UnboundLocalError are unrelated." }, { "left": "issue:43454", "right": "issue:43577", "accept": false, - "reason": "AyaVision weight-tying bug vs Blip2 dtype propagation bug; different models and fixes." + "reason": "AyaVision weight-tying causing garbage text and Blip2 dtype propagation are different model-loading bugs." }, { "left": "issue:43525", "right": "issue:46291", "accept": false, - "reason": "Llama4Config missing pad_token_id is unrelated to the pipelines import failure with kernels v0.15." + "reason": "Missing pad_token_id in Llama4Config is unrelated to pipeline import failing on kernels version/revision requirements." }, { "left": "issue:44568", "right": "issue:45356", "accept": false, - "reason": "Different tokenizer regressions: special tokens not added vs codec handling/fix_mistral_regex warning in Kimi-K2.5." + "reason": "Both are tokenizer regressions, but they concern different tokenizer implementations, models, and failure modes." } ] }, @@ -6993,202 +6928,196 @@ "issue:43540|issue:44560", "issue:44792|issue:44964" ], - "prepared_review_unit_hash": "1f66082a52d6c100eaa61eabb662341467f13a2b0a298d1db2433ab14b6a38e0", + "prepared_review_unit_hash": "1102b6c8863cab3a7efd83da51ccf3c9883996c7d3cb5a6ce50845665e7c6267", "component_budget": { "node_count": 500, "item_count": 500, "soft_pair_count": 1839, - "serialized_chars": 953913, - "estimated_input_tokens": 238479, - "estimated_eval_tokens": 477214 + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 28, - "serialized_chars": 36012, - "estimated_input_tokens": 9003, - "estimated_eval_tokens": 18262 + "serialized_chars": 36020, + "estimated_input_tokens": 9005, + "estimated_eval_tokens": 18266 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is largely heterogeneous; the nearest umbrella is the broad tokenization-regression report, but the suggested soft pairs do not look like true duplicates or mergeable one-fix-fits-both cases.", - "confidence": 0.56, - "canonical_issue_reason": "issue:45701 is the broadest representative in the set: it describes a generic tokenization behavior change and has the most inbound references, while most other items are narrower model- or runtime-specific failures.", + "summary": "The cluster is mostly a set of unrelated issue candidates that only share broad Transformers theme words. I found no strong duplicate PRs, and only one possible soft duplicate is still too weak to accept under strict triage.", + "confidence": 0.91, + "canonical_issue_reason": "issue:45488 is the clearest, most actionable report: it is a concrete regression with a precise title, active discussion, and inbound references, making it a good cluster representative.", "canonical_pr_reason": null, - "best_issue_reason": "issue:45701 is the best cluster anchor because it is umbrella-like and can subsume tokenization regressions more naturally than the model-specific, dependency-specific, or hardware-specific reports.", + "best_issue_reason": "issue:45488 is the strongest overall issue to keep as the canonical example because it describes a specific user-facing regression with enough detail to triage and verify.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:44987", "right": "issue:45538", "accept": false, - "reason": "Different problems: model loading failure for a specific model vs. CLIPTokenizer max-length behavior." + "reason": "Different problems: one is a model-loading failure for a specific repo, the other is a CLIPTokenizer max-length bug." }, { "left": "issue:28282", "right": "issue:34689", "accept": false, - "reason": "Unrelated failures: missing PyTorch import error vs. a Llama 3.2 vision loading regression." + "reason": "Generic missing-PyTorch import error vs a specific Llama 3.2 vision model-loading regression; not the same bug." }, { "left": "issue:45538", "right": "issue:46869", "accept": false, - "reason": "Different subsystems: tokenizer max_length bug vs. YOLO multi-GPU device-map failure." + "reason": "Tokenizer configuration bug vs multi-GPU device-map failure; unrelated failure modes." }, { "left": "issue:44857", "right": "issue:46832", "accept": false, - "reason": "Both are detection-model issues, but they affect different code paths and symptoms; not the same bug." + "reason": "Both are detection-model issues, but one is AMP/CUDA crash in loss and the other is a forward shape/type error; different code paths." }, { "left": "issue:43697", "right": "issue:45357", "accept": false, - "reason": "RT-DETR output differences and Qwen3.5 save_pretrained key regression are separate issues." + "reason": "RT-DETR output regression vs Qwen3.5 save_pretrained key mismatch; different subsystems and symptoms." }, { "left": "issue:43425", "right": "issue:46291", "accept": false, - "reason": "Different dependency breakages with different triggers and fixes." + "reason": "Torch compatibility complaint vs kernels package revision/version requirement; unrelated dependency issues." }, { "left": "issue:33357", "right": "issue:34567", "accept": false, - "reason": "MacOS bus error on CLIP model loading is unrelated to TrainerState token accounting." + "reason": "MacOS bus error in CLIP model loading vs TrainerState token counter not updating; no shared underlying bug." }, { "left": "issue:43404", "right": "issue:45127", "accept": false, - "reason": "Both mention tied weights, but one is a missing tie in Mistral3 and the other is LoRA merge collapse with extended vocab; not the same concrete bug." + "reason": "Both mention tied weights, but one is a missing tie in a specific model class and the other is LoRA merge corruption with extended vocab; not the same fix." }, { "left": "issue:42915", "right": "issue:45538", "accept": false, - "reason": "FineGrainedFP8Config failure is unrelated to CLIPTokenizer max_length." + "reason": "FineGrainedFP8Config failure in Qwen3Moe vs CLIPTokenizer max_length bug; unrelated." }, { "left": "issue:45137", "right": "issue:45405", "accept": false, - "reason": "DeepSpeed ZeRO3 deque crash and an unreleased PEFT version bump are unrelated." + "reason": "DeepSpeed ZeRO3 deque crash vs PEFT version bump/release mismatch; completely different issues." }, { "left": "issue:42175", "right": "issue:43824", "accept": false, - "reason": "Packaging/backend installation issue vs. a model import failure; different causes." + "reason": "Packaging/backend install problem vs missing Qwen2.5-VL symbol import; different dependency layers." }, { "left": "issue:28282", "right": "issue:29127", "accept": false, - "reason": "Missing PyTorch import error is not the same as LayoutLMv3 error-message clarity." + "reason": "Missing PyTorch import error vs LayoutLMv3 error-message clarity issue; not duplicates." }, { "left": "issue:44062", "right": "issue:44991", "accept": false, - "reason": "Different tokenizer bugs: AddedToken kwargs collision vs. a specific tokenizer-loading regression." + "reason": "These are both tokenizer-loading failures, but the evidence is too broad to confirm the same root cause from the titles alone." }, { "left": "issue:44479", "right": "issue:46693", "accept": false, - "reason": "Video-input regression and FlashAttention long-context performance regression are different issues." + "reason": "Video-input regression for Qwen models vs FA2 long-context performance regression; different symptoms and likely different code paths." }, { "left": "issue:43742", "right": "issue:44291", "accept": false, - "reason": "Different model-loading failures with different root causes." + "reason": "MobileLLM key error vs unexpected _is_hf_initialized argument during init_empty_weights; both are load-time errors but not the same bug." }, { "left": "issue:44936", "right": "issue:46821", "accept": false, - "reason": "Trainer evaluate/train lifecycle bug is unrelated to TimesFM window_size handling." + "reason": "Trainer evaluate-after-train failure vs TimesFM window_size attribute error; unrelated components." }, { "left": "issue:43577", "right": "issue:46512", "accept": false, - "reason": "Both concern dtype/state consistency, but one is BLIP2 loading and the other is from_config staleness; not the same bug." + "reason": "Model/QFormer dtype staying float32 vs stale config.dtype after from_config; related theme but different bugs and fixes." }, { "left": "issue:42915", "right": "issue:43749", "accept": false, - "reason": "Different failures: FP8 config handling vs. FSDP CPU RAM efficient loading." + "reason": "Qwen3Moe FP8 config failure vs FSDP CPU RAM efficient loading breakage; different features." }, { "left": "issue:44849", "right": "issue:45701", "accept": false, - "reason": "Qwen3.5 hidden-states bug is unrelated to the broad tokenization regression in 45701." + "reason": "Qwen3.5 output_hidden_states bug vs version-dependent tokenization changes; unrelated." }, { "left": "issue:43452", "right": "issue:43582", "accept": false, - "reason": "GGUF tokenizer/model loading issue and Apple Silicon allocator warmup TypeError are unrelated." + "reason": "gguf_file loader breakage vs Apple Silicon allocator TypeError; different areas." }, { "left": "issue:27957", "right": "issue:28282", "accept": false, - "reason": "Flash Attention 2 compatibility for XLM-Roberta is unrelated to missing-PyTorch import errors." + "reason": "Flash Attention 2 issue in XLMRoberta vs missing-PyTorch import error; unrelated." }, { "left": "issue:24643", "right": "issue:29159", "accept": false, - "reason": "DeepSpeed training weight-shape error and slow/fast tokenizer inconsistency are different bugs." + "reason": "DeepSpeed training RuntimeError vs tokenizer slow/fast inconsistency; no shared code-path problem." }, { "left": "issue:44987", "right": "issue:45907", "accept": false, - "reason": "Model loading regression and list_repo_files revision forwarding are not the same underlying change." + "reason": "Model-loading regression vs list_repo_files revision-forwarding bug; not duplicates." }, { "left": "issue:41720", "right": "issue:43540", "accept": false, - "reason": "Auto device-map CUDA assert and Qwen3OmniMoe video processing are separate failure modes." + "reason": "Auto device-mapping CUDA assert vs video-input processing ValueError in Qwen3OmniMoe; distinct failures." }, { "left": "issue:42175", "right": "issue:44488", "accept": false, - "reason": "Package backend omission and a specific model-loading failure are unrelated." + "reason": "Backend installation packaging issue vs a model-loading failure for a specific repository; different problems." }, { "left": "issue:44568", "right": "issue:45488", "accept": false, - "reason": "Both are tokenizer regressions, but they affect different tokenizers and code paths." - }, - { - "left": "issue:43540", - "right": "issue:44560", - "accept": false, - "reason": "Both are video-related Qwen issues, but the affected models and concrete failures differ." + "reason": "Both are tokenizer regressions, but one is BOS/EOS insertion behavior and the other is ByteLevel pre-tokenizer being overridden; not the same bug." }, { "left": "issue:44792", "right": "issue:44964", "accept": false, - "reason": "Janus image-generation test failure is unrelated to Phi-4 multimodal model loading." + "reason": "Janus image-generation test failure vs Phi-4 multimodal model-loading issue; different models and failure points." } ] }, @@ -7270,8 +7199,8 @@ "issue:44062|issue:45081", "issue:45042|issue:45081", "issue:43582|issue:46821", - "issue:45672|issue:45674", "issue:43742|issue:44589", + "issue:45672|issue:45674", "issue:43756|issue:46399", "issue:43756|issue:45325", "issue:44568|issue:46491", @@ -7289,202 +7218,202 @@ "issue:45357|issue:45920", "issue:44625|issue:44877" ], - "prepared_review_unit_hash": "0159f8acda31d0e2876f5274538ed497e571e8284664cdb5ba819faa54742a28", + "prepared_review_unit_hash": "54d75f45b823f8e8382284e7b026704fb9f63638c384c64c6d9a5d9a768355a5", "component_budget": { "node_count": 500, "item_count": 500, "soft_pair_count": 1839, - "serialized_chars": 953913, - "estimated_input_tokens": 238479, - "estimated_eval_tokens": 477214 + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 27, - "serialized_chars": 35602, - "estimated_input_tokens": 8901, - "estimated_eval_tokens": 18058 + "serialized_chars": 35594, + "estimated_input_tokens": 8899, + "estimated_eval_tokens": 18054 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly heterogeneous false positives. The only convincing duplicate pair is the Qwen video-input regression reports (#44479 and #45381), which appear to describe the same underlying position-id bug. The rest differ by model, subsystem, or failure mode.", - "confidence": 0.82, - "canonical_issue_reason": "#44479 is the broadest and clearest report, covering the video-input regression across multiple Qwen model variants; it also matches the one plausible duplicate pair in the cluster.", + "summary": "Most pairs are clearly unrelated. The only plausible duplicate edge is the Qwen video-input regression pair (44479 \u2194 45381), which appears to describe the same underlying position-id bug across related Qwen VL variants.", + "confidence": 0.74, + "canonical_issue_reason": "issue:44479 is the best canonical issue because it is the broadest, most central report of the Qwen video-input regression and already spans multiple affected Qwen variants.", "canonical_pr_reason": null, - "best_issue_reason": "#44479 is the best global issue to anchor the cluster because it describes the shared concrete regression at a higher level and is more representative than the narrower repro in #45381.", + "best_issue_reason": "issue:44479 is the strongest representative issue in this cluster: it describes a concrete regression with multiple affected models and is the most likely target for consolidation.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:44479", "right": "issue:45381", "accept": true, - "reason": "Same underlying Qwen video-input regression: both describe wrong video position-id handling in the Qwen2.5-VL/Qwen3 family, with #45381 reading like a narrower reproduction of #44479." + "reason": "Both report the same Qwen VL video-input regression around wrong position ids / temporal scaling in v5.3.0; 45381 looks like a narrower manifestation of the broader bug in 44479." }, { "left": "issue:38175", "right": "issue:44898", "accept": false, - "reason": "Different model families and different failures: zero probabilities in SigLIP2 vs non-default-resolution interpolation in Perceiver." + "reason": "Different models and different failure modes: unexpected zero probabilities in SigLIP2 vs Perceiver non-default-resolution failure." }, { "left": "issue:44291", "right": "issue:44821", "accept": false, - "reason": "Unrelated loading/configuration problems: `_is_hf_initialized` with `init_empty_weights` vs `AutoImageProcessor` URL loading." + "reason": "Unrelated load-time errors: unexpected _is_hf_initialized argument vs AutoImageProcessor URL loading." }, { "left": "issue:43540", "right": "issue:44479", "accept": false, - "reason": "Both involve video inputs, but the model families and concrete code paths differ (`Qwen3OmniMoe` processing vs Qwen2.5-VL/Qwen3 position-id regression)." + "reason": "Both involve video inputs, but they affect different Qwen model families and do not look like the same underlying bug." }, { "left": "issue:34567", "right": "issue:36331", "accept": false, - "reason": "Different trainer bugs: token-count tracking vs `compute_loss` signature incompatibility." + "reason": "TrainerState token-count tracking and CustomTrainer compute_loss signature incompatibility are separate training issues." }, { "left": "issue:43257", "right": "issue:43756", "accept": false, - "reason": "Different models and different bugs: DeepSpeed/accelerate weight conversion vs RoPE-layer dropping in SmolLM3." + "reason": "Different subsystems: DeepSpeed/Accelerate MoE weight conversion vs SmolLM3 RoPE layer count regression." }, { "left": "issue:44062", "right": "issue:45081", "accept": false, - "reason": "Tokenizer special-token duplication is unrelated to the Mistral regex patch crash." + "reason": "Tokenizer AddedToken construction error is unrelated to the Mistral regex patch AttributeError." }, { "left": "issue:45042", "right": "issue:45081", "accept": false, - "reason": "Different subsystems: PIL image processor torchvision dependency vs tokenizer regex patching." + "reason": "Image processor torchvision dependency regression is unrelated to tokenizer regex patching." }, { "left": "issue:43582", "right": "issue:46821", "accept": false, - "reason": "Unrelated runtime errors in different areas: Apple Silicon allocator warmup vs TimesFM `window_size`." + "reason": "Apple Silicon allocator TypeError and TimesFM window_size AttributeError are unrelated." }, { - "left": "issue:45672", - "right": "issue:45674", + "left": "issue:43742", + "right": "issue:44589", "accept": false, - "reason": "Both touch 4-bit inference, but one is Gemma4 audio-module `finfo` on uint8 weights and the other is a BitsAndBytes skip-module configuration regression." + "reason": "MobileLLM key error and Float8 storage lookup failure are different loading problems." }, { - "left": "issue:43742", - "right": "issue:44589", + "left": "issue:45672", + "right": "issue:45674", "accept": false, - "reason": "Different loading failures with different causes: MobileLLM key error vs missing float8 storage object." + "reason": "Both mention 4-bit inference, but one is a torch.finfo uint8 audio-module error and the other is a BitsAndBytesConfig skip-module exclusion bug." }, { "left": "issue:43756", "right": "issue:46399", "accept": false, - "reason": "Different Qwen/SmolLM issues: RoPE-layer mismatch vs missing decorator for `apply_rotary_pos_emb` registration." + "reason": "SmolLM3 RoPE-layer issue and Qwen3.5 kernel-registration crash are unrelated." }, { "left": "issue:43756", "right": "issue:45325", "accept": false, - "reason": "Both are Qwen-family regressions, but one is SmolLM3 RoPE-layer handling and the other is Qwen2.5-VL temporal position-id scaling." + "reason": "Both mention RoPE, but the affected models and concrete failures are different." }, { "left": "issue:44568", "right": "issue:46491", "accept": false, - "reason": "Both are tokenizer regressions, but one concerns BOS/EOS insertion and the other leading-space encode/decode behavior." + "reason": "Tokenizer special-token insertion regression and CodeLlama round-trip spacing regression are different tokenizer bugs." }, { "left": "issue:45405", "right": "issue:45600", "accept": false, - "reason": "Completely different issues: a version pin bump vs stale auto-mappings references." + "reason": "Released-version requirement mismatch and removed-config reference in auto_mappings are unrelated CI/release issues." }, { "left": "issue:43716", "right": "issue:46701", "accept": false, - "reason": "Different model stacks and failure modes: Mistral-3 dtype mismatch vs DiffusionGemma pipeline breakage." + "reason": "Different models and pipelines: Mistral-3 dtype mismatch vs DiffusionGemma image-text-to-text pipeline failure." }, { "left": "issue:43577", "right": "issue:45072", "accept": false, - "reason": "Both mention dtype mismatches, but the affected models and concrete paths are different (BLIP-2 QFormer vs SwitchTransformers/TimmWrapperModel)." + "reason": "Both are dtype-related, but they concern different models/code paths and do not look mergeable as one fix." }, { "left": "issue:43531", "right": "issue:44821", "accept": false, - "reason": "Unrelated: Qwen3-MoE sliding-window behavior vs image processor URL loading." + "reason": "Qwen3-MoE sliding_window behavior is unrelated to AutoImageProcessor URL loading." }, { "left": "issue:43906", "right": "issue:46468", "accept": false, - "reason": "No concrete evidence they are the same bug; one is a reproduction of an older issue, the other is an official Nemotron example breakage." + "reason": "An isolated reproduction of a prior issue is not the same as the Nemotron example failure." }, { "left": "issue:24643", "right": "issue:29870", "accept": false, - "reason": "Different training/runtime errors: DeepSpeed weight dimensionality vs BART `generate()` decoder mask assignment." + "reason": "DeepSpeed weight-dimension error and BART generate decoder mask initialization error are unrelated." }, { "left": "issue:44737", "right": "issue:46463", "accept": false, - "reason": "Different code paths entirely: XLNet CPU tensor creation vs T5 decoder-input validation messaging." + "reason": "XLNet CPU-side arange bug and T5 decoder-input validation messaging are unrelated." }, { "left": "issue:43906", "right": "issue:46693", "accept": false, - "reason": "A reproduction of an older issue is not the same as a performance regression in FA2 long-context generation." + "reason": "Prior-issue reproduction and FA2 long-context performance regression are unrelated." }, { "left": "issue:45823", "right": "issue:46463", "accept": false, - "reason": "Different problems: Gemma4 device placement vs T5 error-message wording." + "reason": "Gemma4 device mismatch and T5 decoder validation message are unrelated." }, { "left": "issue:43452", "right": "issue:44488", "accept": false, - "reason": "Both are loading-related, but one is a GGUF/auto-class loading break and the other is a specific model load failure; no clear same root cause." + "reason": "gguf_file loading breakage and sleng-bert load failure are different loader problems." }, { "left": "issue:43697", "right": "issue:46829", "accept": false, - "reason": "Different models and outputs: RTDetrV2 changed predictions vs VideoPrism output field mismatch." + "reason": "RTDetrV2 output drift and VideoPrism hidden_states field mix-up are different model-output bugs." }, { "left": "issue:45357", "right": "issue:45920", "accept": false, - "reason": "Unrelated regressions: Qwen3.5 save_pretrained key mapping vs tokenizer token-ID generation errors." + "reason": "Qwen3.5 save_pretrained key regression and tokenizer token-id regression affect different code paths." }, { "left": "issue:44625", "right": "issue:44877", "accept": false, - "reason": "Different config issues: `num_labels` propagation vs strict config blocking `granite_speech` loading." + "reason": "Qwen3.5 num_labels propagation and granite_speech strict-config loading are separate config issues." } ] }, "evaluator_result": { "accept": true, - "feedback": "The summary is grounded in the packet and stays conservative. The only accepted soft edge, #44479 \u2194 #45381, is reasonably supported by the titles as the same Qwen2.5-VL/Qwen3 video-input position-id regression, while the remaining pair judgments are appropriately cautious false negatives. No PR-mergeability issue applies here." + "feedback": "The cluster summary is mostly grounded in the packet, and the only accepted soft edge (44479 \u2194 45381) is a conservative duplicate call based on strong title overlap within the same Qwen video-input regression family. I did not see evidence of overstated mergeability for any other pair." }, "evaluator_used": true, "retried": false, @@ -7582,200 +7511,197 @@ "issue:45542|issue:46178", "issue:43295|issue:45356" ], - "prepared_review_unit_hash": "2f208cfa7353844ce3906ec5b0684fdb5370d0916c4e517c5501c187b0322705", + "prepared_review_unit_hash": "81b8c520fc3e77676de82ba99032f07922ccf9722ff4b9cb0e5d595da142b8ae", "component_budget": { "node_count": 500, "item_count": 500, "soft_pair_count": 1839, - "serialized_chars": 953913, - "estimated_input_tokens": 238479, - "estimated_eval_tokens": 477214 + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 26, - "serialized_chars": 36198, - "estimated_input_tokens": 9050, - "estimated_eval_tokens": 18356 + "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 set is mostly a grab bag of unrelated bug reports. The only plausible duplicate pair is the v5 tied-weights/save-serialization regression in #44466 and #45005; everything else differs by model, code path, or symptom.", - "confidence": 0.58, - "canonical_issue_reason": "#44466 is the best anchor because it is the most concrete, reproducible regression report in the tied-weights serialization area, and it aligns best with the one plausible duplicate theme in the set.", + "summary": "This cluster is mostly a grab-bag of unrelated bug reports. The soft pairs share superficial vocabulary (loading, tokenizer, config, vision, etc.) but not the same concrete code path, so none look safe to merge as duplicates.", + "confidence": 0.94, + "canonical_issue_reason": "Issue 43749 is the broadest and most discussed loading/regression report here, so it is the closest thing to a cluster anchor, but the set is not a true duplicate cluster.", "canonical_pr_reason": null, - "best_issue_reason": "#44466 is the strongest standalone issue: it names a specific regression, a precise failure mode (`lm_head.weight` serialization), and a clear version comparison.", + "best_issue_reason": "Issue 43749 has the strongest activity and broad applicability among the items, making it the best representative issue in this mixed cluster.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:45663", "right": "issue:46846", "accept": false, - "reason": "Different models and different failure modes: Gemma-4 FSDP2 `KeyError` vs Qwen3.5 TP plan OOM/shape error." + "reason": "Different models and failures: Gemma-4 FSDP2 KeyError vs Qwen3.5 TP-plan OOM/shape error. Only both are distributed-training themed." }, { "left": "issue:43540", "right": "issue:46693", "accept": false, - "reason": "Unrelated issues: Qwen3OmniMoe video input ValueError vs FA2 long-context performance regression." + "reason": "One is a video-input validation error in Qwen3OmniMoe; the other is a long-context FlashAttention performance regression. Different bug classes." }, { "left": "issue:44913", "right": "issue:46121", "accept": false, - "reason": "Both touch rope/config handling, but one is silent config reload defaulting and the other is a type error in rope-key conversion; different bug paths." + "reason": "Both touch RoPE-related code, but one is config reload persistence and the other is JSON-loaded list handling in validation. Not the same defect." }, { "left": "issue:44737", "right": "issue:45588", "accept": false, - "reason": "XLNet CPU `torch.arange` bug and flash-attention `s_aux=None` crash are unrelated code paths." + "reason": "XLNet CPU placement bug and flash-attention None crash are unrelated code paths and symptoms." }, { "left": "issue:28282", "right": "issue:32090", "accept": false, - "reason": "ImportError for missing PyTorch and Trainer GPU broadcast TypeError are unrelated." + "reason": "ImportError for missing PyTorch and Trainer GPU broadcast TypeError are separate environment/runtime issues." }, { "left": "issue:46463", "right": "issue:46899", "accept": false, - "reason": "T5 decoder validation error-message bug is unrelated to Gemma4 multimodal 4-bit dtype casting." + "reason": "T5 decoder-input error-message issue vs Gemma-4 4-bit quantization dtype cast bug. Different models and root causes." }, { "left": "issue:43697", "right": "issue:45216", "accept": false, - "reason": "Different models and symptoms: RTDetrV2 output mismatch vs Qwen3.5 save-pretrained checkpoint corruption." + "reason": "Both are regressions, but one is RTDetrV2 output mismatch and the other is Qwen3.5 checkpoint saving corruption. Not the same code path." }, { "left": "issue:43526", "right": "issue:45479", "accept": false, - "reason": "One is a BeitImageProcessorFast label-reduction bug; the other is a sequence-classification zero-loss issue across models." + "reason": "BeitImageProcessorFast label reduction bug is unrelated to the sequence-classification zero-loss problem." }, { "left": "issue:43720", "right": "issue:44898", "accept": false, - "reason": "BitNet packed-weight loading failure and Perceiver non-default-resolution image-classification failure are different problems." + "reason": "BitNet packed-weight loading and Perceiver interpolation failure are different architectures and loading/inference paths." }, { "left": "issue:43643", "right": "issue:45375", "accept": false, - "reason": "Both involve missing config fields, but one is `trust_remote_code` AutoConfig loading and the other is a specific Qwen3.5MoeVisionConfig field dropped by strict validation." + "reason": "Both mention missing fields, but one is generic AutoConfig with trust_remote_code and the other is a specific Qwen3_5MoeVisionConfig strict-field drop." }, { "left": "issue:43761", "right": "issue:46829", "accept": false, - "reason": "Both mention hidden_states, but one returns `None` and the other puts `last_hidden_state` into the hidden_states slot; different models and opposite symptoms." + "reason": "Both involve output fields, but CLIPVisionModel hidden_states regression and VideoPrism output mis-assignment are separate model-specific bugs." }, { "left": "issue:45200", "right": "issue:46463", "accept": false, - "reason": "Gemma 4 `mm_token_type_ids` defaulting issue is unrelated to T5 decoder input validation messaging." + "reason": "Gemma 4 mm_token_type_ids defaulting and T5 decoder-input validation messaging are unrelated." }, { "left": "issue:44466", "right": "issue:45005", - "accept": true, - "reason": "Both point to the same underlying v5 tied-weights regression around serialization/save-pretrained behavior; #44466 is the more specific instance, while #45005 describes the broader translation-model manifestation." + "accept": false, + "reason": "Both relate loosely to tied weights, but one is device-dependent serialization and the other is translation-model tied-weight handling. Too broad to merge." }, { "left": "issue:44361", "right": "issue:44442", "accept": false, - "reason": "Tokenizer AttributeError on tasks vs AutoTokenizer failing to load a different tokenizer are different failures." + "reason": "Different tokenizers and failures: MLukeTokenizer task AttributeError vs AutoTokenizer loading FastSpeech2ConformerTokenizer." }, { "left": "issue:44493", "right": "issue:46050", "accept": false, - "reason": "Position-id key mismatch and quantization save/load breakage are unrelated." + "reason": "UNEXPECTED Key with position id and sinq quantized save/load breakage are distinct loading/serialization problems." }, { "left": "issue:43653", "right": "issue:44361", "accept": false, - "reason": "BigBird mask-token special-token registration bug and MLukeTokenizer task AttributeError are unrelated tokenizer issues." + "reason": "BigBirdTokenizer special-token decode bug is unrelated to MLukeTokenizer task AttributeError." }, { "left": "issue:41762", "right": "issue:44936", "accept": false, - "reason": "Gemma3 ZeRO-3 load-time index error and trainer.evaluate() post-train failure are different code paths." + "reason": "Gemma3 ZeRO-3 load-time IndexError and trainer.evaluate() after train() are not the same failure mode or subsystem." }, { "left": "issue:43452", "right": "issue:44843", "accept": false, - "reason": "gguf_file breaking tokenizer/model loading and offline-mode `model_info()` misuse are distinct loader bugs." + "reason": "Both affect from_pretrained, but gguf_file support breakage and offline model_info calls are different concrete bugs." }, { "left": "issue:43688", "right": "issue:43749", "accept": false, - "reason": "Aux-loss normalization bug in MoE models and FSDP CPU RAM efficient loading are unrelated." + "reason": "Aux-loss normalization in OLMoE/GPT Oss is unrelated to FSDP CPU RAM-efficient loading." }, { "left": "issue:43819", "right": "issue:44811", "accept": false, - "reason": "DAC latent decoding mismatch and Whisper `skip_special_tokens` handling are different APIs and symptoms." + "reason": "DAC latent reconstruction mismatch and Whisper skip_special_tokens decode behavior are different model APIs and bugs." }, { "left": "issue:38175", "right": "issue:43295", "accept": false, - "reason": "Siglip2 zero-probabilities issue and processor.tokenizer regression are not the same bug." + "reason": "Siglip2 zero-probabilities and processor.tokenizer/image forwarding regression are unrelated issues." }, { "left": "issue:43994", "right": "issue:44898", "accept": false, - "reason": "Siglip2 nonsense outputs and Perceiver resolution/interpolation failure are different model-specific issues." + "reason": "Both are vision-related, but Siglip2 nonsensical outputs and Perceiver interpolation failure are different models and root causes." }, { "left": "issue:43525", "right": "issue:46701", "accept": false, - "reason": "Llama4Config missing `pad_token_id` and DiffusionGemma image-text pipeline failure are unrelated." + "reason": "Llama4Config pad_token_id missing and DiffusionGemma image-text-to-text pipeline breakage are unrelated." }, { "left": "issue:45372", "right": "issue:46170", "accept": false, - "reason": "Gemma 4 processor import dependency issue and Windows ROCm/torch.distributed dependency issue are different environment/import failures." + "reason": "Gemma4 processor import failure via mistral_common and Windows ROCm torch.distributed dependency issues are separate dependency problems." }, { "left": "issue:45542", "right": "issue:46178", "accept": false, - "reason": "TensorBoard-without-TensorFlow backend error and unsupported `tensorflow_text` backend are separate backend-resolution problems." + "reason": "Both concern backend availability, but tensorboard-without-TF vs tensorflow_text backend lookup are not the same bug." }, { "left": "issue:43295", "right": "issue:45356", "accept": false, - "reason": "Processor/tokenizer regression in v4.57.5 and Kimi-K2.5 codec/mistral-regex regression are different bugs." + "reason": "Both are tokenizer regressions, but the reported behaviors and implicated code paths are different." } ] }, - "evaluator_result": { - "accept": true, - "feedback": "Grounded and conservative overall. The only accepted soft pair, #44466/#45005, is supported by the titles and stays appropriately cautious about a shared v5 tied-weights/save_pretrained regression. The remaining rejects do not overstate similarity and the summary does not claim unsupported duplicates." - }, - "evaluator_used": true, + "evaluator_result": null, + "evaluator_used": false, "retried": false, - "accepted_nontrivial_soft_edge": true, + "accepted_nontrivial_soft_edge": false, "error_kind": null, "error_message": null }, @@ -7847,8 +7773,8 @@ "issue:43653|issue:44568", "issue:43646|issue:43749", "issue:43452|issue:43577", - "issue:43688|issue:45704", "issue:43756|issue:46402", + "issue:43688|issue:45704", "issue:44843|issue:44991", "issue:43653|issue:43927", "issue:44464|issue:44743", @@ -7863,209 +7789,203 @@ "issue:44734|issue:46612", "issue:44466|issue:46050", "issue:43504|issue:43720", - "issue:44977|issue:46693", "issue:43504|issue:43550", + "issue:44977|issue:46693", "issue:44464|issue:44977", "issue:41720|issue:44464", "issue:43643|issue:44561", "issue:43577|issue:44554" ], - "prepared_review_unit_hash": "462b92e2a5f1a3be84e4121cb803b00e5a031d217392930f481a049219b15515", + "prepared_review_unit_hash": "dcec38aa04d8b0b0ac8eab02d339ccfa4559e434202d7c224374ba11a4f039b9", "component_budget": { "node_count": 500, "item_count": 500, "soft_pair_count": 1839, - "serialized_chars": 953913, - "estimated_input_tokens": 238479, - "estimated_eval_tokens": 477214 + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 28, - "serialized_chars": 35545, - "estimated_input_tokens": 8887, - "estimated_eval_tokens": 18030 + "serialized_chars": 35538, + "estimated_input_tokens": 8885, + "estimated_eval_tokens": 18026 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is too heterogeneous to treat as a duplicate set: the paired items share only broad themes like tokenization, loading, or generation, but each points to a different concrete code path or model-specific bug. No soft edge is a safe merge candidate.", - "confidence": 0.88, - "canonical_issue_reason": "No single canonical issue stands out because the cluster is a noisy semantic bucket spanning unrelated failures across tokenization, config loading, generation, and model-specific regressions.", - "canonical_pr_reason": "No pull requests are present in the cluster.", - "best_issue_reason": "No issue is a reliable global representative; choosing one would misrepresent the cluster\u2019s mixed topics and distinct root causes.", - "best_pr_reason": "No pull requests are present in the cluster.", + "summary": "This is a loose, heterogeneous set of Transformers issues: tokenization regressions, model-loading failures, generation/cache bugs, and a few unrelated performance/config problems. None of the soft pairs look like the same concrete bug or change.", + "confidence": 0.91, + "canonical_issue_reason": "If a representative issue is needed, issue 44464 is the best anchor: it is a concrete, reproducible generation regression and sits closest to several other generation/cache-related soft matches in this set.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 44464 is the most usable representative because it is specific, actionable, and broadly similar in theme to the other generation-related items without being overly model-specific.", + "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43531", "right": "issue:43742", "accept": false, - "reason": "Both are model-loading related on paper, but one is a Qwen3-MoE sliding_window issue and the other is a MobileLLM key error; different models and failure modes." + "reason": "Different bug classes and models: sliding_window/recurrent-state behavior in Qwen3-MoE vs a key error while loading MobileLLM-125M." }, { "left": "issue:42898", "right": "issue:46491", "accept": false, - "reason": "Both concern tokenizer behavior, but one is a v5 clean_up_tokenization_spaces change and the other is a CodeLlama leading-space round-trip regression; different tokenizer paths." + "reason": "Both are tokenizer regressions, but 42898 is a broad v5 clean_up_tokenization_spaces change while 46491 is a CodeLlama leading-space round-trip bug; different code path and symptom." }, { "left": "issue:29159", "right": "issue:33357", "accept": false, - "reason": "Slow-vs-fast tokenizer inconsistency is unrelated to a MacOS bus error loading a community CLIP model." + "reason": "Slow-vs-fast tokenizer inconsistency is unrelated to the CLIP MacOS bus error." }, { "left": "issue:43653", "right": "issue:44568", "accept": false, - "reason": "Both mention special tokens, but BigBird mask-token registration and mdeberta BOS/EOS add_special_tokens behavior are distinct tokenizer bugs." + "reason": "Both concern special tokens, but one is BigBird mask-token registration and the other is mdeberta BOS/EOS insertion; not the same underlying bug." }, { "left": "issue:43646", "right": "issue:43749", "accept": false, - "reason": "Trust-remote-code config field loss and FSDP RAM-efficient loading are separate subsystems with different causes." + "reason": "AutoConfig trust_remote_code missing fields is unrelated to FSDP_CPU_RAM_EFFICIENT_LOADING being broken." }, { "left": "issue:43452", "right": "issue:43577", "accept": false, - "reason": "GGUF AutoTokenizer/AutoModel loading breakage and Blip2 dtype propagation are unrelated load-time bugs." + "reason": "gguf AutoTokenizer/AutoModel loading failure and BLIP2 dtype propagation are distinct loading issues with different root causes." }, { - "left": "issue:43688", - "right": "issue:45704", + "left": "issue:43756", + "right": "issue:46402", "accept": false, - "reason": "Auxiliary-loss normalization and an apex FusedRMSNorm memory leak are different model internals and not the same defect." + "reason": "Missing RoPE layers in Smollm3 and Sam3 text-encoder weights not loading are different model-specific load bugs." }, { - "left": "issue:43756", - "right": "issue:46402", + "left": "issue:43688", + "right": "issue:45704", "accept": false, - "reason": "Smollm3 RoPE-layer dropout and Sam3Model text-encoder weights not loading are separate architecture-specific issues." + "reason": "Auxiliary-loss normalization and a T5 apex.FusedRMSNorm memory leak are unrelated bugs." }, { "left": "issue:44843", "right": "issue:44991", "accept": false, - "reason": "One is an offline-mode network call in a tokenizer regex patch; the other is a tokenizer-loading regression for a specific model, not the same code path." + "reason": "Both are tokenizer-loading regressions, but one is an offline-mode model_info() call in _patch_mistral_regex and the other is a specific tokenizer load failure for EMBEDDIA/est-roberta." }, { "left": "issue:43653", "right": "issue:43927", "accept": false, - "reason": "Special-token registration and DiaConfig custom token IDs persistence are different bugs in different layers." + "reason": "Both involve token handling, but 43653 is a missing special-token registration and 43927 is custom token IDs being lost across save/load." }, { "left": "issue:44464", "right": "issue:44743", "accept": false, - "reason": "Both involve generation/caching, but compiled-forward chunked generation mismatch and recurrent-state reset with seq_len>1 are distinct problems." + "reason": "Both are generation/cache issues, but 44464 is compiled-forward chunked generation inconsistency while 44743 is recurrent state resetting with cache in modular_qwen3_5." }, { "left": "issue:43688", "right": "issue:45701", "accept": false, - "reason": "Auxiliary-loss normalization and version-dependent tokenization changes are unrelated." + "reason": "Aux-loss normalization is unrelated to a general version-dependent tokenization change." }, { "left": "issue:44484", "right": "issue:44704", "accept": false, - "reason": "save_pretrained shard sizing and AutoProcessor kwargs not reaching cached_file are different API behaviors." + "reason": "save_pretrained shard-size behavior and AutoProcessor kwarg forwarding are different API issues." }, { "left": "issue:45491", "right": "issue:46002", "accept": false, - "reason": "Both mention NaNs, but Gemma3 sliding-window attention NaNs and classification-head NaNs after checkout/init are different symptoms and likely different roots." + "reason": "Both mention NaNs, but Gemma3 mixed-length sliding-window attention on GPU is a different failure mode from NaNs in classification heads after checkout/init." }, { "left": "issue:44479", "right": "issue:44625", "accept": false, - "reason": "A Qwen VL video-input regression and Qwen3.5 num_labels propagation are unrelated model-configuration issues." + "reason": "A video-input regression in Qwen VL models is unrelated to num_labels propagation in Qwen3.5 config." }, { "left": "issue:44821", "right": "issue:46821", "accept": false, - "reason": "AutoImageProcessor URL loading and TimesFM window_size AttributeError have no shared underlying bug." + "reason": "AutoImageProcessor URL loading and TimesFM window_size AttributeError are unrelated." }, { "left": "issue:43688", "right": "issue:45440", "accept": false, - "reason": "Aux-loss normalization and DeepSeekV3MoE divergence from remote implementation are different model-level changes." + "reason": "Aux-loss normalization and native DeepseekV3MoE divergence are different correctness issues in different code paths." }, { "left": "issue:33666", "right": "issue:36010", "accept": false, - "reason": "Qwen2-VL multi-GPU training and a GenerationMixin import error are unrelated." + "reason": "Qwen2-VL multi-GPU training and the GenerationMixin import error are unrelated." }, { "left": "issue:32090", "right": "issue:33666", "accept": false, - "reason": "Trainer gpu_broadcast NoneType and Qwen2-VL multi-GPU training are different distributed-training failures." + "reason": "Trainer GPU broadcast of NoneType and Qwen2-VL multi-GPU training are not the same bug." }, { "left": "issue:44734", "right": "issue:46612", "accept": false, - "reason": "Serve KV-cache continuation indexing and beam-search cache reorder being skipped are both generation-related, but they affect different mechanisms and models." + "reason": "Both involve generation state, but 44734 is a KV-cache continuation indexing crash in serve while 46612 is beam-search cache reorder being skipped for specific models." }, { "left": "issue:44466", "right": "issue:46050", "accept": false, - "reason": "Device-dependent lm_head serialization and sinq quantized save/load breakage are separate persistence bugs." + "reason": "Inconsistent lm_head.weight serialization and broken quantized save/load for sinq are both persistence bugs, but they are different mechanisms." }, { "left": "issue:43504", "right": "issue:43720", "accept": false, - "reason": "Beit legacy-field loading and BitNet packed-weight unpacking are distinct model-loading issues." - }, - { - "left": "issue:44977", - "right": "issue:46693", - "accept": false, - "reason": "A Qwen3.5 flash-attention correctness bug and a long-context FA2 performance regression are not the same issue." + "reason": "Beit legacy-field loading and BitNet packed-weight unpacking during accelerate loading are distinct load-time failures." }, { "left": "issue:43504", "right": "issue:43550", "accept": false, - "reason": "Beit loading via legacy field and Bamba torch.compile SDPA failure are unrelated." + "reason": "A Beit pretrained-preset load issue is unrelated to Bamba torch.compile+SDPA runtime failure." }, { - "left": "issue:44464", - "right": "issue:44977", + "left": "issue:44977", + "right": "issue:46693", "accept": false, - "reason": "Compiled-forward chunked generation inconsistency and Qwen3.5 flash-attention generation failure are different code paths and symptoms." + "reason": "Both mention Qwen3.5 and FA2, but 44977 is a functional generation regression while 46693 is a performance regression after 5.2.0." }, { "left": "issue:41720", "right": "issue:44464", "accept": false, - "reason": "Qwen3 auto device-mapping cudaErrorAssert and compiled-forward generation inconsistency do not share a concrete root cause." + "reason": "cudaErrorAssert on auto device mapping for Qwen3 is unrelated to compiled-forward chunked generation inconsistency." }, { "left": "issue:43643", "right": "issue:44561", "accept": false, - "reason": "Both touch trust_remote_code, but missing fields from AutoConfig and removal of is_torch_fx_available break different parts of the stack." + "reason": "Both touch trust_remote_code, but 43643 is missing fields from AutoConfig and 44561 is a breakage from removing is_torch_fx_available." }, { "left": "issue:43577", "right": "issue:44554", "accept": false, - "reason": "Blip2 dtype propagation and an MPS attention correctness issue are unrelated." + "reason": "BLIP2 dtype propagation and an MPS attention correctness issue with mismatched head dims are unrelated bugs." } ] }, @@ -8142,14 +8062,14 @@ "issue:43611|issue:46846", "issue:43720|issue:43927", "issue:43504|issue:44610", - "issue:33666|issue:36331", "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:44514|issue:45290", "issue:43493|issue:43825", "issue:44554|issue:46365", "issue:43525|issue:43572", @@ -8170,20 +8090,20 @@ "issue:44568|issue:46489", "issue:43278|issue:43701" ], - "prepared_review_unit_hash": "7094f247fb5c7c68396baea3f628fd15cdfa0245da7b016733306f292d538c7f", + "prepared_review_unit_hash": "f4f783095bedc81f5aa4b360f33463f145381a9e0599c8176c83f5b9d1524f4d", "component_budget": { "node_count": 500, "item_count": 500, "soft_pair_count": 1839, - "serialized_chars": 953913, - "estimated_input_tokens": 238479, - "estimated_eval_tokens": 477214 + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 30, - "serialized_chars": 36793, + "serialized_chars": 36794, "estimated_input_tokens": 9199, "estimated_eval_tokens": 18654 }, @@ -8192,192 +8112,192 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is loosely related by Transformers regression/compatibility reports, but the soft pairs are not true duplicates; they cover different models, code paths, and failure modes. Most are distinct tokenizer, processor, loading, or training issues rather than one underlying bug.", - "confidence": 0.66, - "canonical_issue_reason": "Issue #43611 is the broadest representative of the cluster\u2019s recurring v5 compatibility/loading regressions, so it works best as the anchor issue.", + "summary": "This cluster is highly heterogeneous: it mixes unrelated tokenizer, processor, model-loading, training, and performance bugs across many model families. None of the soft-edge pairs look like the same underlying fix target, so I would not merge any of them as duplicates.", + "confidence": 0.96, + "canonical_issue_reason": "No single canonical duplicate stands out; the cluster is too broad and the issues do not share one concrete bug. If forced to pick a representative, issue 43611 is the broadest v5-loading regression, but it is still not a true cluster anchor.", "canonical_pr_reason": null, - "best_issue_reason": "Issue #43611 is the best single issue to represent the cluster because it captures the most general pattern here: Transformers 5.x breaking model loading/initialization paths.", + "best_issue_reason": "Issue 43611 is the most general, user-facing regression among the set and best represents the broad theme of v5 breakage, though it should not be treated as a duplicate hub for the whole cluster.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43611", "right": "issue:46846", "accept": false, - "reason": "Both are v5-era regressions, but one is a base_model_prefix loading issue and the other is a Qwen3.5 tensor-parallel plan bug; different code paths and fixes." + "reason": "Different failures: base_model_prefix loading regression vs missing linear_attn TP plan entries." }, { "left": "issue:43720", "right": "issue:43927", "accept": false, - "reason": "Packed-weight unloading during accelerate loading and DiaConfig token-ID persistence are unrelated failures." + "reason": "BitNet packed-weight loading bug vs DiaConfig token-id persistence bug; unrelated code paths." }, { "left": "issue:43504", "right": "issue:44610", "accept": false, - "reason": "BEiT legacy-field loading and OmDet-Turbo processor size mismatch affect different models and different stages of the pipeline." + "reason": "BEiT legacy-field load failure vs OmDet-Turbo processor image-size mismatch." }, { - "left": "issue:33666", - "right": "issue:36331", + "left": "issue:43606", + "right": "issue:43927", "accept": false, - "reason": "A multi-GPU training question and a CustomTrainer compute_loss signature error are not the same bug." + "reason": "CPU offload device mismatch in Bark vs config token-id loss in Dia; different bugs." }, { - "left": "issue:43606", - "right": "issue:43927", + "left": "issue:33666", + "right": "issue:36331", "accept": false, - "reason": "CPU offload device mismatch in bark-small and DiaConfig save/load token-ID loss are unrelated." + "reason": "Multi-GPU Qwen2-VL training issue vs CustomTrainer signature mismatch; not the same problem." }, { "left": "issue:43278", "right": "issue:43688", "accept": false, - "reason": "Embedding dtype drift in eval and auxiliary-loss normalization are different problems with different symptoms and fixes." + "reason": "Embedding dtype regression vs auxiliary-loss normalization bug; separate model behaviors." }, { "left": "issue:43782", "right": "issue:46693", "accept": false, - "reason": "Qwen3VL weight_only loading failure and FA2 long-context performance regression are distinct issues." + "reason": "Qwen3VL weight_only loading error vs FA2 long-context performance regression." }, { "left": "issue:43262", "right": "issue:43295", "accept": false, - "reason": "Audio apply_chat_template sampling-rate defaulting and processor.tokenizer/image passthrough regression are different processor bugs." + "reason": "Different processor regressions: sampling-rate default vs processor.tokenizer/images compatibility." }, { "left": "issue:43720", "right": "issue:44361", "accept": false, - "reason": "Accelerate packed-weights loading and MLukeTokenizer task AttributeError do not share the same root cause." + "reason": "Packed-weight loading bug vs MLukeTokenizer task AttributeError; unrelated." }, { - "left": "issue:44514", - "right": "issue:45290", + "left": "issue:44964", + "right": "issue:45405", "accept": false, - "reason": "Both involve apply_chat_template, but one is batched padding=False and the other is assistant tool-call/no-content handling; not the same bug." + "reason": "Phi-4 multimodal loading failure vs unreleased PEFT version bump; not duplicates." }, { - "left": "issue:44964", - "right": "issue:45405", + "left": "issue:44514", + "right": "issue:45290", "accept": false, - "reason": "Phi-4 multimodal loading failure and a release/version bump for MIN_PEFT_VERSION are unrelated." + "reason": "Both involve apply_chat_template, but the crashing inputs and failure modes differ." }, { "left": "issue:43493", "right": "issue:43825", "accept": false, - "reason": "SigLIP2 implementation mismatch and a pipeline translation support message are not the same code-path problem." + "reason": "SigLIP2 implementation discrepancy vs pipeline error-message regression; different layers." }, { "left": "issue:44554", "right": "issue:46365", "accept": false, - "reason": "An MPS attention correctness issue and a Trainer import conflict with datasets are unrelated." + "reason": "MPS attention correctness bug vs Trainer import conflict with datasets; unrelated." }, { "left": "issue:43525", "right": "issue:43572", "accept": false, - "reason": "Both are missing config fields, but they concern different models, different fields, and likely different fixes." + "reason": "Both are missing pad-token fields, but they are model-specific config issues with different attributes and fixes." }, { "left": "issue:44977", "right": "issue:46209", "accept": false, - "reason": "Qwen3.5 flash-attention generation behavior and FP8 inference on B200 are separate regressions." + "reason": "Qwen3.5 flash-attention generation bug vs FP8 inference failure on B200; different hardware/code paths." }, { "left": "issue:44625", "right": "issue:45584", "accept": false, - "reason": "Qwen3.5 num_labels propagation and Whisper empty-transcription handling are unrelated bugs." + "reason": "num_labels propagation bug vs Whisper empty-transcription generation bug." }, { "left": "issue:43452", "right": "issue:44451", "accept": false, - "reason": "gguf_file loading breakage and inability to load ScandiBERT are distinct model-loading failures." + "reason": "gguf_file loading regression vs ScandiBERT load failure; not the same root cause." }, { "left": "issue:28218", "right": "issue:33357", "accept": false, - "reason": "Tokenizer extra-space behavior and a MacOS bus error in CLIP are completely different issues." + "reason": "Tokenizer spacing bug vs MacOS bus error in CLIP model; unrelated." }, { "left": "issue:45823", "right": "issue:46050", "accept": false, - "reason": "Gemma4 device-map mismatch and sinq save/load breakage involve different model families and failure modes." + "reason": "Gemma4 device-map mismatch vs sinq save/load quantization breakage." }, { "left": "issue:42796", "right": "issue:44977", "accept": false, - "reason": "A tokenizer BPE-marker bug and a flash-attention generation regression are not the same underlying issue." + "reason": "Tokenizer BPE-marker handling vs Qwen3.5 flash-attention generation; different failures." }, { "left": "issue:43994", "right": "issue:46192", "accept": false, - "reason": "SigLIP2 nonsensical outputs and return_tensors being ignored when text_kwargs is passed are unrelated." + "reason": "SigLIP2 output quality issue vs return_tensors being ignored; unrelated API behavior." }, { "left": "issue:43493", "right": "issue:43844", "accept": false, - "reason": "SigLIP2 HF/JAX discrepancy and a ZeRO-3 gradient explosion are different problems." + "reason": "SigLIP2 implementation discrepancy vs ZeRO-3 gradient increase; not the same bug." }, { "left": "issue:39692", "right": "issue:43295", "accept": false, - "reason": "SigLIP2 documentation errors and the processor.tokenizer/images regression do not match." + "reason": "Documentation example errors vs processor.tokenizer regression; different artifact types and causes." }, { "left": "issue:43278", "right": "issue:43844", "accept": false, - "reason": "Eval dtype drift for embeddings and gradient growth under ZeRO-3 are unrelated training issues." + "reason": "dtype regression in evaluation vs gradient growth under ZeRO-3; separate training issues." }, { "left": "issue:43646", "right": "issue:44466", "accept": false, - "reason": "Custom model initialization breakage and lm_head.weight serialization inconsistency are both v5-related, but not the same concrete bug." + "reason": "Custom model initialization breakage vs device-dependent lm_head serialization inconsistency." }, { "left": "issue:46209", "right": "issue:46710", "accept": false, - "reason": "FP8 inference on B200 and DeepSeek-R1 accuracy regression after upgrade are distinct regressions." + "reason": "FP8 inference failure on B200 vs accuracy regression after upgrade; different symptoms and fixes." }, { "left": "issue:33357", "right": "issue:33666", "accept": false, - "reason": "A MacOS bus error for CLIP and Qwen2-VL multi-GPU training are unrelated." + "reason": "MacOS CLIP bus error vs Qwen2-VL multi-GPU training; no shared code path." }, { "left": "issue:44849", "right": "issue:46209", "accept": false, - "reason": "Qwen3.5 hidden_states behavior and FP8 inference failure are different issues." + "reason": "Qwen3.5 hidden_states bug vs FP8 hardware-specific inference failure." }, { "left": "issue:44568", "right": "issue:46489", "accept": false, - "reason": "mdeberta add_special_tokens regression and DeepSeek-Coder tokenizer output mismatch are separate tokenizer bugs." + "reason": "Special-token insertion bug in mdeberta-v3-base vs DeepSeek-Coder v1 tokenizer regression; same subsystem but not the same concrete bug." }, { "left": "issue:43278", "right": "issue:43701", "accept": false, - "reason": "Embedding dtype changes in eval and resume_from_checkpoint key mismatches are unrelated." + "reason": "Embedding dtype regression vs checkpoint resume key mismatch; unrelated." } ] }, @@ -8451,21 +8371,21 @@ "issue:46858" ], "soft_pairs": [ - "issue:44488|issue:44843", "issue:44792|issue:45405", + "issue:44488|issue:44843", "issue:43927|issue:44361", "issue:44568|issue:45812", "issue:43824|issue:44589", "issue:43575|issue:45405", - "issue:44849|issue:45357", "issue:43504|issue:43606", + "issue:44849|issue:45357", "issue:43257|issue:43299", "issue:45245|issue:46710", "issue:46399|issue:46846", "issue:46821|issue:46832", + "issue:43606|issue:44112", "issue:44493|issue:46285", "issue:44589|issue:45520", - "issue:43606|issue:44112", "issue:43334|issue:43572", "issue:43334|issue:45759", "issue:45072|issue:46832", @@ -8480,202 +8400,202 @@ "issue:44869|issue:46752", "issue:29942|issue:30333" ], - "prepared_review_unit_hash": "b7780bc387e3d79991174c83d19d044a7eeda3f36c9d5ef63ded0ec5267e3b3f", + "prepared_review_unit_hash": "44437a3e15d8804bd818974deeabe902a6cda31379185866587aca655a77df7c", "component_budget": { "node_count": 500, "item_count": 500, "soft_pair_count": 1839, - "serialized_chars": 953913, - "estimated_input_tokens": 238479, - "estimated_eval_tokens": 477214 + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 28, - "serialized_chars": 36302, - "estimated_input_tokens": 9076, - "estimated_eval_tokens": 18408 + "serialized_chars": 36307, + "estimated_input_tokens": 9077, + "estimated_eval_tokens": 18410 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "These issues are mostly unrelated regression reports spanning tokenizers, multimodal configs, quantization/offload, cache behavior, and CI failures. The soft links are largely based on superficial theme overlap rather than the same underlying bug, so none should be merged. No PRs are present in this cluster.", - "confidence": 0.86, - "canonical_issue_reason": "issue:45812 is the broadest, highest-signal user-facing regression here: it affects all Granite models and points to a concrete v4\u2192v5 tokenizer breakage.", + "summary": "This cluster is not a duplicate set: the items span unrelated model-loading, tokenizer, cache/offload, CI, and training issues. The suggested soft matches are mostly superficial keyword overlap (e.g., \"missing\", \"load\", \"OOM\", \"pad_token\") rather than the same concrete bug.", + "confidence": 0.98, + "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "issue:45812 is the best representative issue because it is clearly scoped, has wide impact, and captures the most central repeated theme in the set: a version-regression in model/tokenizer behavior.", + "best_issue_reason": null, "best_pr_reason": null, "soft_edge_verdicts": [ { - "left": "issue:44488", - "right": "issue:44843", + "left": "issue:44792", + "right": "issue:45405", "accept": false, - "reason": "Both are tokenizer/model-loading failures, but one is a specific model not loading and the other is an HF_HUB_OFFLINE regression in _patch_mistral_regex; different code paths and fixes." + "reason": "Different bugs: a janus generation test failure vs an unrelated dependency version pin issue." }, { - "left": "issue:44792", - "right": "issue:45405", + "left": "issue:44488", + "right": "issue:44843", "accept": false, - "reason": "A Janus test failure and a PEFT version bump are unrelated." + "reason": "Both mention loading/tokenizer behavior, but one is a model loading problem and the other is an offline hub call bug in regex patching." }, { "left": "issue:43927", "right": "issue:44361", "accept": false, - "reason": "Different models, different failures, and different likely fixes; not the same bug." + "reason": "Different tokenizer/model failures with different symptoms and code paths; not the same underlying bug." }, { "left": "issue:44568", "right": "issue:45812", "accept": false, - "reason": "Both are tokenizer regressions, but one is missing BOS/EOS addition for a specific tokenizer while the other is wrong token IDs across Granite models; too different to merge." + "reason": "Both involve tokenizer behavior, but one is missing BOS/EOS insertion for a specific model while the other is a broad Granite token-ID regression." }, { "left": "issue:43824", "right": "issue:44589", "accept": false, - "reason": "An import error for a Qwen model and a Float8 storage lookup error are unrelated." + "reason": "One is an import error for a Qwen VL class; the other is a storage backend/type mismatch. No shared failure path." }, { "left": "issue:43575", "right": "issue:45405", "accept": false, - "reason": "A tensor-parallel OOM and an unreleased PEFT version bump are unrelated." + "reason": "Qwen2-57B tensor-parallel OOM is unrelated to a PEFT minimum-version bump." }, { - "left": "issue:44849", - "right": "issue:45357", + "left": "issue:43504", + "right": "issue:43606", "accept": false, - "reason": "Both mention Qwen3.5, but one is output_hidden_states behavior and the other is save_pretrained key mapping; different bug classes." + "reason": "Legacy field loading bug in BEiT is unrelated to Bark CPU offload device-mismatch failures." }, { - "left": "issue:43504", - "right": "issue:43606", + "left": "issue:44849", + "right": "issue:45357", "accept": false, - "reason": "Beit legacy-field loading and Bark CPU-offload device mismatch are unrelated." + "reason": "Different Qwen variants and different failures: output_hidden_states behavior vs incorrect visual encoder keys on save_pretrained." }, { "left": "issue:43257", "right": "issue:43299", "accept": false, - "reason": "Both concern Qwen MoE loading, but one is accelerate/deepspeed weight conversion and the other is a dev-version model-loading breakage; not the same issue." + "reason": "Both are Qwen3 MOE-related, but one is an accelerate+deepspeed conversion path and the other is a transformers dev loading regression for Qwen3VL." }, { "left": "issue:45245", "right": "issue:46710", "accept": false, - "reason": "Category-count runtime error and an accuracy regression are unrelated." + "reason": "Category-count RuntimeError and DeepSeek accuracy regression are distinct problems with no common code path." }, { "left": "issue:46399", "right": "issue:46846", "accept": false, - "reason": "Qwen3.5 kernel registration crash and missing TP-plan entries causing OOM are different code paths." + "reason": "Both are Qwen3.5-related, but one is a missing decorator for a kernel registration and the other is a missing TP plan entry causing OOM/shape errors." }, { "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": "TimesFM window_size attribute error and RT-DETR feature-level mismatch are unrelated model-specific bugs." + }, + { + "left": "issue:43606", + "right": "issue:44112", + "accept": false, + "reason": "Both are CI failures, but they stem from different tests and different device-handling issues." }, { "left": "issue:44493", "right": "issue:46285", "accept": false, - "reason": "Both are structure/field regressions, but one is unexpected position-id keys and the other is CLIP class structure divergence; not the same fix." + "reason": "Position-id key warnings across many models vs CLIP structure inconsistency; different regressions and fixes." }, { "left": "issue:44589", "right": "issue:45520", "accept": false, - "reason": "Float8 storage lookup and flash_attn import handling on Python 3.13 are unrelated." - }, - { - "left": "issue:43606", - "right": "issue:44112", - "accept": false, - "reason": "CPU-offload device mismatch in Bark and a GraniteSpeech CI stale-override failure are not the same underlying bug." + "reason": "Float8 storage lookup failure is unrelated to a flash_attn import-utils KeyError on Python 3.13." }, { "left": "issue:43334", "right": "issue:43572", "accept": false, - "reason": "Both are config-attribute regressions after v5, but they affect different models and different missing fields." + "reason": "Qwen3-VL missing pad_token_id and StableLmConfig pad_token_idx are separate model/config bugs." }, { "left": "issue:43334", "right": "issue:45759", "accept": false, - "reason": "Both involve Qwen multimodal configs, but one is a missing pad_token_id attribute and the other is failing to unwrap text_config; not the same concrete bug." + "reason": "Qwen3-VL pad_token_id AttributeError is unrelated to AutoModelForCausalLM.from_config not unwrapping text_config." }, { "left": "issue:45072", "right": "issue:46832", "accept": false, - "reason": "Dtype mismatches in inference and RT-DETR forward shape handling are unrelated." + "reason": "Dtype mismatch CI failures vs RT-DETR feature-level mismatch; different subsystems and symptoms." }, { "left": "issue:43504", "right": "issue:43653", "accept": false, - "reason": "A BEiT legacy-field loading issue and a BigBird tokenizer special-token registration issue are unrelated." + "reason": "BEiT legacy-field loading bug is unrelated to BigBirdTokenizer special-token registration/empty decode behavior." }, { "left": "issue:43572", "right": "issue:44291", "accept": false, - "reason": "A missing config field and an unexpected init_empty_weights argument are different compatibility regressions." + "reason": "Missing pad_token_idx in StableLmConfig and unexpected _is_hf_initialized arg in init_empty_weights are different version-regression issues." }, { "left": "issue:44464", "right": "issue:46424", "accept": false, - "reason": "Both mention cache/generation, but one is compiled-forward inconsistency and the other is max_cache_length semantics for quantized cache; not mergeable." + "reason": "Chunked generation inconsistency with compiled forward is unrelated to cache length handling for quantized cache only." }, { "left": "issue:43937", "right": "issue:46701", "accept": false, - "reason": "GenerationConfig validation and a DiffusionGemma pipeline failure are unrelated." + "reason": "GenerationConfig validation failure for GLM-5 is unrelated to DiffusionGemma image-text pipeline failure." }, { "left": "issue:46832", "right": "issue:46858", "accept": false, - "reason": "RT-DETR backbone/output mismatch and static-cache slice arithmetic on non-CUDA backends are different problems." + "reason": "RT-DETR backbone/output mismatch and static-cache seq-length tensor bug are separate runtime errors." }, { "left": "issue:43844", "right": "issue:44964", "accept": false, - "reason": "ZeRO-3 gradient growth and Phi-4 multimodal loading failure are unrelated." + "reason": "Gradient explosion with ZeRO-3 and a Phi-4 multimodal loading failure are unrelated problems." }, { "left": "issue:45200", "right": "issue:45229", "accept": false, - "reason": "Gemma 4 token-type defaults and Gemma4 multi-GPU OOM are different issues." + "reason": "Gemma 4 fine-tuning token defaults vs multi-GPU inference OOM are not the same underlying bug." }, { "left": "issue:43873", "right": "issue:46050", "accept": false, - "reason": "Quantization/offload behavior and quantized save/load breakage are related only at a very broad theme level; they are different code paths." + "reason": "Quantization offload behavior and sinq save/load breakage are different quantization issues with different code paths." }, { "left": "issue:44869", "right": "issue:46752", "accept": false, - "reason": "Whisper word-timestamp decoding and empty-list chat templating are unrelated." + "reason": "Whisper timestamp decoding crash on trailing replacement char is unrelated to apply_chat_template on an empty list." }, { "left": "issue:29942", "right": "issue:30333", "accept": false, - "reason": "Flash Attention 2 test failures and MLflow reporting on failed jobs are unrelated." + "reason": "Flash Attention 2 test failures and MLflow job-reporting issues are entirely unrelated." } ] }, @@ -8777,202 +8697,202 @@ "issue:45200|issue:45823", "issue:29870|issue:29942" ], - "prepared_review_unit_hash": "a6b598ee7ca92f97eb8a17f68fbea644c0d0843a99d3be18667f84b0ec628cde", + "prepared_review_unit_hash": "df15d95f8ee061fd431cf0696a05a2350e9ab3a00fbccdb43c89b23e038dd6dc", "component_budget": { "node_count": 500, "item_count": 500, "soft_pair_count": 1839, - "serialized_chars": 953913, - "estimated_input_tokens": 238479, - "estimated_eval_tokens": 477214 + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 28, - "serialized_chars": 35561, - "estimated_input_tokens": 8891, - "estimated_eval_tokens": 18038 + "serialized_chars": 35560, + "estimated_input_tokens": 8890, + "estimated_eval_tokens": 18036 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster is highly heterogeneous: the suggested soft edges are mostly superficial title-level similarities, not the same underlying bug or code path. None of the candidate pairs look safe to merge as duplicates, and there is no clear PR-equivalent here.", - "confidence": 0.95, - "canonical_issue_reason": "Issue 43873 is the most central/open item in this set by activity and inbound references, but the cluster overall does not form a true duplicate group.", + "summary": "This cluster is a loose mix of unrelated Transformers issues across tokenizers, generation, serving, multimodal processors, quantization, and model loading. None of the soft-similarity pairs look like the same underlying bug/change, so I would not merge any of them as duplicates.", + "confidence": 0.86, + "canonical_issue_reason": "No true duplicate hub stands out from the titles alone; if one issue must anchor the cluster, issue:43873 is the broadest active report and has the strongest inbound reference signal, so it is the closest thing to a cluster representative.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 43873 is the best representative only as the most active referenced issue; it is not a strong semantic canonical for the whole cluster.", + "best_issue_reason": "issue:43873 is the best single issue representative because it is broad, currently open, and appears to be the most central runtime/memory-management report in the set.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:45229", "right": "issue:45823", "accept": false, - "reason": "Both are Gemma4-related, but one is CUDA OOM in multi-GPU inference and the other is a device-mismatch forward bug; different failure modes." + "reason": "Both are Gemma4-related, but one is a multi-GPU CUDA OOM and the other is a device-map mismatch in forward; different failure modes and code paths." }, { "left": "issue:43994", "right": "issue:44493", "accept": false, - "reason": "SigLIP2 output correctness and a generic position-id key warning are unrelated symptoms with different code paths." + "reason": "SigLIP2 incorrect outputs vs a position-id key warning across many models; too different to be the same bug." }, { "left": "issue:45588", "right": "issue:46050", "accept": false, - "reason": "One is a flash-attention import/runtime crash with None aux state; the other is save/load breakage for sinq quantization." + "reason": "Flash-attention sinkless-model crash vs sinq save/load corruption; unrelated subsystems and symptoms." }, { "left": "issue:42490", "right": "issue:43696", "accept": false, - "reason": "Local-path import warning and GPT-oss CUDA OOM are unrelated issues." + "reason": "Local-path import warning and GPT-oss OOM are unrelated." }, { "left": "issue:29159", "right": "issue:32090", "accept": false, - "reason": "Tokenizer slow/fast inconsistency is unrelated to Trainer distributed broadcast NoneType handling." + "reason": "Tokenizer slow/fast inconsistency is not the same issue as a Trainer GPU broadcast NoneType error." }, { "left": "issue:44977", "right": "issue:46701", "accept": false, - "reason": "Flash-attention generation issues for Qwen3.5 and a DiffusionGemma pipeline failure are different model families and different bugs." + "reason": "Qwen3.5 flash-attention generation failure vs a DiffusionGemma pipeline output issue; different models and execution paths." }, { "left": "issue:45084", "right": "issue:45520", "accept": false, - "reason": "Template compilation errors and a Python 3.13 flash_attn import KeyError are unrelated." + "reason": "Non-template-node compile error and flash_attn import KeyError are separate import/compilation problems." }, { "left": "issue:28218", "right": "issue:34567", "accept": false, - "reason": "Added-token spacing in tokenizers is unrelated to TrainerState token-count tracking not updating." + "reason": "Added-token spacing behavior and TrainerState token-count tracking are unrelated." }, { "left": "issue:43065", "right": "issue:46369", "accept": false, - "reason": "Dummy Conv2d in Sam3PixelDecoder and SAM3 lite-text memory overflow are different implementation/performance issues." + "reason": "A dummy Conv2d implementation and a memory overflow report on sam3-litetext are not the same defect." }, { "left": "issue:44554", "right": "issue:45446", "accept": false, - "reason": "MPS attention correctness when head dims differ and a PyTorch version check bug in flex_attention are separate problems." + "reason": "MPS attention correctness with mismatched head dims is unrelated to a PyTorch version check bug." }, { "left": "issue:44423", "right": "issue:44734", "accept": false, - "reason": "Continuous-batching multimodal crash and KV-cache continuation indexing failure are different server bugs." + "reason": "Both are serve crashes, but they hit different continuations and different tensor/object errors; not the same underlying bug." }, { "left": "issue:15354", "right": "issue:38175", "accept": false, - "reason": "TorchScript export failure from GeneratorExp and SigLIP2 zero probabilities are unrelated model/export bugs." + "reason": "TorchScript GeneratorExp export failure and SigLIP2 zero probabilities are unrelated." }, { "left": "issue:43493", "right": "issue:43701", "accept": false, - "reason": "SigLIP2 discrepancy and resume_from_checkpoint key mismatch do not share a concrete code path." + "reason": "SigLIP2 HF-vs-JAX discrepancy and resume_from_checkpoint key mismatch do not share a concrete code path." }, { "left": "issue:41720", "right": "issue:46710", "accept": false, - "reason": "Qwen3 auto device mapping CUDA assert and DeepSeek accuracy regression after upgrade are distinct regressions." + "reason": "A800 cudaErrorAssert under auto device mapping is different from an accuracy regression after upgrading Transformers." }, { "left": "issue:45406", "right": "issue:46121", "accept": false, - "reason": "Gemma4Processor missing _tokenizer and rope-validation type handling are unrelated bugs." + "reason": "Gemma4Processor _tokenizer crash in serve and a rope-parameter JSON list type error are unrelated." }, { "left": "issue:44977", "right": "issue:46710", "accept": false, - "reason": "Flash-attention generation failure and accuracy regression after upgrade are not the same bug." + "reason": "Flash-attention generation breakage in Qwen3.5 is not the same as a model accuracy regression after version upgrade." }, { "left": "issue:45072", "right": "issue:45198", "accept": false, - "reason": "bfloat16 dtype mismatch CI failures and Wav2Vec2 save/tokenization failures are different model-specific issues." + "reason": "BF16 dtype mismatches in SwitchTransformers/TimmWrapperModel and Wav2Vec2 save/tokenization failures are unrelated." }, { "left": "issue:43295", "right": "issue:43450", "accept": false, - "reason": "Processor.tokenizer regression and video processor batched-shape bug affect different components and behaviors." + "reason": "Processor.tokenizer regression with image inputs and batched video shape errors are different multimodal bugs." }, { "left": "issue:37428", "right": "issue:43957", "accept": false, - "reason": "Flash-support import error and meta-device model-loading failures are separate loading issues." + "reason": "Missing flash-support import and meta-device model loading failures are not the same issue." }, { "left": "issue:45588", "right": "issue:45759", "accept": false, - "reason": "Flash-attention aux-state crash and AutoModelForCausalLM text_config unwrapping are unrelated." + "reason": "Flash-attention sinkless-model crash and from_config text_config unwrapping are separate processor/config problems." }, { "left": "issue:43540", "right": "issue:45200", "accept": false, - "reason": "Video-input processing in Qwen3OmniMoe and Gemma4 mm_token_type_ids defaulting are different multimodal bugs." + "reason": "Video-input ValueError in Qwen3OmniMoe and missing mm_token_type_ids in Gemma4 are different multimodal preprocessing bugs." }, { "left": "issue:43295", "right": "issue:43873", "accept": false, - "reason": "Gemma4Processor tokenizer access regression and quantization offloading behavior are unrelated." + "reason": "Processor/tokenizer regression and quantization offloading behavior are far apart in both symptom and code path." }, { "left": "issue:44186", "right": "issue:46192", "accept": false, - "reason": "LayoutLMv2Tokenizer NER/padding crash and return_tensors being ignored are different tokenizer API bugs." + "reason": "LayoutLMv2Tokenizer NER/padding crash and return_tensors being ignored are distinct tokenizer API issues." }, { "left": "issue:43572", "right": "issue:46821", "accept": false, - "reason": "StableLmConfig pad_token_idx missing after 5.0 and TimesFM window_size AttributeError are unrelated config issues." + "reason": "StableLmConfig pad_token_idx regression and TimesFM window_size AttributeError are unrelated config bugs." }, { "left": "issue:43295", "right": "issue:44610", "accept": false, - "reason": "Processor._tokenizer regression and OmDet-Turbo image size mismatch are different failures." + "reason": "Tokenizer/processor regression and OmDet-Turbo processor size mismatch are different model-specific preprocessing problems." }, { "left": "issue:45823", "right": "issue:46192", "accept": false, - "reason": "Gemma4 device-mismatch forward bug and return_tensors handling are not the same issue." + "reason": "Gemma4 device-mapping mismatch and return_tensors being ignored are unrelated." }, { "left": "issue:45200", "right": "issue:45823", "accept": false, - "reason": "mm_token_type_ids defaulting and device_map auto forward mismatch are different Gemma4 problems." + "reason": "One is a missing default mm_token_type_ids, the other is a device placement mismatch in forward; not the same bug." }, { "left": "issue:29870", "right": "issue:29942", "accept": false, - "reason": "BART decoder_attention_mask initialization bug and Flash Attention 2 test failures are unrelated." + "reason": "Bart generate decoder_attention_mask bug and Flash Attention 2 test failures are unrelated." } ] }, @@ -9067,196 +8987,196 @@ "issue:43950|issue:46285", "issue:43824|issue:43975", "issue:43452|issue:46512", - "issue:43262|issue:43450", "issue:43334|issue:43525", + "issue:43262|issue:43450", "issue:44442|issue:45198", "issue:45200|issue:46899", "issue:44792|issue:45563" ], - "prepared_review_unit_hash": "8c01b61cf698e9e60390f603a8f4e887d789b7791af6be7f782893e808b8bde0", + "prepared_review_unit_hash": "12676e1e6536853b46f180aea05bc526a8306f7591ff034d825b4ec2982b20db", "component_budget": { "node_count": 500, "item_count": 500, "soft_pair_count": 1839, - "serialized_chars": 953913, - "estimated_input_tokens": 238479, - "estimated_eval_tokens": 477214 + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 26, - "serialized_chars": 35882, - "estimated_input_tokens": 8971, - "estimated_eval_tokens": 18198 + "serialized_chars": 35887, + "estimated_input_tokens": 8972, + "estimated_eval_tokens": 18200 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is a broad mix of unrelated Transformers issues; the soft-similarity pairs mostly share superficial keywords or subsystems but not the same underlying bug. No PRs are present.", - "confidence": 0.96, - "canonical_issue_reason": "issue:44521 is the most representative triage target here: it is open, has the strongest discussion/inbound-reference signal, and describes a concrete user-facing regression with clear multimodal impact.", + "summary": "The cluster is a loose semantic mix of unrelated Transformers regressions and docs/test issues; none of the proposed soft pairs look like true duplicates, so I would not merge any of them. I\u2019d treat issue 44521 as the weakest umbrella representative only because it has the most engagement, but the group is too heterogeneous for a strong canonical duplicate target.", + "confidence": 0.9, + "canonical_issue_reason": "Issue 44521 has the strongest engagement signals (discussion and inbound references) and sits on a high-traffic API path, but the cluster is too heterogeneous to support a truly canonical duplicate issue.", "canonical_pr_reason": null, - "best_issue_reason": "issue:44521 is also the best overall issue candidate because it is active, well-evidenced, and likely to attract follow-up, whereas most other items are narrower or less central.", + "best_issue_reason": "Issue 44521 is the best single representative by impact/engagement, though only weakly so because the cluster spans many unrelated bugs.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43550", "right": "issue:43844", "accept": false, - "reason": "Different bugs: one is a torch.compile/SDPA failure in Bamba, the other is a gradient blow-up during DeepSpeed ZeRO-3 training." + "reason": "Different failures: Bamba torch.compile/SDPA correctness vs DeepSpeed ZeRO-3 gradient growth." }, { "left": "issue:45520", "right": "issue:46178", "accept": false, - "reason": "Both involve import/backend plumbing, but the failures are unrelated: flash_attn import resolution vs tensorflow_text not being in BACKENDS_MAPPING." + "reason": "Unrelated import/backend problems: flash_attn KeyError vs tensorflow_text missing from BACKENDS_MAPPING." }, { "left": "issue:43450", "right": "issue:43575", "accept": false, - "reason": "Different problem types and code paths: batched video processor output shape vs OOM when loading a large model with tensor parallelism." + "reason": "Video processor batched-shape bug vs Qwen2 TP loading OOM; different code paths and symptoms." }, { "left": "issue:43572", "right": "issue:44987", "accept": false, - "reason": "Both are loading/config regressions, but they affect different model families and different failure modes; not the same underlying bug." + "reason": "Both are loading/config regressions, but one is StableLm pad_token_idx and the other is a package/version load failure for a specific model." }, { "left": "issue:44554", "right": "issue:45356", "accept": false, - "reason": "One is an attention correctness issue on MPS, the other is a tokenizer regression/codec warning for Kimi-K2.5." + "reason": "MPS attention correctness issue vs tokenizer codec/regression in Kimi-K2.5; not the same bug." }, { "left": "issue:42490", "right": "issue:43493", "accept": false, - "reason": "Trailing-slash local-path import warnings are unrelated to the SigLIP2 implementation discrepancy." + "reason": "Trailing-slash local path warning is unrelated to the SigLIP2 HF-vs-JAX implementation discrepancy." }, { "left": "issue:37428", "right": "issue:43824", "accept": false, - "reason": "Both are ImportErrors, but they concern different missing symbols/classes in different modules and are not the same defect." + "reason": "Both are import errors, but they involve different missing symbols and different modules/code paths." }, { "left": "issue:43232", "right": "issue:44849", "accept": false, - "reason": "Generation-state update after sync_gpus is unrelated to the Qwen3.5 output_hidden_states bug." + "reason": "Generation kwargs update after sync_gpus is unrelated to the Qwen3.5 output_hidden_states bug." }, { "left": "issue:43295", "right": "issue:43493", "accept": false, - "reason": "Both touch multimodal model behavior, but one is processor.tokenizer/image-handling regression while the other is a SigLIP2 implementation mismatch." + "reason": "Processor.tokenizer/image passthrough regression is a different bug from the SigLIP2 implementation mismatch." }, { "left": "issue:44466", "right": "issue:45127", "accept": false, - "reason": "Serialization of tied lm_head weights is a different issue from LoRA merging with extended vocab causing model collapse." + "reason": "Serialization inconsistency for lm_head.weight is not the same underlying bug as LoRA merge collapse after vocab extension." }, { "left": "issue:39692", "right": "issue:46192", "accept": false, - "reason": "Documentation/example mistakes are not the same as the return_tensors kwarg being silently ignored." + "reason": "Docs/example mistakes and silently ignored return_tensors are different issues." }, { "left": "issue:44521", "right": "issue:45290", "accept": false, - "reason": "Both involve apply_chat_template, but one returns all-zero assistant masks while the other crashes on tool-call messages with no content; different bugs." + "reason": "Both touch apply_chat_template, but one is zeroed assistant masks and the other is a crash on tool-call messages; not the same failure." }, { "left": "issue:43653", "right": "issue:44112", "accept": false, - "reason": "BigBirdTokenizer special-token registration is unrelated to the GraniteSpeech CI stale-device override test." + "reason": "BigBirdTokenizer special-token registration bug vs GraniteSpeech CI device-override test flake; unrelated." }, { "left": "issue:45381", "right": "issue:46209", "accept": false, - "reason": "Video vision_position_ids corruption is unrelated to FP8 inference failure on B200." + "reason": "Qwen2.5-VL video position ids bug is unrelated to FP8 inference failure on B200." }, { "left": "issue:44737", "right": "issue:46894", "accept": false, - "reason": "CPU tensor creation in XLNet positional encoding is a different path from CohereASR's training-loss double-shift bug." + "reason": "XLNet CPU tensor allocation bug vs CohereASR double-shift training-loss bug; different subsystems." }, { "left": "issue:28218", "right": "issue:35141", "accept": false, - "reason": "Tokenization spacing after added tokens is unrelated to embedding resize reinitializing the output embedding." + "reason": "Tokenizer adds extra space after added token, while the other is output embedding reinit after resizing; distinct problems." }, { "left": "issue:43278", "right": "issue:43856", "accept": false, - "reason": "Embedding dtype drift between train/eval is a different issue from MoE training memory inefficiency." + "reason": "Embedding dtype drift in evaluate is not the same as Qwen3 MoE memory inefficiency." }, { "left": "issue:42915", "right": "issue:43824", "accept": false, - "reason": "FineGrainedFP8Config failure in Qwen3Moe is unrelated to the Qwen2.5-VL import error." + "reason": "FineGrainedFP8Config failure for Qwen3Moe and Qwen2_5_VL import error are unrelated." }, { "left": "issue:43950", "right": "issue:46285", "accept": false, - "reason": "Non-persistent buffer corruption on from_pretrained is not the same as CLIP model structural inconsistency." + "reason": "Non-persistent buffer corruption in from_pretrained is a different regression from the CLIP text-model structure change." }, { "left": "issue:43824", "right": "issue:43975", "accept": false, - "reason": "Importing a missing class and incorrect detokenization are distinct bugs with different code paths." + "reason": "ImportError for Qwen2_5_VL is unrelated to DeepSeek detokenization regression." }, { "left": "issue:43452", "right": "issue:46512", "accept": false, - "reason": "gguf_file loading failures are unrelated to from_config(dtype=...) leaving config.dtype stale." + "reason": "gguf_file loading breakage and stale config.dtype after from_config are different lifecycle bugs." }, { - "left": "issue:43262", - "right": "issue:43450", + "left": "issue:43334", + "right": "issue:43525", "accept": false, - "reason": "Audio chat-template sampling-rate defaults and batched video processor shapes are different media-processing bugs." + "reason": "Same symptom class (missing pad_token_id), but different models/configs and no evidence of one shared underlying fix." }, { - "left": "issue:43334", - "right": "issue:43525", + "left": "issue:43262", + "right": "issue:43450", "accept": false, - "reason": "They share a missing pad_token_id symptom, but they affect different config classes/models and likely require separate fixes." + "reason": "Audio apply_chat_template sampling-rate default bug is unrelated to batched video output shape." }, { "left": "issue:44442", "right": "issue:45198", "accept": false, - "reason": "AutoTokenizer loading failure for FastSpeech2ConformerTokenizer is different from Wav2Vec2 save_pretrained/tokenization failure." + "reason": "AutoTokenizer failing FastSpeech2ConformerTokenizer is a different issue than Wav2Vec2 save/tokenization failure." }, { "left": "issue:45200", "right": "issue:46899", "accept": false, - "reason": "Defaulting mm_token_type_ids to zeros is not the same as unsafe dtype casting in multimodal embedders." + "reason": "Missing mm_token_type_ids default and unsafe dtype casting in multimodal embedders are distinct multimodal bugs." }, { "left": "issue:44792", "right": "issue:45563", "accept": false, - "reason": "A failing janus image-generation test is unrelated to a stale paged-generate warning about num_return_sequences." + "reason": "Janus image-generation test failure and stale paged-generate warning are not the same code-path problem." } ] }, @@ -9335,24 +9255,24 @@ "issue:43526|issue:46192", "issue:45200|issue:45325", "issue:43827|issue:43901", - "issue:41720|issue:46209", "issue:43335|issue:43550", + "issue:41720|issue:46209", "issue:44561|issue:46234", "issue:43388|issue:43526", "issue:43819|issue:45083", "issue:44857|issue:45593", "issue:44610|issue:44857", "issue:46424|issue:46858", - "issue:45520|issue:46170", "issue:45759|issue:46285", + "issue:45520|issue:46170", "issue:46512|issue:46821", "issue:43606|issue:45083", "issue:38175|issue:45412", "issue:44206|issue:45584", "issue:43335|issue:43653", "issue:37428|issue:43425", - "issue:29942|issue:32090", "issue:43550|issue:44112", + "issue:29942|issue:32090", "issue:28282|issue:33357", "issue:43611|issue:43931", "issue:44448|issue:44743", @@ -9360,208 +9280,208 @@ "issue:44855|issue:45083", "issue:43278|issue:43825" ], - "prepared_review_unit_hash": "aec120847ba657c617d4a4e61212e972945eca906babab9987464f8a48602f47", + "prepared_review_unit_hash": "9e864900eb2e405c07ec049e758e6b4646bf2ed5b0d2a0ac236515f274ffbbef", "component_budget": { "node_count": 500, "item_count": 500, "soft_pair_count": 1839, - "serialized_chars": 953913, - "estimated_input_tokens": 238479, - "estimated_eval_tokens": 477214 + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 29, - "serialized_chars": 36746, - "estimated_input_tokens": 9187, - "estimated_eval_tokens": 18630 + "serialized_chars": 36749, + "estimated_input_tokens": 9188, + "estimated_eval_tokens": 18632 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Loose cluster of mostly unrelated Transformers issues: v5 compatibility regressions, model-specific loading/shape bugs, processor/tokenizer edge cases, and a few CI/runtime failures. None of the soft pairs look like true duplicates.", - "confidence": 0.77, - "canonical_issue_reason": "Issue 43611 is the broadest, most representative regression in the set: a Transformers v5 loading break that fits the recurring compatibility theme better than the more model-specific reports.", + "summary": "This cluster is heterogeneous: most pairs share only a broad subsystem or theme, not the same concrete bug/change, so all soft edges should be rejected. I\u2019d use issue 43611 as the best representative regression in the set.", + "confidence": 0.62, + "canonical_issue_reason": "issue:43611 is the most broadly impactful concrete regression here: a v5.0 loading break tied to `base_model_prefix`, with clear user-facing impact across multiple model classes.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 43611 is the best single anchor for the cluster because it is a clear, user-facing compatibility regression with broad impact and strong topical overlap with several other v5 breakage reports.", + "best_issue_reason": "Same reason: it is a well-scoped, high-impact runtime regression rather than a narrow model-specific edge case or a docs-only issue.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43819", "right": "issue:44112", "accept": false, - "reason": "Different bugs: DAC latent/STE mismatch vs GraniteSpeech stale device override CI failure." + "reason": "Different bugs: DAC latent reconstruction mismatch vs a GraniteSpeech CI device-override test failure." }, { "left": "issue:44898", "right": "issue:45412", "accept": false, - "reason": "Unrelated failures: Perceiver resize/interpolation issue vs RT-DETR memory release bug." + "reason": "Perceiver interpolation failure and RT-DETR memory leak are unrelated code paths and failure modes." }, { "left": "issue:43526", "right": "issue:46192", "accept": false, - "reason": "Both involve preprocessing/labels, but one is reduce_labels truncation and the other is return_tensors precedence; different defects." + "reason": "Both touch processors, but one is label reduction and the other is `text_kwargs`/`return_tensors` handling." }, { "left": "issue:45200", "right": "issue:45325", "accept": false, - "reason": "Both are multimodal positional-embedding issues, but for different models and different code paths." + "reason": "Different multimodal bugs: missing default `mm_token_type_ids` vs RoPE position scaling for still images." }, { "left": "issue:43827", "right": "issue:43901", "accept": false, - "reason": "Related to v5 pipeline removal, but one is a runtime error message and the other is outdated docs; not the same bug." + "reason": "Both are docs issues, but they concern different APIs and different deprecations." }, { - "left": "issue:41720", - "right": "issue:46209", + "left": "issue:43335", + "right": "issue:43550", "accept": false, - "reason": "Different hardware/model problems: Qwen3 auto device mapping assert vs FP8 inference on B200." + "reason": "SwitchTransformers config bug and Bamba torch.compile/SDPA failure are distinct problems." }, { - "left": "issue:43335", - "right": "issue:43550", + "left": "issue:41720", + "right": "issue:46209", "accept": false, - "reason": "SwitchTransformers sparse-layer config bug is unrelated to Bamba SDPA/torch.compile failure." + "reason": "Different GPU/hardware failures with no shared concrete code path." }, { "left": "issue:44561", "right": "issue:46234", "accept": false, - "reason": "Both are missing helper/import_utils breakages, but they concern different symbols and failure modes." + "reason": "Both involve missing utils in v5, but the missing symbols and breakage mechanisms differ." }, { "left": "issue:43388", "right": "issue:43526", "accept": false, - "reason": "Same broad label-processing area, but one drops tuple elements in gather_for_metrics and the other returns only one label in BeitImageProcessorFast." + "reason": "Metric gathering drops tuple labels; Beit image processor returns only one label. Different layers and symptoms." }, { "left": "issue:43819", "right": "issue:45083", "accept": false, - "reason": "Different helper/pathologies: DAC latent reconstruction vs qwen3_omni_moe feature-length calculation." + "reason": "Different helper/model logic: latent STE mismatch vs feature-length helper behavior." }, { "left": "issue:44857", "right": "issue:45593", "accept": false, - "reason": "Different model bugs: CUDA float16 crash vs loss computation omission when denoising is off." + "reason": "AMP crash in LwDetrImageLoss is unrelated to D-FINE's auxiliary-loss omission." }, { "left": "issue:44610", "right": "issue:44857", "accept": false, - "reason": "Different symptoms and subsystems: OmDet processor size mismatch vs LwDetrImageLoss AMP crash." + "reason": "Processor image-size mismatch and AMP/CUDA loss crash are separate bugs." }, { "left": "issue:46424", "right": "issue:46858", "accept": false, - "reason": "Both touch cache logic, but one is quantized-cache config behavior and the other is a static-cache seq_length tensor bug." + "reason": "Cache-length config handling and static-cache slice arithmetic are different defects." }, { - "left": "issue:45520", - "right": "issue:46170", + "left": "issue:45759", + "right": "issue:46285", "accept": false, - "reason": "Both are import/dependency issues, but one is flash_attn on Python 3.13 and the other is Windows ROCm/torch.distributed." + "reason": "Composite Qwen config unwrapping is unrelated to CLIP structure inconsistency." }, { - "left": "issue:45759", - "right": "issue:46285", + "left": "issue:45520", + "right": "issue:46170", "accept": false, - "reason": "Different model-structure regressions: Qwen multimodal text_config unwrap vs CLIP text-model structure inconsistency." + "reason": "Both are import/platform issues, but they stem from different dependencies and environments." }, { "left": "issue:46512", "right": "issue:46821", "accept": false, - "reason": "Unrelated: stale model.config.dtype after from_config vs TimesFM window_size attribute error." + "reason": "Stale dtype config state and TimesFM `window_size` AttributeError are unrelated." }, { "left": "issue:43606", "right": "issue:45083", "accept": false, - "reason": "Different code paths entirely: CPU offload device mismatch vs feature-extraction output length helper." + "reason": "CPU offload device mismatch is unrelated to qwen3_omni_moe feature-length logic." }, { "left": "issue:38175", "right": "issue:45412", "accept": false, - "reason": "Different issues: unexpected zero probabilities in Siglip2 vs RT-DETR memory not freed on delete." + "reason": "SigLIP2 zero probabilities and RT-DETR memory leakage are unrelated." }, { "left": "issue:44206", "right": "issue:45584", "accept": false, - "reason": "Both are pipeline/model regressions, but one is a feature extractor center arg crash and the other is Whisper empty-transcription handling." + "reason": "Unsupported `center` arg crash and Whisper empty-transcription failure are different audio bugs." }, { "left": "issue:43335", "right": "issue:43653", "accept": false, - "reason": "Different bugs in different components: SwitchTransformers config vs BigBirdTokenizer special-token registration." + "reason": "Sparse-layer creation and BigBird special-token registration are separate issues." }, { "left": "issue:37428", "right": "issue:43425", "accept": false, - "reason": "One is a missing flash-attention symbol; the other is a broader Torch 2.10 incompatibility report." + "reason": "Flash-attention import-symbol failure and generic Torch 2.10 incompatibility are not the same bug." }, { - "left": "issue:29942", - "right": "issue:32090", + "left": "issue:43550", + "right": "issue:44112", "accept": false, - "reason": "Unrelated CI/test failure vs Trainer broadcast NoneType error." + "reason": "Bamba compile/SDPA runtime failure is unrelated to a stale GraniteSpeech CI test." }, { - "left": "issue:43550", - "right": "issue:44112", + "left": "issue:29942", + "right": "issue:32090", "accept": false, - "reason": "Bamba torch.compile/SDPA bug is unrelated to the GraniteSpeech stale device override failure." + "reason": "Flash Attention test failures and Trainer `_gpu_broadcast_one` TypeError are unrelated." }, { "left": "issue:28282", "right": "issue:33357", "accept": false, - "reason": "Different environments and failure modes: PyTorch missing for AutoModel import vs MacOS bus error with community CLIP." + "reason": "PyTorch-missing import error and MacOS bus error in CLIP are different failure modes." }, { "left": "issue:43611", "right": "issue:43931", "accept": false, - "reason": "Both are loading regressions, but one is base_model_prefix handling and the other is Qwen3-VL shape mismatch; not the same underlying issue." + "reason": "Broad v5 loading regression vs a model-specific Qwen3-VL shape mismatch." }, { "left": "issue:44448", "right": "issue:44743", "accept": false, - "reason": "Different model behaviors: Pegasus v4/v5 output change vs recurrent state reset in qwen3_5 cache handling." + "reason": "Pegasus v4/v5 output drift and Qwen recurrent-state reset are unrelated." }, { "left": "issue:43335", "right": "issue:43504", "accept": false, - "reason": "Both involve config/loading, but one is sparse-layer creation and the other is a legacy-field preset load failure." + "reason": "SwitchTransformers sparse-layer config bug vs Beit pretrained-load legacy-field bug." }, { "left": "issue:44855", "right": "issue:45083", "accept": false, - "reason": "Python 3.13 torch.jit parsing error is unrelated to qwen3_omni_moe feature-length arithmetic." + "reason": "Python 3.13 DebertaV2 import parsing failure is unrelated to qwen3_omni_moe helper behavior." }, { "left": "issue:43278", "right": "issue:43825", "accept": false, - "reason": "Different regressions: dtype mismatch between train/eval and a misleading pipeline() error message." + "reason": "Embedding dtype drift in train/eval is unrelated to the v5 pipeline error-message regression." } ] }, @@ -9638,10 +9558,10 @@ "issue:29127|issue:32090", "issue:43335|issue:45083", "issue:43720|issue:44112", - "issue:37428|issue:44351", - "issue:44514|issue:44521", "issue:43335|issue:45593", "issue:43335|issue:44112", + "issue:37428|issue:44351", + "issue:44514|issue:44521", "issue:45290|issue:46326", "issue:45127|issue:45245", "issue:45588|issue:45859", @@ -9656,8 +9576,8 @@ "issue:43606|issue:44857", "issue:28218|issue:32090", "issue:45459|issue:46234", - "issue:39692|issue:44493", "issue:43701|issue:45083", + "issue:39692|issue:44493", "issue:44617|issue:46697", "issue:44206|issue:45823", "issue:32090|issue:33357", @@ -9665,136 +9585,136 @@ "issue:44360|issue:44561", "issue:44857|issue:45072" ], - "prepared_review_unit_hash": "15eb67d311ca0f537e876d1c3484d16340b69c1cb18d13f8b4cda96437fd7dbc", + "prepared_review_unit_hash": "a543cfbcad867e46d9104f1c9badd65c4c065d01f18d0430a1ab32e0e8d2b8b1", "component_budget": { "node_count": 500, "item_count": 500, "soft_pair_count": 1839, - "serialized_chars": 953913, - "estimated_input_tokens": 238479, - "estimated_eval_tokens": 477214 + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 29, - "serialized_chars": 36462, - "estimated_input_tokens": 9116, - "estimated_eval_tokens": 18488 + "serialized_chars": 36466, + "estimated_input_tokens": 9117, + "estimated_eval_tokens": 18490 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "One clear duplicate pair stands out: the two Qwen3.5 MoE from_pretrained errors are the same bug. The other soft pairs share broad subsystems (chat templates, loading, device/dtype issues, tokenizers) but describe different concrete failures and should stay separate.", - "confidence": 0.94, - "canonical_issue_reason": "issue:45310 is the earlier, more complete report of the same Qwen3.5 Moe from_pretrained regression; issue:45478 is a near-identical duplicate.", + "summary": "This cluster is mostly a set of unrelated bug reports. The only clear duplicate pair is the repeated Qwen3.5 Moe `from_pretrained` regression; the rest are similar only at a broad subsystem level and should stay separate.", + "confidence": 0.98, + "canonical_issue_reason": "issue:45310 is the earlier and more detailed report, and issue:45478 repeats the same Qwen3.5 Moe `from_pretrained` error verbatim.", "canonical_pr_reason": null, - "best_issue_reason": "issue:45310 is the strongest issue anchor here because it is the canonical target for the only confirmed duplicate pair and has the clearer regression report.", + "best_issue_reason": "issue:45310 is the strongest canonical issue candidate because it predates the duplicate and has the fuller original discussion context.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:29127", "right": "issue:32090", "accept": false, - "reason": "Different bugs: LayoutLMv3 error messaging vs Trainer GPU broadcast on NoneType." + "reason": "Different failures in different paths: LayoutLMv3 error-message clarity vs Trainer `_gpu_broadcast_one` rejecting `NoneType`." }, { "left": "issue:43335", "right": "issue:45083", "accept": false, - "reason": "Unrelated failures: SwitchTransformers sparse-layer config vs qwen3_omni_moe feature-length helper." + "reason": "Different models and bugs: SwitchTransformers sparse-layer creation vs qwen3_omni_moe feature-length helper behavior." }, { "left": "issue:43720", "right": "issue:44112", "accept": false, - "reason": "Different codepaths: BitNet packed-weight loading vs GraniteSpeech CI device-override test." + "reason": "Both are CI/device-related, but one is BitNet weight unpacking during accelerate loading and the other is a stale GraniteSpeech device override test." }, { - "left": "issue:37428", - "right": "issue:44351", + "left": "issue:43335", + "right": "issue:45593", "accept": false, - "reason": "Different missing imports: flash attention helper vs HybridCache export." + "reason": "SwitchTransformers config bug and D-FINE auxiliary-loss behavior are unrelated." }, { - "left": "issue:44514", - "right": "issue:44521", + "left": "issue:43335", + "right": "issue:44112", "accept": false, - "reason": "Both involve chat templates, but one crashes on batched padding=False and the other returns zero assistant masks; not the same bug." + "reason": "Sparse-layer config bug vs GraniteSpeech stale-device CI failure; not the same code path." }, { - "left": "issue:43335", - "right": "issue:45593", + "left": "issue:37428", + "right": "issue:44351", "accept": false, - "reason": "SwitchTransformers sparse-layer config is unrelated to D-FINE auxiliary-loss behavior." + "reason": "Both are import-related, but they concern different missing symbols and different modules." }, { - "left": "issue:43335", - "right": "issue:44112", + "left": "issue:44514", + "right": "issue:44521", "accept": false, - "reason": "Config bug vs CI stale device-override test; no shared underlying failure." + "reason": "Both involve `apply_chat_template`, but one is batching/padding crash while the other is incorrect assistant masks for multimodal inputs." }, { "left": "issue:45290", "right": "issue:46326", "accept": false, - "reason": "Both touch chat templating, but tool-call/no-content crashes differ from continue_final_message behavior." + "reason": "Different chat-template bugs: tool-call assistant messages with no content vs `continue_final_message` on `mistral_common` tokenizers." }, { "left": "issue:45127", "right": "issue:45245", "accept": false, - "reason": "LoRA/tied-embeddings collapse is unrelated to category-count limits." + "reason": "Extended-vocabulary LoRA merge collapse is unrelated to the 2^24 category limit runtime error." }, { "left": "issue:45588", "right": "issue:45859", "accept": false, - "reason": "Different failures in different subsystems: flash_attention sink handling vs Qwen3.5 MoE CPU-offload rotary embedding." + "reason": "Different failure modes in different components: flash-attention `s_aux=None` crash vs CPU-offload incompatibility in Qwen3_5MoeTextRotaryEmbedding." }, { "left": "issue:34689", "right": "issue:35141", "accept": false, - "reason": "Model-loading regression for Llama 3.2 Vision is unrelated to embedding reinitialization after resize." + "reason": "Model-loading regression for Llama 3.2 Vision is not the same as output embedding reinitialization after resizing embeddings." }, { "left": "issue:42490", "right": "issue:43854", "accept": false, - "reason": "Local-path import warning is not the same as the GLM-4.7-Flash model-loading failure." + "reason": "Local-path import warning and GLM-4.7-Flash loading failure are distinct issues." }, { "left": "issue:46431", "right": "issue:46566", "accept": false, - "reason": "Missing chat template is unrelated to DiffusionGemma weights staying on meta with device_map=auto." + "reason": "Processor missing a chat template is unrelated to DiffusionGemma weights staying on `meta` with `device_map='auto'`." }, { "left": "issue:45310", "right": "issue:45478", "accept": true, - "reason": "Same reported regression: transformers>=5.4.0 causes Qwen3.5 Moe from_pretrained errors in both issues." + "reason": "Same exact Qwen3.5 Moe `from_pretrained` error; `45478` is a duplicate of `45310`." }, { "left": "issue:44448", "right": "issue:44779", "accept": false, - "reason": "Different models and symptoms: Pegasus v4/v5 output drift vs Deepseek tokenizer correctness regression." + "reason": "Separate version-regression reports for different models and tokenization/generation paths." }, { "left": "issue:43819", "right": "issue:44493", "accept": false, - "reason": "DAC latent/STE mismatch is a different bug from unexpected position-id keys across models." + "reason": "DAC latent/forward mismatch and unexpected position-id keys are different bugs." }, { "left": "issue:39692", "right": "issue:43526", "accept": false, - "reason": "Documentation/example errors are unrelated to BeitImageProcessorFast reduce_labels behavior." + "reason": "SigLIP2 documentation example errors are unrelated to BeitImageProcessorFast `reduce_labels` behavior." }, { "left": "issue:26413", @@ -9806,73 +9726,73 @@ "left": "issue:43606", "right": "issue:44857", "accept": false, - "reason": "CPU-offload device mismatch for bark-small is unrelated to float16 AMP crash in LwDetrImageLoss." + "reason": "CPU-offload device mismatch in Bark is unrelated to LwDetrImageLoss AMP CUDA crash." }, { "left": "issue:28218", "right": "issue:32090", "accept": false, - "reason": "Tokenizer added-space bug is unrelated to Trainer NoneType broadcast failure." + "reason": "Tokenizer extra-space behavior is a different bug from Trainer broadcast type handling." }, { "left": "issue:45459", "right": "issue:46234", "accept": false, - "reason": "Tokenizer protobuf error masking is unrelated to the missing falcon_mamba_ssm availability helper." + "reason": "Tokenizer protobuf error masking and missing `is_falcon_mamba_ssm_available` are distinct utility-layer issues." }, { - "left": "issue:39692", - "right": "issue:44493", + "left": "issue:43701", + "right": "issue:45083", "accept": false, - "reason": "SigLIP2 docs/example problems do not match the position-id key regression." + "reason": "Checkpoint key mismatch and qwen3_omni_moe helper behavior do not describe the same bug." }, { - "left": "issue:43701", - "right": "issue:45083", + "left": "issue:39692", + "right": "issue:44493", "accept": false, - "reason": "Resume-from-checkpoint key mismatch is unrelated to qwen3_omni_moe feature-length behavior." + "reason": "Documentation example issues are unrelated to the unexpected position-id key regression." }, { "left": "issue:44617", "right": "issue:46697", "accept": false, - "reason": "CUDA OOM in Sam3Video is unrelated to Swin2SR image over-padding." + "reason": "CUDA OOM in Sam3Video and Swin2SR over-padding are unrelated." }, { "left": "issue:44206", "right": "issue:45823", "accept": false, - "reason": "Unsupported center arg crash in LasrFeatureExtractor is different from Gemma4 PLE device mismatch." + "reason": "Unsupported `center` arg crash and Gemma4 PLE device mismatch are different code paths." }, { "left": "issue:32090", "right": "issue:33357", "accept": false, - "reason": "Trainer broadcast error and Mac CLIP bus error are distinct failures." + "reason": "Trainer `_gpu_broadcast_one` type error is unrelated to the MacOS CLIP bus error." }, { "left": "issue:42490", "right": "issue:43575", "accept": false, - "reason": "Local-path import warning is unrelated to Qwen2-57B-A14B-Instruct tensor-parallel OOM." + "reason": "Import warning for local paths is unrelated to tensor-parallel OOM on Qwen2-57B-A14B-Instruct." }, { "left": "issue:44360", "right": "issue:44561", "accept": false, - "reason": "DSA indexer ReLU discussion is unrelated to removal of is_torch_fx_available breaking trust_remote_code models." + "reason": "DSA indexer discussion and removal of `is_torch_fx_available` breaking trust-remote-code models are unrelated." }, { "left": "issue:44857", "right": "issue:45072", "accept": false, - "reason": "LwDetrImageLoss AMP crash and bfloat16 dtype mismatch in SwitchTransformers/TimmWrapperModel are different concrete bugs." + "reason": "Different dtype/precision failures in different models; not the same underlying bug." } ] }, "evaluator_result": { "accept": true, - "feedback": "The summary is well grounded in the packet. The canonical duplicate pair (issue:45310 / issue:45478) is supported by the identical titles and close timing, and the remaining soft-edge rejections stay conservative by distinguishing concrete symptoms rather than broad subsystem overlap." + "feedback": "The summary is grounded in the packet, and the only accepted pair (issue:45310/issue:45478) is a clear duplicate based on identical titles and highly similar content. The remaining soft-edge verdicts stay conservative by treating broad subsystem similarity as insufficient for duplication." }, "evaluator_used": true, "retried": false, @@ -9974,214 +9894,214 @@ "issue:43278|issue:43575", "issue:44871|issue:44977" ], - "prepared_review_unit_hash": "584163f6c4514e3a021c0c838d4ecac84d5f4802d26026180da39491f7d55009", + "prepared_review_unit_hash": "5a2f54f25b95b5bdd80872bcb8bf6211efca454a03d380bbd7e7d1519a965201", "component_budget": { "node_count": 500, "item_count": 500, "soft_pair_count": 1839, - "serialized_chars": 953913, - "estimated_input_tokens": 238479, - "estimated_eval_tokens": 477214 + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 30, - "serialized_chars": 37081, - "estimated_input_tokens": 9271, - "estimated_eval_tokens": 18798 + "serialized_chars": 37072, + "estimated_input_tokens": 9268, + "estimated_eval_tokens": 18792 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The soft edges are mostly false positives: the paired issues span unrelated bugs across model loading, tokenizer behavior, flash-attention, caching, OOM, serialization, and training/runtime errors. No single canonical issue or PR emerges from this set.", - "confidence": 0.98, - "canonical_issue_reason": null, + "summary": "These reports are largely unrelated, model-specific regressions spanning tokenizer behavior, flash-attention, cache handling, serialization, and packaging. The similarity is mostly thematic, not the same underlying bug/change, so none of the soft pairs should be merged.", + "confidence": 0.95, + "canonical_issue_reason": "issue:45920 is the broadest, most representative standalone bug here: a multi-model v5 tokenizer regression with inbound references and strong discussion activity.", "canonical_pr_reason": null, - "best_issue_reason": null, + "best_issue_reason": "issue:45920 has the best combination of breadth, cross-model impact, and external references, making it the strongest issue to anchor this cluster.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43550", "right": "issue:45083", "accept": false, - "reason": "Different bugs and code paths: torch.compile/SDPA on Bamba-9B-v2 vs a qwen3_omni_moe helper-function length calculation issue." + "reason": "Different failures: Bamba torch.compile/SDPA vs qwen3_omni_moe feature-length helper bug." }, { "left": "issue:27957", "right": "issue:34689", "accept": false, - "reason": "Unrelated failures: XLMRoberta Flash Attention 2 support vs Llama 3.2 90B Vision model loading regression." + "reason": "Flash-Attn2 support request vs Llama 3.2 Vision model-loading regression." }, { "left": "issue:29942", "right": "issue:35141", "accept": false, - "reason": "Different problems: Flash Attention 2 tests failing vs output embedding reinitialization after resizing token embeddings." + "reason": "Failing Flash-Attn2 tests are unrelated to embedding reinitialization after resize." }, { "left": "issue:38175", "right": "issue:44493", "accept": false, - "reason": "Distinct symptoms and subsystems: zero probabilities in SigLIP2 vs unexpected position-id key warnings in many models." + "reason": "Zero probabilities in siglip2 is not the same as unexpected position-id keys across many models." }, { "left": "issue:29870", "right": "issue:32090", "accept": false, - "reason": "Different runtime areas: Bart generate decoder_attention_mask bug vs Trainer GPU broadcast NoneType error." + "reason": "Generate-time decoder mask bug vs Trainer/NoneType broadcast error." }, { "left": "issue:43278", "right": "issue:43329", "accept": false, - "reason": "One is dtype drift between training and eval; the other is undefined variables in multimodal token counting for video." + "reason": "Embedding dtype drift on eval is unrelated to undefined vars in multimodal token counting." }, { "left": "issue:44448", "right": "issue:44977", "accept": false, - "reason": "Different concrete failures: Pegasus v4/v5 output mismatch vs Qwen3.5 flash-attention generation breakage." + "reason": "Different output regression for Pegasus vs Qwen3.5 flash-attention generation crash." }, { "left": "issue:45920", "right": "issue:46192", "accept": false, - "reason": "Both involve tokenization APIs, but one is wrong token IDs regression and the other is return_tensors being ignored with text_kwargs." + "reason": "Tokenizer ID regression vs return_tensors being ignored in text_kwargs." }, { "left": "issue:45290", "right": "issue:45381", "accept": false, - "reason": "Different code paths: chat template tool-call crash vs Qwen2.5-VL video vision_position_ids bug." + "reason": "Chat-template tool-call crash and Qwen2.5-VL video position-id bug affect different paths." }, { "left": "issue:10000", "right": "issue:15354", "accept": false, - "reason": "Completely unrelated: a DistilBERT German model issue vs a torch.jit.script GeneratorExp export problem." + "reason": "German DistilBertModel issue is unrelated to torch.jit.script GeneratorExp export failure." }, { "left": "issue:44977", "right": "issue:45229", "accept": false, - "reason": "Both are generation/inference issues, but flash-attention behavior in Qwen3.5 is unrelated to Gemma4 multi-GPU CUDA OOM." + "reason": "Flash-attention generation problem is unrelated to Gemma4 multi-GPU OOM." }, { "left": "issue:43278", "right": "issue:43381", "accept": false, - "reason": "Different bugs: embedding dtype changes vs gradient checkpointing being disallowed in eval mode." + "reason": "Eval-mode gradient checkpointing is a different bug from BF16-to-FP32 dtype changes." }, { "left": "issue:29870", "right": "issue:33290", "accept": false, - "reason": "No match: Bart generate error is unrelated to deepspeed Adafactor OOM." + "reason": "Decoder attention-mask assignment bug is unrelated to deepspeed adafactor OOM." }, { "left": "issue:43927", "right": "issue:45698", "accept": false, - "reason": "Both touch save/load paths, but one loses custom token IDs and the other loads the wrong custom module after save_pretrained." + "reason": "Custom token IDs lost on save/load is different from loading the wrong custom module after save_pretrained." }, { "left": "issue:44855", "right": "issue:46399", "accept": false, - "reason": "Different failures: Python 3.13 import/JIT parsing problem vs missing decorator in Qwen3.5 kernel registration." + "reason": "Python 3.13 import parsing error is unrelated to Qwen3.5 kernel registration/decorator failure." }, { "left": "issue:45290", "right": "issue:46655", "accept": false, - "reason": "Unrelated processors: chat template tool-call crash vs SmolVLM video padding calculation error." + "reason": "Tool-call chat-template crash is not the same as SmolVLM video padding calculation bug." }, { "left": "issue:40444", "right": "issue:43526", "accept": false, - "reason": "Different multimodal bugs: iterable dataset with multiple images per prompt vs BEiT fast processor label reduction." + "reason": "IterableDataset multi-image finetuning failure is unrelated to BeitImageProcessorFast reduce_labels." }, { "left": "issue:44857", "right": "issue:45588", "accept": false, - "reason": "Different stack traces and causes: AMP float16 crash in LwDetrImageLoss vs flash_attention.py AttributeError on s_aux=None." + "reason": "AMP/CUDA loss crash and flash_attention s_aux=None attribute error are different code paths." }, { "left": "issue:37428", "right": "issue:42175", "accept": false, - "reason": "Unrelated: missing flash_supports_window_size import vs packaging/backend selection missing TensorFlow for pip install '.[torch]'." + "reason": "Missing flash-attention helper import is unrelated to torch-only pip backend packaging." }, { "left": "issue:32090", "right": "issue:33290", "accept": false, - "reason": "Different training failures: unsupported NoneType broadcast vs OOM with Adafactor in deepspeed." + "reason": "Trainer broadcast NoneType error and quantized deepspeed OOM are different symptoms and subsystems." }, { "left": "issue:46421", "right": "issue:46858", "accept": false, - "reason": "Both involve cache/compile behavior, but one is spurious recompilation and the other is device-tensor seq_length breaking slice arithmetic." + "reason": "Chunked-prefill recompilation with lazy cache init is distinct from static-cache seq_length tensor arithmetic." }, { "left": "issue:44561", "right": "issue:44779", "accept": false, - "reason": "Different v5 regressions: removal of is_torch_fx_available breaking trust_remote_code vs DeepSeek tokenizer producing wrong IDs." + "reason": "Removed is_torch_fx_available helper breakage is unrelated to DeepSeek tokenizer regression." }, { "left": "issue:44448", "right": "issue:44466", "accept": false, - "reason": "Both mention v5 behavior, but one is model output drift and the other is inconsistent lm_head.weight serialization tied to device." + "reason": "Pegasus output differences do not match the lm_head.weight serialization/tied-weights issue." }, { "left": "issue:43873", "right": "issue:45229", "accept": false, - "reason": "Different resource issues: quantization offloading not working vs Gemma4 multi-GPU inference CUDA OOM." + "reason": "Quantization/offloading behavior is unrelated to Gemma4 multi-GPU CUDA OOM." }, { "left": "issue:25251", "right": "issue:32090", "accept": false, - "reason": "Different areas: pipeline top_k nesting change vs Trainer GPU broadcast NoneType error." + "reason": "Pipeline top_k nesting bug is unrelated to the Trainer broadcast NoneType error." }, { "left": "issue:43976", "right": "issue:45440", "accept": false, - "reason": "Completely different: Python-version compatibility issue vs divergence of native DeepseekV3MoE from remote implementation." + "reason": "Python-version compatibility bug is unrelated to DeepSeekV3MoE divergence from remote implementation." }, { "left": "issue:42490", "right": "issue:43873", "accept": false, - "reason": "Different bugs: trailing-slash local model import warning vs quantization offloading not behaving as expected." + "reason": "Trailing-slash import warning is not the same as offloading failing with quantization." }, { "left": "issue:44568", "right": "issue:45491", "accept": false, - "reason": "Different model/tokenizer issues: missing BOS/EOS on add_special_tokens vs NaN embeddings with Gemma3 sliding-window attention." + "reason": "Tokenizer add_special_tokens regression is unrelated to Gemma3 sliding-window NaN embeddings." }, { "left": "issue:43278", "right": "issue:43575", "accept": false, - "reason": "Distinct failure modes: BF16-to-FP32 dtype change in eval vs tensor-parallel load OOM for Qwen2-57B." + "reason": "Eval embedding dtype drift is unrelated to tensor-parallel OOM on Qwen2-57B." }, { "left": "issue:44871", "right": "issue:44977", "accept": false, - "reason": "Different issues: eos_token_id mismatch for Gemma-3 vs Qwen3.5 flash-attention generation failure." + "reason": "Gemma3 eos_token_id inconsistency is a different bug from Qwen3.5 flash-attention generation failure." } ] }, @@ -10272,8 +10192,8 @@ "issue:43756|issue:45440", "issue:43653|issue:44442", "issue:29942|issue:30064", - "issue:45372|issue:45406", "issue:44206|issue:44479", + "issue:45372|issue:45406", "issue:43412|issue:45588", "issue:45459|issue:45712", "issue:45491|issue:46693", @@ -10288,226 +10208,226 @@ "issue:45372|issue:45459", "issue:42490|issue:43381" ], - "prepared_review_unit_hash": "c4603b6c675a36347ceabfdab50a405a31716331371d75dca1897c1aeb1d3aec", + "prepared_review_unit_hash": "71cd6015c21faf7a4f6f7e40763cf034a244b7872ef2c0f33546c3eb90daa161", "component_budget": { "node_count": 500, "item_count": 500, "soft_pair_count": 1839, - "serialized_chars": 953913, - "estimated_input_tokens": 238479, - "estimated_eval_tokens": 477214 + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 32, - "serialized_chars": 37294, - "estimated_input_tokens": 9324, - "estimated_eval_tokens": 18904 + "serialized_chars": 37287, + "estimated_input_tokens": 9322, + "estimated_eval_tokens": 18900 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This is a loose topical cluster of mostly unrelated Transformers bugs; the similarity links are not strong enough to treat most pairs as duplicates. The best standalone issue is the AutoTokenizer/tokenizer.json loading regression, which is clear and self-contained.", - "confidence": 0.4, - "canonical_issue_reason": "issue:44462 is a clean, well-scoped loading regression with a concrete repro path, and it is the most representative standalone bug in this otherwise heterogeneous cluster.", + "summary": "This cluster is a grab-bag of mostly unrelated Transformers issues. The soft links are generally based on broad lexical similarity (same model family, same subsystem, or same error class) rather than the same underlying bug, so none should be merged as duplicates.", + "confidence": 0.95, + "canonical_issue_reason": "No clear canonical issue: the items span unrelated model-specific regressions, tokenizer/load bugs, training-state issues, quantization problems, and documentation/examples.", "canonical_pr_reason": null, - "best_issue_reason": "issue:44462 is the strongest general triage candidate: specific, actionable, and centered on a user-visible loading bug rather than an incidental symptom.", + "best_issue_reason": "No issue is representative enough to act as a reliable duplicate target for the whole cluster.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43381", "right": "issue:43701", "accept": false, - "reason": "Different bugs: eval-mode gradient checkpointing vs checkpoint resume key mismatch." + "reason": "Both involve training utilities, but one is eval-mode gradient checkpointing and the other is resume-from-checkpoint key mismatch; different code paths." }, { "left": "issue:45588", "right": "issue:45823", "accept": false, - "reason": "Different code paths: flash-attention sink handling vs Gemma4 PLE device placement." + "reason": "Different failures in Gemma4/flash-attention vs device-map forwarding; same broad area, not the same bug." }, { "left": "issue:33357", "right": "issue:39692", "accept": false, - "reason": "One is a MacOS bus error for a CLIP model; the other is a SigLIP2 doc example mismatch/quantization issue." + "reason": "Both mention vision models, but one is a MacOS CLIP bus error and the other is a SigLIP2 documentation/example issue." }, { "left": "issue:29127", "right": "issue:34567", "accept": false, - "reason": "Unrelated: LayoutLMv3 box validation messaging vs TrainerState token counting update logic." + "reason": "Unrelated problems: LayoutLMv3 error messaging vs TrainerState token-count tracking." }, { "left": "issue:41720", "right": "issue:45229", "accept": false, - "reason": "Both are multi-GPU/CUDA problems, but they are different failure modes and likely different fixes." + "reason": "Both are inference/OOM-like reports, but they affect different models and different failure modes (cuda assert vs CUDA OOM)." }, { "left": "issue:44977", "right": "issue:45563", "accept": false, - "reason": "Flash-attention generation failure is unrelated to a stale warning in paged generate()." + "reason": "Qwen3.5 flash-attention generation bug vs a stale warning in paged generate(); not the same defect." }, { "left": "issue:45198", "right": "issue:45803", "accept": false, - "reason": "Different tokenizer/save issue vs a broad exception-handling bug in Fuyu tensor conversion." + "reason": "Different tokenizer/preprocessing paths and different bugs: save/tokenization failure vs broad exception masking in Fuyu conversion." }, { "left": "issue:44361", "right": "issue:45198", "accept": false, - "reason": "Different tokenizer failures in different models and code paths." + "reason": "Different tokenizer families and unrelated crashes." }, { "left": "issue:44186", "right": "issue:45198", "accept": false, - "reason": "LayoutLMv2 NER/padding crash is unrelated to Wav2Vec2 save_pretrained/tokenization failure." + "reason": "LayoutLMv2 tokenizer padding/NER crash is unrelated to Wav2Vec2 save/tokenization failure." }, { "left": "issue:43653", "right": "issue:44743", "accept": false, - "reason": "Tokenizer special-token registration bug is unrelated to Qwen recurrent-state cache handling." + "reason": "BigBird tokenizer special-token registration vs Qwen3.5 recurrent-state reset; no shared underlying bug." }, { "left": "issue:43611", "right": "issue:46121", "accept": false, - "reason": "Model-loading base_model_prefix regression is not the same as rope-validation list handling." + "reason": "Base-model-prefix loading regression is unrelated to rope-parameter validation and JSON list handling." }, { "left": "issue:29870", "right": "issue:34689", "accept": false, - "reason": "BART generate() decoder mask initialization and Llama 3.2 vision loading are different regressions." + "reason": "BART generation kwargs bug vs Llama 3.2 Vision model loading regression; different subsystems and fixes." }, { "left": "issue:45372", "right": "issue:45588", "accept": false, - "reason": "Gemma4 processor import/dependency failure is distinct from flash-attention sinkless-model crash handling." + "reason": "Gemma4 processor import/dependency issue vs flash-attention AttributeError; same model family, different concrete failure." }, { "left": "issue:42915", "right": "issue:42994", "accept": false, - "reason": "FP8 config failure during Qwen3Moe load is not the same as quantized model saving failure." + "reason": "Both are quantization-related, but one is FineGrainedFP8 loading and the other is quantized-model saving." }, { "left": "issue:43756", "right": "issue:45440", "accept": false, - "reason": "Smollm3 RoPE layer mismatch and DeepseekV3MoE divergence are different model implementation bugs." + "reason": "Different model implementations and different symptoms (RoPE-layer mismatch vs DeepSeekV3MoE divergence)." }, { "left": "issue:43653", "right": "issue:44442", "accept": false, - "reason": "BigBird mask-token registration is unrelated to FastSpeech2Conformer tokenizer loading." + "reason": "BigBird tokenizer special-token bug is unrelated to AutoTokenizer loading FastSpeech2ConformerTokenizer." }, { "left": "issue:29942", "right": "issue:30064", "accept": false, - "reason": "Flash Attention 2 test failures and void segmentation-map processing are unrelated." + "reason": "Flash Attention 2 test failures and void-segmentation-map processing are unrelated." }, { - "left": "issue:45372", - "right": "issue:45406", + "left": "issue:44206", + "right": "issue:44479", "accept": false, - "reason": "Shared Gemma4 context, but one is an import dependency issue while the other is a missing processor attribute in serve." + "reason": "Unsupported center argument in image feature extraction is unrelated to a video-input regression in Qwen VL models." }, { - "left": "issue:44206", - "right": "issue:44479", + "left": "issue:45372", + "right": "issue:45406", "accept": false, - "reason": "Unsupported center argument in LasrFeatureExtractor is unrelated to video-input regression for Qwen VL models." + "reason": "Related Gemma4 area, but one is a mistral_common import failure and the other is a missing _tokenizer attribute in serve; not the same code-path bug." }, { "left": "issue:43412", "right": "issue:45588", "accept": false, - "reason": "Gemma3n export guard/erfinv support and flash-attention aux-state crash are different bugs." + "reason": "Executorch export guard/support issues are unrelated to flash-attention sink-less model crashes." }, { "left": "issue:45459", "right": "issue:45712", "accept": false, - "reason": "Tokenizer exception masking is unrelated to leftover dummy classes in dummy_pt_objects.py." + "reason": "Tokenizer protobuf error handling vs leftover dummy classes; no shared bug." }, { "left": "issue:45491", "right": "issue:46693", "accept": false, - "reason": "Gemma3 NaN embeddings and FA2 long-context performance regression are different symptoms/fixes." + "reason": "Both mention FA2/long context, but one is NaN embeddings and the other is a performance regression." }, { "left": "issue:44265", "right": "issue:45198", "accept": false, - "reason": "torch.export exportability with torch_compilable_check is unrelated to Wav2Vec2 tokenization/save issues." + "reason": "torch.export/torch_compilable_check failure is unrelated to Wav2Vec2 save/tokenization failure." }, { "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 MacOS bus error are unrelated." }, { "left": "issue:43065", "right": "issue:43994", "accept": false, - "reason": "Dummy Conv2d in SAM3 pixel decoder is unrelated to SigLIP2 nonsensical AutoModel/pipeline outputs." + "reason": "Dummy Conv2d placeholder in SAM3 pixel decoder is unrelated to SigLIP2 nonsensical AutoModel/pipeline results." }, { "left": "issue:43065", "right": "issue:45412", "accept": false, - "reason": "Dummy layer cleanup and RT-DETR memory release are different problems." + "reason": "Dummy SAM3 layer cleanup is unrelated to RT-DETR memory not being released." }, { "left": "issue:45245", "right": "issue:45563", "accept": false, - "reason": "Category-count RuntimeError is unrelated to a stale num_return_sequences warning." + "reason": "Category-count limit error and stale generate warning are unrelated." }, { "left": "issue:43575", "right": "issue:43701", "accept": false, - "reason": "Tensor-parallel OOM on Qwen2-57B-A14B is unrelated to checkpoint key mismatch." + "reason": "Tensor-parallel OOM for a large Qwen model is unrelated to checkpoint key mismatch." }, { "left": "issue:43066", "right": "issue:44462", "accept": false, - "reason": "Both concern tokenizers, but one is a wrong decoder-type bug while the other is AutoTokenizer ignoring tokenizer.json." + "reason": "Tokenizer decoder-type migration bug vs AutoTokenizer ignoring tokenizer.json; different loading paths and fixes." }, { "left": "issue:45440", "right": "issue:45593", "accept": false, - "reason": "DeepseekV3MoE implementation divergence is unrelated to D-FINE auxiliary-loss behavior." + "reason": "DeepSeekV3MoE divergence is unrelated to D-FINE auxiliary-loss logic." }, { "left": "issue:45372", "right": "issue:45459", "accept": false, - "reason": "Gemma4 processor import failure and protobuf error masking are separate tokenizer/processor issues." + "reason": "Gemma4 processor import regression is unrelated to protobuf decode-error handling in tokenizers." }, { "left": "issue:42490", "right": "issue:43381", "accept": false, - "reason": "Local-path import warning and eval-mode gradient checkpointing are unrelated." + "reason": "Local-path slash import warning is unrelated to eval-mode gradient checkpointing behavior." } ] }, @@ -10604,222 +10524,222 @@ "issue:45588|issue:45600", "issue:43381|issue:43873", "issue:44514|issue:45381", - "issue:43646|issue:44077", "issue:43761|issue:45859", + "issue:43646|issue:44077", "issue:44977|issue:45245", "issue:44743|issue:44811", "issue:44442|issue:44898", "issue:43262|issue:44206", "issue:42490|issue:44743" ], - "prepared_review_unit_hash": "d794c83d9f492222854316f3c4e8b57ef8f0546de5aef9e1f1b5020417eed637", + "prepared_review_unit_hash": "8d1828c9158d36f878f86610b93a8cc64216a40d85b77f22cacc11c4556c931b", "component_budget": { "node_count": 500, "item_count": 500, "soft_pair_count": 1839, - "serialized_chars": 953913, - "estimated_input_tokens": 238479, - "estimated_eval_tokens": 477214 + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 30, - "serialized_chars": 36569, - "estimated_input_tokens": 9143, - "estimated_eval_tokens": 18542 + "serialized_chars": 36563, + "estimated_input_tokens": 9141, + "estimated_eval_tokens": 18538 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This set is mostly heterogeneous false positives: the soft pairs span different models, subsystems, and failure modes, so none looks like the same underlying bug/change. I would not merge any of the proposed soft edges.", - "confidence": 0.92, - "canonical_issue_reason": "No single canonical issue: these pairs cover distinct bugs across unrelated models/features, so no safe duplicate anchor emerges.", + "summary": "All proposed pairs look like topical false positives rather than true duplicates. They share broad keywords (e.g. Flash Attention, OOM, Qwen, audio) but not the same concrete code path or fix.", + "confidence": 0.96, + "canonical_issue_reason": "No clear canonical issue emerges: the items are semantically mixed, and none of the soft pairs describe the same underlying bug/change strongly enough to anchor a duplicate cluster.", "canonical_pr_reason": null, - "best_issue_reason": "43873 is the most active and referenced issue, but it still doesn't represent the rest of the set well enough to serve as a true canonical.", + "best_issue_reason": "No single issue is a good global representative because the set spans unrelated regressions and feature requests rather than one coherent problem area.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:29127", "right": "issue:35141", "accept": false, - "reason": "LayoutLMv3 error messaging vs token embedding reinit; different code paths and symptoms." + "reason": "Different bugs: LayoutLMv3 error-message clarity vs token embedding reinitialization in post_init." }, { "left": "issue:43421", "right": "issue:45491", "accept": false, - "reason": "Tokenizer post-processor runtime updates vs Gemma3 NaNs in sliding-window attention; unrelated." + "reason": "One is about runtime special-token/post-processor updates; the other is a Gemma3 NaN embedding issue." }, { "left": "issue:43994", "right": "issue:45412", "accept": false, - "reason": "SigLIP2 wrong outputs vs RT-DETR memory leak; different bugs." + "reason": "Different subsystems and symptoms: Siglip2 bad outputs vs RT-DETR memory not being released." }, { "left": "issue:45412", "right": "issue:46369", "accept": false, - "reason": "Both mention memory, but one is GC leak on deletion and the other is model call overflow; not the same issue." + "reason": "Both are memory-related, but one is a GC/memory leak report and the other is model call memory overflow; not the same bug." }, { "left": "issue:43575", "right": "issue:44387", "accept": false, - "reason": "TP loading OOM vs int4 quantization reserved-memory regression; same symptom, different root cause." + "reason": "Both mention OOM, but the causes differ: tensor-parallel loading vs int4 quantization reserved-memory regression." }, { "left": "issue:44442", "right": "issue:45072", "accept": false, - "reason": "Tokenizer load failure vs bfloat16 dtype mismatch in inference; unrelated." + "reason": "Tokenizer loading failure vs dtype mismatch in bfloat16 inference; unrelated failures." }, { "left": "issue:44464", "right": "issue:45584", "accept": false, - "reason": "Compiled-forward chunked generation inconsistency vs Whisper empty-transcription failure after token alignment; different paths." + "reason": "Chunked generation with compiled forward is unrelated to Whisper empty-transcription failure after align_special_tokens." }, { "left": "issue:42915", "right": "issue:43931", "accept": false, - "reason": "Qwen3Moe FP8 config failure vs Qwen3-VL weight-shape mismatch; different models and failures." + "reason": "Qwen3Moe FP8 config failure and Qwen3-VL shape-mismatch loading error are different bugs." }, { "left": "issue:43526", "right": "issue:45834", "accept": false, - "reason": "Label reduction bug in BEiT processor vs long-OCR index error in Kosmos2.5; unrelated." + "reason": "BeitImageProcessorFast label reduction bug is unrelated to Kosmos2.5 long-OCR index errors." }, { "left": "issue:43526", "right": "issue:46693", "accept": false, - "reason": "Label reduction bug vs Flash Attention 2 long-context performance regression; not the same problem." + "reason": "Label reduction bug vs FA2 long-context performance regression; no shared code path." }, { "left": "issue:43824", "right": "issue:44279", "accept": false, - "reason": "Specific import error for Qwen2.5-VL class vs a generic dependency issue; no concrete match." + "reason": "Specific missing import error vs a vague dependency issue; not the same failure." }, { "left": "issue:27957", "right": "issue:29127", "accept": false, - "reason": "Flash Attention 2 support for XLM-Roberta vs LayoutLMv3 error clarity; unrelated." + "reason": "Flash Attention 2 support for XLM-Roberta is unrelated to LayoutLMv3 box-validation messaging." }, { "left": "issue:27957", "right": "issue:33290", "accept": false, - "reason": "Flash Attention 2 support vs Adafactor/DeepSpeed OOM; different subsystems." + "reason": "Flash Attention 2 support/problem vs deepspeed Adafactor OOM; different areas and fixes." }, { "left": "issue:45250", "right": "issue:46640", "accept": false, - "reason": "Both are generic Flash Attention titles, but there is no evidence they describe the same concrete bug/change." + "reason": "Too generic on both sides; they only share the Flash Attention 2 topic, not a concrete shared bug/change." }, { "left": "issue:43994", "right": "issue:46369", "accept": false, - "reason": "SigLIP2 incorrect outputs vs SAM3 LiteText memory overflow; unrelated." + "reason": "Siglip2 incorrect results and SAM3 LiteText memory overflow are unrelated model-specific problems." }, { "left": "issue:33290", "right": "issue:34689", "accept": false, - "reason": "Adafactor OOM in DeepSpeed vs Llama 3.2 Vision load breakage; different failures." + "reason": "OOM in deepspeed Adafactor is not the same as a model-loading regression for Llama 3.2 Vision." }, { "left": "issue:43526", "right": "issue:44186", "accept": false, - "reason": "BEiT reduce_labels bug vs LayoutLMv2Tokenizer crash on padding/truncation; unrelated." + "reason": "Different image/tokenization bugs: label reduction in BeitImageProcessorFast vs LayoutLMv2Tokenizer crashes." }, { "left": "issue:33290", "right": "issue:35141", "accept": false, - "reason": "Optimizer OOM vs embedding reinitialization bug; different code paths." + "reason": "Optimizer OOM and embedding reinitialization are unrelated issues." }, { "left": "issue:43334", "right": "issue:43531", "accept": false, - "reason": "Missing pad_token_id in Qwen3-VL vs sliding_window issue in Qwen3-MoE; different model families." + "reason": "Missing pad_token_id in Qwen3-VL text config is unrelated to a sliding_window bug in Qwen3-MoE." }, { "left": "issue:46121", "right": "issue:46399", "accept": false, - "reason": "Rope-validation TypeError vs missing decorator crash in Qwen3.5 kernels; unrelated." + "reason": "RoPE validation TypeError and rotary-pos-emb registration failure are distinct code paths." }, { "left": "issue:45588", "right": "issue:45600", "accept": false, - "reason": "Flash-attention sink-less AttributeError vs broken auto_mappings references; different causes." + "reason": "Flash-attention sink-less AttributeError is unrelated to stale auto_mappings config references." }, { "left": "issue:43381", "right": "issue:43873", "accept": false, - "reason": "Gradient checkpointing eval-mode limitation vs offloading/quantization bug; unrelated." + "reason": "Gradient checkpointing in eval mode and quantization offloading are different concerns." }, { "left": "issue:44514", "right": "issue:45381", "accept": false, - "reason": "Batched apply_chat_template crash with padding=False vs wrong vision_position_ids for video input; different issues." + "reason": "Batched apply_chat_template padding crash is not the same as wrong vision_position_ids in Qwen2.5-VL video input." }, { - "left": "issue:43646", - "right": "issue:44077", + "left": "issue:43761", + "right": "issue:45859", "accept": false, - "reason": "Transformers 5.0 custom-init breakage vs patchtsmixer post_init policy change; related theme, but not the same fixable bug." + "reason": "CLIP hidden_states regression and Qwen3_5Moe CPU-offload incompatibility are unrelated." }, { - "left": "issue:43761", - "right": "issue:45859", + "left": "issue:43646", + "right": "issue:44077", "accept": false, - "reason": "CLIPVisionModel hidden_states regression vs Qwen3_5Moe CPU-offload incompatibility; unrelated." + "reason": "Generic custom-model initialization break is broader than the specific patchtsmixer post_init policy issue." }, { "left": "issue:44977", "right": "issue:45245", "accept": false, - "reason": "Flash-attention generation regression vs category-count runtime error; no overlap beyond being failures." + "reason": "Qwen3.5 flash-attention generation and a categories-limit runtime error are unrelated." }, { "left": "issue:44743", "right": "issue:44811", "accept": false, - "reason": "Qwen3.5 recurrent-state reset bug vs Whisper skip_special_tokens handling; unrelated." + "reason": "Recurrent-state reset with cache is unrelated to Whisper batch_decode skip_special_tokens handling." }, { "left": "issue:44442", "right": "issue:44898", "accept": false, - "reason": "Tokenizer load failure vs Perceiver resolution/interpolation bug; different components." + "reason": "AutoTokenizer loading failure and Perceiver non-default-resolution failure are different bugs." }, { "left": "issue:43262", "right": "issue:44206", "accept": false, - "reason": "Audio chat-template sample-rate default vs unsupported center arg in feature extractor; different audio code paths." + "reason": "Audio chat-template sampling-rate defaulting and an unsupported center arg in LasrFeatureExtractor are separate issues." }, { "left": "issue:42490", "right": "issue:44743", "accept": false, - "reason": "Trailing-slash local path warning vs Qwen3.5 recurrent-state reset bug; unrelated." + "reason": "Trailing-slash local-model import warning has nothing to do with recurrent state reset in Qwen3." } ] }, @@ -10922,20 +10842,20 @@ "issue:43575|issue:43742", "issue:43606|issue:43701" ], - "prepared_review_unit_hash": "c010b5729a820f4866fddc9b5878fddb6805d6586ad52e93ccd2be5ca8761b35", + "prepared_review_unit_hash": "1a00ed5b77468b8943edc35a61c3a957c8ea10ec0b0a291dfa3de6abccb653a9", "component_budget": { "node_count": 500, "item_count": 500, "soft_pair_count": 1839, - "serialized_chars": 953913, - "estimated_input_tokens": 238479, - "estimated_eval_tokens": 477214 + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 28, - "serialized_chars": 36218, + "serialized_chars": 36217, "estimated_input_tokens": 9055, "estimated_eval_tokens": 18366 }, @@ -10944,210 +10864,180 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "These issues are largely unrelated one-off bugs across different models, loaders, and utilities; I don't see any true duplicate pairs. If forced to anchor the cluster, the most general infrastructure bug is issue 45003.", - "confidence": 0.78, - "canonical_issue_reason": "Issue 45003 is the broadest, most infrastructure-level bug in the set, while the rest are mostly model- or feature-specific regressions with no clear duplicate center.", + "summary": "This cluster is highly heterogeneous: it mixes unrelated bugs in generation, loading, configs, quantization/offloading, tokenizer behavior, docs, and distributed training. None of the soft pairs appears to be the same underlying issue, so I rejected them all.", + "confidence": 0.95, + "canonical_issue_reason": "If a single representative is needed, issue 43701 is the clearest self-contained bug report in the set, but the cluster is too diverse to justify a true canonical duplicate.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 45003 is the best issue anchor because it describes a general code-path safety problem in core loading logic and is more reusable than the model-specific reports.", + "best_issue_reason": "Issue 43701 is the most broadly readable and concrete report, but only as a weak representative; it is not meaningfully connected to the rest of the cluster.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:45003", "right": "issue:45440", "accept": false, - "reason": "Unsafe sys.modules access in modeling_utils is unrelated to DeepseekV3MoE divergence from the remote implementation." + "reason": "Different bug classes and code paths: unsafe sys.modules access vs divergence in DeepseekV3MoE implementation." }, { "left": "issue:43720", "right": "issue:44442", "accept": false, - "reason": "BitNet packed-weight loading during accelerate load is a different failure mode from AutoTokenizer loading FastSpeech2ConformerTokenizer." + "reason": "BitNet accelerate-loading failure and tokenizer-loading failure are unrelated." }, { "left": "issue:42175", "right": "issue:46234", "accept": false, - "reason": "Missing tensorflow in a torch extra and a missing import_utils availability helper are separate packaging/API issues." + "reason": "Package extra dependency coverage vs missing import utility symbol; not the same bug." }, { "left": "issue:27957", "right": "issue:29870", "accept": false, - "reason": "Flash Attention 2 support for XLMRoberta is unrelated to Bart generate() using decoder_attention_mask before assignment." + "reason": "FlashAttention support for XLMRoberta is unrelated to BART generate() decoder_attention_mask initialization." }, { "left": "issue:43493", "right": "issue:43827", "accept": false, - "reason": "SigLIP2 implementation parity and stale pipeline docs after v5 removals are not the same bug." + "reason": "Model implementation discrepancy vs stale docs referencing pipeline(); different issues." }, { "left": "issue:43316", "right": "issue:43525", "accept": false, - "reason": "These are different model config discrepancies affecting different fields and different models." + "reason": "Both are config-related, but they concern different models and different missing/changed fields." }, { "left": "issue:44062", "right": "issue:45907", "accept": false, - "reason": "AddedToken special-argument duplication is unrelated to list_repo_files failing to forward revision." + "reason": "Tokenizer AddedToken argument duplication and list_repo_files revision forwarding are unrelated." }, { "left": "issue:44077", "right": "issue:44479", "accept": false, - "reason": "patchtsmixer post_init validation and the Qwen video-input regression are unrelated." + "reason": "Optional post_init validation for patchtsmixer is not the same as the Qwen video-input regression." }, { "left": "issue:45325", "right": "issue:46697", "accept": false, - "reason": "Qwen2.5-VL rope indexing and Swin2SR over-padding are different preprocessing bugs." + "reason": "Qwen2.5-VL RoPE scaling bug and Swin2SR padding bug affect different preprocessing logic." }, { "left": "issue:43873", "right": "issue:44871", "accept": false, - "reason": "Quantization offloading behavior is unrelated to Gemma-3 eos_token_id config inconsistency." + "reason": "Quantization/offloading behavior and Gemma eos_token_id mismatch are not the same underlying defect." }, { "left": "issue:44855", "right": "issue:45137", "accept": false, - "reason": "A Python 3.13 TorchScript parsing error is not the same as a DeepSpeed ZeRO3 deque underflow." + "reason": "Python 3.13 torch.jit parsing/indentation issue is unrelated to DeepSpeed ZeRO3 deque underflow." }, { "left": "issue:43643", "right": "issue:44913", "accept": false, - "reason": "Missing fields from trust_remote_code AutoConfig and GPTNeoX rotary_pct reload behavior affect different config paths." + "reason": "Remote-code config field loss vs GPTNeoX rotary_pct persistence are different config-loading problems." }, { "left": "issue:43749", "right": "issue:43931", "accept": false, - "reason": "FSDP CPU RAM efficient loading failure is unrelated to Qwen3-VL weight-shape mismatch on load." + "reason": "FSDP CPU RAM-efficient loading breakage and Qwen3-VL shape mismatch are distinct loading failures." + }, + { + "left": "issue:43316", + "right": "issue:45759", + "accept": false, + "reason": "Gemma3TextConfig API discrepancy and composite Qwen multimodal config unwrapping are different config/model-selection bugs." + }, + { + "left": "issue:43761", + "right": "issue:44079", + "accept": false, + "reason": "CLIPVisionModel hidden_states regression and ModelOutput key assignment when None are separate runtime behaviors." + }, + { + "left": "issue:44190", + "right": "issue:44496", + "accept": false, + "reason": "Local dataset loading in a training script is unrelated to AutoConfig failing on an unrecognized model." }, { "left": "issue:44442", "right": "issue:45803", "accept": false, - "reason": "Tokenizer load failure and a bare-except in FuyuBatchFeature tensor conversion are different code paths." + "reason": "Tokenizer loading failure for FastSpeech2Conformer and bare except swallowing interrupts/errors in FuyuBatchFeature are unrelated." }, { "left": "issue:43232", "right": "issue:43388", "accept": false, - "reason": "Generation kwargs synchronization after sync_gpus and metric gathering truncation are unrelated trainer/runtime bugs." + "reason": "Generation kwargs update after sync_gpus and gather_for_metrics tuple-label dropping affect different subsystems." }, { "left": "issue:42907", "right": "issue:45698", "accept": false, - "reason": "Saving dequantized models and loading the wrong custom module after save_pretrained are different save/load problems." + "reason": "Saving dequantized Ministrals/Devstrals and loading the wrong custom module after save_pretrained are distinct save/load issues." }, { "left": "issue:43701", "right": "issue:46506", "accept": false, - "reason": "Checkpoint-resume key mismatch is unrelated to CPU/CUDA divergence in PPDocLayoutV3." + "reason": "Checkpoint-key mismatch and CPU/CUDA output divergence are not the same bug." }, { "left": "issue:42617", "right": "issue:43575", "accept": false, - "reason": "A failure running 3d_parallel.py is not the same as an OOM while loading Qwen2-57B with tensor parallelism." + "reason": "A script launch problem and a tensor-parallel OOM on a specific model are different failures." }, { "left": "issue:27957", "right": "issue:32090", "accept": false, - "reason": "XLMRoberta Flash Attention support and a Trainer GPU broadcast NoneType error are unrelated." + "reason": "FlashAttention2 support for XLMRoberta is unrelated to Trainer GPU broadcast rejecting None." }, { "left": "issue:43550", "right": "issue:43701", "accept": false, - "reason": "torch.compile SDPA divergence in Bamba-9B-v2 is unrelated to resume_from_checkpoint key mismatch." + "reason": "torch.compile SDPA failure in Bamba-9B-v2 is unrelated to resume_from_checkpoint key mismatch." }, { "left": "issue:43575", "right": "issue:43742", "accept": false, - "reason": "Qwen2-57B-A14B tp OOM and MobileLLM-125M key errors are distinct loading failures." + "reason": "Qwen2-57B tensor-parallel OOM and MobileLLM key error do not share a concrete code-path." }, { "left": "issue:43606", "right": "issue:43701", "accept": false, - "reason": "CPU offload device mismatch and resume_from_checkpoint key mismatch are separate issues." + "reason": "CPU offload device mismatch and checkpoint key mismatch are different bugs in different flows." + }, + { + "left": "issue:43761", + "right": "issue:44188", + "accept": false, + "reason": "Hidden_states regression in CLIPVisionModel and attention-kernel divergence under allow_is_bidirectional_skip are unrelated." }, { "left": "issue:44987", "right": "issue:46178", "accept": false, - "reason": "Loading physical-intelligence/fast on transformers>=5.1.0 and tensorflow_text backend registration are unrelated backend/config problems." + "reason": "Version-specific model-loading failure and missing BACKENDS_MAPPING entry for tensorflow_text are not the same issue." }, { "left": "issue:43931", "right": "issue:45216", "accept": false, - "reason": "Qwen3-VL weight-shape mismatch and Qwen3.5 save_pretrained regression do not describe the same bug." - }, - { - "left": "issue:44442", - "right": "issue:45803", - "accept": false, - "reason": "AutoTokenizer failing on one tokenizer class and Fuyu tensor conversion swallowing errors are different failures." - }, - { - "left": "issue:43232", - "right": "issue:43388", - "accept": false, - "reason": "These affect different parts of the training/evaluation stack and do not share the same code path." - }, - { - "left": "issue:42907", - "right": "issue:45698", - "accept": false, - "reason": "One is about dequantized model saving, the other about custom module resolution during reload." - }, - { - "left": "issue:43701", - "right": "issue:46506", - "accept": false, - "reason": "No overlap in mechanism: checkpoint key lookup versus inference divergence across devices." - }, - { - "left": "issue:42617", - "right": "issue:43575", - "accept": false, - "reason": "Parallel script launch failure and tensor-parallel OOM are not mergeable into one fix." - }, - { - "left": "issue:27957", - "right": "issue:32090", - "accept": false, - "reason": "These are unrelated model/runtime issues with no shared concrete failure." - }, - { - "left": "issue:43550", - "right": "issue:43701", - "accept": false, - "reason": "Different symptoms, different subsystems, and no common faulty code path." - }, - { - "left": "issue:43575", - "right": "issue:43742", - "accept": false, - "reason": "Loading a large Qwen model with tp and a key error on MobileLLM are not the same defect." - }, - { - "left": "issue:43606", - "right": "issue:43701", - "accept": false, - "reason": "CPU offload mismatch is unrelated to checkpoint key mismatch." + "reason": "Shape mismatch when loading Qwen3-VL and incorrect save_pretrained output for Qwen3.5 are separate load/save bugs." } ] }, @@ -11242,21 +11132,21 @@ "issue:45803|issue:46170", "issue:44610|issue:46369", "issue:43650|issue:46202", - "issue:43504|issue:45698", "issue:43761|issue:44479", + "issue:43504|issue:45698", "issue:45542|issue:46234", - "issue:43540|issue:44514", "issue:43352|issue:46640", + "issue:43540|issue:44514", "issue:43572|issue:46178" ], - "prepared_review_unit_hash": "d7817e600cea1ef689b7d1cc54ed7362ee36517560407938f79a283c6c822511", + "prepared_review_unit_hash": "d088350cedc2175181464014623910b4e31cfc927d685804f0da835cd54e8bcc", "component_budget": { "node_count": 500, "item_count": 500, "soft_pair_count": 1839, - "serialized_chars": 953913, - "estimated_input_tokens": 238479, - "estimated_eval_tokens": 477214 + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 }, "budget": { "node_count": 48, @@ -11271,174 +11161,174 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is a loose semantic grouping of mostly unrelated Transformers bug reports; there is no true duplicate pair here, so all soft edges should be rejected.", - "confidence": 0.94, - "canonical_issue_reason": null, + "summary": "The cluster is a loose grab-bag of unrelated Transformers issues. None of the soft pairs look like the same concrete bug or change, so all are rejected.", + "confidence": 0.87, + "canonical_issue_reason": "issue:45698 is the strongest standalone bug report here: it is open, has the most discussion, and describes a concrete reproducible regression in from_pretrained/save_pretrained behavior.", "canonical_pr_reason": null, - "best_issue_reason": "issue:45698 is the strongest standalone triage anchor: it is open, highly discussed, and describes a concrete reproducible from_pretrained/save_pretrained loading regression, but it still is not a canonical duplicate for the rest of the cluster.", + "best_issue_reason": "issue:45698 is the best overall issue candidate because it is specific, user-facing, and already has active investigation; the other issues are narrower or clearly unrelated to one another.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:25251", "right": "issue:28282", "accept": false, - "reason": "Different bugs: pipeline top_k output shape vs missing PyTorch backend during AutoModel import." + "reason": "Different bugs: pipeline output shape regression vs missing PyTorch import for AutoModel." }, { "left": "issue:46234", "right": "issue:46291", "accept": false, - "reason": "Unrelated import issues: a missing helper symbol vs kernels package version/revision requirements." + "reason": "One is a missing import_utils symbol; the other is a kernels dependency/version check on importing pipelines." }, { "left": "issue:43295", "right": "issue:43825", "accept": false, - "reason": "Both are v5 regressions, but they affect different code paths: custom processor/tokenizer handling vs pipeline task messaging." + "reason": "Both are v5 regressions, but one is about processor/tokenizer access for custom models and the other is an incorrect pipeline task message." }, { "left": "issue:16998", "right": "issue:22355", "accept": false, - "reason": "Model_max_length behavior for DeBERTa-V3 is unrelated to an onnx import error." + "reason": "Unrelated: model_max_length handling for DeBERTa-V3 vs missing transformers.onnx module." }, { "left": "issue:43295", "right": "issue:44871", "accept": false, - "reason": "Different functionality: processor/tokenizer image handling vs Gemma eos_token_id configuration mismatch." + "reason": "Different model families and failures: processor/tokenizer regression vs Gemma eos_token_id mismatch." }, { "left": "issue:44871", "right": "issue:45584", "accept": false, - "reason": "Different model families and failure modes: Gemma EOS config vs Whisper empty transcription after special-token alignment." + "reason": "Gemma eos token configuration bug vs Whisper empty-transcription failure after token alignment." }, { "left": "issue:44464", "right": "issue:46421", "accept": false, - "reason": "Both mention chunked/compiled execution, but one is output inconsistency and the other is cache lazy-init causing recompilation." + "reason": "Chunked generation inconsistency with compiled forward is not the same as spurious recompilation from cache lazy init." }, { "left": "issue:43316", "right": "issue:43716", "accept": false, - "reason": "API discrepancy in Gemma3TextConfig is unrelated to Mistral-3 image preprocessor dtype mismatch." + "reason": "API discrepancy in Gemma3TextConfig is unrelated to Mistral-3 image preprocessor/model dtype mismatch." }, { "left": "issue:44077", "right": "issue:44534", "accept": false, - "reason": "PatchTSMixer post_init policy and junk-filled non-persistent buffers are separate issues." + "reason": "PatchTSMixer post_init restriction is a model-class API issue, not the non-persistent buffer junk-fill regression." }, { "left": "issue:42175", "right": "issue:43316", "accept": false, - "reason": "Backend packaging/install problem vs a Gemma3 config API discrepancy." + "reason": "Backend packaging/install issue vs config API discrepancy; no shared underlying defect." }, { "left": "issue:44871", "right": "issue:46050", "accept": false, - "reason": "Gemma EOS token config mismatch is unrelated to sinq quantized save/load failure." + "reason": "Gemma eos_token_id inconsistency is distinct from quantized model save/load breakage." }, { "left": "issue:43646", "right": "issue:44913", "accept": false, - "reason": "Custom model init break on v5 is not the same as GPTNeoX rotary_pct not persisting on reload." + "reason": "General custom model initialization breakage is not the same as GPTNeoX rotary_pct not persisting on reload." }, { "left": "issue:45803", "right": "issue:46365", "accept": false, - "reason": "Bare except in tensor conversion is unrelated to Trainer import failure caused by datasets." + "reason": "Bare except swallowing errors in Fuyu conversion is unrelated to Trainer import being blocked by datasets." }, { "left": "issue:44162", "right": "issue:44792", "accept": false, - "reason": "ESM2 model breakage and janus image-generation test failure are different models and code paths." + "reason": "ESM2 breakage and a janus image-generation test failure are different models and failure modes." }, { "left": "issue:43452", "right": "issue:44496", "accept": false, - "reason": "Both are model loading problems, but one is gguf_file handling and the other is missing model_type in config.json." + "reason": "Both concern model loading, but one is gguf_file handling for from_pretrained and the other is missing model_type in config.json." }, { "left": "issue:43404", "right": "issue:43441", "accept": false, - "reason": "Mistral3 lm_head tying bug and Ministral-3 FlashAttention support failure are distinct regressions." + "reason": "lm_head tying in Mistral3ForConditionalGeneration is a different bug from a FlashAttention support failure in Ministral-3." }, { "left": "issue:44242", "right": "issue:46894", "accept": false, - "reason": "Different training-loss bugs: router load-balancing loss omission vs double-shift in CohereASR." + "reason": "Both are loss-related, but they affect different models and mechanisms; the training-loss double-shift bug is not the same as missing load-balancing loss." }, { "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." + "reason": "Cache path collision for local trust_remote_code models is unrelated to max_cache_length only applying to quantized cache." }, { "left": "issue:45803", "right": "issue:46170", "accept": false, - "reason": "Fuyu tensor conversion exception handling is unrelated to Windows ROCm torch.distributed import problems." + "reason": "Different environment failures: Fuyu exception handling vs Windows/ROCm torch.distributed dependency." }, { "left": "issue:44610", "right": "issue:46369", "accept": false, - "reason": "Processor input-size mismatch for OmDet-Turbo is unrelated to SAM3-lite text memory overflow." + "reason": "Processor input-size mismatch for OmDet-Turbo is unrelated to sam3-litetext memory overflow." }, { "left": "issue:43650", "right": "issue:46202", "accept": false, - "reason": "Placeholder text issue and Gemma pipeline roundtrip failure are not the same bug." + "reason": "A malformed issue title about data is unrelated to Gemma pipeline roundtrip failures." }, { - "left": "issue:43504", - "right": "issue:45698", + "left": "issue:43761", + "right": "issue:44479", "accept": false, - "reason": "Both involve loading/pretrained behavior, but one is a legacy field in BEiT semantic segmentation and the other is wrong custom module resolution." + "reason": "CLIPVisionModel hidden_states regression is not the same as the Qwen video-input regression." }, { - "left": "issue:43761", - "right": "issue:44479", + "left": "issue:43504", + "right": "issue:45698", "accept": false, - "reason": "CLIPVision hidden_states regression and Qwen video-input regression are unrelated." + "reason": "Legacy field handling in Beit loading is a different concrete bug from wrong custom module loading after save_pretrained." }, { "left": "issue:45542", "right": "issue:46234", "accept": false, - "reason": "TensorBoard/TensorFlow backend packaging issue vs missing falcon mamba availability helper." + "reason": "TensorBoard/TensorFlow backend packaging issue is unrelated to a missing import_utils helper." }, { - "left": "issue:43540", - "right": "issue:44514", + "left": "issue:43352", + "right": "issue:46640", "accept": false, - "reason": "Video processing failure in Qwen3OmniMoe and batched chat-template padding crash are different code paths." + "reason": "The FlashAttention-2 support complaint is too vague to equate with the NemotronHForCausalLM support error." }, { - "left": "issue:43352", - "right": "issue:46640", + "left": "issue:43540", + "right": "issue:44514", "accept": false, - "reason": "A specific model not supporting FlashAttention 2 yet is not enough to match a generic Flash Attention 2 issue." + "reason": "Qwen3OmniMoe video-processing ValueError and Qwen2_5_VLProcessor batched apply_chat_template crash are different paths." }, { "left": "issue:43572", "right": "issue:46178", "accept": false, - "reason": "StableLmConfig pad_token_idx regression is unrelated to a tensorflow_text backend mapping error." + "reason": "StableLmConfig pad_token_idx regression is unrelated to the tensorflow_text backend mapping error." } ] }, @@ -11514,8 +11404,8 @@ "soft_pairs": [ "issue:44792|issue:45600", "issue:43065|issue:44617", - "issue:44792|issue:44977", "issue:43388|issue:43688", + "issue:44792|issue:44977", "issue:45127|issue:45200", "issue:43673|issue:44743", "issue:41950|issue:43475", @@ -11527,8 +11417,8 @@ "issue:46082|issue:46697", "issue:45632|issue:46617", "issue:44869|issue:46832", - "issue:45245|issue:46858", "issue:44855|issue:45588", + "issue:45245|issue:46858", "issue:43827|issue:43994", "issue:43673|issue:46421", "issue:43931|issue:44060", @@ -11540,196 +11430,190 @@ "issue:25251|issue:27957", "issue:43976|issue:44855" ], - "prepared_review_unit_hash": "29ee9585eace96117b408915ad3620af13f9081390a694d9e6ce1adf3f91bc82", + "prepared_review_unit_hash": "aed5e4635fb2ee774cfb77df6f6442aab4951b77dd104f895847af99376ab2c9", "component_budget": { "node_count": 500, "item_count": 500, "soft_pair_count": 1839, - "serialized_chars": 953913, - "estimated_input_tokens": 238479, - "estimated_eval_tokens": 477214 + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 27, - "serialized_chars": 35947, - "estimated_input_tokens": 8987, - "estimated_eval_tokens": 18230 + "serialized_chars": 35949, + "estimated_input_tokens": 8988, + "estimated_eval_tokens": 18232 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster is highly mixed and the soft pairs are mostly superficial topic matches rather than the same underlying bug. No soft edge is strong enough to merge; the best anchor issue is the broad Transformers v5 non-persistent-buffer regression (43644).", - "confidence": 0.34, - "canonical_issue_reason": "issue:43644 is the broadest and most central bug report in the set, and it has a near-title twin (issue:44534) that looks like the same regression; it is the best candidate to anchor the cluster.", + "summary": "The cluster is mostly heterogeneous; several items are only loosely related by subsystem, but none of the soft-edge pairs look like the same concrete bug or change. The best canonical issue among the set is the non-persistent-buffer regression report.", + "confidence": 0.45, + "canonical_issue_reason": "Issue 43644 is the clearest and earliest report of a concrete regression ('non-persistent buffers filled with junk'), and it is the most canonical-looking candidate in the set.", "canonical_pr_reason": null, - "best_issue_reason": "issue:43644 most clearly represents a reusable root cause affecting core Transformers behavior, unlike the many narrow model-specific or doc-only reports around it.", + "best_issue_reason": "Issue 43644 is the most specific, actionable, and representative report among the issues here; it describes a clear runtime regression rather than a broad symptom or a secondary follow-up.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:44792", "right": "issue:45600", "accept": false, - "reason": "Different failures in unrelated areas: janus image-generation test vs Sam3LiteText auto-mapping/CI breakage." + "reason": "Different bug surfaces: a janus test failure versus a removed Sam3LiteText config reference." }, { "left": "issue:43065", "right": "issue:44617", "accept": false, - "reason": "Both mention SAM 3, but one is about a dummy Conv2d module and the other is a CUDA OOM; not the same bug." + "reason": "Unrelated symptoms: a dummy Conv2d implementation detail versus a CUDA OOM report." }, { - "left": "issue:44792", - "right": "issue:44977", + "left": "issue:43388", + "right": "issue:43688", "accept": false, - "reason": "Unrelated models and code paths: janus image generation test failure vs Qwen3.5 flash-attention generation issue." + "reason": "Different training/eval bugs: metrics label truncation versus auxiliary loss normalization." }, { - "left": "issue:43388", - "right": "issue:43688", + "left": "issue:44792", + "right": "issue:44977", "accept": false, - "reason": "Different training/eval subsystems: metric gathering label trimming vs auxiliary-loss normalization." + "reason": "Same general area of generation, but different models and code paths; not the same bug." }, { "left": "issue:45127", "right": "issue:45200", "accept": false, - "reason": "Different bugs: LoRA merge / tied embeddings collapse vs Gemma 4 mm_token_type_ids defaulting." + "reason": "One is about LoRA merge/tied embeddings, the other about mm_token_type_ids defaults; separate issues." }, { "left": "issue:43673", "right": "issue:44743", "accept": false, - "reason": "Both cache-related, but one is missing cache in chunked prefill and the other is recurrent-state reset with seq_len>1." + "reason": "Both involve cache-like behavior, but one is missing cache in chunked prefill and the other resets recurrent states." }, { "left": "issue:41950", "right": "issue:43475", "accept": false, - "reason": "Video-classification processor selection problem vs SAM 3 vision encoder output attribute error; no shared code-path." + "reason": "Different SAM3 failures: pipeline processor lookup versus a missing vision encoder attribute." }, { "left": "issue:44190", "right": "issue:44964", "accept": false, - "reason": "Different loading problems in different scripts/models; one is local dataset loading, the other is Phi-4 multimodal model loading." + "reason": "Different loading paths: local dataset script failure versus a specific multimodal model load failure." }, { "left": "issue:43525", "right": "issue:43531", "accept": false, - "reason": "Different configuration issues: missing pad_token_id on Llama4Config vs sliding_window behavior in Qwen3-MoE." + "reason": "Unrelated model config/attention bugs: missing pad_token_id versus sliding_window behavior." }, { "left": "issue:44534", "right": "issue:44913", "accept": false, - "reason": "Non-persistent-buffer initialization regression is unrelated to GPTNeoX rotary_pct persistence on reload." + "reason": "Both are configuration/state issues, but non-persistent buffers corruption is not the same as rotary_pct reload behavior." }, { "left": "issue:35141", "right": "issue:42490", "accept": false, - "reason": "Embedding reinitialization bug is unrelated to local-path import warning behavior." + "reason": "Different problems: embedding reinitialization versus a path-with-slash import warning." }, { "left": "issue:44734", "right": "issue:45406", "accept": false, - "reason": "Serve/KV-cache tensor indexing crash is unrelated to Gemma4Processor missing _tokenizer." + "reason": "Separate serve/model failures: KV-cache continuation indexing versus a missing processor attribute." }, { "left": "issue:46082", "right": "issue:46697", "accept": false, - "reason": "Llama head_dim validation and Swin2SR over-padding are unrelated model/config vs image-processing bugs." + "reason": "Completely different domains: head_dim validation versus image padding behavior." }, { "left": "issue:45632", "right": "issue:46617", "accept": false, - "reason": "Both touch trust_remote_code caching, but one is a leaf-name collision and the other is a FileNotFoundError from cache loading; not clearly the same defect." + "reason": "Both touch trust_remote_code caching, but one is a leaf-directory collision and the other is a FileNotFoundError; not the same concrete bug." }, { "left": "issue:44869", "right": "issue:46832", "accept": false, - "reason": "Whisper timestamp decoding crash and RT-DETR feature-level mismatch are unrelated." + "reason": "Unrelated model paths: Whisper timestamp decoding versus RT-DETR feature-level mismatch." }, { - "left": "issue:45245", - "right": "issue:46858", + "left": "issue:44855", + "right": "issue:45588", "accept": false, - "reason": "Category-count limit error is unrelated to the static-cache slice arithmetic/device-tensor issue." + "reason": "Different failure modes and code paths: Python 3.13 JIT parsing versus flash_attention handling of s_aux=None." }, { - "left": "issue:44855", - "right": "issue:45588", + "left": "issue:45245", + "right": "issue:46858", "accept": false, - "reason": "Python 3.13 torch.jit parse failure is unrelated to flash_attention s_aux=None AttributeError." + "reason": "A category-count limit error is unrelated to a device-tensor slice arithmetic bug in cache prefill." }, { "left": "issue:43827", "right": "issue:43994", "accept": false, - "reason": "Docs still referencing pipeline() is not the same as SigLIP2 producing bad outputs with AutoModel/pipeline." + "reason": "One is a docs-removal mismatch, the other is a model/pipeline inference bug; not the same issue." }, { "left": "issue:43673", "right": "issue:46421", "accept": false, - "reason": "Both mention cache and chunked prefill, but one is missing GenerationMixin cache and the other is spurious recompilation from lazy init." + "reason": "Related chunked-prefill/cache area, but missing cache and lazy-init recompilation are distinct concrete regressions." }, { "left": "issue:43931", "right": "issue:44060", "accept": false, - "reason": "Different models and failures: Qwen3-VL weight-shape mismatch vs incorrect tied-weights warning in Qwen3-Next." + "reason": "Different Qwen model issues: weight-shape mismatch versus tied-weights warning contamination." }, { "left": "issue:44242", "right": "issue:46897", "accept": false, - "reason": "Load-balancing loss omission is unrelated to Florence2 training-loss double-shift behavior." + "reason": "Both concern loss computation, but one is missing load-balancing loss and the other is a double-shift training bug." }, { "left": "issue:29870", "right": "issue:35141", "accept": false, - "reason": "Bart generate() UnboundLocalError and output-embedding reinit are different bugs in different code paths." + "reason": "Unrelated bugs: BART generate-time decoder mask initialization versus embedding reinitialization after resize." }, { "left": "issue:44617", "right": "issue:45584", "accept": false, - "reason": "CUDA OOM in SAM 3 video is unrelated to Whisper empty-transcription generation failure." + "reason": "Different models and failures: SAM3 video OOM versus Whisper empty-transcription decoding." }, { "left": "issue:43644", "right": "issue:44913", "accept": false, - "reason": "The non-persistent-buffer regression is unrelated to GPTNeoX rotary_pct persistence on reload." + "reason": "Non-persistent buffer corruption and rotary_pct reload are separate persistence/config bugs." }, { "left": "issue:44933", "right": "issue:45362", "accept": false, - "reason": "Removed image_utils import and Qwen3.5 chat crash are different issues." - }, - { - "left": "issue:25251", - "right": "issue:27957", - "accept": false, - "reason": "Pipeline output nesting change and XLMRoberta Flash Attention 2 support are unrelated." + "reason": "Different domains entirely: image_utils import cleanup versus a Qwen3.5 chat crash." }, { "left": "issue:43976", "right": "issue:44855", "accept": false, - "reason": "Version/Python compatibility issue is unrelated to the Python 3.13 torch.jit indentation parse failure." + "reason": "Separate version-compatibility issues: Python 3.9/3.10 support versus a Python 3.13 IndentationError." } ] }, @@ -11831,212 +11715,209 @@ "issue:43854|issue:44964", "issue:45632|issue:46234" ], - "prepared_review_unit_hash": "36bc3904a9923f272d6e71cb1ece450c8706c527475a2323d2a4eadd7b092340", + "prepared_review_unit_hash": "929d7d544870e43fcf896e34039c982d97152a521ef5a36aa80d6c8763764f66", "component_budget": { "node_count": 500, "item_count": 500, "soft_pair_count": 1839, - "serialized_chars": 953913, - "estimated_input_tokens": 238479, - "estimated_eval_tokens": 477214 + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 28, - "serialized_chars": 35691, - "estimated_input_tokens": 8923, - "estimated_eval_tokens": 18102 + "serialized_chars": 35695, + "estimated_input_tokens": 8924, + "estimated_eval_tokens": 18104 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Most candidate pairs are false positives: they touch different models, processors, or unrelated error paths. Only the two DeepSeek tokenizer regressions plausibly describe the same v5 tokenizer bug and should be clustered together.", - "confidence": 0.86, - "canonical_issue_reason": "issue:46489 is the most specific report: it names DeepSeek-Coder v1, identifies the v5+ tokenizer regression, and explicitly points to the remaining gap in the prior fix.", + "summary": "This cluster is a grab bag of mostly unrelated single-issue bugs across generation, tokenization, multimodal processing, configs, and CI. None of the soft-edge pairs look like true duplicates.", + "confidence": 0.94, + "canonical_issue_reason": "No clear duplicate family emerges from the cluster, so there is no strong canonical issue. If a representative must be picked, issue 45584 is the cleanest self-contained bug report with a specific failure mode and reproducible symptoms.", "canonical_pr_reason": null, - "best_issue_reason": "issue:46489 is the strongest canonical issue candidate because it is detailed, reproducible, and directly anchors the shared tokenizer regression.", + "best_issue_reason": "Issue 45584 is the best representative issue in this set: it has a focused title, a concrete user-facing crash, and enough detail to stand in as a canonical example of the cluster\u2019s overall bug style.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:45563", "right": "issue:45584", "accept": false, - "reason": "Both are generation bugs, but one is about a stale num_return_sequences warning and the other is Whisper failing on empty transcription; different triggers and code paths." + "reason": "Different generation bugs: a stale warning for num_return_sequences vs an empty-transcription crash after align_special_tokens." }, { "left": "issue:43550", "right": "issue:44190", "accept": false, - "reason": "Unrelated subsystems: torch.compile/SDPA on Bamba vs local dataset loading in the image-classification example." + "reason": "Unrelated failures: Bamba torch.compile/SDPA assert vs a dataset-loading script error." }, { "left": "issue:45381", "right": "issue:46369", "accept": false, - "reason": "Different models and failures: Qwen2.5-VL video position ids vs Sam3-LiteText memory overflow." + "reason": "Both touch multimodal models, but one is qwen video vision_position_ids and the other is Sam3 LiteText memory overflow." }, { "left": "issue:44077", "right": "issue:44877", "accept": false, - "reason": "Both involve config handling, but they are different models and distinct validation bugs." + "reason": "Different config problems: optional post_init handling vs strict config loading for granite_speech." }, { "left": "issue:44479", "right": "issue:44734", "accept": false, - "reason": "Video-input regression in Qwen VL models is unrelated to the serve KV-cache continuation tensor-indexing crash." + "reason": "Different code paths: video-input regression in Qwen VL variants vs serve KV-cache continuation indexing." }, { "left": "issue:41720", "right": "issue:45245", "accept": false, - "reason": "Completely different problems: auto device mapping cudaErrorAssert vs category-count limit in another component." + "reason": "CUDA auto device-mapping assert and category-count overflow are unrelated bugs." }, { "left": "issue:43335", "right": "issue:44190", "accept": false, - "reason": "SwitchTransformers sparse-layer construction and image-classification dataset loading are unrelated." + "reason": "SwitchTransformers sparse-layer construction is unrelated to the local dataset-loading script issue." }, { "left": "issue:43065", "right": "issue:44610", "accept": false, - "reason": "Sam3 pixel decoder dummy Conv2d and OmDet-Turbo processor size mismatch are different model-specific issues." + "reason": "Sam3PixelDecoder dummy Conv2d and OmDet-Turbo input-size mismatch are different model/processor bugs." }, { "left": "issue:44464", "right": "issue:45563", "accept": false, - "reason": "Both mention generation, but chunked-generation inconsistency and a stale warning are not the same bug." + "reason": "Different generation issues: inconsistent compiled chunked generation vs a stale warning for num_return_sequences." }, { "left": "issue:39692", "right": "issue:44190", "accept": false, - "reason": "SigLIP2 example errors versus local dataset loading; no shared code path." + "reason": "Documentation example errors are unrelated to the image-classification script failing on local datasets." }, { "left": "issue:45127", "right": "issue:45357", "accept": false, - "reason": "LoRA/extended-vocab collapse is unrelated to Qwen3.5 save_pretrained visual-encoder key handling." + "reason": "LoRA/extended-vocab collapse and incorrect visual encoder keys on save_pretrained are separate save/load problems." }, { "left": "issue:45381", "right": "issue:46655", "accept": false, - "reason": "Both are video-related, but they affect different models and different preprocessing mistakes." + "reason": "Both are video-related, but they affect different models and different failure modes." }, { "left": "issue:44112", "right": "issue:45600", "accept": false, - "reason": "Both are CI-related, but one is a stale test and the other is a broken auto_mappings import reference." + "reason": "Both mention CI, but one is a stale device override test and the other is a broken auto_mappings reference." }, { "left": "issue:44779", "right": "issue:46489", - "accept": true, - "reason": "Both report the same DeepSeek tokenizer regression on transformers v5+, and 46489 explicitly describes the remaining gap in the earlier fix." + "accept": false, + "reason": "Related DeepSeek family, but not the same bug: generic tokenizer regression vs a DeepSeek-Coder v1 gap in a prior fix." }, { "left": "issue:43295", "right": "issue:45276", "accept": false, - "reason": "Processor/tokenizer API regression and Gemma4 resize_token_embeddings propagation are different issues." + "reason": "Processor.tokenizer regression and resize_token_embeddings propagation are different mechanisms." }, { "left": "issue:44617", "right": "issue:46369", "accept": false, - "reason": "Both are SAM3-family memory complaints, but one is video OOM and the other is LiteText memory overflow; not the same concrete bug." + "reason": "Sam3Video OOM and Sam3-LiteText memory overflow are different resource issues in different variants." }, { "left": "issue:43526", "right": "issue:46697", "accept": false, - "reason": "Different image processors and different padding/label logic; no evidence of a shared root cause." + "reason": "ImageProcessorFast label reduction and Swin2SR over-padding are unrelated preprocessing bugs." }, { "left": "issue:45290", "right": "issue:46752", "accept": false, - "reason": "Both involve apply_chat_template crashing, but one is tool-call messages with no content and the other is an empty-list edge case." + "reason": "apply_chat_template crashes in different scenarios: tool-call assistant messages vs an empty list." }, { "left": "issue:44462", "right": "issue:45907", "accept": false, - "reason": "AutoTokenizer ignoring tokenizer.json and list_repo_files not forwarding revision are separate Hub-loading issues." + "reason": "AutoTokenizer ignoring tokenizer.json is unrelated to list_repo_files missing the revision argument." }, { "left": "issue:45276", "right": "issue:45584", "accept": false, - "reason": "Gemma4 embedding resizing and Whisper empty-transcription generation failure are unrelated." + "reason": "Embedding resize propagation and Whisper empty transcription are unrelated bugs." }, { "left": "issue:44737", "right": "issue:45803", "accept": false, - "reason": "XLNet device placement in relative positional encoding is unrelated to Fuyu's bare except handling." + "reason": "XLNet CPU tensor allocation and Fuyu bare except swallowing errors are different issues." }, { "left": "issue:43540", "right": "issue:44734", "accept": false, - "reason": "Qwen3OmniMoe video-input ValueError and serve KV-cache indexing crash are different paths." + "reason": "Qwen3OmniMoe video-input ValueError and serve KV-cache continuation indexing are unrelated." }, { "left": "issue:44938", "right": "issue:45542", "accept": false, - "reason": "Python 3.14 import failure and missing TensorFlow backend detection are different environment issues." + "reason": "Python 3.14 import failure and the tensorboard/TensorFlow backend issue do not share a code path." }, { "left": "issue:25251", "right": "issue:29870", "accept": false, - "reason": "Pipeline top_k nesting change and Bart decoder_attention_mask initialization bug are unrelated." + "reason": "Pipeline top_k nesting and Bart decoder_attention_mask initialization are unrelated bugs." }, { "left": "issue:44521", "right": "issue:45381", "accept": false, - "reason": "assistant_masks being all zero and wrong vision_position_ids are different multimodal bugs." + "reason": "Both are multimodal issues, but assistant_masks being all zeros is not the same as incorrect vision_position_ids." }, { "left": "issue:43673", "right": "issue:46858", "accept": false, - "reason": "Both concern cache/prefill, but one is a missing cache in chunked_prefill and the other is a device-tensor seq-length arithmetic bug." + "reason": "Chunked_prefill cache missing and static-cache seq_length tensor arithmetic are different cache bugs." }, { "left": "issue:43854", "right": "issue:44964", "accept": false, - "reason": "Two different model-loading failures for different repos; no shared underlying change." + "reason": "Two separate model-loading failures for different models and different underlying causes." }, { "left": "issue:45632", "right": "issue:46234", "accept": false, - "reason": "Cache-path collision for trust_remote_code is unrelated to the missing import_utils symbol." + "reason": "Cache-path collision for local models is unrelated to a missing import_utils helper symbol." } ] }, - "evaluator_result": { - "accept": true, - "feedback": "The clustering is well-grounded: the only clearly mergeable pair is the DeepSeek tokenizer regression pair (44779/46489), and the remaining soft pairs are conservatively rejected as different models, code paths, or edge cases. No soft-edge verdict looks overstated from the packet metadata." - }, - "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 }, @@ -12135,220 +12016,220 @@ "issue:43673|issue:44336", "issue:44792|issue:46424" ], - "prepared_review_unit_hash": "7f37892e0099b53f65aa8e6f07303310e0053a0b3c1e5d7bd9a66bb3eb974a16", + "prepared_review_unit_hash": "f693f288b874fc3bd32465a300d4f84d656ee87b4d206492211ca77b29fb743d", "component_budget": { "node_count": 500, "item_count": 500, "soft_pair_count": 1839, - "serialized_chars": 953913, - "estimated_input_tokens": 238479, - "estimated_eval_tokens": 477214 + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 31, - "serialized_chars": 36933, - "estimated_input_tokens": 9234, - "estimated_eval_tokens": 18724 + "serialized_chars": 36931, + "estimated_input_tokens": 9233, + "estimated_eval_tokens": 18722 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This is a heterogeneous bucket of mostly unrelated Transformers issues; every soft pair is a false positive rather than a true duplicate.", - "confidence": 0.98, - "canonical_issue_reason": "No single canonical issue; the cluster spans unrelated bugs, regressions, import errors, config-validation problems, and pipeline/model-specific failures, so any one item would be a poor duplicate target.", + "summary": "This cluster is a loose grab bag of Transformers issues with only superficial topical overlap (imports, configs, multimodal models, caches, etc.). None of the proposed soft pairs look like the same underlying bug or change, so all soft edges should be rejected.", + "confidence": 0.84, + "canonical_issue_reason": "issue:44964 is the most concrete, reproducible, and triage-friendly report in the set: it names a specific model, a specific regression point, and a clear load failure mode.", "canonical_pr_reason": null, - "best_issue_reason": "No issue is especially suitable as a global representative; the highest-signal items are still narrowly scoped to their own model or subsystem and do not unify the cluster.", + "best_issue_reason": "issue:44964 is the best single issue anchor because it is narrowly scoped and actionable, unlike the broader packaging/import/doc or vague regression reports elsewhere in the cluster.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:44246", "right": "issue:45003", "accept": false, - "reason": "Different problems: slow import behavior vs unsafe sys.modules access in modeling_utils." + "reason": "Import-performance complaint vs unsafe sys.modules access: related area, but different bugs and different fixes." }, { "left": "issue:44964", "right": "issue:45600", "accept": false, - "reason": "Different model-loading regression vs removed Sam3LiteText config references breaking CI." + "reason": "Different model families and failure modes: Phi-4 multimodal load regression vs removed Sam3LiteText config refs breaking CI." }, { "left": "issue:43643", "right": "issue:44496", "accept": false, - "reason": "Different config issues: missing fields from trust_remote_code vs unrecognized model lacking model_type." + "reason": "Both are model/config loading errors, but one is missing fields under trust_remote_code and the other is missing model_type for a specific repo." }, { "left": "issue:44112", "right": "issue:44964", "accept": false, - "reason": "CI stale-device test failure is unrelated to loading microsoft/Phi-4-multimodal-instruct." + "reason": "CI test flake in GraniteSpeech is unrelated to the Phi-4 multimodal loading regression." }, { "left": "issue:27957", "right": "issue:35141", "accept": false, - "reason": "FlashAttention 2 support for XLM-Roberta is unrelated to post_init reinitializing tied embeddings after resize." + "reason": "FlashAttention support for XLM-R is unrelated to embedding reinitialization when resizing token embeddings." }, { "left": "issue:42175", "right": "issue:44279", "accept": false, - "reason": "Packaging/backend dependency issue in torch install extras is not the same as a generic transformers dependency problem." + "reason": "Both mention dependencies, but one is a pip extras/backend packaging complaint and the other is a separate general dependency issue." }, { "left": "issue:46082", "right": "issue:46121", "accept": false, - "reason": "LlamaConfig head_dim validation and rope-validation list typing are different config-path bugs." + "reason": "Both are config-related, but they hit different validation paths and different inputs." }, { "left": "issue:36246", "right": "issue:41628", "accept": false, - "reason": "Qwen2.5 VL image-processor export issue is unrelated to AutoImageProcessor import failures." + "reason": "Different missing symbols: Qwen2_5_VLImageProcessor vs AutoImageProcessor." }, { "left": "issue:42796", "right": "issue:44521", "accept": false, - "reason": "Tokenizer BPE marker handling is unrelated to multimodal assistant_mask generation in apply_chat_template." + "reason": "Tokenizer BPE-marker handling and multimodal assistant_mask generation are different code paths and bugs." }, { "left": "issue:36246", "right": "issue:38617", "accept": false, - "reason": "Different modules and failures: Qwen2.5 VL processor export vs missing layer_type_validation symbol." + "reason": "Both are import errors, but for different removed/missing names with no shared concrete fix." }, { "left": "issue:43404", "right": "issue:45216", "accept": false, - "reason": "lm_head tying in Mistral3 generation is unrelated to Qwen3.5 save_pretrained checkpoint regression." + "reason": "Weight-tying bug in a model class is unrelated to a save_pretrained checkpoint regression." }, { "left": "issue:32090", "right": "issue:35141", "accept": false, - "reason": "Trainer _gpu_broadcast_one NoneType error is unrelated to embedding reinitialization after resize_token_embeddings." + "reason": "Trainer broadcast NoneType error is unrelated to post_init reinitializing output embeddings." }, { "left": "issue:44295", "right": "issue:46506", "accept": false, - "reason": "position_ids buffer access bug is unrelated to CPU/CUDA output divergence in PPDocLayoutV3." + "reason": "Buffer access error and CPU/CUDA output divergence are different problems with no clear shared root cause." }, { "left": "issue:41950", "right": "issue:44016", "accept": false, - "reason": "Video-classification pipeline processor lookup bug is unrelated to a notebook syntax error." + "reason": "Video-classification pipeline image-processor lookup is unrelated to a notebook syntax error." }, { "left": "issue:43531", "right": "issue:45600", "accept": false, - "reason": "Qwen3-MoE sliding_window behavior is unrelated to stale auto_mappings config references." + "reason": "Qwen3-MoE sliding-window behavior and auto_mappings CI breakage are unrelated." }, { "left": "issue:43650", "right": "issue:44220", "accept": false, - "reason": "Placeholder data issue is unrelated to _torch_extract_fbank_features behavior." + "reason": "The titles do not describe the same bug; one is too vague to tie to the audio feature extraction issue." }, { "left": "issue:43673", "right": "issue:46424", "accept": false, - "reason": "Generation cache missing in v5.0.0 is unrelated to max_cache_length only affecting quantized cache." + "reason": "Chunked-prefill cache regression and max_cache_length being honored only for Quantized Cache are separate cache behaviors." }, { "left": "issue:43976", "right": "issue:44987", "accept": false, - "reason": "Python version compatibility regression is unrelated to a model-specific load failure for physical-intelligence/fast." + "reason": "Python-version compatibility failure is unrelated to loading a specific model repo." }, { "left": "issue:43295", "right": "issue:45335", "accept": false, - "reason": "Processor.tokenizer regression is unrelated to t5gemma resize_token_embeddings not updating decoder.embed_tokens." + "reason": "Processor/tokenizer regression and t5gemma embedding resize bug are different model pipeline issues." }, { "left": "issue:44016", "right": "issue:44220", "accept": false, - "reason": "Notebook syntax error is unrelated to audio feature extraction." + "reason": "Notebook syntax error is unrelated to the fbank feature extraction bug." }, { "left": "issue:45083", "right": "issue:46506", "accept": false, - "reason": "Qwen3 Omni MoE feature-length helper behavior is unrelated to CPU/CUDA inference differences." + "reason": "Helper output-length logic in qwen3_omni_moe is not the same as a CPU/CUDA parity discrepancy." }, { "left": "issue:42907", "right": "issue:43441", "accept": false, - "reason": "Dequantized Ministral save failure is unrelated to Ministral flash-attention runtime failure." + "reason": "Dequantized checkpoint saving and FlashAttention runtime failures in the Ministral family are distinct bugs." }, { "left": "issue:44683", "right": "issue:45446", "accept": false, - "reason": "Compiled flex_attention on torch>=2.9 is unrelated to the incorrect version check for AuxRequest import." + "reason": "Compiled flex_attention failure on torch>=2.9 and an incorrect version check for AuxRequest import are different flex_attention issues." }, { "left": "issue:43526", "right": "issue:43901", "accept": false, - "reason": "BeitImageProcessorFast reduce_labels bug is unrelated to text-classification docs still mentioning return_all_scores." - }, - { - "left": "issue:43976", - "right": "issue:44279", - "accept": false, - "reason": "Python support regression is unrelated to a generic dependency issue with transformers." + "reason": "Image-processor label reduction bug and pipeline docs mismatch are unrelated." }, { "left": "issue:43381", "right": "issue:45593", "accept": false, - "reason": "Gradient checkpointing eval-mode restriction is unrelated to D-FINE auxiliary-loss handling." + "reason": "Gradient checkpointing eval-mode restriction has nothing to do with D-FINE auxiliary losses." }, { "left": "issue:43479", "right": "issue:45375", "accept": false, - "reason": "Phi4MultimodalConfig default-init bug is unrelated to Qwen3_5MoeVisionConfig silently dropping a field." + "reason": "Both are multimodal config issues, but they concern different models and different missing/default fields." }, { "left": "issue:44496", "right": "issue:45042", "accept": false, - "reason": "Unrecognized model/config loading failure is unrelated to PIL backend image processors requiring torchvision." + "reason": "Unrecognized model/config loading and PIL backend torchvision requirements are different bugs." }, { "left": "issue:43901", "right": "issue:44496", "accept": false, - "reason": "TextClassificationPipeline docs mismatch is unrelated to loading an unrecognized model." + "reason": "Docs behavior mismatch is unrelated to model-type/config import failure." }, { "left": "issue:43673", "right": "issue:44336", "accept": false, - "reason": "Generation cache behavior is unrelated to ANSI codes emitted by loading_report." + "reason": "Cache handling during generation and ANSI codes in loading_report are unrelated." }, { "left": "issue:44792", "right": "issue:46424", "accept": false, - "reason": "Janus image-generation test failure is unrelated to cache max length semantics." + "reason": "Janus image-generation test failure is unrelated to cache length configuration semantics." + }, + { + "left": "issue:44964", + "right": "issue:45003", + "accept": false, + "reason": "Phi-4 model loading regression and unsafe sys.modules access are different code-path problems with different fixes." } ] }, @@ -12423,8 +12304,8 @@ ], "soft_pairs": [ "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", @@ -12451,160 +12332,196 @@ "issue:41950|issue:45823", "issue:44242|issue:45200" ], - "prepared_review_unit_hash": "a823ea0c6b3c692de7c721fa472aca0f2fae6bc320651827186e7956bb0a1069", + "prepared_review_unit_hash": "e85eb4cab8d6019c04994f248c245c0bac98b304af495ac512d62f1251643973", "component_budget": { "node_count": 500, "item_count": 500, "soft_pair_count": 1839, - "serialized_chars": 953913, - "estimated_input_tokens": 238479, - "estimated_eval_tokens": 477214 + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 28, - "serialized_chars": 36207, - "estimated_input_tokens": 9052, - "estimated_eval_tokens": 18360 + "serialized_chars": 36213, + "estimated_input_tokens": 9054, + "estimated_eval_tokens": 18364 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is a grab bag of unrelated regressions across pipelines, tokenizers, loading, serialization, docs, and performance. The soft pairs mostly share only broad subsystem or version-regression similarity, not the same concrete bug/change, so I reject all of them.", - "confidence": 0.56, - "canonical_issue_reason": "issue:45020 is the broadest umbrella-style regression report in the set, centered on recent-version breakage for remote_code models. It is the best single anchor for a heterogeneous cluster, even though it is not a true duplicate of the others.", + "summary": "The cluster is highly heterogeneous: most pairs are only loosely similar by topic (tokenizers, multimodal processors, save/load regressions, pipelines, imports, etc.) but describe different bugs and code paths. No soft edge looks like a true duplicate merge candidate.", + "confidence": 0.33, + "canonical_issue_reason": "If one issue must anchor the cluster, issue 44521 is the strongest candidate: it is a concrete, reproducible multimodal API regression, has the most discussion/inbound references, and is broader than the other items. That said, the set does not form a real duplicate cluster.", "canonical_pr_reason": null, - "best_issue_reason": "issue:45020 is the most general and cluster-representative issue here because it captures the widest class of version-induced model-loading failures. That said, the cluster is too mixed for a strong duplicate canonical.", + "best_issue_reason": "Issue 44521 is the best standalone issue in the set because it is active, well-scoped, and has clear downstream impact on multimodal chat-template behavior. It is still not a duplicate of the others.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:44016", "right": "issue:44062", "accept": false, - "reason": "Notebook syntax error and AddedToken keyword collision are different failures in different code paths." + "reason": "Both involve notebook/tokenizer problems, but one is a notebook syntax error and the other is an AddedToken keyword conflict; different failure modes and fix paths." }, { - "left": "issue:43824", - "right": "issue:44933", + "left": "issue:44496", + "right": "issue:45759", "accept": false, - "reason": "Both are import errors, but they concern different missing symbols/modules and unrelated root causes." + "reason": "Both are model-loading/config issues, but one is an unrecognized model/config problem and the other is about unwraping text_config in AutoModelForCausalLM.from_config; not the same bug." }, { - "left": "issue:44496", - "right": "issue:45759", + "left": "issue:43824", + "right": "issue:44933", "accept": false, - "reason": "One is an unrecognized config/model loading failure; the other is an AutoModelForCausalLM text_config unwrap bug." + "reason": "Both are import errors, but they concern different symbols/modules entirely (Qwen2_5_VLForConditionalGeneration vs image_utils)." }, { "left": "issue:43441", "right": "issue:45588", "accept": false, - "reason": "Both mention FlashAttention, but one is Ministral-specific and the other is an s_aux=None crash in a different integration path." + "reason": "Both touch flash attention, but they hit different code paths: Ministral FlashAttention failure vs flash_attention.py crashing on s_aux=None for sink-less models." }, { "left": "issue:46123", "right": "issue:46369", "accept": false, - "reason": "Partial-batch result dropping and memory overflow are unrelated symptoms with different mechanisms." + "reason": "One is a pipeline batching bug that drops the last partial batch; the other is a model memory overflow. Different layers of the stack." }, { "left": "issue:44327", "right": "issue:46752", "accept": false, - "reason": "QA span decoding top-k bounds error is unrelated to apply_chat_template crashing on an empty list." + "reason": "Both are runtime errors in pipeline utilities, but one is top_k bounds handling in QA span decoding and the other is empty-list handling in apply_chat_template." }, { "left": "issue:44521", "right": "issue:46693", "accept": false, - "reason": "Assistant mask generation bug and long-context FA2 performance regression are different bugs, not one change." + "reason": "Multimodal assistant mask generation and FA2 long-context performance regression are unrelated bugs, even if both affect generation workflows." }, { "left": "issue:44279", "right": "issue:46515", "accept": false, - "reason": "A vague dependency issue and a torch version incompatibility are too generic to treat as the same defect." + "reason": "These are both dependency/environment complaints, but one is a vague transformer dependency issue and the other is a specific torch 2.4.1 incompatibility." }, { "left": "issue:45335", "right": "issue:45584", "accept": false, - "reason": "Gemma/T5 embedding resize behavior and Whisper empty-transcription generation failure are different model-specific issues." + "reason": "Both involve model/token embedding behavior, but one is resize_token_embeddings not propagating in t5gemma while the other is Whisper generation failing after align_special_tokens." }, { "left": "issue:43299", "right": "issue:45216", "accept": false, - "reason": "Qwen3.5/VL model loading regression and save_pretrained serialization regression are related only by family, not the same bug." + "reason": "Both are Qwen3.5 regressions, but one is a loading failure for MoE models and the other is incorrect save_pretrained checkpoint contents." + }, + { + "left": "issue:46082", + "right": "issue:46399", + "accept": false, + "reason": "A config validation issue for Llama head_dim is unrelated to Qwen3.5 kernel registration failures in use_kernels=True." + }, + { + "left": "issue:43295", + "right": "issue:44779", + "accept": false, + "reason": "Both are tokenizer/processor regressions, but one is about processor.tokenizer/images handling and the other is Deepseek tokenizer output correctness." + }, + { + "left": "issue:44075", + "right": "issue:44368", + "accept": false, + "reason": "SGD args being ignored and a tie_word_embeddings warning are unrelated training/config issues." + }, + { + "left": "issue:39692", + "right": "issue:43901", + "accept": false, + "reason": "Both mention documentation, but they cover different pipelines and different problems: SigLIP2 example mismatches vs TextClassificationPipeline return_all_scores docs drift." + }, + { + "left": "issue:43976", + "right": "issue:45405", + "accept": false, + "reason": "Python version support failure and a PEFT minimum-version bump are different dependency-management issues." }, { "left": "issue:43577", "right": "issue:45216", "accept": false, - "reason": "Blip2 dtype propagation and Qwen3.5 save_pretrained are unrelated code paths." + "reason": "Blip2 dtype propagation and Qwen3.5 save_pretrained serialization are different regressions with different affected code paths." }, { "left": "issue:44351", "right": "issue:45278", "accept": false, - "reason": "A specific missing import and a broad import-error report are not the same underlying issue." + "reason": "A missing HybridCache import and a broad set of import errors after upgrade are related only at a very high level; not the same concrete issue." + }, + { + "left": "issue:46082", + "right": "issue:46512", + "accept": false, + "reason": "Both involve config attributes, but one rejects explicit head_dim while the other leaves model.config.dtype stale after from_config(dtype=...)." }, { "left": "issue:42796", "right": "issue:43650", "accept": false, - "reason": "Tokenizer BPE-marker handling and a vague request to add data are unrelated." + "reason": "Tokenizer BPE marker handling and a nonspecific 'ADD THE DATA' report are not the same bug, and the latter is too vague to match." }, { "left": "issue:44779", "right": "issue:45812", "accept": false, - "reason": "Both are tokenizer regressions, but they affect different model families and likely different tokenizer implementations." + "reason": "Both are tokenizer regressions, but they concern different model families and different symptoms (Deepseek incorrect results vs Granite wrong token IDs)." }, { "left": "issue:41950", "right": "issue:45020", "accept": false, - "reason": "Video-classification image-processor lookup is a pipeline bug, not the remote_code model-loading regression described here." + "reason": "A video-classification processor lookup bug and broad remote_code breakage are different compatibility issues." }, { "left": "issue:43299", "right": "issue:45357", "accept": false, - "reason": "Loading failure and incorrect saved visual-encoder keys are different sides of the workflow and not the same bug." + "reason": "Both concern Qwen3.5 save/load behavior, but one is a loading breakage and the other is incorrect visual encoder keys on save_pretrained." }, { "left": "issue:44466", "right": "issue:45276", "accept": false, - "reason": "lm_head serialization inconsistency and resize_token_embeddings not updating embeddings are distinct weight-management bugs." + "reason": "Both involve weight/embedding serialization, but one is device-dependent lm_head.weight serialization while the other is resize_token_embeddings not updating gemma4 submodules." }, { "left": "issue:43299", "right": "issue:43854", "accept": false, - "reason": "Qwen3.5 model loading and missing Qwen2_5_VL import are separate issues with different affected symbols." + "reason": "Different multimodal model families and different failures: Qwen3VL MoE loading vs GLM-4.7-Flash unit-test loading." }, { "left": "issue:43653", "right": "issue:45593", "accept": false, - "reason": "BigBirdTokenizer special-token registration and D-FINE auxiliary-loss behavior are wholly unrelated." + "reason": "BigBirdTokenizer special-token registration and D-FINE auxiliary-loss gating are completely unrelated bugs." }, { "left": "issue:41950", "right": "issue:45823", "accept": false, - "reason": "Pipeline processor lookup and Gemma4 PLE device mismatch are different subsystems and different bugs." + "reason": "The first is a pipeline processor-selection bug; the second is a Gemma4 device placement mismatch during forward. Different subsystems." }, { "left": "issue:44242", "right": "issue:45200", "accept": false, - "reason": "Router-logit loss gating and mm_token_type_ids defaults are separate multimodal/MoE configuration issues." + "reason": "Load-balancing loss gating and Gemma4 mm_token_type_ids defaults are different multimodal training issues with different code paths." } ] }, @@ -12689,9 +12606,9 @@ "issue:43441|issue:45834", "issue:42994|issue:43317", "issue:44623|issue:44749", + "issue:44336|issue:46851", "issue:44261|issue:44263", "issue:45103|issue:45372", - "issue:44336|issue:46851", "issue:45600|issue:45712", "issue:45859|issue:46851", "issue:42175|issue:43976", @@ -12706,20 +12623,20 @@ "issue:43525|issue:43881", "issue:46291|issue:46399" ], - "prepared_review_unit_hash": "f84ea76bf6494bdd348e3f892c7e02047f9313ab14e4c3ac1feb645bc04e85fa", + "prepared_review_unit_hash": "d068d17fb9dd3a5ff0b4f6231d8c000d3bc575c31981429d01e47d250462d336", "component_budget": { "node_count": 500, "item_count": 500, "soft_pair_count": 1839, - "serialized_chars": 953913, - "estimated_input_tokens": 238479, - "estimated_eval_tokens": 477214 + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 27, - "serialized_chars": 35702, + "serialized_chars": 35701, "estimated_input_tokens": 8926, "estimated_eval_tokens": 18108 }, @@ -12728,162 +12645,162 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The soft pairs are mostly superficial theme matches (tokenizers, loading, configs, deepspeed, Qwen/Ministral), but they describe different concrete bugs or code paths. I would not merge any of these pairs as duplicates.", - "confidence": 0.98, - "canonical_issue_reason": "No single issue is a clear duplicate hub; the items span many unrelated regressions and feature bugs.", + "summary": "This cluster is heterogeneous: the soft pairs mostly share only superficial subsystem overlap (tokenizers, loading, generation, multimodal, Deepspeed, etc.) but do not describe the same concrete bug. I would keep a broad, well-discussed issue like import/performance regressions as the best representative issue, but there is no PR in the set.", + "confidence": 0.87, + "canonical_issue_reason": "issue:44246 is the best representative issue here: it has broad user impact, clear reproduction around import-time performance, and higher discussion/inbound activity than most other candidates.", "canonical_pr_reason": null, - "best_issue_reason": "If a representative is needed, issue:43824 is the most visible/high-discussion loading failure, but it still does not subsume the rest of the set.", + "best_issue_reason": "issue:44246 is the strongest general-purpose issue candidate for triage because it is clearly described, has broader scope than model-specific failures, and appears more central than the narrower one-off regressions.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43650", "right": "issue:46209", "accept": false, - "reason": "Completely unrelated: a vague data request vs an FP8 inference failure on B200." + "reason": "Unrelated: a vague \"ADD THE DATA\" issue is not the same as an FP8 inference failure on B200." }, { "left": "issue:44297", "right": "issue:44462", "accept": false, - "reason": "Both involve tokenizer save/load, but one is about saved tokenizer_class metadata and the other about AutoTokenizer ignoring tokenizer.json." + "reason": "Different tokenizer bugs: save_pretrained metadata mismatch vs AutoTokenizer ignoring repository tokenizer.json." }, { "left": "issue:43479", "right": "issue:43525", "accept": false, - "reason": "Different config bugs: multimodal default sub-config initialization vs missing pad_token_id on Llama4Config." + "reason": "Config initialization defaults for multimodal components is unrelated to Llama4 missing pad_token_id." }, { "left": "issue:44683", "right": "issue:45542", "accept": false, - "reason": "Different subsystems and failures: flex_attention on torch>=2.9 vs TensorFlow backend missing because only tensorboard is installed." + "reason": "Different areas and failure modes: torch 2.9 flex_attention vs missing TensorFlow backend installation." }, { "left": "issue:44016", "right": "issue:44291", "accept": false, - "reason": "Notebook syntax error and init_empty_weights argument handling are unrelated bugs." + "reason": "Notebook syntax error is unrelated to init_empty_weights passing an unexpected _is_hf_initialized argument." }, { "left": "issue:44361", "right": "issue:45593", "accept": false, - "reason": "Different model/task issues: MLukeTokenizer AttributeError vs D-FINE auxiliary loss behavior." + "reason": "Tokenizer AttributeError on tasks is not the same as D-FINE auxiliary-loss behavior." }, { "left": "issue:45081", "right": "issue:45907", "accept": false, - "reason": "Mistral tokenizer regex crash vs list_repo_files revision forwarding; no shared code-path bug." + "reason": "Mistral regex patch crash and list_repo_files revision forwarding are different code paths and bugs." }, { "left": "issue:33290", "right": "issue:45656", "accept": false, - "reason": "Both mention deepspeed/optimizer, but one is Adafactor OOM and the other is optimizer.step() being called twice." + "reason": "Both involve Deepspeed/optimizer, but one is OOM with Adafactor and the other is double optimizer stepping." }, { "left": "issue:43441", "right": "issue:45834", "accept": false, - "reason": "Both are model-specific inference bugs, but Ministral FlashAttention failure and Kosmos2.5 OCR index error are different code paths." + "reason": "Different models and issues: FlashAttention failure for Ministral-3 vs OCR index error in Kosmos2.5." }, { "left": "issue:42994", "right": "issue:43317", "accept": false, - "reason": "Quantized model saving and dequantized model loading/offload are distinct stages with different failures." + "reason": "Quantized model saving failure is distinct from loading a dequantized model with gpu+cpu offload." }, { "left": "issue:44623", "right": "issue:44749", "accept": false, - "reason": "Processor save_pretrained missing files vs a performance regression during data filtering; unrelated." + "reason": "Processor save_pretrained missing files is unrelated to a Chinese report about inference/data filtering slowdown." + }, + { + "left": "issue:44336", + "right": "issue:46851", + "accept": false, + "reason": "ANSI-code output in loading_report has nothing to do with Qwen3.5 variable-length position_ids." }, { "left": "issue:44261", "right": "issue:44263", "accept": false, - "reason": "Different model internals: MLA q_a_layernorm config precision issue vs torch.split return handling in GlmMoeDsaIndexer." + "reason": "Numerical precision issue in MLA q_a_layernorm is unrelated to torch.split return values in GlmMoeDsaIndexer." }, { "left": "issue:45103", "right": "issue:45372", "accept": false, - "reason": "auto_docstring annotations crash vs mistral_common ReasoningEffort import error; unrelated bugs." - }, - { - "left": "issue:44336", - "right": "issue:46851", - "accept": false, - "reason": "ANSI output in loading_report is unrelated to Qwen3.5 variable-length position_ids behavior." + "reason": "Auto-docstring future-annotations crash and ReasoningEffort import failure are different bugs in different modules." }, { "left": "issue:45600", "right": "issue:45712", "accept": false, - "reason": "Both touch generated mapping/dummy artifacts, but one is stale auto_mappings references and the other is leftover dummy classes in dummy_pt_objects." + "reason": "Broken auto_mappings references and leftover dummy_pt_objects cleanup are separate maintenance issues." }, { "left": "issue:45859", "right": "issue:46851", "accept": false, - "reason": "Both are Qwen3.5 issues, but CPU-offload incompatibility and variable-length position_ids are different problems." + "reason": "CPU offload incompatibility in rotary embedding is not the same as variable-length input handling." }, { "left": "issue:42175", "right": "issue:43976", "accept": false, - "reason": "Package/backend installation issue vs Python version compatibility; not the same bug." + "reason": "Packaging TensorFlow omission is unrelated to a Python 3.9+/3.10+ compatibility break." }, { "left": "issue:44623", "right": "issue:45397", "accept": false, - "reason": "Processor save_pretrained missing files and gemma-4 zero3 from_pretrained are different failures." + "reason": "Missing files on processor save_pretrained is not the same as gemma-4 zero3 from_pretrained loading." }, { "left": "issue:44671", "right": "issue:44779", "accept": false, - "reason": "Different regressions: CamemBERT masked-LM predictions vs Deepseek tokenizer output correctness." + "reason": "Incorrect masked LM predictions in CamemBERT and incorrect Deepseek tokenizer outputs are separate model/tokenizer regressions." }, { "left": "issue:43723", "right": "issue:44162", "accept": false, - "reason": "Tokenizer loading issue and ESM2 model breakage are unrelated." + "reason": "AutoTokenizer loading failure and ESM2 breakage are different model-loading problems." }, { "left": "issue:43901", "right": "issue:44246", "accept": false, - "reason": "Docs mismatch for return_all_scores vs intermittent slow import; no shared underlying code bug." + "reason": "Docs mismatch for return_all_scores is unrelated to intermittent slow imports." }, { "left": "issue:43673", "right": "issue:43761", "accept": false, - "reason": "Generation cache missing in chunked_prefill vs CLIPVision hidden_states regression; different models and paths." + "reason": "GenerationMixin cache missing during chunked_prefill and CLIPVision hidden_states regression are different components and symptoms." }, { "left": "issue:43824", "right": "issue:43976", "accept": false, - "reason": "ImportError for a Qwen2.5-VL class is unrelated to Python 3.9/3.10 compatibility." + "reason": "Missing Qwen2_5_VL import and Python version support issues are not the same bug." }, { "left": "issue:45542", "right": "issue:45800", "accept": false, - "reason": "TensorFlow backend packaging bug vs torch/transformers version incompatibility; not duplicates." + "reason": "Missing TensorFlow backend from a torch install is distinct from a torch/transformers version incompatibility." }, { "left": "issue:42796", "right": "issue:43441", "accept": false, - "reason": "Same model family, but tokenizer BPE-marker handling and FlashAttention failure are separate defects." + "reason": "Tokenizer BPE marker handling and FlashAttention failure are different classes of bugs." }, { "left": "issue:43525", @@ -12895,7 +12812,7 @@ "left": "issue:46291", "right": "issue:46399", "accept": false, - "reason": "Both mention kernels, but one is dependency version/revision requirements and the other is a registration/decorator crash in Qwen3.5." + "reason": "kernels v0.15 version/revision requirement and Qwen3.5 rotary embedding registration crash are different integration failures." } ] }, @@ -12970,16 +12887,16 @@ "soft_pairs": [ "issue:44188|issue:44336", "issue:44016|issue:45823", - "issue:44589|issue:45362", "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:44246|issue:44496", "issue:45083|issue:46851", + "issue:44246|issue:44496", "issue:46399|issue:46851", "issue:45561|issue:45563", "issue:45103|issue:46399", @@ -12987,219 +12904,219 @@ "issue:46614|issue:46665", "issue:43540|issue:46752", "issue:45072|issue:46829", - "issue:43531|issue:45362", "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:44242|issue:45593", "issue:45588|issue:46851", "issue:45003|issue:46821", "issue:42886|issue:46132", "issue:44451|issue:46752" ], - "prepared_review_unit_hash": "e0d98802b6840352d080baeb393f4e8139a05b1cd4ab6e16883c41336ac9eb0e", + "prepared_review_unit_hash": "a9476120dcbdbde30727f0907ed815a5295a9a36b9a7af73fed4b5397b3ca748", "component_budget": { "node_count": 500, "item_count": 500, "soft_pair_count": 1839, - "serialized_chars": 953913, - "estimated_input_tokens": 238479, - "estimated_eval_tokens": 477214 + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 29, - "serialized_chars": 35847, - "estimated_input_tokens": 8962, - "estimated_eval_tokens": 18180 + "serialized_chars": 35841, + "estimated_input_tokens": 8961, + "estimated_eval_tokens": 18178 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster is mostly a set of unrelated Transformers bug reports across different models, subsystems, and failure modes. I would not merge any of the soft-edge pairs; they are at best superficially similar by keywords or broad area.", - "confidence": 0.87, - "canonical_issue_reason": "Issue 46851 is the best-scoped concrete bug report in the set, with a clear failure mode and reproducible model-specific behavior. There is no true duplicate-heavy canonical issue in this cluster.", + "summary": "This cluster is a loose collection of unrelated Transformers issues that only share superficial keywords (models, config, loading, position_ids, etc.). I do not see any true duplicate pairs among the soft edges.", + "confidence": 0.95, + "canonical_issue_reason": "Issue 46851 is the best representative of the cluster only in the sense that it is a concrete, reproducible runtime regression with a clear code path. But the cluster is not a duplicate set, so no single issue is a true canonical for the whole group.", "canonical_pr_reason": null, - "best_issue_reason": "46851 is the most actionable single issue here: specific title, clear runtime symptom, and narrow scope. It is still not a duplicate target for the others.", + "best_issue_reason": "46851 has the clearest, most actionable bug report and a specific model-family regression; it is the strongest stand-alone issue in this mixed cluster.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:44188", "right": "issue:44336", "accept": false, - "reason": "Different subsystems and symptoms: attention kernel divergence under torch.compile vs a docstring/formatting issue." + "reason": "Completely different subsystems: attention kernel behavior under torch.compile vs terminal ANSI handling in loading_report." }, { "left": "issue:44016", "right": "issue:45823", "accept": false, - "reason": "Notebook syntax error is unrelated to Gemma4 device placement/forward mismatch." + "reason": "Notebook syntax error and Gemma4 device_map forward mismatch are unrelated bug classes." }, { - "left": "issue:44589", - "right": "issue:45362", + "left": "issue:44336", + "right": "issue:44373", "accept": false, - "reason": "Storage backend lookup failure and Qwen3.5 chat crash are unrelated code paths." + "reason": "ANSI escaping in loading_report and a wrong docstring for position_ids are not the same underlying issue." }, { - "left": "issue:44336", - "right": "issue:44373", + "left": "issue:44589", + "right": "issue:45362", "accept": false, - "reason": "ANSI output formatting is unrelated to a wrong docstring for position_ids." + "reason": "Float8 storage lookup failure is unrelated to a Qwen3.5 chat crash." }, { "left": "issue:43525", "right": "issue:43643", "accept": false, - "reason": "Both mention missing fields, but one is a specific Llama4Config attribute bug and the other is a trust_remote_code config-construction inconsistency." + "reason": "Both involve config fields, but one is a missing pad_token_id attribute and the other is trust_remote_code returning incomplete AutoConfig fields; different failures and fixes." }, { "left": "issue:46655", "right": "issue:46726", "accept": false, - "reason": "Both involve vision processors, but one is a padding calculation bug in video processing and the other is an error path for missing vision backend support." + "reason": "Video padding calculation bug vs missing vision backend reporting; same area, but distinct code paths and symptoms." }, { "left": "issue:45325", "right": "issue:46851", "accept": false, - "reason": "Both involve position_ids, but they affect different Qwen model families and different logic." + "reason": "Both mention position_ids, but they affect different Qwen model families and different logic (RoPE scaling vs variable-length input handling)." }, { "left": "issue:41628", "right": "issue:43976", "accept": false, - "reason": "Import error for AutoImageProcessor is unrelated to a Python version compatibility regression." + "reason": "Importing AutoImageProcessor and Python version compatibility are unrelated." }, { "left": "issue:44297", "right": "issue:45397", "accept": false, - "reason": "Tokenizer serialization mismatch is unrelated to the Gemma-4 zero3 loading failure." + "reason": "Tokenizer save_pretrained metadata mismatch is unrelated to Gemma4 zero3 from_pretrained behavior." }, { "left": "issue:43784", "right": "issue:45542", "accept": false, - "reason": "Sentence-transformers import NameError and missing TensorFlow backend handling are different dependency/import bugs." + "reason": "Both are import-time backend issues, but one is missing nn for sentence-transformers and the other is undefined tf backend when only tensorboard is installed; not the same bug." }, { - "left": "issue:44246", - "right": "issue:44496", + "left": "issue:45083", + "right": "issue:46851", "accept": false, - "reason": "Slow import and unrecognized model/config parsing are not the same issue." + "reason": "Different Qwen3 submodels and different helper paths; not the same position_ids bug." }, { - "left": "issue:45083", - "right": "issue:46851", + "left": "issue:44246", + "right": "issue:44496", "accept": false, - "reason": "Different Qwen model variants and different helper behavior; not the same underlying bug." + "reason": "Moe load-balancing loss gating and unrecognized model_type are unrelated." }, { "left": "issue:46399", "right": "issue:46851", "accept": false, - "reason": "Kernel registration decorator failure is unrelated to variable-length position_ids handling." + "reason": "Kernel registration decorator crash is a separate issue from variable-length position_ids handling." }, { "left": "issue:45561", "right": "issue:45563", "accept": false, - "reason": "A pytest-xdist race in testing utilities is unrelated to a stale generate() warning." + "reason": "A pytest-xdist file race and a stale generation warning are unrelated." }, { "left": "issue:45103", "right": "issue:46399", "accept": false, - "reason": "Auto-docstring annotation parsing and apply_rotary_pos_emb registration are unrelated." + "reason": "Auto-docstring annotation parsing crash and Qwen kernel registration are unrelated tooling/runtime bugs." }, { "left": "issue:43526", "right": "issue:45003", "accept": false, - "reason": "BEiT image processor label reduction bug is unrelated to unsafe sys.modules access in modeling_utils." + "reason": "BEiT reduce_labels behavior and unsafe sys.modules access are unrelated." }, { "left": "issue:46614", "right": "issue:46665", "accept": false, - "reason": "Tokenizer phonemizer backend state bug is unrelated to from_pretrained behavior inside Python classes." + "reason": "Stale phonemizer language state and inconsistent from_pretrained behavior are different bugs." }, { "left": "issue:43540", "right": "issue:46752", "accept": false, - "reason": "Video-input processing error in Qwen3OmniMoe and empty-list chat-template crash are distinct failures." + "reason": "Qwen3OmniMoe video input validation and empty-list chat template handling are unrelated." }, { "left": "issue:45072", "right": "issue:46829", "accept": false, - "reason": "dtype mismatch in inference and miswired hidden_states output are different model/runtime bugs." + "reason": "bfloat16 dtype mismatch issues are unrelated to VideoPrism hidden_states/output packaging." }, { - "left": "issue:43531", - "right": "issue:45362", + "left": "issue:44295", + "right": "issue:45397", "accept": false, - "reason": "Sliding-window bug in Qwen3-MoE is not the same as a chat crash in Qwen3.5." + "reason": "Reading a bufferized position_ids tensor and Gemma4 zero3 loading are different failure modes." }, { - "left": "issue:44295", - "right": "issue:45397", + "left": "issue:43531", + "right": "issue:45362", "accept": false, - "reason": "Position_ids buffer reading error and Gemma-4 zero3 loading are unrelated." + "reason": "Both are Qwen-family regressions, but one is sliding_window handling in Qwen3-MoE and the other is chat behavior in Qwen3.5; not the same concrete bug." }, { "left": "issue:42491", "right": "issue:43299", "accept": false, - "reason": "These both mention Qwen MoE, but one is a hf4.x/hf5.x LoRA compatibility problem and the other is a loading regression in 5.0.0.dev0." + "reason": "Both involve Qwen3 MoE compatibility, but one is LoRA reuse across major versions and the other is model loading breakage in dev0; different root causes." }, { "left": "issue:44246", "right": "issue:44871", "accept": false, - "reason": "Slow import is unrelated to an eos_token_id configuration inconsistency." + "reason": "Load-balancing loss gating and Gemma-3 eos_token_id inconsistency are unrelated." }, { - "left": "issue:44242", - "right": "issue:45593", + "left": "issue:44336", + "right": "issue:44485", "accept": false, - "reason": "Both concern losses, but one is MoE load-balancing loss gating and the other is D-FINE auxiliary losses when denoising is off." + "reason": "Terminal ANSI output bug and GLM-5 RoPE implementation discussion are unrelated." }, { - "left": "issue:44336", - "right": "issue:44485", + "left": "issue:44242", + "right": "issue:45593", "accept": false, - "reason": "Docstring/ANSI output issue is unrelated to GLM-5 RoPE implementation." + "reason": "Both mention auxiliary losses, but they are different models and different gating conditions; too broad to merge." }, { "left": "issue:45588", "right": "issue:46851", "accept": false, - "reason": "Flash-attention sink-less model crash and variable-length position_ids are different code paths." + "reason": "flash_attention sink-less model crash is unrelated to Qwen3.5 variable-length position_ids." }, { "left": "issue:45003", "right": "issue:46821", "accept": false, - "reason": "Unsafe sys.modules access is unrelated to a TimesFM window_size AttributeError." + "reason": "Unsafe sys.modules access and TimesFM window_size AttributeError are unrelated." }, { "left": "issue:42886", "right": "issue:46132", "accept": false, - "reason": "Offline cache tokenization failure is unrelated to AttentionInterface.register changing function behavior." + "reason": "Offline tokenizer cache loading and AttentionInterface.register behavior are unrelated." }, { "left": "issue:44451", "right": "issue:46752", "accept": false, - "reason": "Model-loading failure for ScandiBERT and empty-list chat-template crash are not the same bug." + "reason": "ScandiBERT loading failure and empty-list chat template crash are unrelated." } ] }, @@ -13300,211 +13217,211 @@ "issue:44242|issue:44857", "issue:43526|issue:46897", "issue:44038|issue:46506", - "issue:29870|issue:30333", - "issue:44987|issue:45459" + "issue:44987|issue:45459", + "issue:29870|issue:30333" ], - "prepared_review_unit_hash": "383b091939776c5c8348c7cdbeec50b999eac595254cdae5ba8068060012461f", + "prepared_review_unit_hash": "878fac6693f8e40a18b5696fb830cb80d80b6407edfc4b04ade152691b94bacb", "component_budget": { "node_count": 500, "item_count": 500, "soft_pair_count": 1839, - "serialized_chars": 953913, - "estimated_input_tokens": 238479, - "estimated_eval_tokens": 477214 + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 29, - "serialized_chars": 36555, - "estimated_input_tokens": 9139, - "estimated_eval_tokens": 18534 + "serialized_chars": 36559, + "estimated_input_tokens": 9140, + "estimated_eval_tokens": 18536 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The set is heterogeneous with no true duplicate cluster; the soft edges mostly connect issues that share a subsystem name but not the same underlying bug. I would keep the candidate links rejected.", - "confidence": 0.88, - "canonical_issue_reason": "issue:44479 is the strongest standalone bug report in the set: it describes a concrete regression, names multiple affected model families, and is framed as an actionable code-path failure.", + "summary": "This cluster is highly heterogeneous: most items are distinct bug reports across unrelated subsystems, and none of the soft pairs are strong enough to treat as duplicates. The closest themes are video-processing regressions and strict/config handling, but they still point to different root causes and code paths.", + "confidence": 0.22, + "canonical_issue_reason": "issue:44479 is the broadest and most representative issue here: it describes a concrete regression affecting multiple Qwen video models in v5.3.0, which is the closest thing to a shared bug theme in the set.", "canonical_pr_reason": null, - "best_issue_reason": "issue:44479 is the best representative issue overall because it is specific, reproducible, and broad enough to capture a real regression across related models.", + "best_issue_reason": "issue:44479 is the best issue candidate because it covers a widespread, user-visible regression with multiple affected models and a clear version-specific breakage; the others are narrower or unrelated.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43550", "right": "issue:46506", "accept": false, - "reason": "Both are runtime discrepancies, but one is a Bamba SDPA/torch.compile failure and the other is a CPU-vs-CUDA output mismatch in PPDocLayoutV3; different bugs." + "reason": "Different models, different failure modes, and different subsystems; both mention runtime issues but not the same concrete bug." }, { "left": "issue:44779", "right": "issue:45292", "accept": false, - "reason": "Tokenizer corruption in Deepseek v5 is unrelated to resize_token_embeddings not propagating to output embeddings." + "reason": "Tokenizer correctness regression and resize_token_embeddings behavior are unrelated code paths." }, { "left": "issue:36246", "right": "issue:42757", "accept": false, - "reason": "Different import errors from different modules: Qwen2.5 VL image processor vs is_offline_mode in huggingface_hub." + "reason": "Import errors from different symbols/modules with no shared underlying defect." }, { "left": "issue:36246", "right": "issue:44351", "accept": false, - "reason": "Both are import failures, but they concern different symbols and code paths; no evidence of the same underlying breakage." + "reason": "Both are import errors, but for different APIs and causes; not the same bug." }, { "left": "issue:41950", "right": "issue:44220", "accept": false, - "reason": "Video-classification using image processors and _torch_extract_fbank_features() are separate processing paths with different symptoms." + "reason": "Video-classification processor lookup and fbank feature extraction are separate processing bugs." }, { "left": "issue:43901", "right": "issue:45003", "accept": false, - "reason": "Docs/behavior mismatch for return_all_scores is unrelated to unsafe sys.modules access in modeling_utils." + "reason": "Documentation mismatch for a pipeline setting is unrelated to unsafe sys.modules access." }, { "left": "issue:43257", "right": "issue:45137", "accept": false, - "reason": "Both involve DeepSpeed/large-model loading, but one is missing Qwen3 MOE weight conversion and the other is a zero3 deque underflow; different failures." + "reason": "Both involve DeepSpeed, but one is MoE weight conversion and the other is a ZeRO3 deque/indexing crash." }, { "left": "issue:45003", "right": "issue:45561", "accept": false, - "reason": "Unsafe sys.modules access is unrelated to the pytest-xdist captured_info.txt race." + "reason": "Unsafe module lookup and pytest-xdist file race are unrelated." }, { "left": "issue:44496", "right": "issue:45003", "accept": false, - "reason": "An unrecognized model/config loading failure is not the same as unsafe sys.modules lookups." + "reason": "Model config loading failure and sys.modules access are not the same issue." }, { "left": "issue:43540", "right": "issue:46655", "accept": false, - "reason": "Both touch video inputs, but one is a Qwen3OmniMoe ValueError and the other is a SmolVLM padding bug; different code paths." + "reason": "Both are video-input bugs, but in different models with different failure mechanisms." }, { "left": "issue:44162", "right": "issue:45200", "accept": false, - "reason": "ESM2 breakage and Gemma 4 mm_token_type_ids defaulting are unrelated model-specific issues." + "reason": "ESM2 breakage and Gemma 4 token-type defaults are unrelated." }, { "left": "issue:45375", "right": "issue:45759", "accept": false, - "reason": "Both involve Qwen multimodal configs, but one is a missing config field dropped by @strict and the other is AutoModelForCausalLM.from_config not unwrapping text_config; distinct bugs." + "reason": "Both are Qwen multimodal config issues, but one is strict-field dropping and the other is AutoModelForCausalLM text_config unwrapping." }, { "left": "issue:43262", "right": "issue:44246", "accept": false, - "reason": "Audio chat-template sampling-rate handling is unrelated to an import-time performance complaint." + "reason": "Audio chat-template sampling-rate defaulting and import latency are unrelated." }, { "left": "issue:44479", "right": "issue:46655", "accept": false, - "reason": "Both are video-related regressions, but the affected models and failure mechanisms differ." + "reason": "Both are video regressions, but they affect different model families and appear to stem from different processor bugs." }, { "left": "issue:44295", "right": "issue:44749", "accept": false, - "reason": "Registering position_ids as a buffer causing an error is unrelated to a general performance slowdown report." + "reason": "A runtime position_ids error and a performance regression during filtering are not the same bug." }, { "left": "issue:43475", "right": "issue:46369", "accept": false, - "reason": "Same SAM3 family, but one is a missing attribute in SAM 3 Video and the other is a memory overflow in sam3-litetext; not the same bug." + "reason": "SAM 3 video encoder attribute error and SAM 3 LiteText memory overflow are distinct failures." }, { "left": "issue:45859", "right": "issue:46399", "accept": false, - "reason": "Both are Qwen3.5-related, but CPU offload incompatibility and missing decorator registration are different concrete failures." + "reason": "Both mention Qwen3.5 rotary/embed-related code, but one is CPU offload incompatibility and the other is a kernel registration decorator crash." }, { "left": "issue:45137", "right": "issue:45600", "accept": false, - "reason": "Zero3 deque underflow and broken auto_mappings references are unrelated." + "reason": "DeepSpeed ZeRO3 deque error and auto_mappings CI breakage are unrelated." }, { "left": "issue:36246", "right": "issue:43596", "accept": false, - "reason": "Importing a missing image processor and an IndexError in BertModel with ZeRO3 init are unrelated." + "reason": "Different import/runtime errors with no shared root cause." }, { "left": "issue:44877", "right": "issue:45375", "accept": false, - "reason": "Both are strict-config complaints, but they affect different models and different missing-field behavior." + "reason": "Both involve strict config handling, but they concern different models and different outcomes." }, { "left": "issue:43295", "right": "issue:45292", "accept": false, - "reason": "Processor.tokenizer/image passing regression is unrelated to resize_token_embeddings propagation." + "reason": "Processor API regression and token-embedding resize behavior are unrelated." }, { "left": "issue:45292", "right": "issue:45584", "accept": false, - "reason": "Embedding resize/output-embedding propagation has nothing to do with Whisper empty-transcription generation failure." + "reason": "Token embedding resizing and Whisper empty-transcription generation are different problems." }, { "left": "issue:44466", "right": "issue:45335", "accept": false, - "reason": "Both mention tied weights/embeddings, but one is serialization inconsistency by device and the other is resize_token_embeddings not updating decoder.embed_tokens; not the same bug." + "reason": "Serialization inconsistency for lm_head.weight and decoder.embed_tokens resize behavior are not the same underlying bug." }, { "left": "issue:44623", "right": "issue:46726", "accept": false, - "reason": "Missing processor.save_pretrained files and AutoProcessor's unrecognized image processor error have different root causes." + "reason": "Missing files on save_pretrained and unrecognized image processor on missing backend are different processor issues." }, { "left": "issue:44242", "right": "issue:44857", "accept": false, - "reason": "MoE load-balancing loss omission is unrelated to an AMP/CUDA crash in LwDetrImageLoss." + "reason": "MoE load-balancing loss omission and AMP/CUDA crash in image loss are unrelated." }, { "left": "issue:43526", "right": "issue:46897", "accept": false, - "reason": "Label reduction bug in BeitImageProcessorFast and Florence2's training-loss double-shift issue are different model-specific problems." + "reason": "Label reduction bug in BeitImageProcessorFast and Florence2 training-loss shift issue are not the same change." }, { "left": "issue:44038", "right": "issue:46506", "accept": false, - "reason": "Qwen3-VL-Moe on transformers 5.0 and PPDocLayoutV3 CPU/CUDA divergence are unrelated." + "reason": "Qwen3-VL-Moe bug and PPDocLayoutV3 CPU/CUDA discrepancy are unrelated." }, { - "left": "issue:29870", - "right": "issue:30333", + "left": "issue:44987", + "right": "issue:45459", "accept": false, - "reason": "BART generate() unbound-local error and MLFlow not reporting failed jobs are unrelated operational bugs." + "reason": "Model loading failure for a specific repo and tokenizer protobuf error masking are different bugs." }, { - "left": "issue:44987", - "right": "issue:45459", + "left": "issue:29870", + "right": "issue:30333", "accept": false, - "reason": "Loading physical-intelligence/fast on transformers>=5.1.0 is unrelated to protobuf-related tokenizer error masking." + "reason": "Bart generate() UnboundLocalError and MLflow job-status reporting are unrelated." } ] }, @@ -13607,124 +13524,124 @@ "issue:43065|issue:46123", "issue:44297|issue:44623" ], - "prepared_review_unit_hash": "692fea03121cfdaca1eb0aae7424001fd507c153b2595f780aae9eab27c48580", + "prepared_review_unit_hash": "7f378e615ded3dc5f9655bdb81d40ef945a63c31f18af692ec24323c99fd2721", "component_budget": { "node_count": 500, "item_count": 500, "soft_pair_count": 1839, - "serialized_chars": 953913, - "estimated_input_tokens": 238479, - "estimated_eval_tokens": 477214 + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 29, - "serialized_chars": 35965, - "estimated_input_tokens": 8992, - "estimated_eval_tokens": 18240 + "serialized_chars": 35960, + "estimated_input_tokens": 8990, + "estimated_eval_tokens": 18236 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is heterogeneous: mostly unrelated bug reports across model loading, tokenizers, video, cache, and compile paths. No PRs are present. The most coherent anchor is the StaticCache/get_seq_length device-tensor bug, but none of the soft pairs look like true duplicates or mergeable PR-equivalents.", - "confidence": 0.17, - "canonical_issue_reason": "Issue #46858 is the clearest, most actionable report in the recurring cache/get_seq_length theme, with a concrete failure mode on non-CUDA backends and torch.compile.", + "summary": "This cluster is highly heterogeneous; the soft pairs are mostly superficially similar but not the same underlying bug or change. No PRs are present, so the best anchor is an issue, not a PR.", + "confidence": 0.41, + "canonical_issue_reason": "issue:46858 is the best single anchor: it is a concrete, code-path-specific bug with a clear failure mode and is the closest thing here to a reusable bug family (cache length / static cache behavior). The rest of the cluster is largely unrelated.", "canonical_pr_reason": null, - "best_issue_reason": "Issue #46858 is the best anchor because it is specific, reproducible, and tied to a concrete code path; it is more representative than the broader or more ambiguous reports nearby.", + "best_issue_reason": "issue:46858 is the strongest issue candidate because it is explicit, actionable, and narrowly scoped, whereas most other items are broader feature regressions, warnings, or unrelated model-specific failures.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43352", "right": "issue:45250", "accept": false, - "reason": "Both mention flash attention, but one is a model-specific support error for NemotronHForCausalLM and the other is a generic topic; not the same bug." + "reason": "Both mention Flash Attention 2.0, but one is a model-specific unsupported-path error and the other is a generic topic issue; not the same bug." }, { "left": "issue:42199", "right": "issue:46851", "accept": false, - "reason": "DETR cardinality/background-class behavior and Qwen3.5 variable-length position_ids are unrelated code paths." + "reason": "DET R cardinality/background-class handling and Qwen3.5 variable-length position_ids are unrelated code paths and symptoms." }, { "left": "issue:43526", "right": "issue:46894", "accept": false, - "reason": "BeitImageProcessorFast label reduction and CohereASR training-loss shifting are different subsystems and failures." + "reason": "BEiT label reduction and CohereASR loss shifting are different model families and different failure modes." }, { "left": "issue:44060", "right": "issue:44077", "accept": false, - "reason": "A tied-weights warning in Qwen3-Next and a patchtsmixer post_init policy issue are not the same underlying problem." + "reason": "One is a tied-weights warning in Qwen3-Next; the other is a patchtsmixer post_init policy issue. Same broad theme only at a high level." }, { "left": "issue:44368", "right": "issue:44977", "accept": false, - "reason": "Both touch Qwen3.5, but one is a tied-embeddings warning and the other is a flash-attention generation failure." + "reason": "A tied-embeddings warning is not the same as a flash-attention generation regression in Qwen3.5." }, { "left": "issue:43653", "right": "issue:44162", "accept": false, - "reason": "BigBirdTokenizer special-token decode behavior and ESM2 breakage are unrelated; no shared concrete bug." + "reason": "Tokenizer special-token registration and ESM2 breakage are different subsystems and different root causes." }, { "left": "issue:45412", "right": "issue:46123", "accept": false, - "reason": "RT-DETR garbage-collection memory release and MaskGenerationPipeline partial-batch dropping are different bugs." + "reason": "RT-DETR memory cleanup and MaskGenerationPipeline batching logic are unrelated bugs." }, { "left": "issue:44479", "right": "issue:46170", "accept": false, - "reason": "A video-input regression in Qwen models and a Windows/ROCm torch.distributed import issue do not share a code path." + "reason": "Video-input regression in Qwen models and Windows/ROCm import dependency failures are unrelated." }, { "left": "issue:42886", "right": "issue:45229", "accept": false, - "reason": "Offline tokenizer cache loading and Gemma4 multi-GPU OOM are unrelated failures." + "reason": "Offline tokenizer cache loading and Gemma4 multi-GPU CUDA OOM are different classes of failure." }, { "left": "issue:45397", "right": "issue:45656", "accept": false, - "reason": "Gemma-4 zero3 loading and deepspeed double optimizer stepping are different problems." + "reason": "Gemma-4 ZeRO3 loading and double optimizer stepping under deepspeed are not the same concrete bug." }, { "left": "issue:43065", "right": "issue:43408", "accept": false, - "reason": "Dummy Conv2d in Sam3PixelDecoder vs. a sam3_video/sam3_tracker type warning are not the same issue." + "reason": "SAM3 pixel-decoder implementation detail and sam3_video vs sam3_tracker model-type warning are distinct issues." }, { "left": "issue:43475", "right": "issue:43540", "accept": false, - "reason": "Missing fpn_position_embeddings in SAM3 Video and a Qwen3OmniMoe video-input ValueError are unrelated." + "reason": "SAM3 vision encoder output attribute error and Qwen3OmniMoe video-input ValueError are unrelated." }, { "left": "issue:44877", "right": "issue:46170", "accept": false, - "reason": "Strict granite_speech config loading and a Windows ROCm import failure are distinct bugs." + "reason": "Strict config loading for granite_speech and ROCm/Windows distributed import errors do not share the same bug." }, { "left": "issue:46170", "right": "issue:46832", "accept": false, - "reason": "One is a platform import/dependency problem; the other is an RTDetr forward shape/feature-level mismatch." + "reason": "A platform/dependency import issue is not the same as an RT-DETR feature-level mismatch in forward." }, { "left": "issue:44368", "right": "issue:45563", "accept": false, - "reason": "Two different stale warnings in different features; not the same underlying defect." + "reason": "These are both warnings, but they concern different generation settings and different code paths." }, { "left": "issue:43723", @@ -13736,79 +13653,79 @@ "left": "issue:44995", "right": "issue:45200", "accept": false, - "reason": "GlmMoeDsa stale cache on second forward and Gemma 4 mm_token_type_ids defaults are unrelated." + "reason": "GlmMoeDsa stale cache crash and Gemma 4 mm_token_type_ids defaults are different model behaviors." }, { "left": "issue:43931", "right": "issue:44368", "accept": false, - "reason": "Qwen3-VL weight-shape mismatch and a tied-embeddings warning are separate model-loading issues." + "reason": "Qwen3-VL weight-shape mismatch and tied-embeddings warnings are unrelated." }, { "left": "issue:44188", "right": "issue:46202", "accept": false, - "reason": "torch.compile attention-kernel divergence and Gemma pipeline roundtrip behavior are unrelated." + "reason": "torch.compile attention-kernel divergence and Gemma token roundtrip failure are not the same issue." }, { "left": "issue:44483", "right": "issue:46202", "accept": false, - "reason": "Chat/completions request rejection and a Gemma tokenizer roundtrip bug are different layers and failures." + "reason": "Chat-completions API rejection and Gemma pipeline roundtrip failure are different regressions." }, { "left": "issue:46829", "right": "issue:46858", "accept": false, - "reason": "VideoPrism output field mix-up and cache.get_seq_length tensor arithmetic are not the same bug." + "reason": "VideoPrism output field mix-up and cache.get_seq_length returning a device tensor are different bugs." }, { "left": "issue:45987", "right": "issue:46082", "accept": false, - "reason": "StaticCache return-type contract and LlamaConfig head_dim validation are unrelated." + "reason": "A cache return-type contract bug and a LlamaConfig head_dim validation issue are unrelated." }, { "left": "issue:43906", "right": "issue:46202", "accept": false, - "reason": "An isolated reproduction issue is not the same as the Gemma pipeline roundtrip failure." + "reason": "An isolated reproduction of another issue is not enough to treat it as the same bug as the Gemma roundtrip failure." }, { "left": "issue:44995", "right": "issue:45588", "accept": false, - "reason": "A stale indexer cache crash and a flash_attention.py s_aux=None AttributeError are different code paths." + "reason": "A stale cache crash in GlmMoeDsa and a flash_attention.py None-handling crash are separate code paths." }, { "left": "issue:44188", "right": "issue:46421", "accept": false, - "reason": "Both mention compile/cache concepts, but one is attention-kernel divergence and the other is lazy cache initialization causing recompilation." + "reason": "Both mention compile/cache behavior, but one is attention-kernel divergence and the other is lazy cache init causing recompiles; not the same fix." }, { "left": "issue:45685", "right": "issue:46431", "accept": false, - "reason": "MPS histogram support and missing chat templates are unrelated." + "reason": "MPS histogram integer support and missing chat template validation are unrelated." }, { "left": "issue:45103", "right": "issue:46858", "accept": false, - "reason": "Auto-docstring annotation handling and cache.get_seq_length slice arithmetic are unrelated issues." + "reason": "auto_docstring annotation handling and cache length arithmetic are completely different concerns." }, { "left": "issue:43065", "right": "issue:46123", "accept": false, - "reason": "SAM3 pixel decoder implementation and partial-batch handling in MaskGenerationPipeline are not the same bug." + "reason": "SAM3 pixel-decoder internals and partial-batch masking logic do not describe the same defect." }, { "left": "issue:44297", "right": "issue:44623", "accept": false, - "reason": "Both involve save_pretrained, but tokenizer_class metadata mismatch and missing processor files are different concrete failures." + "reason": "Tokenizer save_pretrained metadata mismatch and processor save_pretrained missing files are both persistence-related, but they are different failure modes and artifacts." } ] }, @@ -13898,9 +13815,9 @@ "issue:28282|issue:30990", "issue:41628|issue:44933", "issue:43475|issue:46821", - "issue:45310|issue:45375", - "issue:44964|issue:45137", "issue:45230|issue:45397", + "issue:44964|issue:45137", + "issue:45310|issue:45375", "issue:43986|issue:46726", "issue:44869|issue:45356", "issue:44038|issue:44623", @@ -13912,20 +13829,20 @@ "issue:43723|issue:46326", "issue:45406|issue:46202" ], - "prepared_review_unit_hash": "f70d75bc3e652bfc64e49a53475072a64cd05b3d66283f53ff3e127911828a6f", + "prepared_review_unit_hash": "f8827e362c7661cdd0dcc697950e8ba2fdb86000c6f81eeef3ed42af8b8a0730", "component_budget": { "node_count": 500, "item_count": 500, "soft_pair_count": 1839, - "serialized_chars": 953913, - "estimated_input_tokens": 238479, - "estimated_eval_tokens": 477214 + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 29, - "serialized_chars": 36054, + "serialized_chars": 36055, "estimated_input_tokens": 9014, "estimated_eval_tokens": 18284 }, @@ -13934,186 +13851,186 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is noisy: the items mostly describe different Transformers regressions across unrelated models/subsystems. I did not find any soft pair that clearly refers to the same concrete bug or change, so I rejected them all as duplicates.", - "confidence": 0.83, - "canonical_issue_reason": "issue:46726 is the clearest, most actionable representative of the set: a specific AutoProcessor failure when a vision backend is missing, with a well-defined user-facing error path.", + "summary": "This cluster is mostly a loose grab-bag of Transformers v5 regressions and model-specific loading/behavior issues. The soft links are generally only broad topical similarity, not the same underlying bug or change, so they should stay separate.", + "confidence": 0.38, + "canonical_issue_reason": "Issue 43646 is the broadest, most representative regression report in the set: it frames a general Transformers 5.0 compatibility break around custom model initialization, which matches the cluster\u2019s overall theme better than the narrower model-specific reports.", "canonical_pr_reason": null, - "best_issue_reason": "issue:46726 is the best single issue to stand in for this cluster because it is concrete, narrowly scoped, and representative of the processor/vision-backend failure mode seen in the nearby issues.", + "best_issue_reason": "Issue 43646 is the best single issue representative because it is general, clearly regression-oriented, and spans the same v5 compatibility class as many of the other reports.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:44361", "right": "issue:44749", "accept": false, - "reason": "Different problems: tokenizer AttributeError vs. a performance regression after upgrade." + "reason": "Different failures: MLukeTokenizer task-time AttributeError vs a performance regression after upgrading Transformers. Same library, but not the same bug path." }, { "left": "issue:43994", "right": "issue:46726", "accept": false, - "reason": "Both touch vision/image flows, but one is incorrect outputs and the other is a missing-backend error; not the same bug." + "reason": "Both involve vision/image tooling, but one reports bad outputs from AutoModel/pipeline and the other is a missing-backend error in AutoProcessor. Different problems and fixes." }, { "left": "issue:44749", "right": "issue:45656", "accept": false, - "reason": "Unrelated: data filtering slowdown vs. DeepSpeed optimizer-step duplication." + "reason": "A data-processing slowdown after upgrade is unrelated to DeepSpeed causing optimizer.step to run twice." }, { "left": "issue:43701", "right": "issue:44749", "accept": false, - "reason": "Checkpoint resume key mismatch is unrelated to the slowdown report." + "reason": "Checkpoint key mismatch and slowdown after upgrade are unrelated symptoms with different code paths." }, { "left": "issue:44871", "right": "issue:46132", "accept": false, - "reason": "Gemma eos-token config inconsistency is unrelated to AttentionInterface behavior changes." + "reason": "Gemma eos_token_id inconsistency is unrelated to AttentionInterface.register changing callable behavior." }, { "left": "issue:44623", "right": "issue:45656", "accept": false, - "reason": "Processor save files issue is unrelated to deepspeed optimizer stepping twice." + "reason": "Missing files from processor.save_pretrained is a serialization bug, not a DeepSpeed optimizer-step issue." }, { "left": "issue:43646", "right": "issue:44162", "accept": false, - "reason": "AutoConfig missing fields with trust_remote_code is unrelated to ESM2 being broken." + "reason": "Custom model initialization breakage and ESM2 being broken are both broad regressions, but they are distinct model-specific failures." }, { "left": "issue:43441", "right": "issue:46752", "accept": false, - "reason": "Different code paths: FlashAttention failure vs. empty-chat-template IndexError." + "reason": "FlashAttention failure for Ministral-3 is unrelated to apply_chat_template crashing on an empty list." }, { "left": "issue:45375", "right": "issue:45478", "accept": false, - "reason": "Same model family, but one is a missing config field and the other is a from_pretrained loading error; not clearly the same concrete bug." + "reason": "Both are Qwen3.5/MoE related, but one is a missing config field under @strict while the other is a generic from_pretrained regression; not clearly the same concrete bug." }, { "left": "issue:42994", "right": "issue:43482", "accept": false, - "reason": "Quantized model saving failure is distinct from GGUF loading failure." + "reason": "Quantized model saving failure and GGUF loading failure are different quantization pathways with different failure modes." }, { "left": "issue:43475", "right": "issue:43643", "accept": false, - "reason": "SAM3 vision encoder attribute error is unrelated to AutoConfig trust_remote_code field loss." + "reason": "SAM 3 Video missing attribute and AutoConfig trust_remote_code field loss are unrelated issues." }, { "left": "issue:43232", "right": "issue:43673", "accept": false, - "reason": "Both are generation-related, but they affect different internals and symptoms." + "reason": "Both are generation-related, but sync_gpus kwarg handling and chunked_prefill cache availability are different bugs." }, { "left": "issue:44188", "right": "issue:46851", "accept": false, - "reason": "Attention-kernel divergence under torch.compile is unrelated to Qwen3.5 variable-length position_ids handling." + "reason": "Attention kernel divergence under torch.compile is unrelated to Qwen3.5 variable-length position_ids handling." }, { "left": "issue:28282", "right": "issue:30990", "accept": false, - "reason": "PyTorch import missing is not the same as a Sentence Transformers loading hang." + "reason": "PyTorch-not-found import error and Sentence Transformers hanging on load are different issues." }, { "left": "issue:41628", "right": "issue:44933", "accept": false, - "reason": "Both are import-related, but the missing AutoImageProcessor export and a nonexistent image_utils import are not the same bug." + "reason": "Both are import/export problems, but they concern different missing symbols and different modules; too coarse to merge." }, { "left": "issue:43475", "right": "issue:46821", "accept": false, - "reason": "SAM3 vision encoder attribute error is unrelated to TimesFM window_size handling." + "reason": "SAM 3 Video attribute error and TimesFM window_size AttributeError are different model-specific failures." }, { - "left": "issue:45310", - "right": "issue:45375", + "left": "issue:45230", + "right": "issue:45397", "accept": false, - "reason": "Both mention Qwen3.5 MoE, but one is a generic from_pretrained error and the other is a specific config-field omission; insufficient to merge." + "reason": "A generic bug report and Gemma-4 zero3 from_pretrained failure are unrelated." }, { "left": "issue:44964", "right": "issue:45137", "accept": false, - "reason": "Phi-4 multimodal loading failure is unrelated to DeepSpeed ZeRO3 deque exhaustion." + "reason": "Phi-4 multimodal loading failure and DeepSpeed ZeRO3 deque underflow are unrelated subsystems." }, { - "left": "issue:45230", - "right": "issue:45397", + "left": "issue:45310", + "right": "issue:45375", "accept": false, - "reason": "A placeholder bug report does not identify the same bug as gemma-4 zero3 from_pretrained." + "reason": "Both touch Qwen3.5/MoE, but one is a generic from_pretrained error and the other is a specific missing deepstack_visual_indexes field; not enough evidence they are the same underlying defect." }, { "left": "issue:43986", "right": "issue:46726", "accept": false, - "reason": "Both involve missing vision backends, but one is a video-model AutoProcessor crash and the other is an image-processor error message; not clearly the same concrete failure." + "reason": "AutoProcessor crash without torchvision and AutoProcessor reporting an unrecognized image processor are different missing-dependency / error-message behaviors." }, { "left": "issue:44869", "right": "issue:45356", "accept": false, - "reason": "Whisper timestamp decode crash is unrelated to Kimi tokenizer codec handling." + "reason": "Whisper timestamp decode crash and Kimi-K2.5 tokenizer codec regression are unrelated tokenizer issues." }, { "left": "issue:44038", "right": "issue:44623", "accept": false, - "reason": "Qwen3-VL-Moe loading issues are unrelated to processor.save_pretrained missing files." + "reason": "Qwen3-VL-Moe from_pretrained failures and processor.save_pretrained missing files are different operations and failure modes." }, { "left": "issue:39692", "right": "issue:46726", "accept": false, - "reason": "Documentation example errors and missing-backend AutoProcessor handling are different issues." + "reason": "Documentation example errors for SigLIP2 and a runtime missing-vision-backend crash are not the same bug." }, { "left": "issue:46463", "right": "issue:46877", "accept": false, - "reason": "T5 decoder-input validation messaging is unrelated to auto_docstring stdout diagnostics." + "reason": "Incorrect T5 decoder-input validation messaging and auto_docstring emitting diagnostics at import time are unrelated docstring/lint behaviors." }, { "left": "issue:44373", "right": "issue:46851", "accept": false, - "reason": "A docstring bug for position_ids is not the same as a runtime variable-length-input failure." + "reason": "A wrong docstring for position_ids is unrelated to a runtime bug in variable-length inputs using position_ids." }, { "left": "issue:42994", "right": "issue:43525", "accept": false, - "reason": "Quantized saving failure is unrelated to Llama4Config missing pad_token_id." + "reason": "Quantized model saving failure and Llama4Config missing pad_token_id are different issues." }, { "left": "issue:42371", "right": "issue:45092", "accept": false, - "reason": "TF32 settings guidance is unrelated to InternVL2 meta-initialization incompatibility." + "reason": "TF32 settings guidance and InternVL2 meta-initialization incompatibility are unrelated." }, { "left": "issue:43723", "right": "issue:46326", "accept": false, - "reason": "Tokenizer loading regression and continue_final_message behavior are different bugs." + "reason": "Tokenizer loading in v5 and continue_final_message behavior for mistral_common are different tokenizer bugs." }, { "left": "issue:45406", "right": "issue:46202", "accept": false, - "reason": "Gemma4Processor missing _tokenizer is unrelated to Gemma pipeline roundtrip on \u2581." + "reason": "Gemma4Processor missing _tokenizer and Gemma pipeline roundtrip failure on \u2581 are different bugs in different layers." } ] }, @@ -14218,20 +14135,20 @@ "issue:44623|issue:45310", "issue:42673|issue:46693" ], - "prepared_review_unit_hash": "8814bb0b6c7e4fa06d64dbe6caa5d66c9f9f83365b9183b1ea96fbcab637a7c9", + "prepared_review_unit_hash": "950da8a2926f00589bfc9e633d9d5b7ebb65c3add4a1fe373138a7466d403e88", "component_budget": { "node_count": 500, "item_count": 500, "soft_pair_count": 1839, - "serialized_chars": 953913, - "estimated_input_tokens": 238479, - "estimated_eval_tokens": 477214 + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 30, - "serialized_chars": 36540, + "serialized_chars": 36539, "estimated_input_tokens": 9135, "estimated_eval_tokens": 18526 }, @@ -14240,198 +14157,198 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster remains mostly heterogeneous. The only clear duplicate candidate is the exact-title repeat of the Qwen3ForCausalLM VRAM leak report (issue:42673 / issue:43704); the other soft edges are best treated as separate bugs or regressions.", - "confidence": 0.84, - "canonical_issue_reason": "issue:42673 is the best anchor because it is the more detailed and better-referenced report in the exact-title Qwen3ForCausalLM VRAM leak pair; issue:43704 matches the same title but is a shorter later repeat.", + "summary": "This cluster is mostly a grab bag of unrelated issue reports. The clearest duplicate pair is the repeated Qwen3ForCausalLM VRAM leak report (#42673/#43704); the rest do not look like the same underlying bug.", + "confidence": 0.95, + "canonical_issue_reason": "#42673 is the best canonical issue because it is the older report in the clearest duplicate pair with #43704, and it has more discussion activity plus the same title/symptom.", "canonical_pr_reason": null, - "best_issue_reason": "issue:42673 is the strongest global issue candidate because it has the richest discussion and references for the repeated Qwen3ForCausalLM VRAM leak report.", + "best_issue_reason": "#42673 is the strongest representative issue in this cluster: it matches #43704 on the Qwen3ForCausalLM VRAM leak symptom and has the richer discussion context.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43873", "right": "issue:44945", "accept": false, - "reason": "Different problems: quantization/offloading behavior vs incorrect outputs under pipeline parallelism." + "reason": "Quantization/offloading failure vs pipeline-parallel output correctness; different code paths and symptoms." }, { "left": "issue:44945", "right": "issue:45753", "accept": false, - "reason": "Pipeline-parallel output corruption and a Qwen3_5 image infinite loop are unrelated code paths." + "reason": "Pipeline-parallel output issue vs image-specific infinite loop; not the same bug." }, { "left": "issue:43784", "right": "issue:46399", "accept": false, - "reason": "Both are model import/runtime issues, but they involve different symbols, triggers, and fixes." + "reason": "Import-time NameError in sentence-transformers integration vs Qwen3.5 kernel registration crash; unrelated failures." }, { "left": "issue:41628", "right": "issue:44351", "accept": false, - "reason": "Both are import errors, but for different exported names and likely different missing-export causes." + "reason": "Different missing symbols (`AutoImageProcessor` vs `HybridCache`) and likely separate API breakages." }, { "left": "issue:44857", "right": "issue:46877", "accept": false, - "reason": "A CUDA float16 loss crash and auto_docstring stdout noise are unrelated." + "reason": "CUDA/AMP loss crash vs docstring lint logging at import time; no shared bug." }, { "left": "issue:45593", "right": "issue:46877", "accept": false, - "reason": "Aux-loss behavior in D-FINE is unrelated to docstring diagnostics at import time." + "reason": "Training loss/auxiliary-loss behavior vs auto_docstring stdout noise; unrelated." }, { "left": "issue:42673", "right": "issue:43704", "accept": true, - "reason": "Exact title match for the same Qwen3ForCausalLM VRAM leak report; the packet supports treating 43704 as a repeat of 42673's bug report." + "reason": "Same title and same symptom: Qwen3ForCausalLM VRAM leak when used in multiple dataloader threads; clearly the same underlying bug." }, { "left": "issue:44038", "right": "issue:45310", "accept": false, - "reason": "Both mention Qwen3 family loading, but the model variants and failure contexts differ enough to treat as separate bugs." + "reason": "Different Qwen MoE models, different version constraints, and different failure modes." }, { "left": "issue:36010", "right": "issue:41628", "accept": false, - "reason": "Importing GenerationMixin and importing AutoImageProcessor are distinct API breakages." + "reason": "Separate import errors for different exported symbols; not one bug." }, { "left": "issue:44222", "right": "issue:45310", "accept": false, - "reason": "FP8 save_pretrained failure is unrelated to Qwen3.5 from_pretrained loading errors." + "reason": "FP8 save_pretrained failure vs MoE from_pretrained error; different concrete code paths." }, { "left": "issue:43278", "right": "issue:45137", "accept": false, - "reason": "Embedding dtype drift during eval and a DeepSpeed ZeRO3 deque IndexError are unrelated." + "reason": "Embedding dtype drift during eval vs DeepSpeed ZeRO3 deque IndexError; unrelated." }, { "left": "issue:46202", "right": "issue:46701", "accept": false, - "reason": "Both involve Gemma, but one is a tokenizer roundtrip issue and the other is a pipeline/output wiring problem." + "reason": "Gemma token roundtrip issue vs DiffusionGemma pipeline failure; too different to merge." }, { "left": "issue:42994", "right": "issue:43824", "accept": false, - "reason": "Quantized model saving failure and an import error for Qwen2_5_VLForConditionalGeneration are unrelated." + "reason": "Quantized model saving failure vs missing Qwen2_5_VL class import; unrelated." }, { "left": "issue:45081", "right": "issue:45103", "accept": false, - "reason": "Mistral tokenizer regex patching and auto_docstring annotation handling are separate subsystems." + "reason": "Tokenizer regex patch crash vs auto_docstring annotation handling crash; different subsystems." }, { "left": "issue:42831", "right": "issue:44222", "accept": false, - "reason": "FineGrainedFP8 accuracy regression and FP8 save_pretrained failure are different bugs." + "reason": "FP8 accuracy regression vs FP8 model saving bug; same broad area but not the same bug." }, { "left": "issue:43232", "right": "issue:44792", "accept": false, - "reason": "Generation kwargs propagation after sync_gpus and a janus image-generation test failure are unrelated." + "reason": "Generation kwargs/sync_gpus bug vs Janus image-generation test failure; not the same failure." }, { "left": "issue:43504", "right": "issue:45561", "accept": false, - "reason": "A legacy config field breaking pretrained loading and an xdist race on captured_info.txt do not share a code path." + "reason": "Legacy-field model loading issue vs pytest-xdist race on a temp file; unrelated." }, { "left": "issue:42175", "right": "issue:44933", "accept": false, - "reason": "Missing TensorFlow in a torch extra install and a missing image_utils import are unrelated." + "reason": "Backend extra packaging problem vs missing image_utils import; different packaging/API issues." }, { "left": "issue:45561", "right": "issue:45600", "accept": false, - "reason": "A test file race condition and stale auto_mappings references are separate CI issues." + "reason": "File-race in patched testing utils vs stale auto_mappings references; separate CI/test problems." }, { "left": "issue:43976", "right": "issue:44933", "accept": false, - "reason": "Python version support breakage and a missing import from image_utils are unrelated." + "reason": "Python version compatibility issue vs missing import from image_utils; unrelated." }, { "left": "issue:44496", "right": "issue:45561", "accept": false, - "reason": "Model config loading failure and pytest-xdist file races are unrelated." + "reason": "Unrecognized model/config loading failure vs xdist file race; no shared code path." }, { "left": "issue:30333", "right": "issue:33290", "accept": false, - "reason": "MLflow reporting after failed training and a DeepSpeed OOM with Adafactor are not the same bug." + "reason": "MLflow reporting failure vs DeepSpeed OOM; different runtime issues." }, { "left": "issue:45405", "right": "issue:46877", "accept": false, - "reason": "A version pin/release mismatch is unrelated to auto_docstring import-time diagnostics." + "reason": "Dependency version bump complaint vs docstring import-time output; unrelated." }, { "left": "issue:30990", "right": "issue:35141", "accept": false, - "reason": "Sentence Transformers loading hang and token embedding reinitialization are different issues." + "reason": "Loading hang in Sentence Transformers vs embedding reinitialization after resize; different bugs." }, { "left": "issue:42898", "right": "issue:44568", "accept": false, - "reason": "Tokenizer cleanup behavior and missing BOS/EOS insertion are distinct tokenizer regressions." + "reason": "Tokenization cleanup behavior change vs add_special_tokens/BOS-EOS regression; related domain but different behavior." }, { "left": "issue:42175", "right": "issue:45712", "accept": false, - "reason": "TensorFlow backend packaging and leftover dummy classes are unrelated maintenance issues." + "reason": "Torch extra packaging issue vs leftover dummy classes leaking into dir(); different maintenance problems." }, { "left": "issue:43479", "right": "issue:46829", "accept": false, - "reason": "Multimodal config initialization and VideoPrism output-field mismatch are unrelated." + "reason": "Multimodal config default-init bug vs hidden_states field mismatch in video classification output; unrelated." }, { "left": "issue:46077", "right": "issue:46123", "accept": false, - "reason": "Causal-mask construction in Gemma4 and pipeline final-batch handling are different code paths." + "reason": "Bidirectional attention mask bug vs final partial batch dropping results; different pipelines." }, { "left": "issue:44623", "right": "issue:45310", "accept": false, - "reason": "Processor save_pretrained missing files and Qwen3.5 from_pretrained errors are unrelated." + "reason": "processor.save_pretrained missing files vs Qwen3.5 MoE from_pretrained error; not the same bug." }, { "left": "issue:42673", "right": "issue:46693", "accept": false, - "reason": "Qwen3ForCausalLM VRAM leakage in dataloader threads is unrelated to the FA2 long-context performance regression." + "reason": "VRAM leak in multi-threaded Qwen3ForCausalLM loading vs FA2 long-context performance regression; unrelated performance problems." } ] }, "evaluator_result": { "accept": false, - "feedback": "Rejecting because the rationale is not fully grounded in the packet: issue:42673 is marked closed, not open, and the packet only supports that issue:42673 and issue:43704 share the same title, not that they have the same full bug report. The rest of the soft-edge verdicts are generally conservative, but these factual overstatements are enough to reject." + "feedback": "The duplicate-pair judgment is mostly conservative, but the canonical issue reason overstates the packet in two places: it claims this is the \"only true duplicate pair\" and says #42673 has \"more references,\" neither of which is directly supported by the packet (both issues show 0 inbound references). Please soften those claims to what is actually evidenced, e.g. older issue with more discussion activity and matching title/symptom." }, "evaluator_used": true, "retried": true, @@ -14525,221 +14442,221 @@ "issue:22355|issue:28282", "issue:44060|issue:45125", "issue:44075|issue:46231", - "issue:44933|issue:45712", "issue:44360|issue:46762", + "issue:44933|issue:45712", "issue:42831|issue:44623", "issue:44297|issue:45656", "issue:42796|issue:43723", "issue:44222|issue:44623" ], - "prepared_review_unit_hash": "f7e8b0a3558a0a67ed2418d849302789b10e339f27ba74da6e4bfac98bef0082", + "prepared_review_unit_hash": "fae5594fb13d5113b23540c680f434d8fbb0e67d0c6a75ff44b6b66cdff312fd", "component_budget": { "node_count": 500, "item_count": 500, "soft_pair_count": 1839, - "serialized_chars": 953913, - "estimated_input_tokens": 238479, - "estimated_eval_tokens": 477214 + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 30, - "serialized_chars": 36256, - "estimated_input_tokens": 9064, - "estimated_eval_tokens": 18384 + "serialized_chars": 36263, + "estimated_input_tokens": 9066, + "estimated_eval_tokens": 18388 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is highly heterogeneous: it mixes unrelated tokenizer regressions, docs/example bugs, model-loading/import errors, optimizer issues, and multimodal model-specific problems. No soft pair looks like the same underlying fix, so all candidate duplicate edges should be rejected.", - "confidence": 0.97, - "canonical_issue_reason": "No single true canonical issue exists for this mixed cluster; if one must be chosen, issue 45812 is the strongest representative because it is a concrete, high-signal functional regression with external references.", + "summary": "No strong duplicate group emerges here: the items are mostly unrelated Transformers issues spanning tokenizers, multimodal configs, docs, model loading, and training utilities. The closest theme is general v5/tokenizer regression, but the soft pairs are still too different to merge as duplicates.", + "confidence": 0.86, + "canonical_issue_reason": "issue:42898 is the most umbrella-like representative of the recurring tokenizer/v5 regression theme, with a broad user-facing behavior change rather than a narrow model-specific symptom.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 45812 is the best overall anchor: it reports a broad, user-visible tokenizer regression with clear impact and more evidence than the smaller docs/import outliers.", + "best_issue_reason": "issue:42898 is the best single issue to anchor this noisy cluster because it is broad, clearly stated, and closest to the shared regression theme across several nearby candidates.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43479", "right": "issue:44869", "accept": false, - "reason": "Both are bug reports in transformers, but one is a multimodal config defaulting issue and the other is a Whisper decoding crash; different code paths and fixes." + "reason": "Different bugs: default multimodal config initialization vs a Whisper word-timestamp decode crash." }, { "left": "issue:43723", "right": "issue:44568", "accept": false, - "reason": "Both involve tokenizers, but one is an AutoTokenizer loading regression and the other is add_special_tokens behavior for a specific tokenizer; not the same bug." + "reason": "Both are v5/tokenizer-related, but one is an AutoTokenizer loading failure and the other is missing BOS/EOS insertion for a specific tokenizer." }, { "left": "issue:44016", "right": "issue:46231", "accept": false, - "reason": "A notebook syntax error and a documentation parameter-name mismatch are distinct documentation issues, not one underlying change." + "reason": "A notebook syntax error and a docs parameter-name mismatch are unrelated documentation issues." }, { "left": "issue:42898", "right": "issue:45812", "accept": false, - "reason": "Both are tokenizer regressions, but clean_up_tokenization_spaces behavior and Granite token ID generation are separate failure modes." + "reason": "Both involve tokenizers, but one is a general clean_up_tokenization_spaces v5 behavior change and the other is Granite token IDs being wrong." }, { "left": "issue:44625", "right": "issue:46877", "accept": false, - "reason": "Config propagation in Qwen3.5 and docstring import-time stdout noise are unrelated defects." + "reason": "Qwen3.5 label propagation and auto_docstring stdout logging are unrelated subsystems." }, { "left": "issue:42898", "right": "issue:44779", "accept": false, - "reason": "These are different tokenizer regressions affecting different models and behaviors, not one shared code-path bug." + "reason": "Both are tokenizer regressions, but they affect different models and different failure modes." }, { "left": "issue:39692", "right": "issue:46909", "accept": false, - "reason": "Both are broken examples, but they concern different documentation pages and different error types." + "reason": "Different broken documentation examples for different models; not the same underlying change." }, { "left": "issue:43479", "right": "issue:44079", "accept": false, - "reason": "One is multimodal config initialization, the other is ModelOutput key assignment; unrelated internals." + "reason": "Multimodal config defaults and ModelOutput key assignment are unrelated code paths." }, { "left": "issue:43874", "right": "issue:45092", "accept": false, - "reason": "Both touch vision/multimodal loading, but one is a missing processor method and the other is a remote-code/meta-init incompatibility; not the same bug." + "reason": "A missing image-processor method and remote-code/meta-init incompatibility are distinct multimodal loading issues." }, { "left": "issue:43784", "right": "issue:43824", "accept": false, - "reason": "Both are import errors, but they stem from different missing symbols and different packages/models." + "reason": "Different import errors from different causes: missing nn import vs missing Qwen2.5-VL export." }, { "left": "issue:27957", "right": "issue:30333", "accept": false, - "reason": "Flash Attention 2 support for XLMRoberta is unrelated to MLflow reporting on failed training jobs." + "reason": "Flash Attention model support and MLflow reporting on failed jobs are unrelated." }, { "left": "issue:44734", "right": "issue:45103", "accept": false, - "reason": "A server KV-cache indexing crash and an auto_docstring annotations crash are separate subsystems with different fixes." + "reason": "Serve/KV-cache indexing crash and auto_docstring annotation handling are unrelated." }, { "left": "issue:30333", "right": "issue:34689", "accept": false, - "reason": "MLflow job-status reporting and Llama 3.2 model loading are unrelated issues." + "reason": "MLflow failure reporting and Llama model loading regression are different problems." }, { "left": "issue:42222", "right": "issue:45200", "accept": false, - "reason": "VitPose model breakage and Gemma 4 mm_token_type_ids defaults are different multimodal-model bugs." + "reason": "VitPose model breakage and Gemma 4 mm_token_type_ids defaults are different model-specific bugs." }, { "left": "issue:42898", "right": "issue:43295", "accept": false, - "reason": "Both are v5 tokenizer/processor regressions, but one is token cleanup and the other is processor.tokenizer/image passing behavior." + "reason": "Both are v5 regressions involving tokenizer/processor behavior, but the concrete failures are different." }, { "left": "issue:42898", "right": "issue:45356", "accept": false, - "reason": "Both are tokenizer regressions, but Kimi-K2.5 codec handling differs from clean_up_tokenization_spaces behavior." + "reason": "Tokenizer regressions in v5, but one is clean_up_tokenization_spaces behavior and the other is Kimi codec/regex handling." }, { "left": "issue:44295", "right": "issue:45656", "accept": false, - "reason": "Reading position_ids after buffer registration and double optimizer.step under deepspeed are unrelated." + "reason": "Reading a registered buffer and deepspeed double optimizer steps are unrelated." }, { "left": "issue:43116", "right": "issue:44190", "accept": false, - "reason": "Different example scripts with different failures: empty multi-label outputs versus loading a local dataset." + "reason": "Two different example scripts with different failures; not the same bug." }, { "left": "issue:30333", "right": "issue:35141", "accept": false, - "reason": "MLflow reporting on failure and embedding reinitialization in post_init are unrelated bugs." + "reason": "Failed MLflow reporting and embedding reinitialization after resize are unrelated." }, { "left": "issue:45405", "right": "issue:45712", "accept": false, - "reason": "A dependency-version bump and leftover dummy-class cleanup are not the same issue." + "reason": "A dependency version bump problem and leftover dummy class cleanup are not the same issue." }, { "left": "issue:44222", "right": "issue:44297", "accept": false, - "reason": "Both mention save_pretrained, but one is FP8 MoE serialization and the other is tokenizer config metadata; different artifacts and causes." + "reason": "Both mention save_pretrained, but one is FP8 MoE model saving and the other is tokenizer config metadata." }, { "left": "issue:22355", "right": "issue:28282", "accept": false, - "reason": "Missing submodule import and missing PyTorch dependency are different import-time failures." + "reason": "Import errors, but one is missing transformers.onnx and the other is missing PyTorch entirely." }, { "left": "issue:44060", "right": "issue:45125", "accept": false, - "reason": "Tied-weights warning logic and missing _tp_plan for tensor parallelism are separate Qwen model issues." + "reason": "Different Qwen model bugs: a wrong tied-weights warning versus missing tensor-parallel plan metadata." }, { "left": "issue:44075", "right": "issue:46231", "accept": false, - "reason": "An optimizer implementation bug and a documentation typo are not the same underlying change." + "reason": "Optimizer behavior and documentation wording mismatch are unrelated." }, { - "left": "issue:44933", - "right": "issue:45712", + "left": "issue:44360", + "right": "issue:46762", "accept": false, - "reason": "A nonexistent image_utils import and leftover dummy classes leaking into dir(transformers) are different import-surface problems." + "reason": "A DSA indexer implementation concern and Minimax attention numerical differences are different algorithmic issues." }, { - "left": "issue:44360", - "right": "issue:46762", + "left": "issue:44933", + "right": "issue:45712", "accept": false, - "reason": "The DSA indexer/ReLU discussion and Minimax numerical differences are unrelated." + "reason": "Both touch exports/symbol visibility, but one is a nonexistent import and the other is leftover dummy classes in a different module." }, { "left": "issue:42831", "right": "issue:44623", "accept": false, - "reason": "FineGrainedFP8 accuracy issues and processor.save_pretrained missing files are distinct bugs." + "reason": "Fine-grained FP8 accuracy and processor.save_pretrained file omission are unrelated." }, { "left": "issue:44297", "right": "issue:45656", "accept": false, - "reason": "Tokenizer metadata serialization and deepspeed double-step behavior do not share a concrete code path." + "reason": "Tokenizer save metadata and deepspeed optimizer stepping are unrelated code paths." }, { "left": "issue:42796", "right": "issue:43723", "accept": false, - "reason": "A specific tokenizer BPE-marker bug and a generic AutoTokenizer loading regression are different failures." + "reason": "Both are tokenizer-related, but one is BPE marker handling for a specific model and the other is a generic AutoTokenizer loading issue." }, { "left": "issue:44222", "right": "issue:44623", "accept": false, - "reason": "Both involve saving, but one is FP8 MoE weights and the other is processor file export; not mergeable as one fix." + "reason": "Both involve persistence APIs, but one is FP8 MoE model saving and the other is processor file output; not the same bug." } ] }, @@ -14829,8 +14746,8 @@ "issue:43232|issue:45834", "issue:44295|issue:44623", "issue:43475|issue:46305", - "issue:46209|issue:46431", "issue:44263|issue:44360", + "issue:46209|issue:46431", "issue:46305|issue:46655", "issue:44263|issue:44373", "issue:30990|issue:38175", @@ -14840,217 +14757,217 @@ "issue:42907|issue:46050", "issue:44279|issue:46328", "issue:45381|issue:46305", - "issue:44483|issue:45406", - "issue:44261|issue:45092" + "issue:44261|issue:45092", + "issue:44483|issue:45406" ], - "prepared_review_unit_hash": "973baf1c28250ac64d0d0fd10ce17284eaa6e5f07606c92e94fd4589b44de977", + "prepared_review_unit_hash": "fcb8986b7140d2eae39939c643fc793a6d2518b1657b67aaeeb3924c1c3f4555", "component_budget": { "node_count": 500, "item_count": 500, "soft_pair_count": 1839, - "serialized_chars": 953913, - "estimated_input_tokens": 238479, - "estimated_eval_tokens": 477214 + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 30, - "serialized_chars": 35668, - "estimated_input_tokens": 8917, - "estimated_eval_tokens": 18090 + "serialized_chars": 35671, + "estimated_input_tokens": 8918, + "estimated_eval_tokens": 18092 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is a loose mix of mostly unrelated Transformers v5 regression and model-specific bug reports. The soft-similarity edges mostly reflect shared vocabulary (save/load, config, vision, tokenizer) rather than the same underlying defect, so none should be merged.", - "confidence": 0.77, - "canonical_issue_reason": "issue:43644 is the broadest concrete regression in the set: non-persistent buffers being filled with junk in Transformers 5.0, which has clear system-level impact and best matches the cluster\u2019s common v5-regression theme.", + "summary": "This set is a loose collection of mostly unrelated Transformers issues. The soft edges are driven by surface-level similarity around v5 regressions, config/save/load problems, and model-specific failures, but the underlying bugs are distinct in almost every pair.", + "confidence": 0.92, + "canonical_issue_reason": "issue:45071 is the best cluster anchor because it is the broadest library-level regression here (core PretrainedConfig/type-checking behavior in v5), whereas most other issues are narrower model- or feature-specific failures.", "canonical_pr_reason": null, - "best_issue_reason": "issue:43644 is the strongest representative issue here because it describes a concrete, cross-cutting runtime regression rather than a narrow model-specific symptom.", + "best_issue_reason": "issue:45071 is the strongest representative issue for the cluster\u2019s apparent theme of v5 regression breakage: it is user-facing, cross-cutting, and more general than the model-specific or one-off API errors in the rest of the set.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:44448", "right": "issue:46132", "accept": false, - "reason": "Different failures: Pegasus output regression vs AttentionInterface.register behavior change." + "reason": "Both mention v4/v5 behavior, but one is Pegasus output drift and the other is AttentionInterface.register semantics; different code paths and fixes." }, { "left": "issue:45245", "right": "issue:46132", "accept": false, - "reason": "Unrelated bugs: category-limit RuntimeError vs attention registration semantics." + "reason": "Category-count overflow is a data/encoding limit issue, not an attention-interface behavior change." }, { "left": "issue:43479", "right": "issue:43644", "accept": false, - "reason": "Config defaults when None is a different code path from junk non-persistent buffers." + "reason": "Both involve config/init behavior, but one is multimodal config defaults and the other is non-persistent buffer initialization; not the same bug." }, { "left": "issue:41628", "right": "issue:42757", "accept": false, - "reason": "Both are import errors, but for different symbols/modules and likely different dependency causes." + "reason": "Two separate import errors from different packages/symbols; only superficially similar as dependency breakage." }, { "left": "issue:44373", "right": "issue:44485", "accept": false, - "reason": "Docstring correction is not the same as a RoPE implementation bug." + "reason": "Wrong docstring for position_ids is documentation-only; GLM-5 RoPE is a real implementation issue." }, { "left": "issue:44297", "right": "issue:45310", "accept": false, - "reason": "Tokenizer metadata mismatch on save_pretrained is distinct from Qwen3.5 from_pretrained/MoE loading failure." + "reason": "One is tokenizer save_pretrained metadata mismatch, the other is a from_pretrained loading error; related area but different failures." }, { "left": "issue:46132", "right": "issue:46710", "accept": false, - "reason": "Behavior of a registered function is unrelated to an accuracy regression after upgrade." + "reason": "Both are regressions, but one is an attention registration behavior change and the other is an accuracy regression in a specific model after upgrade." }, { "left": "issue:45071", "right": "issue:46082", "accept": false, - "reason": "PretrainedConfig type-checking regression and LlamaConfig head_dim validation are separate config issues." + "reason": "Both touch config validation, but one is type-checking behavior and the other is explicit head_dim acceptance rules; not the same defect." }, { "left": "issue:44222", "right": "issue:44484", "accept": false, - "reason": "FP8 save_pretrained bug is not the same as the max_shard_size default question." + "reason": "FP8 moe save_pretrained corruption and max_shard_size defaults are unrelated save_pretrained concerns." }, { "left": "issue:43723", "right": "issue:44977", "accept": false, - "reason": "Tokenizer loading failure and flash-attention generation failure are different bugs." + "reason": "Tokenizer loading failure and flash-attention generation failure affect different subsystems and different models." }, { "left": "issue:43122", "right": "issue:46209", "accept": false, - "reason": "Tokenization regression is unrelated to FP8 inference failure on B200." + "reason": "Tokenization drift across versions and FP8 inference failure on B200 are unrelated issues." }, { "left": "issue:44263", "right": "issue:45092", "accept": false, - "reason": "GlmMoeDsaIndexer split behavior and remote-code meta-init incompatibility are different problems." + "reason": "torch.split return values in a DSA indexer are unrelated to remote-code checkpoint incompatibility with meta initialization." }, { "left": "issue:44190", "right": "issue:46082", "accept": false, - "reason": "Dataset-loading script failure is unrelated to LlamaConfig head_dim rejection." + "reason": "A local dataset loading script failure is not the same bug as LlamaConfig head_dim validation." }, { "left": "issue:43232", "right": "issue:46424", "accept": false, - "reason": "Generation kwargs sync bug is not the same as cache max_length handling." + "reason": "Generation kwargs update after sync_gpus and cache max length handling are different generation/cache behaviors." }, { "left": "issue:43232", "right": "issue:45834", "accept": false, - "reason": "Generation state update bug and Kosmos2.5 long-input index error are different code paths." + "reason": "The generation bug and the Kosmos2.5 OCR index error are unrelated." }, { "left": "issue:44295", "right": "issue:44623", "accept": false, - "reason": "Reading a registered buffer and processor save_pretrained missing files are unrelated." + "reason": "A position_ids buffer access error is not the same as processor.save_pretrained missing files." }, { "left": "issue:43475", "right": "issue:46305", "accept": false, - "reason": "SAM3 missing attribute and MoonshineStreaming off-by-one masking do not match." + "reason": "SAM3 missing an attribute and MoonshineStreaming off-by-one masking are different model-specific bugs." }, { - "left": "issue:46209", - "right": "issue:46431", + "left": "issue:44263", + "right": "issue:44360", "accept": false, - "reason": "FP8 inference on B200 is unrelated to missing chat template support." + "reason": "Both mention the DSA indexer, but one is about split return values and the other about a missing ReLU; different fixes." }, { - "left": "issue:44263", - "right": "issue:44360", + "left": "issue:46209", + "right": "issue:46431", "accept": false, - "reason": "Both involve DSA indexer code, but one is torch.split return values and the other is a missing ReLU; not the same bug." + "reason": "FP8 inference on B200 and missing chat template are unrelated issues." }, { "left": "issue:46305", "right": "issue:46655", "accept": false, - "reason": "Sliding-window off-by-one and video padding miscalculation are different defects." + "reason": "Both are video-processing bugs, but one is a sliding-window off-by-one and the other is a padding calculation error; not the same defect." }, { "left": "issue:44263", "right": "issue:44373", "accept": false, - "reason": "Indexer return values vs a wrong docstring are not the same underlying issue." + "reason": "Indexer split behavior and a wrong docstring for position_ids are unrelated." }, { "left": "issue:30990", "right": "issue:38175", "accept": false, - "reason": "Sentence Transformers loading hang and SigLIP2 zero probabilities are unrelated." + "reason": "Sentence Transformers loading hangs and SigLIP2 zero probabilities are different model-level problems." }, { "left": "issue:42831", "right": "issue:45310", "accept": false, - "reason": "FineGrainedFP8 accuracy loss and Qwen3.5 MoE from_pretrained error are distinct." + "reason": "FP8 accuracy regression and Qwen3.5 from_pretrained failure are distinct issues." }, { "left": "issue:45092", "right": "issue:45357", "accept": false, - "reason": "Meta-init incompatibility for old InternVL2 checkpoints is different from incorrect visual encoder keys on save_pretrained." + "reason": "InternVL2 meta-init incompatibility and Qwen3.5 save_pretrained key mismatch involve different models and different failure modes." }, { "left": "issue:46369", "right": "issue:46909", "accept": false, - "reason": "Memory overflow in sam3-litetext is unrelated to a broken documentation example." + "reason": "SAM3 memory overflow and a broken DIA documentation example are unrelated." }, { "left": "issue:42907", "right": "issue:46050", "accept": false, - "reason": "Dequantized save failure and sinq save/load breakage are both quantization-related but not the same concrete bug." + "reason": "Both involve save/load of quantized models, but one is dequantized Ministrals/Devstrals saving and the other is sinq round-trip breakage; different quantization paths." }, { "left": "issue:44279", "right": "issue:46328", "accept": false, - "reason": "General dependency issue is unrelated to SequenceFeatureExtractor.pad performance behavior." + "reason": "A dependency problem and a NumPy-to-list performance issue are unrelated." }, { "left": "issue:45381", "right": "issue:46305", "accept": false, - "reason": "Wrong vision_position_ids and sliding-window masking are different model behaviors." + "reason": "Both concern video position/masking, but they affect different models and code paths; not the same underlying bug." }, { - "left": "issue:44483", - "right": "issue:45406", + "left": "issue:44261", + "right": "issue:45092", "accept": false, - "reason": "Chat-completions request rejection and Gemma4Processor missing _tokenizer are separate serving failures." + "reason": "MLA q_a_layernorm epsilon/config precision issue is unrelated to InternVL2 meta-initialization incompatibility." }, { - "left": "issue:44261", - "right": "issue:45092", + "left": "issue:44483", + "right": "issue:45406", "accept": false, - "reason": "MLA q_a_layernorm precision issue is unrelated to InternVL2 meta-initialization incompatibility." + "reason": "A chat/completions request rejection and a Gemma4Processor missing _tokenizer attribute are different serving failures." } ] }, @@ -15151,225 +15068,212 @@ "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:41628|issue:45278" + "issue:43976|issue:44945", + "issue:42907|issue:43937" ], - "prepared_review_unit_hash": "7c3ea2ff26ebbd5e576055db34f2e07eea06cd11e2006583fc3bbd58f212a976", + "prepared_review_unit_hash": "64bc82c6a86de145aab6f5b4c345da8a7d8f30c26aeec00acdc0db8852a7f9ed", "component_budget": { "node_count": 500, "item_count": 500, "soft_pair_count": 1839, - "serialized_chars": 953913, - "estimated_input_tokens": 238479, - "estimated_eval_tokens": 477214 + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 }, "budget": { "node_count": 48, "item_count": 48, - "soft_pair_count": 31, - "serialized_chars": 36355, - "estimated_input_tokens": 9089, - "estimated_eval_tokens": 18434 + "soft_pair_count": 30, + "serialized_chars": 36050, + "estimated_input_tokens": 9013, + "estimated_eval_tokens": 18282 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Most candidates are only loosely related by broad subsystem (tokenizers/imports/model loading) rather than the same underlying bug. I rejected all soft edges as duplicates and selected the broad Transformers 5 regression issue as the best canonical issue for this mixed cluster.", - "confidence": 0.74, - "canonical_issue_reason": "issue:45278 is the broadest, most representative regression here: a version-upgrade breakage causing multiple import failures, which makes it a better umbrella canonical than the more narrowly scoped bugs.", + "summary": "This cluster is heterogeneous: most items are distinct regressions or feature-specific bugs, with a few superficially similar import/tokenizer issues but no clear duplicate pair. The best umbrella issue is the broad version-upgrade import breakage report.", + "confidence": 0.82, + "canonical_issue_reason": "issue:45278 is the broadest and most representative issue in the set, covering multiple import failures after upgrading transformers, unlike the more narrow or unrelated reports.", "canonical_pr_reason": null, - "best_issue_reason": "issue:45278 has the clearest cross-cutting impact and best fits as a cluster anchor for widespread import/API breakage in the Transformers 5 transition.", + "best_issue_reason": "issue:45278 is the most useful global issue candidate because it describes a broader, user-visible regression pattern that can subsume several import-related symptoms.", "best_pr_reason": null, "soft_edge_verdicts": [ - { - "left": "issue:22355", - "right": "issue:27957", - "accept": false, - "reason": "Different problems: missing module import versus a Flash Attention 2 model-specific issue." - }, { "left": "issue:44829", "right": "issue:46002", "accept": false, - "reason": "Both concern model behavior, but one is FA3 training degeneration and the other is NaNs after checkout/init; not the same code path." + "reason": "Different problems: flash_attention_3 training collapse vs NaNs after checkout/init." }, { "left": "issue:43644", "right": "issue:46305", "accept": false, - "reason": "Unrelated bugs: non-persistent buffer corruption versus a streaming mask off-by-one." + "reason": "Unrelated bugs: non-persistent buffer junk values vs a sliding-window off-by-one in audio masking." }, { "left": "issue:42831", "right": "issue:44038", "accept": false, - "reason": "Different models and failure modes: FP8 accuracy regression versus Qwen3-VL-Moe compatibility/loading issue." + "reason": "Different symptoms and code paths: FineGrainedFP8 accuracy issue vs Qwen3-VL-Moe bug." }, { "left": "issue:41628", "right": "issue:43824", "accept": false, - "reason": "Both are import errors, but they are for different missing symbols and likely different breakages." + "reason": "Both are import errors, but for different missing symbols and likely different export/regression causes." }, { "left": "issue:43408", "right": "issue:46402", "accept": false, - "reason": "Different SAM3 issues: wrong model type warning versus missing text encoder weights." + "reason": "Different SAM3-related issues: model-type warning vs missing text encoder weights." }, { "left": "issue:42831", "right": "issue:45753", "accept": false, - "reason": "No shared underlying bug; FP8 accuracy and an infinite-loop image case are unrelated." + "reason": "Accuracy regression in FP8 is not the same as an infinite-loop bug on a specific image." }, { "left": "issue:22355", "right": "issue:28218", "accept": false, - "reason": "Import error and tokenizer spacing behavior are unrelated." + "reason": "Module import failure vs tokenizer spacing behavior; no shared bug." }, { "left": "issue:39401", "right": "issue:44568", "accept": false, - "reason": "Both are tokenizer bugs, but they affect different tokenizers and different behaviors (offset mapping vs special tokens)." + "reason": "Both tokenizer bugs, but one is offset_mapping for Qwen3 and the other is BOS/EOS insertion for mdeberta-v3-base." }, { "left": "issue:42886", "right": "issue:43066", "accept": false, - "reason": "Offline cache loading and decoder-type mismatch are different tokenizer issues." + "reason": "Offline cache loading failure is unrelated to a wrong tokenizer decoder type." }, { "left": "issue:42886", "right": "issue:42994", "accept": false, - "reason": "Tokenizer cache loading is unrelated to quantized model saving failure." + "reason": "Tokenizer cache/offline loading and quantized model saving are different failure modes." }, { "left": "issue:42757", "right": "issue:45278", "accept": false, - "reason": "Both involve imports, but one is a specific huggingface_hub symbol issue and the other is broad Transformers 5 breakage." + "reason": "44257-like single missing import is not the same concrete bug as the broader multi-import regression; too broad to merge as duplicates." }, { "left": "issue:27957", "right": "issue:30990", "accept": false, - "reason": "Model Flash Attention support and Sentence Transformers loading hang are not the same bug." + "reason": "Flash Attention 2 on XLMRoberta and Sentence Transformers loading hang are unrelated." }, { "left": "issue:44188", "right": "issue:44373", "accept": false, - "reason": "Attention-kernel divergence and a docstring typo are unrelated." + "reason": "Attention-kernel divergence and a wrong docstring are different kinds of issues." }, { "left": "issue:43408", "right": "issue:46369", "accept": false, - "reason": "Both mention SAM3, but one is a class mismatch warning and the other is a memory overflow; not mergeable as one fix." + "reason": "SAM3 model-type warning is unrelated to 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." + "reason": "Generation kwargs handling after sync_gpus is unrelated to D-FINE auxiliary-loss behavior." }, { "left": "issue:42796", "right": "issue:43122", "accept": false, - "reason": "Both are tokenization-related, but one is a model-specific BPE marker bug and the other is a version-to-version tokenization change." + "reason": "Ministral BPE marker handling differs from a version-to-version tokenization drift report." }, { "left": "issue:43646", "right": "issue:46002", "accept": false, - "reason": "Custom model initialization breakage and NaNs in classification heads are different defects." + "reason": "Custom model initialization breakage is not the same as NaNs in classification heads." }, { "left": "issue:45381", "right": "issue:46909", "accept": false, - "reason": "Video position IDs bug and a broken documentation example are unrelated." + "reason": "Qwen2.5-VL video position IDs and a broken documentation example for DIA are unrelated." }, { "left": "issue:43540", "right": "issue:46909", "accept": false, - "reason": "Qwen3OmniMoe video-processing failure is unrelated to a DIA docs example." + "reason": "Video-input ValueError in Qwen3OmniMoe is unrelated to a documentation example bug." }, { "left": "issue:30990", "right": "issue:33357", "accept": false, - "reason": "Sentence Transformers loading hang and MacOS bus error are different runtime failures." + "reason": "Loading hang for Sentence Transformers and a MacOS bus error in CLIP are different failures." }, { "left": "issue:42907", "right": "issue:44945", "accept": false, - "reason": "Dequantized model saving and pipeline-parallelism output correctness are unrelated." + "reason": "Dequantized Ministral save failure is unrelated to incorrect outputs under pipeline parallelism." }, { "left": "issue:43576", "right": "issue:43784", "accept": false, - "reason": "Broken env command and a sentence-transformers import NameError are different regressions." + "reason": "Broken env command and a NameError importing sentence-transformers are different issues." }, { "left": "issue:44295", "right": "issue:44297", "accept": false, - "reason": "Position_ids buffer access and tokenizer_class metadata mismatch are unrelated." + "reason": "Position_ids buffer read error is unrelated to tokenizer_class mismatch on save_pretrained." }, { "left": "issue:42371", "right": "issue:44263", "accept": false, - "reason": "TF32 settings guidance and torch.split return handling in GLM-MoE are unrelated." + "reason": "TF32 settings warning and torch.split return values in GlmMoeDsaIndexer do not share a bug." }, { "left": "issue:39401", "right": "issue:44488", "accept": false, - "reason": "Different tokenizer failures for different models; not the same underlying bug." + "reason": "Qwen3 offset_mapping bug and failure to load cjvt/sleng-bert are different tokenizer/model issues." }, { "left": "issue:42907", "right": "issue:44964", "accept": false, - "reason": "Quantized model saving failure and Phi-4 multimodal loading failure are unrelated." - }, - { - "left": "issue:43976", - "right": "issue:44945", - "accept": false, - "reason": "Python version compatibility and pipeline-parallel output corruption are not the same issue." + "reason": "Quantized model saving failure is unrelated to a multimodal model load failure." }, { "left": "issue:46172", "right": "issue:46762", "accept": false, - "reason": "Different models and different numerical discrepancies; not one shared bug." + "reason": "DeepSeek-V4 comb transpose bug and Minimax attention numerical drift are not the same code-path problem." }, { - "left": "issue:42907", - "right": "issue:43937", + "left": "issue:43976", + "right": "issue:44945", "accept": false, - "reason": "Quantized saving and invalid GenerationConfig are unrelated." + "reason": "Python version compatibility issue is unrelated to pipeline-parallelism output corruption." }, { - "left": "issue:41628", - "right": "issue:45278", + "left": "issue:42907", + "right": "issue:43937", "accept": false, - "reason": "Specific AutoImageProcessor import failure and broad import regressions overlap only at a high level, not the same underlying defect." + "reason": "Quantized save failure is unrelated to invalid GenerationConfig handling for GLM-5." } ] }, @@ -15398,6 +15302,7 @@ "issue:29159", "issue:36296", "issue:39401", + "issue:41628", "issue:42175", "issue:42757", "issue:42907", @@ -15411,7 +15316,6 @@ "issue:43756", "issue:43784", "issue:43824", - "issue:43937", "issue:44038", "issue:44060", "issue:44117", @@ -15431,6 +15335,7 @@ "issue:44945", "issue:45200", "issue:45216", + "issue:45278", "issue:45290", "issue:45478", "issue:45600", @@ -15443,8 +15348,9 @@ ], "soft_pairs": [ "issue:43756|issue:44060", - "issue:43784|issue:45600", + "issue:41628|issue:45278", "issue:44117|issue:44704", + "issue:43784|issue:45600", "issue:39401|issue:44521", "issue:44704|issue:45290", "issue:44625|issue:44829", @@ -15469,211 +15375,210 @@ "issue:42175|issue:44938", "issue:43493|issue:45753", "issue:46231|issue:46877", - "issue:44188|issue:46328", - "issue:43937|issue:44485" + "issue:44188|issue:46328" ], - "prepared_review_unit_hash": "365ed03fd78b85f2e009f61feedcd4d2b83573d36826932c9d268a95c1234952", + "prepared_review_unit_hash": "d8638ad84b6d7b2e8de3643f450eb3f02de8483ca5b73400b546873670c8db49", "component_budget": { "node_count": 500, "item_count": 500, "soft_pair_count": 1839, - "serialized_chars": 953913, - "estimated_input_tokens": 238479, - "estimated_eval_tokens": 477214 + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 }, "budget": { - "node_count": 47, - "item_count": 47, + "node_count": 48, + "item_count": 48, "soft_pair_count": 29, - "serialized_chars": 35674, - "estimated_input_tokens": 8919, - "estimated_eval_tokens": 18094 + "serialized_chars": 36237, + "estimated_input_tokens": 9060, + "estimated_eval_tokens": 18376 }, "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 links are broad topical similarities (tokenizers, save/load, model-family bugs, docs), but none look like the same underlying defect or a PR that should be merged together.", - "confidence": 0.89, - "canonical_issue_reason": "No single canonical issue fits well: the items span unrelated tokenizer, model-loading, training, packaging, and documentation bugs, and the soft edges are only thematic.", + "summary": "This cluster is mostly a grab bag of unrelated Transformers regressions and import/behavior issues. None of the soft pairs look like the same underlying bug or change, so all are rejected. The broadest representative issue is the general import-regression report.", + "confidence": 0.82, + "canonical_issue_reason": "issue:45278 is the best umbrella-style representative because it describes a broad upgrade-related import regression, which is the closest thing to a cluster-wide theme here.", "canonical_pr_reason": null, - "best_issue_reason": "If a representative anchor is needed, issue:43122 is the closest because it is a generic tokenizer-regression report, but it still does not subsume the rest of the cluster.", + "best_issue_reason": "issue:45278 is the most general and high-signal issue in the set, but it still only covers one narrow family of failures rather than the whole cluster.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43756", "right": "issue:44060", "accept": false, - "reason": "Different models and different problems: RoPE layer count vs tied-weights warning." + "reason": "Different models and failure modes: RoPE layer counting vs a tied-weights warning in Qwen3-Next." }, { - "left": "issue:43784", - "right": "issue:45600", + "left": "issue:41628", + "right": "issue:45278", "accept": false, - "reason": "Both are import-related, but they fail in different places and involve different code paths." + "reason": "Both are import problems, but they are different symbols and likely different causes; too broad to merge." }, { "left": "issue:44117", "right": "issue:44704", "accept": false, - "reason": "Tokenizer mapping fallback bug vs AutoProcessor/cached_file kwargs handling; not the same defect." + "reason": "One is TOKENIZER_MAPPING_NAMES returning None; the other is AutoProcessor not forwarding kwargs. Different code paths." + }, + { + "left": "issue:43784", + "right": "issue:45600", + "accept": false, + "reason": "Sentence-transformers import NameError is unrelated to stale auto_mappings entries for Sam3LiteText." }, { "left": "issue:39401", "right": "issue:44521", "accept": false, - "reason": "Both involve tokenization, but one is offset mapping for Qwen3 and the other is assistant mask generation for multimodal chat templates." + "reason": "Tokenizer offset_mapping bug vs multimodal assistant_mask generation bug; not the same underlying defect." }, { "left": "issue:44704", "right": "issue:45290", "accept": false, - "reason": "Different APIs and symptoms: cached_file kwargs forwarding vs chat template crashing on tool-call messages." + "reason": "Cached-file kwargs forwarding and chat-template tool-call crashing are separate API paths." }, { "left": "issue:44625", "right": "issue:44829", "accept": false, - "reason": "Config propagation bug vs flash-attention training degeneration; unrelated code paths." + "reason": "Config propagation for num_labels is unrelated to a flash_attention_3 training degeneration bug." }, { "left": "issue:44360", "right": "issue:44485", "accept": false, - "reason": "Completely unrelated: DSA indexer behavior vs GLM-5 RoPE implementation." + "reason": "DSA indexer/ReLU discussion and GLM-5 RoPE implementation are different model/kernel issues." }, { "left": "issue:43408", "right": "issue:43475", "accept": false, - "reason": "Same broad SAM 3 area, but one is a model-type warning and the other is a missing output attribute; not the same bug." + "reason": "Both mention SAM3, but one is a model-type warning and the other is a missing output attribute." }, { "left": "issue:43122", "right": "issue:44521", "accept": false, - "reason": "Tokenizer version regression vs multimodal chat-template assistant masks; only broad tokenizer-related similarity." + "reason": "Tokenizer version drift is not the same as assistant_masks becoming all zero for multimodal inputs." }, { "left": "issue:44038", "right": "issue:44297", "accept": false, - "reason": "Qwen3-VL-Moe regression vs tokenizer save_pretrained metadata mismatch; different failures." + "reason": "Qwen3-VL-Moe import/compatibility issue vs tokenizer save_pretrained metadata mismatch; different bugs." }, { "left": "issue:43122", "right": "issue:46326", "accept": false, - "reason": "Both are tokenizer regressions, but they affect different tokenizer implementations and different APIs." + "reason": "Different tokenizer regressions affecting different behaviors and APIs." }, { "left": "issue:42907", "right": "issue:43452", "accept": false, - "reason": "Different model-loading/saving issues with no shared concrete code path." + "reason": "Saving dequantized model weights and gguf_file loading breakage are distinct workflows." }, { "left": "issue:44038", "right": "issue:44295", "accept": false, - "reason": "Unrelated: model/version-specific Qwen3-VL-Moe issue vs position_ids buffer access bug." + "reason": "Qwen3-VL-Moe regression and position_ids buffer access error do not describe the same bug." }, { "left": "issue:45216", "right": "issue:45478", "accept": false, - "reason": "Same model family and version context, but one is save_pretrained corruption and the other is from_pretrained failure; not clearly one bug." + "reason": "Related family and model name, but save_pretrained regression and from_pretrained error are different concrete failures." }, { "left": "issue:22355", "right": "issue:29159", "accept": false, - "reason": "Unrelated: missing transformers.onnx module vs slow/fast tokenizer inconsistency." + "reason": "Missing transformers.onnx import is unrelated to slow/fast tokenizer inconsistency." }, { "left": "issue:39401", "right": "issue:46614", "accept": false, - "reason": "Both tokenizer bugs, but different tokenizer classes and different state handling." + "reason": "Qwen3 offset_mapping and stale phonemizer language state are unrelated tokenization features." }, { "left": "issue:44038", "right": "issue:44222", "accept": false, - "reason": "Only a loose model-family overlap; save_pretrained FP8 issue is not the same as the Qwen3-VL-Moe regression." + "reason": "Both involve Qwen3-VL-Moe, but one is import/model wiring and the other is FP8 save_pretrained behavior." }, { "left": "issue:43824", "right": "issue:44351", "accept": false, - "reason": "Different missing symbols from transformers; same symptom type, different underlying issues." + "reason": "Different missing imports from different modules and symbols." }, { "left": "issue:43116", "right": "issue:45200", "accept": false, - "reason": "Different example/training behaviors with no shared concrete failure mode." + "reason": "Example-script multi-label output bug and Gemma 4 token-type defaults are separate issues." }, { "left": "issue:43122", "right": "issue:44779", "accept": false, - "reason": "Both are tokenizer regressions, but they concern different models and different tokenizer logic." + "reason": "Both are tokenizer regressions, but for different models and symptoms; not mergeable as one fix." }, { "left": "issue:42757", "right": "issue:44351", "accept": false, - "reason": "Both are import errors, but they are for different symbols and likely different dependency/version problems." + "reason": "Both are import errors, but one is huggingface_hub and the other is transformers; different broken exports." }, { "left": "issue:36296", "right": "issue:43116", "accept": false, - "reason": "Tensor parallel training bug vs example script returning empty classification results; unrelated." + "reason": "Tensor parallel training and multi-label classification output are unrelated." }, { "left": "issue:43278", "right": "issue:44945", "accept": false, - "reason": "Dtype mismatch in evaluation vs pipeline-parallel output corruption; different subsystems." + "reason": "Dtype mismatch during evaluate is unrelated to incorrect pipeline-parallel outputs." }, { "left": "issue:22355", "right": "issue:29127", "accept": false, - "reason": "Completely unrelated issues: missing onnx module vs layoutlmv3 error-message clarity." + "reason": "Missing module import and LayoutLMv3 error-message clarity are unrelated." }, { "left": "issue:42175", "right": "issue:44938", "accept": false, - "reason": "Packaging/backend selection bug vs Python 3.14 load failure; different causes." + "reason": "Backend packaging omission and Python 3.14 load failure are different installation/runtime problems." }, { "left": "issue:43493", "right": "issue:45753", "accept": false, - "reason": "Different models and failures; only a broad multimodal-model similarity." + "reason": "SigLIP2 implementation discrepancy and Qwen3_5 image loop are unrelated model-specific bugs." }, { "left": "issue:46231", "right": "issue:46877", "accept": false, - "reason": "Docs/utility-related only at a high level; one is an API doc mismatch and the other is noisy import-time diagnostics." + "reason": "Documentation parameter-name mismatch and import-time doc-lint stdout output are separate docs tooling issues." }, { "left": "issue:44188", "right": "issue:46328", "accept": false, - "reason": "Entirely unrelated: torch.compile attention-kernel divergence vs feature extractor padding inefficiency." - }, - { - "left": "issue:43937", - "right": "issue:44485", - "accept": false, - "reason": "Same model family, but GenerationConfig validation and RoPE implementation are different code paths and not the same bug." + "reason": "torch.compile attention-kernel branching and a tokenizer phonemizer padding issue are unrelated." } ] }, @@ -15705,23 +15610,23 @@ "issue:42371", "issue:42673", "issue:42898", - "issue:43064", "issue:43066", "issue:43116", "issue:43441", "issue:43493", "issue:43611", "issue:43618", - "issue:43673", "issue:43825", "issue:43827", "issue:43874", + "issue:43937", "issue:43976", "issue:43992", "issue:44038", "issue:44075", "issue:44263", "issue:44355", + "issue:44485", "issue:44610", "issue:44625", "issue:44704", @@ -15747,6 +15652,7 @@ "issue:46909" ], "soft_pairs": [ + "issue:43937|issue:44485", "issue:38617|issue:45278", "issue:42898|issue:46614", "issue:43066|issue:44779", @@ -15765,213 +15671,212 @@ "issue:43992|issue:44704", "issue:44075|issue:44936", "issue:43116|issue:43825", - "issue:43066|issue:46710", "issue:28348|issue:46640", + "issue:43066|issue:46710", "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:33453|issue:39401" ], - "prepared_review_unit_hash": "96b8e9d7cfc97f9360151816654f93c4117353b95567db6bba1cdcb283979388", + "prepared_review_unit_hash": "fac701cfef7f999166e3036fc9d2fe69d6e223945ca70344007483add8b44e17", "component_budget": { "node_count": 500, "item_count": 500, "soft_pair_count": 1839, - "serialized_chars": 953913, - "estimated_input_tokens": 238479, - "estimated_eval_tokens": 477214 + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 28, - "serialized_chars": 35743, - "estimated_input_tokens": 8936, - "estimated_eval_tokens": 18128 + "serialized_chars": 35637, + "estimated_input_tokens": 8910, + "estimated_eval_tokens": 18076 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The set is heterogeneous and does not form a single duplicate cluster. All soft-similarity pairs are false positives: they share broad themes like v5 regressions, tokenizers, or loading, but not the same concrete bug or code path.", + "summary": "All soft-edge candidates look like false positives: they share broad themes (tokenizers, Qwen/GLM, FlashAttention, upgrade regressions) but point to different concrete bugs, models, or code paths. No pull requests are present.", "confidence": 0.88, - "canonical_issue_reason": "If forced to choose one issue as the closest umbrella, issue:45278 is the broadest and most representative v5 regression report (many import errors after upgrade). That said, the overall set is too diverse to treat as one duplicate family.", + "canonical_issue_reason": "issue:45278 is the best umbrella-style anchor: it describes a broad, reproducible import regression after upgrade with multiple failing imports, making it the most representative issue in this set.", "canonical_pr_reason": null, - "best_issue_reason": "issue:45278 is the best single issue to keep as the canonical tracker because it is broad, upgrade-focused, and likely to absorb related import-regression reports; the other issues are either narrower, model-specific, or unrelated.", + "best_issue_reason": "issue:45278 is the strongest global issue candidate because it is a clear, high-signal regression report with broad impact and an obvious upgrade trigger, so it is the most useful triage anchor.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:43937", + "right": "issue:44485", + "accept": false, + "reason": "Same GLM-5 family, but one is GenerationConfig validation and the other is RoPE implementation; different failure points and not the same bug." + }, { "left": "issue:38617", "right": "issue:45278", "accept": false, - "reason": "Both concern imports after v5 upgrades, but one is a specific missing symbol while the other is a broad import-regression report; not the same bug." + "reason": "Both are import-related, but 38617 is a specific missing symbol while 45278 is a broad upgrade regression with many import errors; not the same underlying issue." }, { "left": "issue:42898", "right": "issue:46614", "accept": false, - "reason": "Tokenizer cleanup behavior and stale phonemizer language state are different tokenizer bugs with different code paths." + "reason": "Tokenization-space behavior change and stale phonemizer backend state are unrelated code paths." }, { "left": "issue:43066", "right": "issue:44779", "accept": false, - "reason": "Both mention tokenizer regressions in v5, but one is about decoder type and the other about DeepSeek tokenization correctness; different defects." + "reason": "Both involve tokenizer regressions, but one is decoder-type selection and the other is DeepSeek output correctness; different concrete bugs." }, { "left": "issue:43611", "right": "issue:46002", "accept": false, - "reason": "Model loading via base_model_prefix and NaNs in classification heads are unrelated problems." + "reason": "Model loading/base_model_prefix breakage and NaNs in classification heads are distinct problems." }, { "left": "issue:44038", "right": "issue:45478", "accept": false, - "reason": "Both involve Qwen3.x/Moe loading, but they target different model families and failure modes; not the same concrete issue." + "reason": "They involve different Qwen3 variants and different failures: VL-Moe bug vs from_pretrained error for Qwen3.5 MoE." }, { "left": "issue:43976", "right": "issue:45446", "accept": false, - "reason": "Python version support regression and a flex_attention import guard are unrelated." + "reason": "Python version compatibility regression and a wrong PyTorch version gate for AuxRequest import are not the same issue." }, { "left": "issue:44355", "right": "issue:46231", "accept": false, - "reason": "Compiled Python file errors and a documentation parameter-name mismatch are not the same bug." + "reason": "Compiled Python file errors and a docs parameter-name mismatch are unrelated." }, { "left": "issue:43116", "right": "issue:43493", "accept": false, - "reason": "A classification example returning empty results and a SigLIP2 implementation discrepancy are separate issues." + "reason": "Example script multi-label output bug and SigLIP2 implementation discrepancy are different components and symptoms." }, { "left": "issue:42673", "right": "issue:45072", "accept": false, - "reason": "VRAM leakage in dataloader threads and bfloat16 dtype mismatches are different runtime problems." + "reason": "VRAM leak across dataloader threads and bfloat16 dtype mismatches are separate runtime failures." }, { "left": "issue:45357", "right": "issue:45478", "accept": false, - "reason": "Both are Qwen3.5-related, but one is save_pretrained key handling and the other is from_pretrained loading; different code paths." + "reason": "Both are Qwen3.5 MoE-related, but one is save_pretrained key serialization and the other is from_pretrained loading; not the same concrete code-path problem." }, { "left": "issue:44625", "right": "issue:46002", "accept": false, - "reason": "num_labels propagation and NaNs in classification heads are unrelated model issues." + "reason": "Propagating num_labels into text config and NaNs in classification heads are different defects." }, { "left": "issue:42371", "right": "issue:43874", "accept": false, - "reason": "TF32 behavior settings and a missing multimodal-token helper method are unrelated." + "reason": "TF32 API guidance/regression and a missing image-processor method are unrelated." }, { "left": "issue:45083", "right": "issue:45446", "accept": false, - "reason": "Feature-extractor length math in qwen3_omni_moe and a PyTorch version check in flex_attention are unrelated." + "reason": "Helper-length computation behavior and a PyTorch version check bug are not the same underlying change." }, { "left": "issue:44610", "right": "issue:46909", "accept": false, - "reason": "Processor image-size mismatch and a broken documentation example are not the same underlying bug." + "reason": "Processor over-padding and a broken documentation example are different kinds of issues." }, { "left": "issue:43618", "right": "issue:46829", "accept": false, - "reason": "Both involve output fields, but they affect different model outputs and different attributes; too broad to merge." + "reason": "Both are output-field problems, but they affect different model outputs and different fields; not mergeable as one bug." }, { "left": "issue:43992", "right": "issue:44704", "accept": false, - "reason": "Missing embed_tokens.weight on load and kwargs not forwarded to cached_file are distinct loading issues." + "reason": "Missing embed_tokens.weight and incomplete kwargs forwarding to cached_file are separate loader issues." }, { "left": "issue:44075", "right": "issue:44936", "accept": false, - "reason": "Optimizer argument handling and trainer.evaluate() failure after train() are unrelated." + "reason": "SGD argument handling and trainer.evaluate failing after train are unrelated." }, { "left": "issue:43116", "right": "issue:43825", "accept": false, - "reason": "Example script classification behavior and pipeline() error-message docs are different problems." + "reason": "Classification example behavior and pipeline translation error messaging are different subsystems." }, { - "left": "issue:43066", - "right": "issue:46710", + "left": "issue:28348", + "right": "issue:46640", "accept": false, - "reason": "Tokenizer decoder-type regression and an accuracy regression on DeepSeek-R1 are not the same concrete defect." + "reason": "Both mention FlashAttention 2, but one is GPT2LMHeadModel support and the other is a generic flash-attention issue; different model/code paths." }, { - "left": "issue:28348", - "right": "issue:46640", + "left": "issue:43066", + "right": "issue:46710", "accept": false, - "reason": "FlashAttention 2 support for GPT2LMHeadModel is a feature request; the other is a separate Flash Attention bug report." + "reason": "Tokenizer decoder-type regression and DeepSeek accuracy regression after upgrade are not the same bug." }, { "left": "issue:44263", "right": "issue:46762", "accept": false, - "reason": "A torch.split return-value issue in one model is unrelated to numerical differences in a different attention implementation." + "reason": "A torch.split return-value bug in GlmMoeDsaIndexer is different from Minimax MSA attention numerical differences." }, { "left": "issue:46614", "right": "issue:46752", "accept": false, - "reason": "A stale phonemizer backend and an empty-list chat-template crash are unrelated tokenizer/utility bugs." + "reason": "Stale phonemizer backend state and an empty-list crash in apply_chat_template are unrelated." }, { "left": "issue:44995", "right": "issue:46697", "accept": false, - "reason": "A stale indexer-cache crash and image over-padding are different subsystems and failure modes." + "reason": "Second-forward stale cache crash and image over-padding are different code paths." }, { "left": "issue:45092", "right": "issue:46762", "accept": false, - "reason": "Meta-initialization incompatibility for remote-code checkpoints and Minimax attention numerical drift are unrelated." + "reason": "Old InternVL2 meta-init incompatibility and Minimax attention numerical discrepancies are unrelated." }, { "left": "issue:42222", "right": "issue:43441", "accept": false, - "reason": "A vitpose import/breakage issue and a Ministral FlashAttention failure are different model-specific bugs." + "reason": "VitPose model breakage and Ministral-3 FlashAttention failure are different model-specific regressions." }, { "left": "issue:43827", "right": "issue:45446", "accept": false, - "reason": "Docs still referencing pipeline() and a flex_attention import version-check bug are unrelated." + "reason": "Documentation still referencing pipeline() and a PyTorch version check for flex_attention are not the same issue." }, { "left": "issue:33453", "right": "issue:39401", "accept": false, - "reason": "Tokenizer-loading regression and Qwen3 offset_mapping errors are both tokenizer-related but not the same bug." - }, - { - "left": "issue:43064", - "right": "issue:43673", - "accept": false, - "reason": "FSDP2 optimizer-state corruption and missing GenerationMixin cache in chunked_prefill are unrelated training vs generation issues." + "reason": "Generic tokenizer loading regression and Qwen3 offset_mapping are both tokenizer-related but describe different failures." } ] }, @@ -15999,10 +15904,10 @@ "issue:39401", "issue:41628", "issue:42222", - "issue:42371", "issue:42617", "issue:42673", "issue:42796", + "issue:43064", "issue:43122", "issue:43408", "issue:43450", @@ -16011,7 +15916,6 @@ "issue:43937", "issue:43986", "issue:44117", - "issue:44261", "issue:44291", "issue:44380", "issue:44410", @@ -16045,11 +15949,12 @@ "issue:46752" ], "soft_pairs": [ + "issue:43064|issue:43673", "issue:42673|issue:42796", "issue:43937|issue:45563", "issue:44380|issue:46077", - "issue:45356|issue:46614", "issue:43673|issue:46328", + "issue:45356|issue:46614", "issue:42673|issue:46655", "issue:42222|issue:46752", "issue:44291|issue:45071", @@ -16065,218 +15970,217 @@ "issue:44829|issue:45216", "issue:45071|issue:46512", "issue:44117|issue:45081", - "issue:38617|issue:41628", "issue:43408|issue:43742", + "issue:38617|issue:41628", "issue:43122|issue:46298", "issue:42617|issue:45341", "issue:45362|issue:46328", "issue:42222|issue:46178", "issue:44484|issue:45310", - "issue:44877|issue:46231", - "issue:42371|issue:44261" + "issue:44877|issue:46231" ], - "prepared_review_unit_hash": "1223dc637c36a7915d373144ff0eaf6ddbf754abb380b9d7cf8d22abc744639c", + "prepared_review_unit_hash": "50670bdb5f784d2654cf074465b15fd3ab392faeb023185430ab437c1a922e2f", "component_budget": { "node_count": 500, "item_count": 500, "soft_pair_count": 1839, - "serialized_chars": 953913, - "estimated_input_tokens": 238479, - "estimated_eval_tokens": 477214 + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 }, "budget": { - "node_count": 48, - "item_count": 48, + "node_count": 47, + "item_count": 47, "soft_pair_count": 29, - "serialized_chars": 36394, - "estimated_input_tokens": 9099, - "estimated_eval_tokens": 18454 + "serialized_chars": 35864, + "estimated_input_tokens": 8966, + "estimated_eval_tokens": 18188 }, "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 bug reports (tokenizers, model loading, generation, vision/video processors, config/version checks). The soft links are similarity-only and do not look like duplicates or mergeable PR-equivalents.", - "confidence": 0.89, - "canonical_issue_reason": "issue:44117 is the broadest, most core-framework representative: it describes a general `from_pretrained`/token mapping assumption bug rather than a model-specific edge case.", + "summary": "Loose cluster of mostly unrelated transformers issues that only share broad theme similarity (v5 regressions, tokenization, generation, model-loading, and training). None of the soft pairs look like the same underlying bug/change, so they should stay separate.", + "confidence": 0.82, + "canonical_issue_reason": "If forced to pick one issue as the umbrella representative, issue #43122 is the broadest and most central: it describes a cross-version tokenizer regression with a clear repro, and it is closer to the other tokenizer-focused reports than the model-specific or utility-specific bugs. That said, this is still a noisy mixed cluster, not a true duplicate set.", "canonical_pr_reason": null, - "best_issue_reason": "issue:44117 is a good global representative because it targets a central loading path and has a clear, actionable failure mode affecting multiple tokenizer families.", + "best_issue_reason": "Issue #43122 is the best single issue to retain because it is the most general high-signal regression in the set and the most plausible umbrella for nearby tokenizer complaints; the rest are narrower, unrelated, or touch different code paths.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:43064", + "right": "issue:43673", + "accept": false, + "reason": "Different failures: optimizer-state mismatch under FSDP2+PEFT+cpu_ram_efficient_loading vs missing GenerationMixin cache during chunked prefill." + }, { "left": "issue:42673", "right": "issue:42796", "accept": false, - "reason": "Both are tokenizer-related, but one is a VRAM leak in multi-threaded inference and the other is a BPE-marker handling bug; different code paths and symptoms." + "reason": "One is a VRAM leak in multi-threaded dataloader use; the other is tokenizer BPE-marker handling. Same broad area, different bug." }, { "left": "issue:43937", "right": "issue:45563", "accept": false, - "reason": "GenerationConfig validation and a stale paged-generate warning are separate generation issues with different failure modes." + "reason": "GenerationConfig validation error vs stale warning text for num_return_sequences in paged generate; not the same defect." }, { "left": "issue:44380", "right": "issue:46077", "accept": false, - "reason": "GPT2 attention scaling under SDPA/FlashAttention is unrelated to Gemma4 using the wrong attention mask type." + "reason": "Different attention bugs: GPT2 attention scaling ignored under SDPA/FlashAttention vs Gemma4 building causal masks when bidirectional attention is requested." }, { - "left": "issue:45356", - "right": "issue:46614", + "left": "issue:43673", + "right": "issue:46328", "accept": false, - "reason": "These both mention tokenizer behavior, but one is a Kimi-K2.5 codec regression while the other is stale language backend state in Wav2Vec2 phonemization." + "reason": "Generation cache missing in chunked prefill vs SequenceFeatureExtractor.pad inefficiency; unrelated code paths." }, { - "left": "issue:43673", - "right": "issue:46328", + "left": "issue:45356", + "right": "issue:46614", "accept": false, - "reason": "Chunked-prefill cache handling and a SequenceFeatureExtractor padding inefficiency are unrelated performance/behavior bugs." + "reason": "Tokenizer codec/BPE regression vs stale phonemizer backend after language switch; different tokenizer subsystems and symptoms." }, { "left": "issue:42673", "right": "issue:46655", "accept": false, - "reason": "VRAM leak in dataloader threads vs incorrect video padding calculation are distinct issues in different subsystems." + "reason": "VRAM leak in Qwen3ForCausalLM threads vs video processor padding calculation error; not the same bug." }, { "left": "issue:42222", "right": "issue:46752", "accept": false, - "reason": "VitPose model breakage and empty-list `apply_chat_template` crashes are unrelated failures." + "reason": "VitPose model breakage vs apply_chat_template crashing on empty list; unrelated failures." }, { "left": "issue:44291", "right": "issue:45071", "accept": false, - "reason": "Unexpected `_is_hf_initialized` argument during `init_empty_weights` is a loading bug, while the other is config type checking in v5.4.0; different causes." + "reason": "Both are v5 regressions, but one is an unexpected _is_hf_initialized argument during init_empty_weights and the other is PretrainedConfig type checking. Different issues." }, { "left": "issue:45405", "right": "issue:45446", "accept": false, - "reason": "A PEFT version bump issue and an AuxRequest PyTorch version check bug are unrelated dependency/version problems." + "reason": "PEFT minimum-version bump metadata problem vs incorrect torch version guard for AuxRequest import; unrelated." }, { "left": "issue:44117", "right": "issue:45588", "accept": false, - "reason": "Tokenizer mapping `None` handling and flash_attention crash on `s_aux=None` affect different APIs and code paths." + "reason": "TOKENIZER_MAPPING_NAMES None handling vs flash_attention crash on s_aux=None; different layers of the stack." }, { "left": "issue:44466", "right": "issue:46002", "accept": false, - "reason": "Serialization of tied weights depends on device; NaNs in classification heads after checkout/init are a different model-state bug." + "reason": "Serialization of lm_head.weight depending on device vs NaNs in classification heads after checkout/init; different failure modes." }, { "left": "issue:44410", "right": "issue:45310", "accept": false, - "reason": "Missing projections in qwen3next layer 0 is a model-architecture issue; Qwen3.5 from_pretrained error is a separate loading regression." + "reason": "Missing qwen3next projections and Qwen3.5 from_pretrained errors are both model-family issues, but not the same concrete code-path bug." }, { "left": "issue:39401", "right": "issue:44779", "accept": false, - "reason": "Both are tokenizer regressions, but one is Qwen3 offset mapping and the other is Deepseek tokenization output changes; not the same bug." + "reason": "Both concern tokenizer regressions, but one is wrong offset_mapping and the other is incorrect DeepSeek tokenization results. Similar theme, not the same bug." }, { "left": "issue:43937", "right": "issue:44977", "accept": false, - "reason": "Invalid GenerationConfig and Qwen3.5 flash-attention generation failure are different generation-path problems." + "reason": "GLM-5 invalid GenerationConfig and Qwen3.5 flash-attention generation problems are distinct model-specific issues." }, { "left": "issue:45341", "right": "issue:46328", "accept": false, - "reason": "A testing_utils bug and a padding performance issue are not the same defect." + "reason": "A tiny testing_utils bug vs a padding performance issue; no shared underlying defect." }, { "left": "issue:43450", "right": "issue:43986", "accept": false, - "reason": "Video processor batch-shape bug and AutoProcessor crash without torchvision are related to video tooling but are distinct failure modes." + "reason": "Video processor shape bug vs confusing crash loading a video model without torchvision; different root causes." }, { "left": "issue:45341", "right": "issue:45561", "accept": false, - "reason": "Both touch testing utilities, but one is a generic bug and the other is a pytest-xdist race on a specific file; not duplicate." + "reason": "Testing_utils bug vs pytest-xdist race on captured_info.txt; different concurrency and file-handling problems." }, { "left": "issue:44829", "right": "issue:45216", "accept": false, - "reason": "FlashAttention-3 training degeneration and Qwen3.5 save_pretrained regression are different problems in different paths." + "reason": "Degenerate training with flash_attention_3 vs Qwen3.5 save_pretrained checkpoint regression; different code paths." }, { "left": "issue:45071", "right": "issue:46512", "accept": false, - "reason": "Type-checking regression for `PretrainedConfig` and stale `model.config.dtype` after `from_config` are separate config issues." + "reason": "PreTrainedConfig type-checking breakage vs stale model.config.dtype after from_config(dtype=...); related only by version timing." }, { "left": "issue:44117", "right": "issue:45081", "accept": false, - "reason": "Tokenizer mapping `None` handling and `_patch_mistral_regex` backend attribute crash are both loading-time tokenizer bugs, but they occur in different functions with different root causes." + "reason": "Tokenizer mapping fallback bug vs _patch_mistral_regex backend_tokenizer AttributeError; different tokenizer plumbing bugs." }, { - "left": "issue:38617", - "right": "issue:41628", + "left": "issue:43408", + "right": "issue:43742", "accept": false, - "reason": "Import errors for two different symbols (`layer_type_validation` vs `AutoImageProcessor`) are unrelated API breakages." + "reason": "Model-type warning mismatch vs key error loading facebook/MobileLLM-125M; not the same defect." }, { - "left": "issue:43408", - "right": "issue:43742", + "left": "issue:38617", + "right": "issue:41628", "accept": false, - "reason": "A sam3_video/sam3_tracker type mismatch warning and a key error while loading MobileLLM-125M are different model-loading issues." + "reason": "Both are import errors, but for different missing exports and different call sites; not duplicates." }, { "left": "issue:43122", "right": "issue:46298", "accept": false, - "reason": "Tokenization changes across versions and a noisy logging message do not describe the same underlying bug." + "reason": "Tokenization regression vs noisy logging message; no shared underlying bug." }, { "left": "issue:42617", "right": "issue:45341", "accept": false, - "reason": "Failure to run `3d_parallel.py` and a testing_utils bug are unrelated." + "reason": "Script execution failure vs a small testing_utils bug; unrelated." }, { "left": "issue:45362", "right": "issue:46328", "accept": false, - "reason": "Qwen3.5 chat crash and a SequenceFeatureExtractor padding inefficiency are unrelated." + "reason": "Qwen3.5 chat crash vs SequenceFeatureExtractor.pad inefficiency; different components." }, { "left": "issue:42222", "right": "issue:46178", "accept": false, - "reason": "VitPose model breakage and backend registration error for `tensorflow_text` are separate subsystem issues." + "reason": "VitPose model breakage vs backend mapping rejecting tensorflow_text; unrelated." }, { "left": "issue:44484", "right": "issue:45310", "accept": false, - "reason": "max_shard_size default question and Qwen3.5 from_pretrained regression are not duplicates." + "reason": "Question about shard size default vs Qwen3.5 MoE from_pretrained error; not the same bug." }, { "left": "issue:44877", "right": "issue:46231", "accept": false, - "reason": "A strict config loader issue for `granite_speech` and a doc typo in `set_training()` are not the same bug." - }, - { - "left": "issue:42371", - "right": "issue:44261", - "accept": false, - "reason": "TF32 settings API guidance and an MLA q_a_layernorm epsilon precision issue are unrelated." + "reason": "Strict config loading failure vs a documentation parameter-name typo; not duplicates." } ] }, @@ -16349,6 +16253,7 @@ "issue:46894" ], "soft_pairs": [ + "issue:42371|issue:44261", "issue:43937|issue:44496", "issue:39401|issue:46209", "issue:44410|issue:45478", @@ -16361,15 +16266,15 @@ "issue:43066|issue:43122", "issue:44263|issue:44485", "issue:43097|issue:44493", - "issue:43931|issue:44410", "issue:43673|issue:45341", + "issue:43931|issue:44410", "issue:41628|issue:45071", "issue:42222|issue:45542", "issue:45071|issue:45542", "issue:44483|issue:45464", "issue:44484|issue:45478", - "issue:44263|issue:46172", "issue:43874|issue:44261", + "issue:44263|issue:46172", "issue:44360|issue:46172", "issue:27957|issue:43352", "issue:43931|issue:44038", @@ -16380,227 +16285,230 @@ "issue:43986|issue:45341", "issue:44261|issue:44485" ], - "prepared_review_unit_hash": "fbbd07c8f4b71f86c35b2b3e1ad4cdde078068043a927aa9bc74ff25a2dd0d8c", + "prepared_review_unit_hash": "c81712ba752605d86db3ab09da09478366df739be8919be1f3d5e016bf2128d4", "component_budget": { "node_count": 500, "item_count": 500, "soft_pair_count": 1839, - "serialized_chars": 953913, - "estimated_input_tokens": 238479, - "estimated_eval_tokens": 477214 + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 }, "budget": { "node_count": 47, "item_count": 47, - "soft_pair_count": 30, - "serialized_chars": 35564, - "estimated_input_tokens": 8891, - "estimated_eval_tokens": 18038 + "soft_pair_count": 31, + "serialized_chars": 35867, + "estimated_input_tokens": 8967, + "estimated_eval_tokens": 18190 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is highly heterogeneous: most items are distinct regression reports across imports, tokenization, model loading, attention, and training behavior. A few pairs are superficially similar, but I would not merge them as duplicates without stronger code-path evidence.", - "confidence": 0.31, - "canonical_issue_reason": "issue:45278 is the broadest representative of the set\u2019s repeated version-upgrade regression theme, especially import failures after upgrading Transformers; it is still only a partial fit because the cluster spans many unrelated bug classes.", + "summary": "This cluster is mostly a grab bag of unrelated Transformers regression reports, with a few loose v5/tokenizer/import similarities. The broadest issue is the generic upgrade/import-regression report; only the tokenizer-v5 pair looks like a plausible duplicate.", + "confidence": 0.42, + "canonical_issue_reason": "issue:45278 is the broadest, most umbrella-style report in the set: it describes widespread import failures after upgrading to v5.5.0, which best matches the cluster\u2019s overall regression theme better than the many model-specific bugs.", "canonical_pr_reason": null, - "best_issue_reason": "issue:45278 is the most generally representative issue among these mixed regression reports, but the overall set is too diverse to treat as one tight duplicate cluster.", + "best_issue_reason": "issue:45278 is the best single representative issue because it is the most general and most likely to subsume other compatibility/import regressions in this cluster.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:42371", + "right": "issue:44261", + "accept": false, + "reason": "Different problems: TF32 configuration behavior vs an MLA layernorm epsilon precision bug." + }, { "left": "issue:43937", "right": "issue:44496", "accept": false, - "reason": "Different failure classes: invalid GenerationConfig vs unrecognized model/model_type missing." + "reason": "One is a GenerationConfig validation error; the other is an unrecognized model/config loading failure." }, { "left": "issue:39401", "right": "issue:46209", "accept": false, - "reason": "Tokenizer offset_mapping bug and FP8 inference failure are unrelated." + "reason": "Tokenizer offset-mapping bug and FP8 inference failure are unrelated symptoms and code paths." }, { "left": "issue:44410", "right": "issue:45478", "accept": false, - "reason": "Missing Qwen3Next projections and a Qwen3.5-Moe from_pretrained error are distinct bugs." + "reason": "Both mention Qwen3-family models, but one is missing projections in a layer while the other is a from_pretrained loading error." }, { "left": "issue:42371", "right": "issue:46617", "accept": false, - "reason": "TF32 settings warning is unrelated to a cache FileNotFoundError." + "reason": "TF32 settings are unrelated to cache-loading FileNotFoundError behavior." }, { "left": "issue:44261", "right": "issue:46762", "accept": false, - "reason": "Both are numerical issues, but they involve different models and different attention/layernorm code paths." + "reason": "Layernorm epsilon precision mismatch is not the same underlying issue as Minimax attention numerical differences." }, { "left": "issue:43208", "right": "issue:46894", "accept": false, - "reason": "xLSTM training blockers and a CohereASR double-shift loss bug are not the same bug." + "reason": "Both are training bugs, but they involve different models and different failure mechanisms." }, { "left": "issue:44933", "right": "issue:45278", "accept": false, - "reason": "Both are import-related, but one is a missing symbol and the other is a broad upgrade regression; not the same concrete issue." + "reason": "Both are import-related, but one is a specific missing export and the other is a broad upgrade regression report; not the same bug." }, { "left": "issue:38617", "right": "issue:44220", "accept": false, - "reason": "ImportError for layer_type_validation is unrelated to _torch_extract_fbank_features()." + "reason": "An ImportError for a missing symbol is not the same as the fbank feature extraction issue." }, { "left": "issue:44075", "right": "issue:45656", "accept": false, - "reason": "Ignored SGD args and double optimizer steps under deepspeed are different optimizer bugs." + "reason": "SGD argument handling and double optimizer stepping under deepspeed are different optimizer bugs." }, { "left": "issue:43066", "right": "issue:43122", - "accept": false, - "reason": "Both concern tokenizer behavior in v5, but the reported symptoms and likely code paths are different enough to keep separate." + "accept": true, + "reason": "Both point to the same v5 tokenizer regression: changed decoder/tokenization behavior after the 4.57.x -> 5.0 transition." }, { "left": "issue:44263", "right": "issue:44485", "accept": false, - "reason": "torch.split return values and GLM-5 RoPE implementation are unrelated." + "reason": "One is a torch.split/indexing bug in a MoE indexer; the other is a RoPE implementation issue." }, { "left": "issue:43097", "right": "issue:44493", "accept": false, - "reason": "Removed tie_embeddings_and_encoder_decoder and unexpected position_id keys are different problems." + "reason": "Removing tie_embeddings_and_encoder_decoder and unexpected position-id keys are different configuration/model-loading regressions." }, { - "left": "issue:43931", - "right": "issue:44410", + "left": "issue:43673", + "right": "issue:45341", "accept": false, - "reason": "Weight-shape mismatch in Qwen3-VL-30B and missing projections in qwen3next are different architecture issues." + "reason": "Chunked-prefill GenerationMixin cache behavior and a testing_utils bug do not share the same failure mode." }, { - "left": "issue:43673", - "right": "issue:45341", + "left": "issue:43931", + "right": "issue:44410", "accept": false, - "reason": "Generation cache during chunked_prefill is unrelated to a testing_utils bug." + "reason": "Different Qwen3-related model-loading failures: weight-shape mismatch vs missing attention projections." }, { "left": "issue:41628", "right": "issue:45071", "accept": false, - "reason": "AutoImageProcessor import failure and PreTrainedConfig type checking breakage are unrelated." + "reason": "Missing AutoImageProcessor import and PreTrainedConfig type-checking breakage are unrelated API regressions." }, { "left": "issue:42222", "right": "issue:45542", "accept": false, - "reason": "vitpose model breakage and a TensorBoard-without-TensorFlow backend error are unrelated." + "reason": "Vitpose model breakage and tensorboard/TensorFlow backend detection are unrelated." }, { "left": "issue:45071", "right": "issue:45542", "accept": false, - "reason": "Type-checking regression and undefined tf backend error are different bugs." + "reason": "These are different subsystems: config type checking versus backend initialization." }, { "left": "issue:44483", "right": "issue:45464", "accept": false, - "reason": "Both hit chat/completions, but one is a general request rejection and the other is a model-specific streaming failure." + "reason": "Both involve chat/completions, but one is request acceptance in v5.3 and the other is streaming inference on a specific model." }, { "left": "issue:44484", "right": "issue:45478", "accept": false, - "reason": "max_shard_size default behavior is unrelated to the Qwen3.5-Moe from_pretrained error." + "reason": "Saving shard size defaults and Qwen3.5 from_pretrained loading errors are unrelated." }, { - "left": "issue:44263", - "right": "issue:46172", + "left": "issue:43874", + "right": "issue:44261", "accept": false, - "reason": "GlmMoeDsaIndexer split semantics and deepseek-v4 transpose behavior are unrelated." + "reason": "Missing image-processor method and MLA layernorm epsilon precision are different bugs." }, { - "left": "issue:43874", - "right": "issue:44261", + "left": "issue:44263", + "right": "issue:46172", "accept": false, - "reason": "Missing get_number_of_image_patches and missing rms_norm_eps are different bugs." + "reason": "Torch split return handling in GlmMoeDsaIndexer is unrelated to deepseek-v4 transpose behavior." }, { "left": "issue:44360", "right": "issue:46172", "accept": false, - "reason": "DSA indexer missing ReLU and deepseek-v4 dtype/transpose logic are unrelated." + "reason": "A missing ReLU in the DSA indexer and a dtype/transpose issue are not the same bug." }, { "left": "issue:27957", "right": "issue:43352", "accept": false, - "reason": "Flash Attention 2 support issues for different models are not enough to treat as the same concrete bug." + "reason": "Both mention Flash Attention 2, but one is a model support gap and the other is a model-specific incompatibility." }, { "left": "issue:43931", "right": "issue:44038", "accept": false, - "reason": "Qwen3-VL weight mismatch and Qwen3-VL-Moe load failure are distinct model-loading issues." + "reason": "Both are Qwen3-VL-family reports, but the concrete failures differ: shape mismatch versus a separate model bug." }, { "left": "issue:43012", "right": "issue:43577", "accept": false, - "reason": "bf16 compilation warning and BLIP2 dtype mismatch are separate precision/dtype bugs." + "reason": "A compile-time precision warning and a load-time dtype mismatch are related only loosely, not the same code-path bug." }, { "left": "issue:43441", "right": "issue:44829", "accept": false, - "reason": "Both mention flash attention, but they affect different models and different failure modes." + "reason": "FlashAttention 2/3 issues in different models and different symptoms are not the same fix." }, { "left": "issue:43097", "right": "issue:43452", "accept": false, - "reason": "Removed tie_embeddings support and gguf_file breakage are unrelated." + "reason": "A removed config option and a gguf_file from_pretrained regression are separate issues." }, { "left": "issue:45003", "right": "issue:45341", "accept": false, - "reason": "Unsafe sys.modules access and a testing_utils bug do not describe the same issue." + "reason": "Unsafe sys.modules access and a testing_utils bug are unrelated." }, { "left": "issue:43986", "right": "issue:45341", "accept": false, - "reason": "AutoProcessor/video-model import crash without torchvision is unrelated to testing_utils." + "reason": "Video AutoProcessor crashing without torchvision and a testing_utils bug do not share the same underlying cause." }, { "left": "issue:44261", "right": "issue:44485", "accept": false, - "reason": "LayerNorm epsilon precision and GLM-5 RoPE implementation are different code paths." - }, - { - "left": "issue:42371", - "right": "issue:46617", - "accept": false, - "reason": "TF32 API guidance and cache loading FileNotFoundError are not the same bug." + "reason": "Layernorm epsilon precision and GLM-5 RoPE implementation are different numerical issues, not one bug." } ] }, - "evaluator_result": null, - "evaluator_used": false, + "evaluator_result": { + "accept": true, + "feedback": "The evaluation is grounded in the packet and stays conservative. Using issue:45278 as the umbrella representative for the broad import-regression cluster is reasonable, and the only accepted soft pair (43066/43122) is well supported by the v5 tokenizer regression theme. The rejected pairs are mostly defensible as distinct bugs or at least not clearly mergeable into a single fix." + }, + "evaluator_used": true, "retried": false, - "accepted_nontrivial_soft_edge": false, + "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null }, @@ -16681,8 +16589,8 @@ "issue:43986|issue:45372", "issue:43097|issue:46050", "issue:43986|issue:44484", - "issue:46655|issue:46664", "issue:42890|issue:46298", + "issue:46655|issue:46664", "issue:46172|issue:46710", "issue:38617|issue:44351", "issue:46298|issue:46752", @@ -16695,14 +16603,14 @@ "issue:44704|issue:45103", "issue:44380|issue:46421" ], - "prepared_review_unit_hash": "f71c1e78177e7c732f27087afa92819df810e5f0df32530785d855879cf0bca2", + "prepared_review_unit_hash": "50e9ffb6950379da37eaf38b4682bcc445ee31f3fe550c56264a748e5ca1982e", "component_budget": { "node_count": 500, "item_count": 500, "soft_pair_count": 1839, - "serialized_chars": 953913, - "estimated_input_tokens": 238479, - "estimated_eval_tokens": 477214 + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 }, "budget": { "node_count": 47, @@ -16717,180 +16625,180 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster is highly heterogeneous: it mixes unrelated import errors, model-specific regressions, training bugs, processor issues, and logging/test problems. None of the soft pairs look like the same underlying bug or change.", - "confidence": 0.97, - "canonical_issue_reason": "No single canonical issue stands out because the items are not a duplicate set; they cover many unrelated failure modes and subsystems.", + "summary": "All soft pairs look like false positives: they share broad subsystems or similar error wording, but not the same concrete bug or change. No pull requests are present in the set.", + "confidence": 0.81, + "canonical_issue_reason": "Issue 43873 is the strongest anchor: it is still open, has the most discussion/inbound references, and describes a substantive user-impacting bug with a clear reproduction surface.", "canonical_pr_reason": null, - "best_issue_reason": "No global best issue is a strong representative here; the cluster is too mixed for one issue to credibly summarize the whole set.", + "best_issue_reason": "Issue 43873 is the best overall issue candidate because it is detailed, high-signal, and central enough to attract follow-up reports.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:45753", "right": "issue:46369", "accept": false, - "reason": "Different failures: Qwen3_5 infinite-loop behavior vs a Trainer import problem involving datasets." + "reason": "Different failures and models: infinite looping on a specific image vs memory overflow in sam3-litetext." }, { "left": "issue:44117", "right": "issue:44987", "accept": false, - "reason": "Tokenizer mapping None handling is unrelated to a model-loading failure for physical-intelligence/fast." + "reason": "One is a generic tokenizer mapping assumption bug; the other is a specific model-loading failure. Similar loading context, but not the same defect." }, { "left": "issue:43874", "right": "issue:44263", "accept": false, - "reason": "Same GLM naming family, but one is an image-processor method gap and the other is a tensor split/indexer issue." + "reason": "Multimodal token counting AttributeError and a torch.split return-value issue are separate code paths." }, { "left": "issue:45440", "right": "issue:46172", "accept": false, - "reason": "Both involve DeepSeek models, but the bugs are different code paths: native implementation divergence vs a dtype/transpose issue." + "reason": "Both are DeepSeek-related, but one is native-vs-remote divergence and the other is a specific dtype/transpose bug." }, { "left": "issue:43531", "right": "issue:45341", "accept": false, - "reason": "Qwen3-MoE sliding-window behavior is unrelated to a testing utility bug." + "reason": "Qwen3-MoE sliding-window behavior and a testing utility bug are unrelated." }, { "left": "issue:43673", "right": "issue:46123", "accept": false, - "reason": "Chunked-prefill cache initialization and partial-batch mask pipeline behavior are distinct problems." + "reason": "Cache initialization during chunked prefill and a final-batch pipeline flag bug are different behaviors." }, { "left": "issue:43097", "right": "issue:45588", "accept": false, - "reason": "A removed config field is unrelated to a flash-attention crash on sink-less models." + "reason": "A removed config option and a flash-attention crash on sink-less models are distinct issues." }, { "left": "issue:38617", "right": "issue:42757", "accept": false, - "reason": "Both are import errors, but for different symbols and modules; not the same bug." + "reason": "Both are import errors, but they involve different missing symbols from different modules and likely different dependency breaks." }, { "left": "issue:44038", "right": "issue:44410", "accept": false, - "reason": "Both mention Qwen, but one is a Transformers 5.0 compatibility issue and the other is missing projections in qwen3next." + "reason": "Same broad model family, but one is a load-time bug and the other is missing projections in layer 0." }, { "left": "issue:44410", "right": "issue:45375", "accept": false, - "reason": "Different model/config fields and different symptoms; not one underlying defect." + "reason": "Missing attention projections and a missing config field are separate defects in different parts of the model stack." }, { "left": "issue:43208", "right": "issue:46897", "accept": false, - "reason": "xLSTM training blockers and a Florence2 loss-shift bug are unrelated." + "reason": "xLSTM training blockers and a Florence2 loss-shift bug do not describe the same underlying problem." }, { "left": "issue:43064", "right": "issue:44315", "accept": false, - "reason": "Optimizer-state corruption with FSDP2/PEFT is a different bug from Liger Kernel not applying under model_init." + "reason": "Optimizer state corruption under FSDP2/PEFT is unrelated to Liger Kernel not being applied with model_init." }, { "left": "issue:43986", "right": "issue:45372", "accept": false, - "reason": "A missing torchvision dependency path is unrelated to a mistral_common import breakage." + "reason": "Different dependency failures: missing torchvision vs missing ReasoningEffort from mistral_common." }, { "left": "issue:43097", "right": "issue:46050", "accept": false, - "reason": "A removed embedding-tie config is unrelated to save/load breakage for a quantized model." + "reason": "One concerns a removed embedding-tie option; the other is save/load corruption for quantized models." }, { "left": "issue:43986", "right": "issue:44484", "accept": false, - "reason": "AutoProcessor video-model loading and max_shard_size default behavior are unrelated." + "reason": "Processor import crash and max_shard_size semantics are unrelated." }, { - "left": "issue:46655", - "right": "issue:46664", + "left": "issue:42890", + "right": "issue:46298", "accept": false, - "reason": "Both are SmolVLM video-processor issues, but one is padding calculation and the other is num_frames semantics." + "reason": "A flaky test due to missing set_seed and a noisy logging message are not the same bug." }, { - "left": "issue:42890", - "right": "issue:46298", + "left": "issue:46655", + "right": "issue:46664", "accept": false, - "reason": "Missing set_seed in a test file is unrelated to a noisy logging message." + "reason": "Both hit SmolVLM video processing, but one is padding calculation and the other is num_frames semantics; related area, different bug." }, { "left": "issue:46172", "right": "issue:46710", "accept": false, - "reason": "A DeepSeek-V4 tensor operation bug is not the same as an accuracy regression after upgrading Transformers." + "reason": "A specific tensor transformation bug in DeepSeek-v4 is not the same as a broader accuracy regression after upgrade." }, { "left": "issue:38617", "right": "issue:44351", "accept": false, - "reason": "Both are import errors, but they concern different exported names and different failure sites." + "reason": "Both are import-name failures, but they refer to different symbols and modules, so they are separate breakages." }, { "left": "issue:46298", "right": "issue:46752", "accept": false, - "reason": "Logging noise and an empty-list chat-template crash are unrelated." + "reason": "Logging noise and an empty-list IndexError in apply_chat_template are unrelated." }, { "left": "issue:44062", "right": "issue:44117", "accept": false, - "reason": "AddedToken keyword duplication is a tokenizer-construction bug, not the TOKENIZER_MAPPING_NAMES None issue." + "reason": "AddedToken keyword duplication and TOKENIZER_MAPPING_NAMES returning None are different tokenizer issues." }, { "left": "issue:43097", "right": "issue:43873", "accept": false, - "reason": "Config-field removal and quantization/offloading behavior are different problems." + "reason": "A removed config field and a quantization/offloading malfunction do not share the same code-path problem." }, { "left": "issue:43012", "right": "issue:46821", "accept": false, - "reason": "A precision warning from bfloat16 compilation is unrelated to a TimesFM window_size AttributeError." + "reason": "A bfloat16 warning during compilation and a window_size AttributeError are unrelated." }, { "left": "issue:44117", "right": "issue:46701", "accept": false, - "reason": "Tokenizer mapping fallback logic is unrelated to the DiffusionGemmaGenerationOutput pipeline failure." + "reason": "Tokenizer-mapping assumptions and a DiffusionGemma pipeline failure are different bugs." }, { "left": "issue:43502", "right": "issue:43992", "accept": false, - "reason": "Both are loading-related, but local_files_only network leakage and a missing embed_tokens weight are different bugs." + "reason": "local_files_only network leakage and a missing embed_tokens weight are separate from-pretrained problems." }, { "left": "issue:44936", "right": "issue:46365", "accept": false, - "reason": "Trainer.evaluate failing after train is unrelated to failing to import Trainer because of datasets." + "reason": "trainer.evaluate() after train() and importing Trainer conflicting with datasets are unrelated." }, { "left": "issue:44704", "right": "issue:45103", "accept": false, - "reason": "Passing kwargs to cached_file and an auto-docstring crash under future annotations are unrelated." + "reason": "cached_file kwarg forwarding and auto_docstring crashing under future annotations are different subsystems." }, { "left": "issue:44380", "right": "issue:46421", "accept": false, - "reason": "GPT2 attention scaling under SDPA/FlashAttention and chunked-prefill recompilation are different code-path issues." + "reason": "Attention scaling being ignored and lazy cache initialization causing recompilation are distinct issues." } ] }, @@ -16996,220 +16904,220 @@ "issue:44038|issue:44230", "issue:43756|issue:44485" ], - "prepared_review_unit_hash": "478d79b1f06c7efb40e801cc4d04089dd4cf9219ffc2e6ae4d371f747bf62602", + "prepared_review_unit_hash": "2d423037086e011d074020447f48145678ad2da55ddae6092b940622724de9f0", "component_budget": { "node_count": 500, "item_count": 500, "soft_pair_count": 1839, - "serialized_chars": 953913, - "estimated_input_tokens": 238479, - "estimated_eval_tokens": 477214 + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 31, - "serialized_chars": 36745, - "estimated_input_tokens": 9187, - "estimated_eval_tokens": 18630 + "serialized_chars": 36744, + "estimated_input_tokens": 9186, + "estimated_eval_tokens": 18628 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is a loose collection of mostly unrelated Transformers issues; the soft-similarity pairs are broad matches, not true duplicates. I found no PRs, and no pair is close enough to merge as the same underlying bug/change.", - "confidence": 0.9, - "canonical_issue_reason": "issue:43502 is the cleanest anchor: a concrete, user-facing bug with a narrow symptom ('local_files_only=True' still triggers network calls) and an obvious reproduction path.", + "summary": "This cluster is heterogeneous: the soft pairs mostly share only broad Transformers surface areas, not the same concrete bug. I rejected every soft edge and picked a single nominal issue representative from the most actionable load-path regression.", + "confidence": 0.42, + "canonical_issue_reason": "Issue 43502 is the most concrete, user-facing regression in the set, with a narrow loading/offline code path and a clear reproduction. The cluster itself is not a true duplicate group, so this is only a nominal representative.", "canonical_pr_reason": null, - "best_issue_reason": "issue:43502 is the best representative issue overall because it is specific, actionable, and clearly scoped; it is more triage-friendly than the more model- or environment-specific reports.", + "best_issue_reason": "Issue 43502 is the strongest standalone report here: clear symptom, specific API path, and a plausible localized fix surface. It is the best fit if one issue must represent the set.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:44938", "right": "issue:46178", "accept": false, - "reason": "Different failures: Python 3.14 import/load breakage vs invalid backend mapping for tensorflow_text." + "reason": "Both are import/loading-related, but one is a Python 3.14 load failure and the other is an invalid backend mapping error for tensorflow_text; different bugs." }, { "left": "issue:43986", "right": "issue:44117", "accept": false, - "reason": "Different code paths: AutoProcessor video-model import crash vs tokenizer mapping returning None in from_pretrained." + "reason": "Different failure modes: missing torchvision handling in AutoProcessor versus TOKENIZER_MAPPING_NAMES returning None in from_pretrained." }, { "left": "issue:42907", "right": "issue:44655", "accept": false, - "reason": "Both concern saving, but one is dequantized model saving and the other is Pipeline.save_pretrained; not the same bug." + "reason": "One is about failing to save dequantized Ministral/Devstral models; the other is pipeline object save_pretrained behavior. Not the same bug." }, { "left": "issue:42907", "right": "issue:45753", "accept": false, - "reason": "Completely different symptoms: save failure vs Qwen3_5 infinite generation loop on a specific image." + "reason": "Save/dequantization failure versus an infinite loop on a specific image in Qwen3_5; unrelated symptoms and code paths." }, { "left": "issue:43825", "right": "issue:44655", "accept": false, - "reason": "Different functionality: pipeline task-support error message vs inability to save Pipeline objects." + "reason": "Pipeline translation-task warning and pipeline save_pretrained are separate issues; same subsystem only at a very broad level." }, { "left": "issue:43012", "right": "issue:43550", "accept": false, - "reason": "Different problems: a warning emitted during compile vs a Bamba SDPA compile failure." + "reason": "One is a compile-time precision warning for bfloat16 models; the other is a Bamba SDPA torch.compile failure. Different concrete bugs." }, { "left": "issue:45712", "right": "issue:46298", "accept": false, - "reason": "One is leftover dummy classes leaking into imports/checks; the other is a noisy logging message." + "reason": "Repo hygiene around dummy classes leaking without torch is unrelated to a noisy logging message." }, { "left": "issue:43208", "right": "issue:43550", "accept": false, - "reason": "Different models and failure modes: xLSTM training bugs vs Bamba SDPA/torch.compile failure." + "reason": "xLSTM training bugs and Bamba torch.compile/SDPA failure are distinct model-specific problems." }, { "left": "issue:43012", "right": "issue:44368", "accept": false, - "reason": "Both mention warnings, but about unrelated causes: bfloat16 compile warning vs tie_word_embeddings warning in Qwen3.5 LoRA fine-tuning." + "reason": "A PyTorch precision warning on compile is not the same as a tie_word_embeddings warning during Qwen3.5 LoRA fine-tuning." }, { "left": "issue:43992", "right": "issue:44484", "accept": false, - "reason": "Different area entirely: missing embed_tokens weight in UMT5 loading vs save_pretrained shard-size question." + "reason": "Missing embed_tokens.weight in UMT5Encoder.from_pretrained is unrelated to the default max_shard_size question." }, { "left": "issue:43352", "right": "issue:43976", "accept": false, - "reason": "Nemotron Flash Attention support vs a Python version compatibility problem; unrelated." + "reason": "Flash Attention support in NemotronH and Python-version compatibility for Transformers 5.1.0 are unrelated." }, { "left": "issue:43352", "right": "issue:44380", "accept": false, - "reason": "Different attention-related issues: unsupported Flash Attention 2.0 vs GPT2 attention scaling being ignored under SDPA/FlashAttention." + "reason": "Nemotron Flash Attention support and GPT2 attention scaling being ignored are different model/backend bugs." }, { "left": "issue:43066", "right": "issue:43906", "accept": false, - "reason": "One is a tokenizer decoder type regression; the other is an isolated reproduction of a separate issue." + "reason": "Wrong tokenizer decoder type in v5 is not the same as an isolated reproduction of another issue." }, { "left": "issue:44162", "right": "issue:46172", "accept": false, - "reason": "Different model families and failures: ESM2 breakage vs a DeepSeek dtype/transpose bug." + "reason": "ESM2 breakage and a deepseek-v4 dtype/transpose bug are different model-specific failures." }, { "left": "issue:43097", "right": "issue:46298", "accept": false, - "reason": "Removed config attribute warning vs generic noisy logging; not the same defect." + "reason": "Removal of tie_embeddings_and_encoder_decoder and a noisy logging message are unrelated." }, { "left": "issue:44380", "right": "issue:46132", "accept": false, - "reason": "Attention scaling ignored in GPT2 vs registered function behavior changing in AttentionInterface.register." + "reason": "GPT2 SDPA/FlashAttention scaling is unrelated to AttentionInterface.register changing the registered function behavior." }, { "left": "issue:45584", "right": "issue:46365", "accept": false, - "reason": "Whisper empty transcription failure vs importing Trainer being affected by datasets installation." + "reason": "Whisper empty-transcription generation failure and an import conflict with datasets are unrelated." }, { "left": "issue:44077", "right": "issue:46365", "accept": false, - "reason": "PatchTSMixer post_init validation vs Trainer import conflict with datasets; unrelated." + "reason": "patchtsmixer post_init validation and Trainer import issues caused by datasets are not the same bug." }, { "left": "issue:44222", "right": "issue:45478", "accept": false, - "reason": "Related to MoE and fp8, but one is save_pretrained and the other is from_pretrained; distinct code paths and symptoms." + "reason": "FP8 save_pretrained problems for moe models are distinct from a Qwen3.5 MoE from_pretrained error." }, { "left": "issue:36683", "right": "issue:43582", "accept": false, - "reason": "Gemma3Config missing vocab_size vs Apple Silicon caching_allocator_warmup TypeError; no real overlap." + "reason": "Gemma3Config missing vocab_size and an Apple Silicon caching_allocator_warmup TypeError are unrelated." }, { "left": "issue:44380", "right": "issue:45479", "accept": false, - "reason": "GPT2 attention scaling bug vs degenerate zero-loss classification behavior; unrelated." + "reason": "GPT2 attention scaling under SDPA/FlashAttention is a different bug from zero-loss classification with num_labels=1." }, { "left": "issue:42175", "right": "issue:46365", "accept": false, - "reason": "Packaging/backend-selection issue vs Trainer import conflict with datasets; not the same root cause." + "reason": "TensorFlow not being included in the torch extra and Trainer import failing with datasets are distinct dependency issues." }, { "left": "issue:42757", "right": "issue:43824", "accept": false, - "reason": "Different import errors from different packages/classes: is_offline_mode vs Qwen2_5_VLForConditionalGeneration." + "reason": "Missing is_offline_mode import and missing Qwen2_5_VLForConditionalGeneration export are different import regressions." }, { "left": "issue:43064", "right": "issue:44380", "accept": false, - "reason": "Distributed training optimizer-state bug vs attention scaling behavior; unrelated." + "reason": "FSDP2/PEFT optimizer state corruption and GPT2 attention scaling are unrelated training/backend issues." }, { "left": "issue:42490", "right": "issue:43012", "accept": false, - "reason": "Local-path slash warning vs bfloat16 compile warning; different subsystems and symptoms." + "reason": "Trailing-slash local model loading warning is unrelated to the bfloat16 compile precision warning." }, { "left": "issue:43208", "right": "issue:43526", "accept": false, - "reason": "xLSTM training blockers vs BeitImageProcessorFast reduce_labels bug; unrelated." + "reason": "xLSTM training bugs and BeitImageProcessorFast reduce_labels returning one label are different code paths." }, { "left": "issue:44483", "right": "issue:44514", "accept": false, - "reason": "OpenAI-style chat/completions API rejection vs batched apply_chat_template crash; different interfaces and failures." + "reason": "A v5 chat/completions request acceptance problem is not the same as a batched apply_chat_template crash." }, { "left": "issue:43618", "right": "issue:44380", "accept": false, - "reason": "CLIPOutput attentions missing vs GPT2 attention scaling ignored; not the same bug." + "reason": "CLIPOutput attentions no longer assigned and GPT2 attention scaling ignored are separate model behavior bugs." }, { "left": "issue:38617", "right": "issue:43502", "accept": false, - "reason": "ImportError for layer_type_validation vs local_files_only network requests; different failure modes." + "reason": "Importing layer_type_validation and making API requests despite local_files_only=True are both load-path symptoms, but they are clearly different defects." }, { "left": "issue:44038", "right": "issue:44230", "accept": false, - "reason": "Different Qwen/MoE issues: generic 5.0 breakage for Qwen3-VL-Moe vs fp8 inference support." + "reason": "Qwen3-VL-Moe breakage and fp8 support for Qwen3-VL/Qwen3.5 MoE inference are different feature/compatibility issues." }, { "left": "issue:43756", "right": "issue:44485", "accept": false, - "reason": "RoPE layer-count mismatch in Smollm3 vs GLM-5 RoPE implementation discussion; not the same concrete change." + "reason": "Smollm3 RoPE layer mismatch and GLM-5 RoPE implementation discussion are not the same concrete bug." } ] }, @@ -17318,14 +17226,14 @@ "issue:43867|issue:45375", "issue:42890|issue:45561" ], - "prepared_review_unit_hash": "51eaa8e35d89e86f2e489c91f27010224c351b4757f75e13d3fee9df3839e7bc", + "prepared_review_unit_hash": "3f6f3809204bc985243ec8b22433072cc9bcf6afed97c77f091751b0ee3b7e5e", "component_budget": { "node_count": 500, "item_count": 500, "soft_pair_count": 1839, - "serialized_chars": 953913, - "estimated_input_tokens": 238479, - "estimated_eval_tokens": 477214 + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 }, "budget": { "node_count": 48, @@ -17340,216 +17248,216 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly heterogeneous regression/issues across different models, APIs, and runtime environments. None of the soft pairs look like the same underlying bug or change closely enough to merge.", - "confidence": 0.24, - "canonical_issue_reason": "issue:45278 is the broadest umbrella-style report here ('many import errors after update') and is the closest thing to a generic regression tracker, though it is not a true duplicate center for the rest of the set.", + "summary": "The cluster is mostly a set of unrelated regression/compatibility issues across different models, loaders, and utilities. None of the soft-edge pairs look like true duplicates under a strict same-bug standard, though a few are superficially adjacent (especially CLIP and Gemma/Qwen model-family regressions).", + "confidence": 0.83, + "canonical_issue_reason": "issue:43618 is the clearest, most concrete regression report in the set: it names a specific model/output field and points to a focused code path, making it a strong canonical issue anchor.", "canonical_pr_reason": null, - "best_issue_reason": "issue:45278 is the most broadly framed issue, making it the least specific candidate to stand in for a mixed cluster; however, the items are too unrelated for a strong canonical choice.", + "best_issue_reason": "issue:43618 is the best standalone issue candidate because it is narrowly scoped, actionable, and has a clear symptom-to-code-path mapping compared with the broader or more ambiguous reports.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:42994", "right": "issue:44655", "accept": false, - "reason": "Both involve save_pretrained, but one is quantized model saving and the other is Pipeline serialization; different objects and failure modes." + "reason": "Both mention saving, but one is quantized model saving and the other is pipeline save_pretrained; different objects and likely different fixes." }, { "left": "issue:43618", "right": "issue:43761", "accept": false, - "reason": "Both are CLIP/v5 regression reports, but one is about attentions being missing and the other about hidden_states being None; separate output-path bugs." + "reason": "Same CLIP family, but one is missing attentions and the other is missing hidden_states; related regression theme, not clearly the same bug." }, { "left": "issue:43299", "right": "issue:43867", "accept": false, - "reason": "Both are model-loading failures, but they point to different symptoms and likely different serialization/code-path issues." + "reason": "Both concern model loading regressions, but they affect different Qwen/serialization scenarios and do not point to the same code path." }, { "left": "issue:36010", "right": "issue:42757", "accept": false, - "reason": "Both are import errors, but one is GenerationMixin and the other is is_offline_mode from huggingface_hub; unrelated dependency breakages." + "reason": "Different import failures from different dependencies; no shared underlying defect is evident." }, { "left": "issue:44625", "right": "issue:46365", "accept": false, - "reason": "Qwen config propagation and Trainer import conflict are unrelated; they do not share a concrete bug or fix." + "reason": "A Qwen config propagation bug versus a Trainer import problem; unrelated subsystems and failure modes." }, { "left": "issue:45276", "right": "issue:46077", "accept": false, - "reason": "Same model family, but one is embedding resizing and the other is attention-mask construction; distinct code paths." + "reason": "Both are Gemma4 issues, but resizing embeddings and bidirectional attention masks are separate behaviors and code paths." }, { "left": "issue:44355", "right": "issue:46877", "accept": false, - "reason": "Compiled-file import errors and import-time docstring lint output are unrelated problems." + "reason": "Compiled-file import errors versus import-time docstring lint output; unrelated." }, { "left": "issue:43352", "right": "issue:43937", "accept": false, - "reason": "Flash Attention support gating and invalid GenerationConfig validation are different model/runtime issues." + "reason": "Different model-specific generation/attention failures; not the same concrete bug." }, { "left": "issue:44483", "right": "issue:46369", "accept": false, - "reason": "Chat/completions request handling and SAM LiteText memory overflow are unrelated." + "reason": "API request validation versus memory overflow in a model forward path; clearly different problems." }, { "left": "issue:36683", "right": "issue:43475", "accept": false, - "reason": "Missing Gemma3 vocab_size and missing SAM3 vision output attribute are separate model-specific attribute bugs." + "reason": "Different missing-attribute errors on different model families (Gemma3 vs SAM 3 Video)." }, { "left": "issue:43611", "right": "issue:44655", "accept": false, - "reason": "Base_model_prefix loading regression and Pipeline save_pretrained failure affect different flows." + "reason": "Base_model_prefix loading regression and pipeline save_pretrained failure are separate serialization issues." }, { "left": "issue:43901", "right": "issue:44655", "accept": false, - "reason": "One is a docs/behavior mismatch for TextClassificationPipeline; the other is a save_pretrained serialization issue." + "reason": "Docs mismatch for TextClassificationPipeline versus pipeline persistence; not the same bug." }, { "left": "issue:43867", "right": "issue:45216", "accept": false, - "reason": "Both concern loading/saving checkpoints, but one is a sorted state_dict load error and the other is incorrect Qwen3.5 save_pretrained output." + "reason": "One is loading failure with sorted state_dict, the other is incorrect checkpoint saving; related area but different directions and likely different fixes." }, { "left": "issue:43976", "right": "issue:44938", "accept": false, - "reason": "Both are Python-version compatibility reports, but for different versions and likely different breakpoints." + "reason": "Both are Python-version compatibility issues, but for different interpreter versions and different failure modes." }, { "left": "issue:36683", "right": "issue:44077", "accept": false, - "reason": "A missing config field and an optimizer-argument bug are unrelated." + "reason": "Different model/config attribute bug versus optimizer argument handling; unrelated." }, { "left": "issue:44625", "right": "issue:46077", "accept": false, - "reason": "Qwen3.5 label propagation and Gemma4 bidirectional attention masking are distinct issues." + "reason": "Qwen config propagation and Gemma4 attention-mask construction are unrelated." }, { "left": "issue:46077", "right": "issue:46082", "accept": false, - "reason": "Both are Gemma4 config/behavior issues, but one is attention masking and the other is head_dim validation; not the same bug." + "reason": "Gemma4 attention masking versus LlamaConfig head_dim validation; different code paths." }, { "left": "issue:43618", "right": "issue:45216", "accept": false, - "reason": "Different regressions: missing attentions vs incorrect checkpoint saving." + "reason": "CLIP output regression versus Qwen3.5 save_pretrained regression; different models and bug classes." }, { "left": "issue:43208", "right": "issue:46829", "accept": false, - "reason": "xLSTM training blockers and VideoPrism output wiring are unrelated." + "reason": "xLSTM training blockers versus VideoPrism output formatting; no shared bug." }, { "left": "issue:36683", "right": "issue:43525", "accept": false, - "reason": "Different config attribute omissions for different model families." + "reason": "Missing config fields in different model families; not the same defect." }, { "left": "issue:36683", "right": "issue:44625", "accept": false, - "reason": "One missing attribute is vocab_size; the other is num_labels propagation. Similar pattern, but not the same concrete bug." + "reason": "Both are config-related, but they affect different models and different missing fields." }, { "left": "issue:29870", "right": "issue:36296", "accept": false, - "reason": "Bart generation UnboundLocalError and tensor-parallel training bug are unrelated." + "reason": "Bart generate-time UnboundLocalError versus tensor-parallel training bug; unrelated." }, { "left": "issue:43992", "right": "issue:45310", "accept": false, - "reason": "Both involve from_pretrained loading, but they affect different models with different missing-weight behavior." + "reason": "Both involve Qwen3.5 from_pretrained, but one is a missing encoder weight and the other is a broader model loading error; not clearly the same issue." }, { "left": "issue:43064", "right": "issue:44492", "accept": false, - "reason": "Distributed optimizer-state bug and cache-strategy typo are unrelated." + "reason": "Optimizer-state corruption in distributed training versus a typo in cache strategies; unrelated." }, { "left": "issue:43010", "right": "issue:45278", "accept": false, - "reason": "Cache update no_grad behavior and broad import failures are different classes of bug." + "reason": "no_grad decoration for cache updates versus broad import errors after upgrade; different fixes and code paths." }, { "left": "issue:44075", "right": "issue:46365", "accept": false, - "reason": "SGD arguments ignored and Trainer import failure are unrelated." + "reason": "Optimizer argument handling versus Trainer import failure; unrelated." }, { "left": "issue:44908", "right": "issue:46851", "accept": false, - "reason": "Scheduler kwargs bug and Qwen3.5 variable-length position_ids failure are unrelated." + "reason": "Scheduler kwargs bug versus variable-length position_ids handling; different components." }, { "left": "issue:29870", "right": "issue:33453", "accept": false, - "reason": "Bart generate() UnboundLocalError and tokenizer-loading regression are different failures." + "reason": "Bart generation mask bug versus tokenizer loading regression; different subsystems." }, { "left": "issue:44938", "right": "issue:45071", "accept": false, - "reason": "Python 3.14 import failure and PretrainedConfig type-checking regression are separate compatibility issues." + "reason": "Python 3.14 load failure versus PretrainedConfig type-checking regression; different issues." }, { "left": "issue:44355", "right": "issue:44855", "accept": false, - "reason": "Compiled-file import errors and Python 3.13 JIT parsing failure are unrelated runtime/import bugs." + "reason": "Compiled Python import errors versus a Python 3.13 @torch.jit.script parsing problem; unrelated." }, { "left": "issue:43867", "right": "issue:46202", "accept": false, - "reason": "A load-model state_dict sorting problem and a Gemma pipeline roundtrip issue are different serialization flows." + "reason": "State_dict loading failure versus Gemma pipeline roundtrip on a specific token; different failure modes." }, { "left": "issue:36683", "right": "issue:43064", "accept": false, - "reason": "Missing config field and FSDP2/PEFT optimizer state corruption are unrelated." + "reason": "A config attribute bug versus distributed optimizer state corruption; unrelated." }, { "left": "issue:43867", "right": "issue:45375", "accept": false, - "reason": "Generic load-model error and missing Qwen3_5MoeVisionConfig field are distinct schema/serialization bugs." + "reason": "Different Qwen/Gemma serialization issues with different missing fields and paths." }, { "left": "issue:42890", "right": "issue:45561", "accept": false, - "reason": "Both touch test infrastructure, but one is flaky seeding in a model test and the other is an xdist race on a captured-info file." + "reason": "Missing set_seed in tests versus xdist race on a captured info file; test stability issues, but not the same bug." } ] }, @@ -17654,17 +17562,17 @@ "issue:45125|issue:45375", "issue:46664|issue:46726", "issue:43010|issue:43232", - "issue:43502|issue:44222", - "issue:43502|issue:44961" + "issue:43502|issue:44961", + "issue:43502|issue:44222" ], - "prepared_review_unit_hash": "d15b6cf886fd389c3ea01a4a79e20398e9e388a842e411c19ac98dee6a49bdc0", + "prepared_review_unit_hash": "c4565b6fd063ed56f5fbfd45c72718c85d2e246c36114b58ac4eb80e260733ef", "component_budget": { "node_count": 500, "item_count": 500, "soft_pair_count": 1839, - "serialized_chars": 953913, - "estimated_input_tokens": 238479, - "estimated_eval_tokens": 477214 + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 }, "budget": { "node_count": 48, @@ -17679,217 +17587,214 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is not a true duplicate set: it mixes unrelated import, tokenizer, tensor-parallel, caching, and model-specific regression reports. No PRs are present; all soft pairs should be rejected.", - "confidence": 0.28, - "canonical_issue_reason": "No single canonical issue stands out because the cluster is heterogeneous; if forced, issue 43502 is the broadest user-facing loader/cache regression, but it does not unify the rest of the items.", + "summary": "This cluster is mostly a grab bag of unrelated Transformers issues: import errors, tokenizer/model loading regressions, tensor-parallel problems, and model-specific bugs. The only clear soft duplicate is the import-time slowdown pair around `import transformers` / `import *` in models.", + "confidence": 0.83, + "canonical_issue_reason": "Issue 41669 is the clearest, most actionable root-cause report: it identifies a specific import-time regression from `import *` in models and matches the only near-duplicate slow-import report.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 43502 is the best single issue to keep as a representative, concrete regression report about offline/loading behavior, but it should not absorb the other disparate bugs.", + "best_issue_reason": "Best overall representative because it is concrete, scoped to one code path, and describes a fixable performance regression rather than a vague symptom or unrelated model-specific bug.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43010", "right": "issue:45850", "accept": false, - "reason": "Both mention regressions after v5, but one is about cache/update no_grad semantics and the other is a vague broad-breaking-models report; not the same bug." + "reason": "Different problems: `@torch.no_grad` on cache/layer updates vs a broad complaint that v5 breaks many models." }, { "left": "issue:41669", "right": "issue:44246", - "accept": false, - "reason": "Import-star cleanup causing startup cost and intermittent import slowness are related only broadly; they are different performance complaints with different fixes." + "accept": true, + "reason": "Both describe the same underlying issue: slow `transformers` import. 41669 gives the specific root cause (`import *` in models), while 44246 is the generic symptom report." }, { "left": "issue:36296", "right": "issue:43575", "accept": false, - "reason": "Both touch tensor parallelism, but one is a training bug and the other is OOM on loading a specific model; not the same concrete failure path." + "reason": "Both mention tensor parallelism, but one is a training bug and the other is an OOM while loading a specific model; too different to treat as the same bug." }, { "left": "issue:43010", "right": "issue:43976", "accept": false, - "reason": "Decorating cache updates with no_grad is unrelated to the Python version compatibility problem." + "reason": "Unrelated: generation/cache mutation behavior vs Python version compatibility." }, { "left": "issue:43618", "right": "issue:44077", "accept": false, - "reason": "CLIP attentions assignment and patchtsmixer post_init policy are unrelated model/API issues." + "reason": "Different model issues with different symptoms and code paths." }, { "left": "issue:43992", "right": "issue:45478", "accept": false, - "reason": "Missing UMT5 embed weights and Qwen3.5 Moe from_pretrained failures are different model-loading bugs." + "reason": "Both are `from_pretrained` failures, but they involve different models and different failure modes." }, { "left": "issue:43867", "right": "issue:45357", "accept": false, - "reason": "A state_dict sorting load error and incorrect visual encoder keys are distinct checkpoint-loading regressions." + "reason": "Different regression classes: state-dict loading order issue vs incorrect visual encoder keys during save." }, { "left": "issue:28282", "right": "issue:33453", "accept": false, - "reason": "PyTorch missing ImportError and tokenizer-loading regression are unrelated import/runtime issues." + "reason": "Completely different issues: missing PyTorch dependency vs tokenizer loading regression." }, { "left": "issue:43618", "right": "issue:46285", "accept": false, - "reason": "Both involve CLIP, but one is about attentions output and the other is a structural API inconsistency; not the same underlying change." + "reason": "`CLIPOutput.attentions` assignment bug is unrelated to the CLIP text model structural inconsistency." }, { "left": "issue:44360", "right": "issue:45440", "accept": false, - "reason": "A DSA indexer ReLU discussion is unrelated to DeepSeekV3MoE implementation divergence." + "reason": "Different architectures and failures: DSA indexer activation issue vs DeepSeekV3 implementation divergence." }, { "left": "issue:42890", "right": "issue:44964", "accept": false, - "reason": "Seed-related flaky tests and a Phi-4 multimodal loading failure are unrelated." + "reason": "Test flakiness due to missing `set_seed()` is not the same as a model load failure for Phi-4 multimodal." }, { "left": "issue:42890", "right": "issue:44112", "accept": false, - "reason": "Test nondeterminism from missing set_seed is unrelated to the stale device override CI failure." + "reason": "Both involve tests/CI, but the failures are different and not the same underlying bug." }, { "left": "issue:42890", "right": "issue:45600", "accept": false, - "reason": "Tokenization test seeding issues do not match removed Sam3LiteText config references in auto_mappings." + "reason": "Test seed instability and broken auto-mappings reference are unrelated." }, { "left": "issue:42898", "right": "issue:46664", "accept": false, - "reason": "Tokenizer-space cleanup behavior and video num_frames semantics are separate behavior changes." + "reason": "Tokenizer cleanup behavior change and video processor `num_frames` semantics are unrelated." }, { "left": "issue:43502", "right": "issue:44062", "accept": false, - "reason": "Offline API-request leakage and AddedToken keyword collision are unrelated loader/tokenizer bugs." + "reason": "`local_files_only=True` still making requests is unrelated to `AddedToken(... special=...)` argument duplication." }, { "left": "issue:44734", "right": "issue:46617", "accept": false, - "reason": "KV-cache continuation indexing in serve and trust_remote_code cache FileNotFoundError are different code paths." + "reason": "KV-cache continuation crash in serve is unrelated to cache-loading `FileNotFoundError` for trust-remote-code models." }, { "left": "issue:43723", "right": "issue:44556", "accept": false, - "reason": "Tokenizer loading regression in v5 and checkpoint reload-after-upgrade failures are both load-related but not the same bug." + "reason": "Related only at a very high level; one is tokenizer loading and the other is checkpoint reload after version upgrade, not clearly the same bug." }, { "left": "issue:44222", "right": "issue:45125", "accept": false, - "reason": "FP8 save_pretrained behavior and missing tensor-parallel plan are separate model serialization vs parallelism issues." + "reason": "FP8 save-pretrained behavior and missing `_tp_plan` for tensor parallelism are different issues." }, { "left": "issue:44387", "right": "issue:44556", "accept": false, - "reason": "Int4 quantization memory regression and reload failure after upgrade are different problems." + "reason": "Int4 memory regression/OOM and checkpoint reload compatibility are different failure modes." }, { "left": "issue:43824", "right": "issue:46617", "accept": false, - "reason": "A missing import/export for Qwen2_5_VL and a cache FileNotFoundError are unrelated." + "reason": "Missing exported model class vs cache `FileNotFoundError` are unrelated." }, { "left": "issue:43502", "right": "issue:44485", "accept": false, - "reason": "local_files_only API leakage is unrelated to GLM-5 RoPE implementation discussion." + "reason": "`local_files_only` network calls and GLM-5 RoPE implementation are unrelated." }, { "left": "issue:43874", "right": "issue:46617", "accept": false, - "reason": "Missing image-processor method and trust_remote_code cache lookup failure are distinct issues." + "reason": "Missing image-processor method causing multimodal token counting failure is unrelated to cache file lookup failure." }, { "left": "issue:43576", "right": "issue:46132", "accept": false, - "reason": "Broken env command and AttentionInterface.register behavior change are unrelated CLI/API regressions." - }, - { - "left": "issue:43010", - "right": "issue:43065", - "accept": false, - "reason": "no_grad decoration for cache updates and a dummy Conv2d in Sam3PixelDecoder are not the same bug." + "reason": "Broken CLI `transformers env` and `AttentionInterface.register` behavior change are different issues." }, { "left": "issue:45125", "right": "issue:45478", "accept": false, - "reason": "Both concern Qwen3.5 MoE, but one is specifically missing _tp_plan and the other is a broader from_pretrained error; not enough evidence to merge." + "reason": "Both concern Qwen3.5 MoE, but one is tensor-parallel metadata and the other is a broader `from_pretrained` error; not enough to merge as one bug." }, { "left": "issue:30333", "right": "issue:33453", "accept": false, - "reason": "MLFlow reporting after failed training and tokenizer loading regression are unrelated." + "reason": "MLflow job-reporting failure and tokenizer loading regression are unrelated." }, { "left": "issue:44355", "right": "issue:44749", "accept": false, - "reason": "Compiled Python file errors and post-upgrade data filtering slowdown are different symptoms and fixes." + "reason": "Compiled Python-file errors and a performance regression after version upgrade are unrelated." }, { "left": "issue:42548", "right": "issue:44351", "accept": false, - "reason": "Different missing public imports from transformers; both are ImportErrors, but not the same symbol/path regression." + "reason": "Different import errors for different symbols; same package, but not the same underlying bug." }, { "left": "issue:45125", "right": "issue:45375", "accept": false, - "reason": "Missing tensor-parallel plan and missing vision config field are unrelated Qwen3.5 issues." + "reason": "Missing tensor-parallel plan and missing vision config field are different Qwen3.5-MoE issues." }, { "left": "issue:46664", "right": "issue:46726", "accept": false, - "reason": "Video processor num_frames semantics and AutoProcessor backend-missing image-processor handling are different multimodal issues." + "reason": "Video processor `num_frames` semantics and AutoProcessor backend-missing error are unrelated." }, { "left": "issue:43010", "right": "issue:43232", "accept": false, - "reason": "Cache update no_grad decoration and generation kwargs update after sync_gpus are different generation-path bugs." + "reason": "Cache/layer `no_grad` concern is unrelated to generation kwargs update after `sync_gpus`." }, { "left": "issue:43502", - "right": "issue:44222", + "right": "issue:44961", "accept": false, - "reason": "Offline loading/API request leakage and FP8 save_pretrained behavior are unrelated." + "reason": "`local_files_only` network requests and the unrelated `racoon` issue are not duplicates." }, { "left": "issue:43502", - "right": "issue:44961", + "right": "issue:44222", "accept": false, - "reason": "A concrete loading bug and a non-descriptive unrelated issue title do not indicate the same bug." + "reason": "Network requests despite `local_files_only` is unrelated to FP8 MoE save/pretrained behavior." } ] }, - "evaluator_result": null, - "evaluator_used": false, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet, and the one accepted soft duplicate (issue:41669 vs issue:44246) is conservative and well supported by the titles. The remaining soft-edge rejections are also appropriately cautious; no overstatement or mergeability issue is apparent." + }, + "evaluator_used": true, "retried": false, - "accepted_nontrivial_soft_edge": false, + "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null }, @@ -17991,238 +17896,238 @@ "issue:44393|issue:45753", "issue:43576|issue:44556" ], - "prepared_review_unit_hash": "ade34ac0ce8f78d0374bcd0a979a044170311391b43eb8f408697076a1340f9a", + "prepared_review_unit_hash": "61fdb544773a2c6576db7c378a82481f9c0003e5c4bb77143bc1c4dd4ac1c301", "component_budget": { "node_count": 500, "item_count": 500, "soft_pair_count": 1839, - "serialized_chars": 953913, - "estimated_input_tokens": 238479, - "estimated_eval_tokens": 477214 + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 34, - "serialized_chars": 36919, - "estimated_input_tokens": 9230, - "estimated_eval_tokens": 18716 + "serialized_chars": 36924, + "estimated_input_tokens": 9231, + "estimated_eval_tokens": 18718 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is largely heterogeneous: it mixes unrelated loading/import regressions, docs issues, model-specific bugs, and a few broad v5 breakage reports. Under strict duplicate triage, none of the soft edges are strong enough to merge.", - "confidence": 0.62, - "canonical_issue_reason": "Issue 45850 is the broadest umbrella-style report in the set and best matches the cluster\u2019s repeated theme of v5 regressions affecting many models, even though the overall cluster is still too mixed to treat as one duplicate group.", + "summary": "The items are mostly a heterogeneous mix of unrelated regressions, import errors, docs issues, and model-specific bugs. None of the soft-similarity pairs look like the same underlying defect/change, so all candidate duplicate merges should be rejected.", + "confidence": 0.94, + "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "Issue 44556 is the most actionable standalone issue here: it states a concrete regression, clear version upgrade context, and a reproducible loading failure, making it the best issue to anchor follow-up triage.", + "best_issue_reason": "issue:43576 is the clearest concrete regression report in the set and the most plausible representative issue, but the cluster is too heterogeneous to treat it as a true canonical duplicate target.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:30990", "right": "issue:31795", "accept": false, - "reason": "SentenceTransformers loading hang and a forward-docs confusion are different problems in different layers." + "reason": "Different problems: a loading hang vs documentation confusion in `model.forward`." }, { "left": "issue:43827", "right": "issue:44961", "accept": false, - "reason": "Docs about pipeline removals are unrelated to the unrelated placeholder issue titled 'racoon'." + "reason": "One is a docs cleanup after pipeline removals; the other is an unrelated issue with no matching bug signal." }, { "left": "issue:30333", "right": "issue:36296", "accept": false, - "reason": "MLflow failure reporting and tensor-parallel training are distinct behaviors with no shared code-path evidence." + "reason": "MLFlow reporting on training failure is unrelated to tensor-parallel training behavior." }, { "left": "issue:43299", "right": "issue:44556", "accept": false, - "reason": "Both are loading regressions, but they target different model families and different version contexts without a concrete shared failure mode." + "reason": "Both involve v5 regressions, but one is Qwen3VL MoE loading and the other is checkpoint reload compatibility; not the same bug." }, { "left": "issue:29870", "right": "issue:31795", "accept": false, - "reason": "A generation-time UnboundLocalError is not the same as a documentation clarity issue." + "reason": "A BART generation `UnboundLocalError` is not the same as a documentation inconsistency." }, { "left": "issue:44355", "right": "issue:45071", "accept": false, - "reason": "Compiled-file errors and PretrainedConfig type-checking regressions are separate failures." + "reason": "Compiled-file execution errors are different from `PretrainedConfig` type-checking breakage." }, { "left": "issue:33453", "right": "issue:34689", "accept": false, - "reason": "Tokenizer-loading regression and Llama 3.2 Vision model-loading breakage are different symptoms and likely different root causes." + "reason": "Both are loading regressions, but they affect different model-loading paths and likely different causes." }, { "left": "issue:42757", "right": "issue:43576", "accept": false, - "reason": "A missing dependency import and a broken CLI env command are not the same bug." + "reason": "Importing `is_offline_mode` from `huggingface_hub` is a separate import regression from the broken `transformers env` command." }, { "left": "issue:43576", "right": "issue:45071", "accept": false, - "reason": "The env command regression is unrelated to the PretrainedConfig typing failure." + "reason": "A CLI command regression and a config type-checking regression are different failure modes." }, { "left": "issue:44315", "right": "issue:44360", "accept": false, - "reason": "Liger Kernel application during model_init and a DSA-indexer ReLU discussion are unrelated subsystems." + "reason": "`model_init` losing Liger Kernel application and a DSA/ReLU discussion are unrelated." }, { "left": "issue:31795", "right": "issue:38617", "accept": false, - "reason": "Documentation confusion and an ImportError for layer_type_validation do not share an underlying bug." + "reason": "Docs about `input_ids`/`past_key_values` do not match an `ImportError` for `layer_type_validation`." }, { "left": "issue:43576", "right": "issue:43824", "accept": false, - "reason": "A broken env command and a missing Qwen2.5-VL export are different issues." + "reason": "Broken environment CLI and missing `Qwen2_5_VLForConditionalGeneration` export are distinct issues." }, { "left": "issue:44961", "right": "issue:45468", "accept": false, - "reason": "The titles provide no evidence of the same model, feature, or failing code path." + "reason": "The placeholder 'racoon' issue is not the same as the Gemma-4 audio positional encoding bug." }, { "left": "issue:44485", "right": "issue:44961", "accept": false, - "reason": "GLM-5 RoPE implementation and an unrelated 'racoon' report do not match." + "reason": "GLM-5 RoPE implementation discussion is unrelated to the placeholder issue." }, { "left": "issue:44351", "right": "issue:46617", "accept": false, - "reason": "A top-level import export problem and a cache FileNotFoundError are different loading failures." + "reason": "A missing `HybridCache` import is different from a cache-file `FileNotFoundError` when loading from cache." }, { "left": "issue:44908", "right": "issue:44961", "accept": false, - "reason": "Scheduler kwargs handling is unrelated to the unrelated 'racoon' issue." + "reason": "Scheduler kwargs being ignored has nothing to do with the placeholder issue." }, { "left": "issue:43576", "right": "issue:44938", "accept": false, - "reason": "The v5 env command bug and Python 3.14 loading failure are distinct regressions." + "reason": "Both mention v5 breakage, but one is the `transformers env` command and the other is Python 3.14 load failure." }, { "left": "issue:36296", "right": "issue:43901", "accept": false, - "reason": "Tensor-parallel training and pipeline docs behavior are not the same underlying change." + "reason": "Tensor-parallel training bugs are unrelated to `return_all_scores` documentation mismatch." }, { "left": "issue:46143", "right": "issue:46144", "accept": false, - "reason": "Both mention RoFormer, but one is about kwargs propagation and the other about attention-interface usage; that is not enough to prove one concrete duplicate bug." + "reason": "Missing `**kwargs` forwarding and RoFormer attention-interface usage are related broadly, but not the same concrete bug." }, { "left": "issue:43475", "right": "issue:46617", "accept": false, - "reason": "A SAM3 vision output attribute error and a cache FileNotFoundError are unrelated failures." + "reason": "A missing SAM3 vision output attribute is unrelated to trust-remote-code cache loading failures." }, { "left": "issue:44222", "right": "issue:44230", "accept": false, - "reason": "Saving FP8 MoE artifacts and FP8 inference support are different code paths and different problems." + "reason": "One is about FP8 `save_pretrained`, the other about FP8 inference support; different code paths." }, { "left": "issue:41628", "right": "issue:42548", "accept": false, - "reason": "These are separate missing-symbol import errors for different top-level exports." + "reason": "`AutoImageProcessor` and `PreTrainedModel` import errors are separate symbol/export problems." }, { "left": "issue:44315", "right": "issue:46209", "accept": false, - "reason": "Model-init kernel application and FP8 inference on B200 are unrelated issues." + "reason": "Liger Kernel application with `model_init` is unrelated to FP8 inference on B200." }, { "left": "issue:44485", "right": "issue:46664", "accept": false, - "reason": "RoPE implementation discussion and video num_frames semantics do not share a code path." + "reason": "GLM-5 RoPE implementation and SmolVLM `num_frames` semantics are different model-specific issues." }, { "left": "issue:43673", "right": "issue:44492", "accept": false, - "reason": "Chunked-prefill cache missing and a cache-strategy typo are clearly different bugs." + "reason": "A functional cache-missing bug is not the same as a typo in cache strategy text." }, { "left": "issue:45362", "right": "issue:45850", "accept": false, - "reason": "A specific Qwen3.5 chat crash is not the same as a broad 'many models break' report." + "reason": "A Qwen3.5-35B chat crash is much more specific than a broad complaint that v5 breaks many models." }, { "left": "issue:44273", "right": "issue:45542", "accept": false, - "reason": "Lazy loading behavior and the TensorBoard-without-TF backend error are unrelated." + "reason": "Lazy loading problems and TensorBoard/TensorFlow backend detection are different defects." }, { "left": "issue:44623", "right": "issue:45230", "accept": false, - "reason": "A concrete missing-files bug in processor.save_pretrained is not the same as a generic bug report." + "reason": "Missing files from `processor.save_pretrained()` is not the same as a generic bug report." }, { "left": "issue:44360", "right": "issue:46664", "accept": false, - "reason": "A DSA indexer ReLU discussion and SmolVLM video processor semantics are unrelated." + "reason": "A DSA/ReLU discussion is unrelated to SmolVLM video-frame semantics." }, { "left": "issue:44038", "right": "issue:45125", "accept": false, - "reason": "Both touch Qwen3 MoE, but one is a broad v5/model-loading complaint and the other is a specific missing _tp_plan tensor-parallel issue; not enough to merge as one duplicate." + "reason": "A Qwen3-VL loading bug and missing tensor-parallel plan for Qwen3_5Moe are separate model-family issues." }, { "left": "issue:44273", "right": "issue:45003", "accept": false, - "reason": "Lazy loading problems and unsafe sys.modules access are different failure modes." + "reason": "Lazy loading not working and unsafe `sys.modules` access are different loading-path bugs." }, { "left": "issue:44393", "right": "issue:45071", "accept": false, - "reason": "Qwen3-VL 2D bbox output issues and PretrainedConfig type checking are unrelated." + "reason": "Qwen3-VL 2D bbox output errors are unrelated to `PretrainedConfig` type-checking failures." }, { "left": "issue:44393", "right": "issue:45753", "accept": false, - "reason": "A bounding-box output error and an infinite-loop image-specific bug are different symptoms and likely different roots." + "reason": "Both touch Qwen3-VL/Qwen3.5, but the symptoms differ: bbox output errors vs an infinite loop on a specific image." }, { "left": "issue:43576", "right": "issue:44556", "accept": false, - "reason": "The env command regression and checkpoint reloading regression are both v5-related but not the same concrete bug." + "reason": "The v5 `transformers env` CLI break and checkpoint reload incompatibility are distinct regressions." } ] }, @@ -18336,268 +18241,268 @@ "issue:41669|issue:43704", "issue:31795|issue:34689" ], - "prepared_review_unit_hash": "dca001733e9b5ca7f9b4bc708d6affee9f26f214da5bc2453b00fe72585456aa", + "prepared_review_unit_hash": "b256177e7a3e7fb4bafee74583a503cc8fb6994ec6c762550612f040eeb298a5", "component_budget": { "node_count": 500, "item_count": 500, "soft_pair_count": 1839, - "serialized_chars": 953913, - "estimated_input_tokens": 238479, - "estimated_eval_tokens": 477214 + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 40, - "serialized_chars": 38540, - "estimated_input_tokens": 9635, - "estimated_eval_tokens": 19526 + "serialized_chars": 38535, + "estimated_input_tokens": 9634, + "estimated_eval_tokens": 19524 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster is mostly heterogeneous: the soft matches are driven by loose lexical overlap around imports, version regressions, caching, and model-loading issues, but they do not look like the same underlying bug/change. No PRs are present.", - "confidence": 0.84, - "canonical_issue_reason": "No strong canonical issue: the cluster spans many unrelated reports. If forced, issue:45850 is only the broadest umbrella for version-regression complaints, but it is too vague to serve as a true duplicate target.", + "summary": "This cluster is mostly a loose collection of Transformers v5 regression reports around model loading, import behavior, and config/API compatibility. The soft pairs are semantically related in places, but none are strong enough to treat as the same underlying bug/change.", + "confidence": 0.78, + "canonical_issue_reason": "45850 is the broadest umbrella in the set: it summarizes the general v5 breakage theme affecting many models and code paths.", "canonical_pr_reason": null, - "best_issue_reason": "Issue:45850 is the closest thing to an anchor because it is the broadest cross-model v5 breakage report, but it is still too generic to unify this cluster cleanly.", + "best_issue_reason": "45278 is the clearest representative of the recurring concrete pattern here: upgrade-related import/loading regressions across multiple models.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:41669", "right": "issue:43673", "accept": false, - "reason": "Different problems: import/package-lazy-loading performance vs generation cache missing in chunked prefill." + "reason": "Both relate to import/loading behavior, but one is about `import *` slowing imports and the other is a missing cache object during chunked prefill; different bugs." }, { "left": "issue:43525", "right": "issue:45850", "accept": false, - "reason": "Specific missing pad_token_id/config bug vs a vague broad v5 breakage complaint." + "reason": "`pad_token_id` missing in one config class is a specific attribute regression; the other is a broad complaint about many models breaking after v5." }, { "left": "issue:44556", "right": "issue:45092", "accept": false, - "reason": "Both involve loading older checkpoints, but the concrete failure modes differ (general v5 reload breakage vs remote-code/meta-init incompatibility)." + "reason": "Both are v5 loading incompatibilities, but one is checkpoint reload after upgrade and the other is remote-code/meta-init incompatibility for a specific model family." }, { "left": "issue:43572", "right": "issue:45850", "accept": false, - "reason": "Specific StableLmConfig padding-token regression vs a generic \u201cmany models break\u201d report." + "reason": "A missing `pad_token_idx` in one config is not the same underlying issue as a general 'many models break' regression report." }, { "left": "issue:44492", "right": "issue:45850", "accept": false, - "reason": "Cache-strategy typo/cleanup issue is unrelated to the broad v5 model-breakage complaint." + "reason": "A typo in cache-strategy docs/labels is unrelated to the broad v5 model-breakage report." }, { "left": "issue:45278", "right": "issue:45850", "accept": false, - "reason": "Generic upgrade/import breakage, but not the same concrete bug as the broad \u201ctoo many models\u201d report." + "reason": "Related theme, but one is a concrete import-error regression after upgrade while the other is a generic umbrella complaint; not the same bug." }, { "left": "issue:44230", "right": "issue:46095", "accept": false, - "reason": "FP8 inference support request vs a DeepSeek weight-conversion question; different features and code paths." + "reason": "Different requests entirely: FP8 inference support for Qwen models versus a DeepSeekV4 weight-conversion question." }, { "left": "issue:42371", "right": "issue:44492", "accept": false, - "reason": "TF32 API guidance vs a cache-strategy typo; no shared underlying bug." + "reason": "TF32 API guidance and a cache-strategy typo are unrelated issues." }, { "left": "issue:41669", "right": "issue:46218", "accept": false, - "reason": "Import slowness from lazy-loading/import-* usage vs an actual missing-module import error." + "reason": "`import *` performance complaints are not the same as a missing `Qwen3_5ForCausalLM` import/export error." }, { "left": "issue:43408", "right": "issue:44315", "accept": false, - "reason": "Wrong model-type warning vs Liger Kernel not being applied through model_init; unrelated behavior." + "reason": "Different model/config problems: a wrong model-type warning versus Liger Kernel not being applied with `model_init`." }, { "left": "issue:42890", "right": "issue:44908", "accept": false, - "reason": "Test seeding flakiness in SAM HQ vs inverse_sqrt scheduler kwargs being ignored." + "reason": "One is missing `set_seed()` in tests; the other is a scheduler ignoring kwargs. No shared underlying bug." }, { "left": "issue:44246", "right": "issue:44273", "accept": false, - "reason": "Both mention import/lazy-loading, but one is general slowness and the other is a specific lazy-loading malfunction; too weak to call the same bug." + "reason": "Both mention loading/lazy-loading, but one is intermittent import slowness and the other is a general lazy-loading failure; too broad to merge." }, { "left": "issue:43576", "right": "issue:44908", "accept": false, - "reason": "Broken transformers env command vs scheduler kwarg handling; different subsystems." + "reason": "`transformers env` is broken in v5, while the other is a scheduler keyword-argument bug; different code paths." }, { "left": "issue:43352", "right": "issue:44393", "accept": false, - "reason": "Flash Attention 2 support gap vs Qwen3-VL bounding-box output errors; unrelated." + "reason": "Flash Attention 2 support gating for Nemotron is unrelated to a Qwen3-VL bounding-box output bug." }, { "left": "issue:43976", "right": "issue:44393", "accept": false, - "reason": "Python-version compatibility regression vs Qwen3-VL 2D bbox output issue; not the same bug." + "reason": "A Python version compatibility issue is not the same as a Qwen3-VL hallucination/output problem." }, { "left": "issue:44908", "right": "issue:45071", "accept": false, - "reason": "Scheduler kwargs ignored vs PretrainedConfig type-checking breakage; different code paths." + "reason": "A scheduler kwargs bug and a `PretrainedConfig` type-checking regression are separate problems." }, { "left": "issue:44295", "right": "issue:45230", "accept": false, - "reason": "Position_ids buffer access error vs a generic bug report with no concrete overlap." + "reason": "Reading `position_ids` after buffer registration is a different failure mode than an unspecified bug report." }, { "left": "issue:44297", "right": "issue:45230", "accept": false, - "reason": "Tokenizer save_pretrained mismatch vs a generic bug report; no shared failure mode." + "reason": "Tokenizer save metadata mismatch is a specific serialization bug, not the same as the generic bug report." }, { "left": "issue:44704", "right": "issue:44908", "accept": false, - "reason": "Both involve kwargs, but one is processor/cached_file forwarding and the other is scheduler kwargs; not the same bug." + "reason": "`AutoProcessor.from_pretrained` not forwarding kwargs to `cached_file` is unrelated to an inverse-sqrt scheduler parameter bug." }, { "left": "issue:44492", "right": "issue:46328", "accept": false, - "reason": "Cache-strategy typo vs SequenceFeatureExtractor.pad performance regression; unrelated." + "reason": "A typo in cache strategies is not the same as `SequenceFeatureExtractor.pad` converting arrays inefficiently." }, { "left": "issue:41669", "right": "issue:44351", "accept": false, - "reason": "Import-* performance issue vs missing HybridCache symbol import error; different problems." + "reason": "Import-star cleanup and a `HybridCache` import failure are distinct import-related problems." }, { "left": "issue:41669", "right": "issue:45341", "accept": false, - "reason": "Import performance issue vs a small testing_utils bug; unrelated." + "reason": "A broad import-performance issue and a small testing-utils bug are not the same underlying change." }, { "left": "issue:43976", "right": "issue:45850", "accept": false, - "reason": "Python-version compatibility regression is not the same as the broad generic \u201cmany models break\u201d complaint." + "reason": "Python-version incompatibility is more specific than the broad 'many models break' v5 complaint, and they do not point to the same defect." }, { "left": "issue:44393", "right": "issue:45478", "accept": false, - "reason": "Different Qwen3.5/Qwen3-VL failures with no concrete shared code-path." + "reason": "Both are Qwen3.5/Qwen3-VL-related, but the symptoms are different model errors with no evidence of a shared root cause." }, { "left": "issue:43976", "right": "issue:46218", "accept": false, - "reason": "Python-version compatibility regression vs a missing Qwen3_5ForCausalLM import; not the same issue." + "reason": "A Python support regression and a missing module import are unrelated." }, { "left": "issue:44077", "right": "issue:44861", "accept": false, - "reason": "PatchTSMixer post_init policy change vs tied-weight-key AttributeError; different underlying bugs." + "reason": "One is an API/validation rule change for `post_init`; the other is an AttributeError in tied-weight key handling." }, { "left": "issue:31795", "right": "issue:33453", "accept": false, - "reason": "Documentation confusion about forward args vs tokenizer loading regression; unrelated." + "reason": "Documentation confusion about forward args is unrelated to a tokenizer-loading regression." }, { "left": "issue:43576", "right": "issue:44273", "accept": false, - "reason": "Broken env command vs lazy-loading malfunction; too broad and not the same concrete bug." + "reason": "A broken CLI environment command and a lazy-loading failure are too different to be duplicates." }, { "left": "issue:44315", "right": "issue:44829", "accept": false, - "reason": "Liger Kernel application bug vs flash_attention_3 degenerate training; different implementation problems." + "reason": "Liger Kernel integration not applied with `model_init` is a separate issue from degenerate training under flash_attention_3." }, { "left": "issue:44360", "right": "issue:45468", "accept": false, - "reason": "DSA indexer ReLU discussion vs Gemma-4 audio positional encoding bug; unrelated." + "reason": "A missing ReLU in the DSA indexer and Gemma-4 positional encoding are unrelated model-implementation issues." }, { "left": "issue:43519", "right": "issue:46655", "accept": false, - "reason": "Different multimodal processor math bugs in different models and code paths." + "reason": "Both are processor math bugs, but one is timestamp calculation and the other is video padding calculation; different code paths." }, { "left": "issue:44273", "right": "issue:44908", "accept": false, - "reason": "Lazy-loading issue vs scheduler kwargs bug; no shared underlying defect." + "reason": "Lazy loading failures and scheduler kwargs handling are unrelated." }, { "left": "issue:44655", "right": "issue:44861", "accept": false, - "reason": "Pipeline save_pretrained problem vs tied-weight-key AttributeError; unrelated." + "reason": "Saving pipeline objects and tied-weight key crashes are separate serialization/internal-state bugs." }, { "left": "issue:45230", "right": "issue:45310", "accept": false, - "reason": "Generic bug report vs Qwen3.5 MoE from_pretrained failure; too vague to match." + "reason": "An empty bug report does not identify the same issue as the Qwen3.5 MoE from_pretrained error." + }, + { + "left": "issue:43576", + "right": "issue:46218", + "accept": false, + "reason": "A broken environment CLI command and a missing Qwen3_5 model import are distinct failures." }, { "left": "issue:43519", "right": "issue:44485", "accept": false, - "reason": "Qwen3VL timestamp math vs GLM-5 RoPE implementation; different model components." + "reason": "Qwen3-VL timestamp math and GLM-5 RoPE implementation are both model-specific, but not the same bug." }, { "left": "issue:45468", "right": "issue:46894", "accept": false, - "reason": "Gemma-4 audio positional encoding vs CohereASR training-loss double-shift; unrelated." - }, - { - "left": "issue:43519", - "right": "issue:44908", - "accept": false, - "reason": "Processor timestamp calculation vs scheduler kwargs bug; different subsystems." + "reason": "Gemma-4 audio positional encoding and CohereASR training-loss shifting are different model/training bugs." }, { "left": "issue:41669", "right": "issue:43704", "accept": false, - "reason": "Import-time/lazy-loading performance issue vs VRAM leak in multithreaded dataloaders." + "reason": "Import-star cleanup and a VRAM leak in multi-dataloader-thread use are unrelated." }, { "left": "issue:31795", "right": "issue:34689", "accept": false, - "reason": "Documentation ambiguity vs a concrete model-loading regression; not the same bug." + "reason": "Forward-API documentation confusion is not the same as a model-loading failure for Llama 3.2 Vision." } ] }, @@ -18672,8 +18577,8 @@ "issue:46018|issue:46172", "issue:43704|issue:44485", "issue:44485|issue:45468", - "issue:43502|issue:43519", "issue:42371|issue:43704", + "issue:43502|issue:43519", "issue:45850|issue:46218", "issue:43519|issue:44623", "issue:44263|issue:45468", @@ -18714,298 +18619,298 @@ "issue:37934|issue:43668", "issue:42936|issue:43668" ], - "prepared_review_unit_hash": "ae0d602cd65bef3a4b5d684bd45b8f4a8d5cf7f122aa5bb29bca20150c305c78", + "prepared_review_unit_hash": "d608f88c678707decacc54b86967c736f1ba99820bec13aacec1a26fcd2f60ee", "component_budget": { "node_count": 500, "item_count": 500, "soft_pair_count": 1839, - "serialized_chars": 953913, - "estimated_input_tokens": 238479, - "estimated_eval_tokens": 477214 + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 }, "budget": { "node_count": 46, "item_count": 46, "soft_pair_count": 44, - "serialized_chars": 38876, - "estimated_input_tokens": 9719, - "estimated_eval_tokens": 19694 + "serialized_chars": 38877, + "estimated_input_tokens": 9720, + "estimated_eval_tokens": 19696 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a grab-bag of unrelated model-specific regressions, import errors, tokenizer complaints, and one broad v5-breakage report. None of the soft pairs look like the same concrete bug/change, so I would not merge any of them.", - "confidence": 0.77, - "canonical_issue_reason": "issue:45850 is the broadest umbrella report in the set (\"Since 5.0 version it breaks too many models\") and is the closest thing to a cluster anchor for the many v5 regression-style complaints.", + "summary": "This cluster is noisy and does not contain any clear duplicate issue/PR groupings. The soft pairs mostly share broad vocabulary (v5, tokenizer, attention, import, loading) but describe different models, subsystems, and failure modes. I rejected all soft edges and picked the most representative concrete issue as a weak anchor.", + "confidence": 0.58, + "canonical_issue_reason": "issue:31515 is the most concrete and broadly framed user-facing regression in the set, and it has the strongest soft connectivity, but there is still no true duplicate cluster here.", "canonical_pr_reason": null, - "best_issue_reason": "issue:45850 is the most representative issue for this cluster\u2019s recurring upgrade/regression theme; the other issues are either narrow model-specific bugs or unrelated one-offs.", + "best_issue_reason": "issue:31515 is the best single issue anchor because it is specific, actionable, and closest to a general underlying runtime/loading problem, even though it does not truly duplicate the others.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:46018", "right": "issue:46172", "accept": false, - "reason": "Both mention DeepSeek-V4, but they describe different bugs: shared-expert gating vs a comb.transpose weight path." + "reason": "Different models and bugs: DeepSeek-V4 gating vs RoFormer attention implementation." }, { "left": "issue:43704", "right": "issue:44485", "accept": false, - "reason": "Different models and different failure modes: Qwen3 VRAM leak vs GLM-5 RoPE implementation." + "reason": "VRAM leak in dataloader threads is unrelated to GLM-5 RoPE math." }, { "left": "issue:44485", "right": "issue:45468", "accept": false, - "reason": "Gemma-4 audio positional encoding and GLM-5 RoPE are unrelated code paths." + "reason": "Both are model-specific, but the bugs are different: RoPE implementation vs audio positional encoding." }, { - "left": "issue:43502", - "right": "issue:43519", + "left": "issue:42371", + "right": "issue:43704", "accept": false, - "reason": "Local-files-only network requests and Qwen3VL timestamp calculation are unrelated bugs." + "reason": "TF32 configuration API behavior is unrelated to a multi-thread VRAM leak." }, { - "left": "issue:42371", - "right": "issue:43704", + "left": "issue:43502", + "right": "issue:43519", "accept": false, - "reason": "TF32 API settings vs a Qwen3 VRAM leak; no shared concrete bug." + "reason": "local_files_only network requests and Qwen3VL timestamp calculation are different code paths." }, { "left": "issue:45850", "right": "issue:46218", "accept": false, - "reason": "Both are broad v5 complaints, but one is a generic breakage report and the other is a specific Qwen3_5 import failure." + "reason": "A vague version-wide breakage report is not the same bug as a specific Qwen3_5 import failure." }, { "left": "issue:43519", "right": "issue:44623", "accept": false, - "reason": "Timestamp calculation in Qwen3VL is unrelated to processor.save_pretrained missing files." + "reason": "Processor timestamp math and save_pretrained missing files are unrelated." }, { "left": "issue:44263", "right": "issue:45468", "accept": false, - "reason": "torch.split return values in GlmMoeDsaIndexer is unrelated to Gemma-4 audio positional encoding." + "reason": "torch.split return handling is unrelated to Gemma-4 audio positional encoding." }, { "left": "issue:44261", "right": "issue:45468", "accept": false, - "reason": "MLA q_a_layernorm eps precision issue and Gemma-4 positional encoding are different bugs." + "reason": "Missing rms_norm_eps in MLA q_a_layernorm does not match Gemma-4 audio positional encoding." }, { "left": "issue:43298", "right": "issue:46095", "accept": false, - "reason": "Audio continuous batching support and DeepSeek-V4 weight conversion are different feature requests/issues." + "reason": "Continuous batching for audio models and DeepSeek-V4 weight conversion are unrelated." }, { "left": "issue:43519", "right": "issue:43704", "accept": false, - "reason": "Qwen3VL timestamp math and Qwen3 VRAM leak are separate bugs in different areas." + "reason": "Timestamp calculation and VRAM leak in dataloader threads are different issues." }, { "left": "issue:44351", "right": "issue:46218", "accept": false, - "reason": "Both are import errors, but they involve different missing symbols/modules and likely different root causes." + "reason": "Both are import errors, but for different missing symbols/modules and different underlying causes." }, { "left": "issue:43352", "right": "issue:46218", "accept": false, - "reason": "Nemotron flash-attn support and Qwen3_5 import failure are different model-specific issues." + "reason": "Flash Attention support for NemotronH is unrelated to importing Qwen3_5ForCausalLM." }, { "left": "issue:42913", "right": "issue:43066", "accept": false, - "reason": "Both are tokenizer-v5 complaints, but one is a general behavior change and the other is a decoder-type mismatch; not enough evidence of the same bug." + "reason": "Both mention tokenizer v5, but one is a broad behavior regression and the other is a specific decoder-type bug." }, { "left": "issue:28348", "right": "issue:45250", "accept": false, - "reason": "GPT2-specific flash-attention support and a generic flash-attention topic are not the same concrete issue." + "reason": "Both mention Flash Attention 2.0, but they target different models and different requests." }, { "left": "issue:44908", "right": "issue:44998", "accept": false, - "reason": "inverse_sqrt scheduler kwargs bug is unrelated to the generic \"Unemployment\" issue." + "reason": "Scheduler kwargs handling is unrelated to the nonsensical 'Unemployment' issue." }, { "left": "issue:44492", "right": "issue:44998", "accept": false, - "reason": "Cache-strategy typo and the generic \"Unemployment\" issue are unrelated." + "reason": "A cache-strategy typo is not related to the unrelated 'Unemployment' issue." }, { "left": "issue:43618", "right": "issue:44998", "accept": false, - "reason": "CLIPOutput attentions regression is unrelated to the generic \"Unemployment\" issue." + "reason": "CLIPOutput attentions regression is unrelated to the unrelated 'Unemployment' issue." }, { "left": "issue:44998", "right": "issue:45850", "accept": false, - "reason": "A meaningless generic issue is not the same as the broad v5 breakage report." + "reason": "No relation between the unrelated 'Unemployment' issue and a broad v5 breakage report." }, { "left": "issue:44938", "right": "issue:44998", "accept": false, - "reason": "Python 3.14 load failure and the generic \"Unemployment\" issue are unrelated." + "reason": "Python 3.14 loading failure is unrelated to the unrelated 'Unemployment' issue." }, { "left": "issue:44861", "right": "issue:45290", "accept": false, - "reason": "Tied-weight-key crash and chat-template tool-call crash are different code paths." + "reason": "Tied-weight key crash and chat template tool-call crash are different failures." }, { "left": "issue:31515", "right": "issue:38617", "accept": false, - "reason": "Slow from_pretrained checkpoint loading is unrelated to the layer_type_validation import error." + "reason": "Slow from_pretrained loading and missing layer_type_validation import are unrelated." }, { "left": "issue:43572", "right": "issue:44861", "accept": false, - "reason": "StableLm pad_token_idx regression and tied-weight-key crash are different issues." + "reason": "Missing pad_token_idx in config is unrelated to tied-weight key handling." }, { "left": "issue:43525", "right": "issue:44861", "accept": false, - "reason": "Llama4 pad_token_id AttributeError is unrelated to tied-weight-key handling." + "reason": "Missing pad_token_id in Llama4Config is unrelated to tied-weight key handling." }, { "left": "issue:28348", "right": "issue:43352", "accept": false, - "reason": "Both concern flash attention, but for different models and different requests; not the same change." + "reason": "Both concern Flash Attention 2.0, but they are different models and different kinds of support requests." }, { "left": "issue:44230", "right": "issue:44623", "accept": false, - "reason": "Qwen3.5 fp8 inference support and processor.save_pretrained missing files are unrelated." + "reason": "FP8 support for Qwen3-VL/Qwen3.5 MoE is unrelated to processor.save_pretrained missing files." }, { "left": "issue:29870", "right": "issue:31515", "accept": false, - "reason": "BART generate() decoder_attention_mask bug is unrelated to slow checkpoint loading." + "reason": "Bart generate decoder mask crash and from_pretrained slowdown are unrelated." }, { "left": "issue:30333", "right": "issue:31515", "accept": false, - "reason": "MLflow reporting on failed jobs and slow checkpoint loading are unrelated." + "reason": "MLflow reporting on failed jobs is unrelated to checkpoint loading slowness." }, { "left": "issue:31515", "right": "issue:36296", "accept": false, - "reason": "Checkpoint loading speed and tensor-parallel training are different problems." + "reason": "from_pretrained performance and tensor-parallel training bug are different problems." }, { "left": "issue:31515", "right": "issue:31795", "accept": false, - "reason": "Checkpoint loading performance and documentation confusion about forward inputs are unrelated." + "reason": "Loading performance and documentation confusion are not the same issue." }, { "left": "issue:44230", "right": "issue:45310", "accept": false, - "reason": "FP8 inference support and Qwen3.5 Moe from_pretrained failure are different Qwen3.5 issues with different root causes." + "reason": "FP8 support and a from_pretrained error for Qwen3.5 MoE are not the same concrete bug." }, { "left": "issue:43298", "right": "issue:44230", "accept": false, - "reason": "Audio continuous batching and Qwen3.5 fp8 inference support are unrelated." + "reason": "Continuous batching for audio models is unrelated to FP8 support for Qwen3-VL/Qwen3.5 MoE." }, { "left": "issue:43717", "right": "issue:46144", "accept": false, - "reason": "Mamba init_weights behavior and RoFormer attention-interface usage are different architectures and bugs." + "reason": "Mamba init_weights usage is unrelated to RoFormer not using the attention interface." }, { "left": "issue:42936", "right": "issue:46018", "accept": false, - "reason": "Mask2Former ignore_value handling is unrelated to DeepSeek shared-expert gating." + "reason": "Mask2former ignore_value handling is unrelated to DeepSeek-V4 gating." }, { "left": "issue:43668", "right": "issue:44031", "accept": false, - "reason": "ModernBERT norm_eps typing and tokenizer regex warnings are unrelated." + "reason": "A config type hint bug is unrelated to tokenizer regex warning spam." }, { "left": "issue:42936", "right": "issue:43240", "accept": false, - "reason": "Mask2Former ignore_value and kwargs not passed to loss are different bugs." + "reason": "Ignored mask value handling and kwargs propagation to loss are different bugs." }, { "left": "issue:43240", "right": "issue:46144", "accept": false, - "reason": "Loss-function kwargs forwarding and RoFormer attention interface are unrelated." + "reason": "Loss kwargs propagation is unrelated to RoFormer attention interface use." }, { "left": "issue:42936", "right": "issue:44031", "accept": false, - "reason": "Mask2Former ignore_value and tokenizer regex warnings do not share a concrete code path." + "reason": "Mask2former ignore_value and tokenizer regex warnings are unrelated." }, { "left": "issue:42936", "right": "issue:46144", "accept": false, - "reason": "Mask2Former ignore_value and RoFormer attention implementation are unrelated." + "reason": "Mask2former ignore_value handling is unrelated to RoFormer attention interface usage." }, { "left": "issue:42936", "right": "issue:43813", "accept": false, - "reason": "Mask2Former ignore_value and the orig_conversion.quantization_operation typo are unrelated." + "reason": "A missing ignore_value use site is unrelated to a typo in quantization_operation." }, { "left": "issue:43240", "right": "issue:43668", "accept": false, - "reason": "Loss kwargs forwarding and ModernBERT type-hint correctness are unrelated." + "reason": "Loss kwargs propagation and a config type-hint bug are unrelated." }, { "left": "issue:43668", "right": "issue:43813", "accept": false, - "reason": "ModernBERT norm_eps typing and a quantization-operation typo are unrelated." + "reason": "Incorrect type hint and typo in quantization_operation are unrelated." }, { "left": "issue:37934", "right": "issue:43668", "accept": false, - "reason": "Both mention normalization, but one asks about Llama4TextL2Norm behavior and the other is just a ModernBERT type-hint issue." + "reason": "Llama4TextL2Norm/RMS norm question is unrelated to a ModernBERT type-hint issue." }, { "left": "issue:42936", "right": "issue:43668", "accept": false, - "reason": "Mask2Former ignore_value and ModernBERT norm_eps typing are unrelated." + "reason": "Mask2former ignore_value and ModernBERT norm_eps typing are unrelated." } ] }, @@ -19038,12 +18943,12 @@ "issue:10519|issue:11005", "issue:11005|issue:11307" ], - "prepared_review_unit_hash": "ee6c45ba78eab54d2e60dc82ce25e27a01eb054a5b9f8e67638cff752efcec18", + "prepared_review_unit_hash": "1c745a834526bfbbd579ab48506844f5b1d94c4370237f758623c5e5f215598c", "component_budget": { "node_count": 3, "item_count": 3, "soft_pair_count": 3, - "serialized_chars": 2876, + "serialized_chars": 2875, "estimated_input_tokens": 719, "estimated_eval_tokens": 1694 }, @@ -19051,7 +18956,7 @@ "node_count": 3, "item_count": 3, "soft_pair_count": 3, - "serialized_chars": 2876, + "serialized_chars": 2875, "estimated_input_tokens": 719, "estimated_eval_tokens": 1694 }, @@ -19060,30 +18965,30 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These three issues are not duplicates: one asks for a token truncation API change, one asks for ReduceLROnPlateau-like scheduler behavior, and one asks for Wav2Vec2 word boundary time offsets. They touch unrelated features and code paths.", - "confidence": 0.98, - "canonical_issue_reason": "Issue #10519 is the clearest, most concrete feature request in the set, with a specific API behavior change and well-defined scope.", + "summary": "These three issues are not duplicates: one asks for a truncation-from-beginning option, one asks for ReduceLROnPlateau-like behavior, and one asks for Wav2Vec2 word boundary timestamps. They target different features and code paths.", + "confidence": 0.96, + "canonical_issue_reason": "No canonical issue: the items are about unrelated requests, so there is no single underlying bug or feature to consolidate.", "canonical_pr_reason": null, - "best_issue_reason": "#10519 is the best representative issue because it is explicit and narrowly scoped; the other two are unrelated feature requests rather than variants of the same problem.", + "best_issue_reason": "Issue #10519 is the most concrete and actionable of the set, with a clear feature request and specific truncation behavior, even though it is not duplicate of the others.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:10519", "right": "issue:11307", "accept": false, - "reason": "Different feature areas: truncation behavior vs Wav2Vec2 word-offset extraction. No shared code-path or underlying bug is evident." + "reason": "Different domains: token truncation behavior vs Wav2Vec2 word offset extraction. No shared underlying bug or change." }, { "left": "issue:10519", "right": "issue:11005", "accept": false, - "reason": "One is about token truncation options, the other about learning-rate scheduling behavior. These are unrelated requests." + "reason": "One is about tokenizer truncation direction; the other is about learning-rate scheduling functionality. These are unrelated feature requests." }, { "left": "issue:11005", "right": "issue:11307", "accept": false, - "reason": "Scheduler functionality and Wav2Vec2 timing offsets are unrelated; they do not describe the same underlying issue." + "reason": "ReduceLROnPlateau-like training control and Wav2Vec2 timing offsets are unrelated and would not be merged as one issue." } ] }, @@ -19115,12 +19020,12 @@ "issue:13244|issue:16618", "issue:15980|issue:16618" ], - "prepared_review_unit_hash": "c6e63eaa8ff3fe98c1e5f488ea7216b7c17a29d93ff26680cf0da2c5bd415092", + "prepared_review_unit_hash": "1c7c6c2b77cdc567735227dcb3a7a69e721f7082446520adda4dfb574dc6744f", "component_budget": { "node_count": 3, "item_count": 3, "soft_pair_count": 2, - "serialized_chars": 2479, + "serialized_chars": 2477, "estimated_input_tokens": 620, "estimated_eval_tokens": 1496 }, @@ -19128,7 +19033,7 @@ "node_count": 3, "item_count": 3, "soft_pair_count": 2, - "serialized_chars": 2479, + "serialized_chars": 2477, "estimated_input_tokens": 620, "estimated_eval_tokens": 1496 }, @@ -19137,24 +19042,24 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These three issues do not look like duplicates. #13244 is about Tapas tokenization differing from TensorFlow, #15980 is about a bad error message when downloading a private model while logged out, and #16618 is a separate model-loading failure for bert-base-uncased. The soft similarities appear to be broad model-related noise rather than the same bug.", - "confidence": 0.92, - "canonical_issue_reason": "No single canonical duplicate is evident; the reports describe different underlying problems rather than one shared bug.", + "summary": "This cluster is mostly a loose semantic group around model/tokenization topics, but the Tapas tokenization issue is unrelated to the model-loading reports. The two model-loading issues are only superficially similar and do not look like the same underlying bug.", + "confidence": 0.72, + "canonical_issue_reason": "Issue #16618 is the broadest and most representative of the only plausible overlap in the cluster, but the duplication evidence is weak.", "canonical_pr_reason": null, - "best_issue_reason": "#16618 is the strongest representative of the cluster because it is the most direct model-loading failure report; #15980 is mainly an error-message/auth issue and #13244 is unrelated tokenization behavior.", + "best_issue_reason": "#16618 is the best issue candidate only because it is the closest to the cluster's shared model-loading theme; #15980 is a narrower auth/error-message report and #13244 is unrelated.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:13244", "right": "issue:16618", "accept": false, - "reason": "Tapas tokenization mismatch and a failed bert-base-uncased load are different code paths and different bugs." + "reason": "Tapas tokenization vs. loading bert-base-uncased are different code paths and different user problems; this is not the same bug." }, { "left": "issue:15980", "right": "issue:16618", "accept": false, - "reason": "One is a private-model auth/error-message problem; the other is a general load failure for a public model, so they are not the same underlying issue." + "reason": "One is specifically about the error message when downloading a private model while logged out; the other is a generic inability to load bert-base-uncased. Same general area, but not the same underlying issue." } ] }, @@ -19184,12 +19089,12 @@ "soft_pairs": [ "issue:26350|issue:33260" ], - "prepared_review_unit_hash": "175e701cd348584a602b9d595ed7b1207cb54e56d2624bf5363c091b09e1a5b7", + "prepared_review_unit_hash": "75c2f88b8a6c6012032f577cfdb1814882bcf520a49bef3c26f30f166547807f", "component_budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 1590, + "serialized_chars": 1589, "estimated_input_tokens": 398, "estimated_eval_tokens": 1052 }, @@ -19197,7 +19102,7 @@ "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 1590, + "serialized_chars": 1589, "estimated_input_tokens": 398, "estimated_eval_tokens": 1052 }, @@ -19206,18 +19111,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These are two unrelated community-contribution feature requests: one for Flash Attention 2 support, the other for GGUF support. They target different capabilities and architectures, so they should not be clustered as duplicates.", - "confidence": 0.98, - "canonical_issue_reason": null, + "summary": "These are not duplicates: one requests Flash Attention 2 support for more architectures, the other requests GGUF support for more architectures. They target different features and code paths.", + "confidence": 0.97, + "canonical_issue_reason": "No single canonical issue fits here; the two issues are distinct feature requests with different underlying implementation work.", "canonical_pr_reason": null, - "best_issue_reason": "No single canonical issue fits because the two issues request different features in different areas of the project.", + "best_issue_reason": "Neither issue is a better canonical for the other, since they do not describe the same bug or change.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:26350", "right": "issue:33260", "accept": false, - "reason": "Different feature requests: Flash Attention 2 support vs GGUF support. They do not describe the same underlying bug or change." + "reason": "Different enhancement requests (Flash Attention 2 vs GGUF support); not the same underlying change or bug." } ] }, @@ -19276,184 +19181,184 @@ "issue:44370|issue:46226", "issue:46222|issue:46226" ], - "prepared_review_unit_hash": "64eb3fed569d01e3e7d44f183306e6f040e3d480f307e7a02cfa1f3e5edf76f5", + "prepared_review_unit_hash": "1a401dd5788e7fb5821128907bd5c740e2fa01cf59b774f1a476c09d6211d8c3", "component_budget": { "node_count": 8, "item_count": 8, "soft_pair_count": 24, - "serialized_chars": 13524, - "estimated_input_tokens": 3381, - "estimated_eval_tokens": 7018 + "serialized_chars": 13529, + "estimated_input_tokens": 3383, + "estimated_eval_tokens": 7022 }, "budget": { "node_count": 8, "item_count": 8, "soft_pair_count": 24, - "serialized_chars": 13524, - "estimated_input_tokens": 3381, - "estimated_eval_tokens": 7018 + "serialized_chars": 13529, + "estimated_input_tokens": 3383, + "estimated_eval_tokens": 7022 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "This cluster mixes a few near-identical i18n translation template issues with several distinct language-specific requests, plus two Thai-titled issues that are only clearly duplicates of each other. I would keep the generic placeholder translation issue as the representative and reject most cross-language edges.", - "confidence": 0.74, - "canonical_issue_reason": "Issue 43834 is the clearest representative of the repeated generic i18n translation-docs template: it has the placeholder title, is the earliest of the near-identical placeholder issues, and matches the same workflow as 44365 and 44370.", + "summary": "This cluster is not one true duplicate set: most issues are separate language-specific documentation translation requests, while only the repeated placeholder-template issues and the identical Thai pair look like real duplicates.", + "confidence": 0.77, + "canonical_issue_reason": "issue:43834 is the best representative of the duplicate-like subgroup because it is the earliest of the near-identical placeholder-title issues and sits in the densest duplicate neighborhood.", "canonical_pr_reason": null, - "best_issue_reason": "43834 is the best issue to anchor the cluster because it is the earliest clean example of the repeated placeholder translation request and is closest to the other near-identical placeholder issues.", + "best_issue_reason": "issue:43834 is the strongest global representative for the cluster\u2019s duplicate core, but only for the placeholder-template subgroup, not for the distinct language-specific requests.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:27088", "right": "issue:44370", "accept": false, - "reason": "Different language-specific translation requests; the English/Turkish-vs-placeholder titles suggest distinct targets, not one duplicate issue." + "reason": "Different language request (Turkish vs generic placeholder template); similar body structure is not enough to treat as the same underlying issue." }, { "left": "issue:27088", "right": "issue:44365", "accept": false, - "reason": "Same as above: this is a Turkish translation request, while 44365 is a generic placeholder issue." + "reason": "Turkish translation request is distinct from the generic placeholder-template issue." }, { "left": "issue:27088", "right": "issue:43834", "accept": false, - "reason": "Turkish-specific request versus generic placeholder template; not the same underlying issue." + "reason": "Different concrete translation request; same broad docs/i18n theme only." }, { "left": "issue:27088", "right": "issue:38435", "accept": false, - "reason": "Turkish and Romanian translation requests are separate language tasks, not duplicates." + "reason": "Turkish and Romanian translation requests are separate language tasks, not the same underlying issue." }, { "left": "issue:27088", "right": "issue:28936", "accept": false, - "reason": "Turkish and Spanish translation requests are different concrete translation targets." + "reason": "Turkish and Spanish translation requests are distinct by target language." }, { "left": "issue:28936", "right": "issue:43834", "accept": false, - "reason": "Spanish-specific request versus generic placeholder issue; no evidence they are the same report." + "reason": "Spanish translation request is not the same issue as the generic placeholder-template ticket." }, { "left": "issue:28936", "right": "issue:44365", "accept": false, - "reason": "Spanish translation request and generic placeholder issue are not the same underlying task." + "reason": "Different target language; not a duplicate underlying change/request." }, { "left": "issue:28936", "right": "issue:38435", "accept": false, - "reason": "Spanish and Romanian translation requests are distinct language-specific issues." + "reason": "Spanish and Romanian translation requests are separate items." }, { "left": "issue:28936", "right": "issue:44370", "accept": false, - "reason": "Spanish-specific request versus generic placeholder issue; too little to treat as duplicate." + "reason": "Spanish translation request does not match the generic placeholder-template issue." }, { "left": "issue:38435", "right": "issue:44370", "accept": false, - "reason": "Romanian request and generic placeholder issue are not clearly the same underlying issue." + "reason": "Romanian translation request is separate from the placeholder-template ticket." }, { "left": "issue:38435", "right": "issue:44365", "accept": false, - "reason": "Romanian translation request versus generic placeholder issue; different concrete target." + "reason": "Different language-specific request; same subsystem is too broad to merge." }, { "left": "issue:38435", "right": "issue:43834", "accept": false, - "reason": "Romanian-specific translation request is distinct from the generic placeholder issue." + "reason": "Romanian request and generic placeholder-template issue are not the same underlying issue." }, { "left": "issue:38435", "right": "issue:46222", "accept": false, - "reason": "Different titles and no explicit shared target; the similarity looks like template overlap, not the same issue." + "reason": "Romanian translation request is unrelated to the Thai-titled issue." }, { "left": "issue:38435", "right": "issue:46226", "accept": false, - "reason": "Different titles and no explicit shared target; not enough to merge as one issue." + "reason": "Romanian translation request is unrelated to the Thai-titled issue." }, { "left": "issue:43834", "right": "issue:44370", "accept": true, - "reason": "Same placeholder title and near-identical body length/content strongly indicate duplicate submissions of the same translation-docs request." + "reason": "Same generic placeholder title and near-identical issue shape; these look like duplicated copies of the same template-based issue." }, { "left": "issue:43834", "right": "issue:44365", "accept": true, - "reason": "Same placeholder title and nearly identical body suggest the same underlying template issue." + "reason": "Same placeholder-template issue with no meaningful distinguishing information; likely duplicate." }, { "left": "issue:43834", "right": "issue:46222", "accept": false, - "reason": "The titles are completely different, and the similarity alone is not enough to prove the same underlying issue." + "reason": "Placeholder-template issue and Thai-text issue do not look like the same underlying request." }, { "left": "issue:43834", "right": "issue:46226", "accept": false, - "reason": "Different visible issue text; the overlap looks generic rather than a true duplicate." + "reason": "Placeholder-template issue and Thai-text issue are not the same concrete issue." }, { "left": "issue:44365", "right": "issue:44370", "accept": true, - "reason": "These are effectively the same placeholder translation issue with near-identical content." + "reason": "Same placeholder title and extremely similar body profile; plausible duplicate of one template issue." }, { "left": "issue:44365", "right": "issue:46222", "accept": false, - "reason": "Different titles and no explicit shared language target; not clearly the same issue." + "reason": "Generic placeholder-template issue vs Thai-text issue; not the same request." }, { "left": "issue:44365", "right": "issue:46226", "accept": false, - "reason": "Different visible content; similarity is not sufficient for a duplicate merge." + "reason": "Generic placeholder-template issue vs Thai-text issue; not a duplicate." }, { "left": "issue:44370", "right": "issue:46222", "accept": false, - "reason": "Different titles and no explicit issue target make this too uncertain to merge." + "reason": "Generic placeholder-template issue and Thai-text issue appear unrelated." }, { "left": "issue:44370", "right": "issue:46226", "accept": false, - "reason": "Different visible issue text; does not look like the same underlying report." + "reason": "Generic placeholder-template issue and Thai-text issue appear unrelated." }, { "left": "issue:46222", "right": "issue:46226", "accept": true, - "reason": "These two are identical in title and body length and were opened a day apart, so they look like duplicate submissions of the same issue." + "reason": "These have the same title and appear to be exact copies of the same issue." } ] }, "evaluator_result": { "accept": true, - "feedback": "The summary is grounded in the packet and stays conservative. The analyst correctly treats 43834/44365/44370 as the placeholder-template cluster, keeps the language-specific issues separate, and only marks 46222/46226 as a clear duplicate pair. No soft-edge verdict looks overstated from the provided evidence." + "feedback": "The summary stays grounded in the packet: it distinguishes the clearly different language-specific translation requests from the near-identical placeholder-template issues and the identical Thai pair. The soft-edge verdicts are conservative and do not overclaim duplicates beyond what the titles/structure support." }, "evaluator_used": true, "retried": false, @@ -19536,250 +19441,274 @@ "issue:42503|issue:46653", "issue:45027|issue:46653" ], - "prepared_review_unit_hash": "e882f2b4f97bd5ebd12e1519287a1e13355e94e494826c5e63254fb80ba76610", + "prepared_review_unit_hash": "0927151adf127f16702dd2d443533d3dc2fd11e8886287482f7d722bff4b1a7e", "component_budget": { "node_count": 19, "item_count": 19, "soft_pair_count": 40, - "serialized_chars": 24672, - "estimated_input_tokens": 6168, - "estimated_eval_tokens": 12592 + "serialized_chars": 24676, + "estimated_input_tokens": 6169, + "estimated_eval_tokens": 12594 }, "budget": { "node_count": 19, "item_count": 19, "soft_pair_count": 40, - "serialized_chars": 24672, - "estimated_input_tokens": 6168, - "estimated_eval_tokens": 12592 + "serialized_chars": 24676, + "estimated_input_tokens": 6169, + "estimated_eval_tokens": 12594 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "All soft links are high-level similarity matches between distinct model requests or feature proposals, not the same underlying issue. The titles point to different models/capabilities, so every soft edge is rejected.", - "confidence": 0.81, - "canonical_issue_reason": "If a representative issue is needed, #45322 is the clearest stand-alone request: explicit model name, concrete scope, and a well-formed proposal. It is not a duplicate of the others.", + "summary": "This cluster is mostly a pile of unrelated model-addition requests. The similarity signal is driven by generic wording like \"Add/Support model\" rather than a shared underlying bug or code change, so none of the soft edges look mergeable as duplicates.", + "confidence": 0.97, + "canonical_issue_reason": "Issue 43671 is the best representative anchor here because it is a clear, active model-support request and sits near the densest soft-similarity area, but the overall set is too heterogeneous to justify true duplicate canonization.", "canonical_pr_reason": null, - "best_issue_reason": "#45322 is the most suitable issue representative because it is specific and self-contained, unlike the more generic or broader requests in the set.", + "best_issue_reason": "Issue 43671 is the most useful global representative: it is concrete, active, and thematically closest to the only obvious mini-cluster (Qwen3 voice support). Even so, it should not be treated as a duplicate of the other requests.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:36718", "right": "issue:41115", "accept": false, - "reason": "RoMa keypoint matcher vs MiniCPM3 model architecture; unrelated requests." + "reason": "Different requested artifacts: a RoMa keypoint matcher versus a MiniCPM3 model architecture. Shared \"add model\" phrasing only." }, { "left": "issue:36817", "right": "issue:41211", "accept": false, - "reason": "EuroBert config request vs DEIMv2 model request; different models." + "reason": "Both are model/config requests, but EuroBert config and DEIMv2 are unrelated model additions with no shared code-path change." }, { "left": "issue:36817", "right": "issue:43671", "accept": false, - "reason": "EuroBert config vs Qwen3-TTS support; unrelated model families." + "reason": "EuroBert config work is unrelated to Qwen3-TTS support; same broad request pattern, different model and modality." }, { "left": "issue:36817", "right": "issue:36879", "accept": false, - "reason": "EuroBert config vs RF-DETR model; different targets." + "reason": "EuroBert config and RF-DETR are distinct model-support requests with different targets and implementation work." }, { "left": "issue:36879", "right": "issue:41211", "accept": false, - "reason": "RF-DETR and DEIMv2 are distinct model additions." + "reason": "RF-DETR and DEIMv2 are separate detection-model additions, not the same underlying bug or change." }, { "left": "issue:36879", "right": "issue:43909", "accept": false, - "reason": "RF-DETR vision model vs LFM2.5 Audio support; different domains." + "reason": "RF-DETR is a detection model; LFM2.5 Audio 1.5B is an audio model. Different feature requests." }, { "left": "issue:36879", "right": "issue:44121", "accept": false, - "reason": "RF-DETR model request vs weight-sparse transformer request; not the same change." + "reason": "RF-DETR and a weight-sparse transformer request are different model families and different implementation work." }, { "left": "issue:36879", "right": "issue:43837", "accept": false, - "reason": "RF-DETR object detection vs Qwen3-ASR support; unrelated model requests." + "reason": "RF-DETR versus Qwen3-ASR are unrelated model-support requests across different domains." }, { "left": "issue:36879", "right": "issue:43671", "accept": false, - "reason": "RF-DETR model vs Qwen3-TTS support; different capabilities." + "reason": "RF-DETR is a vision detector request, not the same change as Qwen3-TTS support." }, { "left": "issue:37015", "right": "issue:43837", "accept": false, - "reason": "NeoBERT request vs Qwen3-ASR support; no concrete overlap." + "reason": "NeoBERT and Qwen3-ASR are unrelated models; no shared code-path problem." }, { "left": "issue:37015", "right": "issue:44121", "accept": false, - "reason": "NeoBERT model addition vs weight-sparse transformer request; distinct work." + "reason": "NeoBERT support and a weight-sparse transformer are different model additions, not duplicates." }, { "left": "issue:37015", "right": "issue:43909", "accept": false, - "reason": "NeoBERT vs LFM2.5 Audio; different model families." + "reason": "NeoBERT and LFM2.5 Audio 1.5B are unrelated requests." }, { "left": "issue:36879", "right": "issue:37015", "accept": false, - "reason": "RF-DETR and NeoBERT are unrelated model additions." + "reason": "RF-DETR and NeoBERT are different model-support issues in different domains." }, { "left": "issue:37015", "right": "issue:42503", "accept": false, - "reason": "NeoBERT vs ModernVBERT; similar naming only, different model requests." + "reason": "NeoBERT and ModernVBERT models are separate model-addition requests, not one concrete change." }, { "left": "issue:41115", "right": "issue:43837", "accept": false, - "reason": "MiniCPM3 architecture request vs Qwen3-ASR support; unrelated." + "reason": "MiniCPM3 architecture support and Qwen3-ASR support are different model requests." }, { "left": "issue:41115", "right": "issue:44121", "accept": false, - "reason": "MiniCPM3 model architecture vs weight-sparse transformer; different changes." + "reason": "MiniCPM3 architecture work is not the same as adding a weight-sparse transformer model." }, { "left": "issue:41115", "right": "issue:42503", "accept": false, - "reason": "MiniCPM3 vs ModernVBERT; different models and scopes." + "reason": "MiniCPM3 and ModernVBERT are distinct models with different integration work." }, { "left": "issue:41115", "right": "issue:45027", "accept": false, - "reason": "MiniCPM3 vs Voxtral TTS support; unrelated models." + "reason": "MiniCPM3 versus Voxtral-4B-TTS support are unrelated model requests." }, { "left": "issue:41211", "right": "issue:43671", "accept": false, - "reason": "DEIMv2 vs Qwen3-TTS; no shared concrete code path." + "reason": "DEIMv2 and Qwen3-TTS support are different model additions, not the same change." }, { "left": "issue:42503", "right": "issue:43837", "accept": false, - "reason": "ModernVBERT models vs Qwen3-ASR support; different model requests." + "reason": "ModernVBERT and Qwen3-ASR are separate model-support tasks." }, { "left": "issue:42503", "right": "issue:43671", "accept": false, - "reason": "ModernVBERT vs Qwen3-TTS; unrelated." + "reason": "ModernVBERT models are unrelated to Qwen3-TTS support." }, { "left": "issue:42503", "right": "issue:42738", "accept": false, - "reason": "ModernVBERT models vs BERT-like models with RoPE; not the same request." + "reason": "ModernVBERT and BERT-like models with RoPE are broadly related in wording only; they are not the same concrete request." }, { "left": "issue:42738", "right": "issue:45027", "accept": false, - "reason": "BERT-like RoPE support vs Voxtral TTS support; different domains." + "reason": "RoPE-related BERT-like models and Voxtral TTS support are unrelated." }, { "left": "issue:42738", "right": "issue:46653", "accept": false, - "reason": "BERT-like RoPE request vs unrelated person-only issue; no duplicate basis." + "reason": "The second item is not even a model request, so these cannot be duplicates." }, { "left": "issue:43671", "right": "issue:43837", "accept": false, - "reason": "Qwen3-TTS vs Qwen3-ASR are related by family name only; separate modalities and changes." + "reason": "Both are Qwen3 support requests, but TTS and ASR are different modalities and would not plausibly merge into one PR." }, { "left": "issue:43671", - "right": "issue:44121", + "right": "issue:45027", "accept": false, - "reason": "Qwen3-TTS support vs weight-sparse transformer request; unrelated." + "reason": "Qwen3-TTS and Voxtral-4B-TTS are different model integrations, not the same underlying change." }, { "left": "issue:43671", - "right": "issue:45027", + "right": "issue:45177", "accept": false, - "reason": "Qwen3-TTS vs Voxtral TTS are both audio, but different model-specific support tasks." + "reason": "Qwen3-TTS support and DeepSeek-OCR2 are unrelated model requests." }, { "left": "issue:43671", - "right": "issue:45177", + "right": "issue:44121", "accept": false, - "reason": "Qwen3-TTS support vs DeepSeek-OCR2; different model families." + "reason": "Qwen3-TTS support is not the same change as adding a weight-sparse transformer model." }, { "left": "issue:43837", "right": "issue:43909", "accept": false, - "reason": "Qwen3-ASR support vs LFM2.5 Audio; different model additions." + "reason": "Qwen3-ASR and LFM2.5 Audio 1.5B are both audio-related, but they are different model support requests." }, { "left": "issue:43837", "right": "issue:45322", "accept": false, - "reason": "Qwen3-ASR support vs EUPE by Meta; unrelated." + "reason": "Qwen3-ASR support and EUPE are unrelated model integrations." }, { "left": "issue:43837", "right": "issue:44121", "accept": false, - "reason": "Qwen3-ASR support vs weight-sparse transformer request; not the same bug/change." + "reason": "Qwen3-ASR support is unrelated to the weight-sparse transformer request." + }, + { + "left": "issue:43909", + "right": "issue:44205", + "accept": false, + "reason": "LFM2.5 Audio 1.5B and SAM3-LiteText are different model families and domains." + }, + { + "left": "issue:43909", + "right": "issue:46645", + "accept": false, + "reason": "Audio model support and a Granite SWA+sink option are not the same code-path change." + }, + { + "left": "issue:44121", + "right": "issue:45322", + "accept": false, + "reason": "Weight-sparse transformer support and EUPE are separate model requests with no concrete overlap." }, { "left": "issue:43837", "right": "issue:45027", "accept": false, - "reason": "Qwen3-ASR support vs Voxtral TTS support; different speech models." + "reason": "Qwen3-ASR and Voxtral-4B-TTS are different speech-model support requests." + }, + { + "left": "issue:37015", + "right": "issue:45027", + "accept": false, + "reason": "NeoBERT and Voxtral TTS support are unrelated requests." }, { "left": "issue:43837", "right": "issue:45177", "accept": false, - "reason": "Qwen3-ASR support vs DeepSeek-OCR2; unrelated." + "reason": "Qwen3-ASR and DeepSeek-OCR2 are different model integrations." }, { "left": "issue:45322", "right": "issue:46133", "accept": false, - "reason": "EUPE request vs TIPSv2; both vision-language but distinct model proposals." + "reason": "EUPE and TIPSv2 are distinct vision-language model requests; same broad area, not the same change." }, { "left": "issue:42503", "right": "issue:46653", "accept": false, - "reason": "ModernVBERT models vs a person-named issue; no duplicate relation." + "reason": "ModernVBERT models and \"Mirko Privitera\" are not even comparable as the same artifact." }, { "left": "issue:45027", "right": "issue:46653", "accept": false, - "reason": "Voxtral TTS support vs a person-named issue; unrelated." + "reason": "Voxtral-4B-TTS support and \"Mirko Privitera\" are unrelated; no duplicate relationship." } ] }, @@ -19809,12 +19738,12 @@ "soft_pairs": [ "issue:38740|issue:46443" ], - "prepared_review_unit_hash": "8c49c6b3dbc519cb2adb44b4a025d999031cc3a9c226a7cc7bf20d0bf5ff05a9", + "prepared_review_unit_hash": "3cdb9c76e66faae3acf6dd34157f4a7f78e469e1a970923dd8ab3257dd93b582", "component_budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 1505, + "serialized_chars": 1506, "estimated_input_tokens": 377, "estimated_eval_tokens": 1010 }, @@ -19822,7 +19751,7 @@ "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 1505, + "serialized_chars": 1506, "estimated_input_tokens": 377, "estimated_eval_tokens": 1010 }, @@ -19831,18 +19760,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "The two issues are unrelated: one is a docs request for adding Pruna as an optimization framework, and the other is a Vision RoPE refactor. The soft similarity signal is not enough to treat them as duplicates or the same underlying change.", - "confidence": 0.99, + "summary": "These two issues are unrelated: one is a docs request to add Pruna as an optimization framework, the other is a Vision RoPE refactor. There is no duplicate relationship or shared code path.", + "confidence": 0.98, "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": null, + "best_issue_reason": "No canonical issue is appropriate here; the issues describe different topics and different parts of the project.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:38740", "right": "issue:46443", "accept": false, - "reason": "Different topics, different subsystems, and no shared targets/files or explicit linkage; the titles/body suggest unrelated work rather than the same bug or change." + "reason": "Not the same underlying bug or change: one is a documentation request about Pruna, the other is a Vision RoPE refactor in model code." } ] }, @@ -19872,12 +19801,12 @@ "soft_pairs": [ "issue:41084|issue:43037" ], - "prepared_review_unit_hash": "cd58e139aaee1761f800db61e6b38b62a888e5b64894ab73472c495b6218916d", + "prepared_review_unit_hash": "31b3d78ee4bc5e7a83396d91ede3d3fff2e984a35754c6a68f9d5705070ae2af", "component_budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 1479, + "serialized_chars": 1480, "estimated_input_tokens": 370, "estimated_eval_tokens": 996 }, @@ -19885,7 +19814,7 @@ "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 1479, + "serialized_chars": 1480, "estimated_input_tokens": 370, "estimated_eval_tokens": 996 }, @@ -19894,18 +19823,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "The two issues appear unrelated: one is about block decoding, while the other requests DeepSeek v3.2 model support. I would not cluster them as duplicates.", - "confidence": 0.97, - "canonical_issue_reason": null, + "summary": "These two issues are not duplicates: one is about block decoding behavior, the other is about DeepSeek v3.2 model support.", + "confidence": 0.98, + "canonical_issue_reason": "No canonical issue: the topics are unrelated and do not describe the same underlying bug or feature request.", "canonical_pr_reason": null, - "best_issue_reason": null, + "best_issue_reason": "No single best issue for this cluster, because the pair is heterogeneous rather than a duplicate set.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:41084", "right": "issue:43037", "accept": false, - "reason": "Different problems and likely different code paths: block decoding is a low-level decoding behavior issue, while DeepSeek v3.2 support is a model-compatibility request." + "reason": "Different requests: block decoding vs. DeepSeek v3.2 support. They do not point to the same code-path or change." } ] }, @@ -19935,7 +19864,7 @@ "soft_pairs": [ "issue:41306|issue:41826" ], - "prepared_review_unit_hash": "68435cabcbc4a3aef9c0e4b96dca919a034dfbfc4700a72bbb4cf2cf309e5121", + "prepared_review_unit_hash": "3ba5a29fad91e31c7cd5dd19c6889b497cdce05790702486a446d0f61a583285", "component_budget": { "node_count": 2, "item_count": 2, @@ -19957,18 +19886,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "The two issues are both about Tiled-* features, but they describe different changes: one is about a `from_pretrained()` flag for TiledFusedLogitsLoss/FLCE, the other is about integrating TiledMLP for lower memory use. They do not look like duplicates.", - "confidence": 0.96, - "canonical_issue_reason": "No canonical duplicate candidate: the issues target different components and different requested behavior.", + "summary": "These are not duplicates: one is a specific API/flag behavior request for `from_pretrained()` with TiledFusedLogitsLoss/FLCE, while the other is a separate feature request to integrate TiledMLP for lower memory usage.", + "confidence": 0.95, + "canonical_issue_reason": "Issue 41306 is the more concrete, narrowly scoped report and better represents the cluster than the separate TiledMLP feature request.", "canonical_pr_reason": null, - "best_issue_reason": "#41306 is the more concrete and implementation-oriented issue, with a specific API surface (`from_pretrained()`) and clearer actionable scope.", + "best_issue_reason": "41306 is the best issue to surface because it describes a specific, actionable behavior change with a clear target API.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:41306", "right": "issue:41826", "accept": false, - "reason": "Both mention the Tiled family, but one is about FLCE/from_pretrained() and the other is about TiledMLP memory footprint; these are different code paths and changes." + "reason": "Different underlying requests: one concerns a FLCE flag in `from_pretrained()`, the other concerns integrating TiledMLP for memory savings. They do not appear to be the same bug or change." } ] }, @@ -19998,12 +19927,12 @@ "soft_pairs": [ "issue:43598|issue:43600" ], - "prepared_review_unit_hash": "fc49d24bd6c087ec38ae94b7b212efe9abeec705431e9fa57d33c11c0e6a228f", + "prepared_review_unit_hash": "0549078a92ee910dc9ef85b71802bbf8f4046f0836e07e1641b8402ef0c7cb36", "component_budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 1584, + "serialized_chars": 1583, "estimated_input_tokens": 396, "estimated_eval_tokens": 1048 }, @@ -20011,7 +19940,7 @@ "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 1584, + "serialized_chars": 1583, "estimated_input_tokens": 396, "estimated_eval_tokens": 1048 }, @@ -20020,18 +19949,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These two issues are both about Trainer extensibility, but they describe different problems: one is about `remove_unused_column` behavior, the other about FLOPS estimation for custom trainers. They are not the same underlying bug or change.", - "confidence": 0.93, - "canonical_issue_reason": null, + "summary": "These two issues are thematically related around custom trainers and Trainer behavior, but they describe different underlying problems: one is about `remove_unused_column` and customizability, the other is about FLOPS estimation for custom trainers. I would not cluster them as duplicates.", + "confidence": 0.86, + "canonical_issue_reason": "Issue 43598 is the better representative because it states a concrete Trainer behavior change (`remove_unused_column`) and a broader customizability concern, while the other issue is a separate FLOPS-estimation edge case.", "canonical_pr_reason": null, - "best_issue_reason": "`issue:43598` is the more concrete and self-contained request, focused on a specific Trainer behavior (`remove_unused_column`). `issue:43600` is a separate metric/accounting concern.", + "best_issue_reason": "43598 is the more central and actionable issue in this pair; 43600 is narrower and concerns a different code path.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43598", "right": "issue:43600", "accept": false, - "reason": "They share a broad Trainer/customization theme, but one concerns column removal customizability while the other concerns FLOPS estimation for custom trainers; these are distinct code paths and not a duplicate bug." + "reason": "Related theme, but not the same bug/change: one is about column removal/customization in Trainer, the other about FLOPS accounting for custom trainers. They could be fixed separately and would not naturally merge into one issue." } ] }, @@ -20061,12 +19990,12 @@ "soft_pairs": [ "issue:43684|issue:44405" ], - "prepared_review_unit_hash": "3439505cea281dbfee3554e601e1c8a1742bc265c88d552d5e518e50b81bc73c", + "prepared_review_unit_hash": "ccc0f405441b4d4f9fe12e5dabfafe05510b4637cda319e1d7c6b6a834be9540", "component_budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 1612, + "serialized_chars": 1611, "estimated_input_tokens": 403, "estimated_eval_tokens": 1062 }, @@ -20074,7 +20003,7 @@ "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 1612, + "serialized_chars": 1611, "estimated_input_tokens": 403, "estimated_eval_tokens": 1062 }, @@ -20083,18 +20012,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "The two issues are related only at a broad model-support level, but they request different capabilities for different Qwen model families and classes. They should not be treated as duplicates.", - "confidence": 0.95, - "canonical_issue_reason": "Issue 43684 is the broader umbrella request about registering Qwen3-Omni in AutoModel/AutoModelForConditionalGeneration, so it is the closest thing to a cluster anchor. Issue 44405 is a separate, narrower request for AutoModelForSequenceClassification support for Qwen3.5.", + "summary": "The two issues are about different Qwen model families and different AutoModel entry points, so they do not look like duplicates.", + "confidence": 0.96, + "canonical_issue_reason": "Issue #43684 is the broader registration request for Qwen3-Omni across AutoModel and AutoModelForConditionalGeneration, so it is the more central issue in this pair.", "canonical_pr_reason": null, - "best_issue_reason": "43684 is the better representative issue because it targets the core registration gap rather than a distinct downstream task-specific API support request.", + "best_issue_reason": "Issue #43684 is the better representative because it targets a core model-registration gap rather than a single task-specific head.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43684", "right": "issue:44405", "accept": false, - "reason": "Different model families and different APIs: Qwen3-Omni registration vs Qwen3.5 sequence classification support. Shared theme is only generic model support, not the same underlying bug or change." + "reason": "Different underlying requests: Qwen3-Omni model registration for AutoModel/AutoModelForConditionalGeneration vs. AutoModelForSequenceClassification support for Qwen3.5. They concern different configs and different model heads, so they should not be merged as the same bug." } ] }, @@ -20131,12 +20060,12 @@ "issue:43845|issue:43859", "issue:43846|issue:43859" ], - "prepared_review_unit_hash": "305e88e11d0fc4c339752b0b5756e0715a85444a9e6f5d57857916e5b0911a2b", + "prepared_review_unit_hash": "a2a96ad7f6b8a935fe151f73e45175f342d4388cf7fa7d412839d3455ba059f8", "component_budget": { "node_count": 4, "item_count": 4, "soft_pair_count": 6, - "serialized_chars": 4265, + "serialized_chars": 4266, "estimated_input_tokens": 1067, "estimated_eval_tokens": 2390 }, @@ -20144,7 +20073,7 @@ "node_count": 4, "item_count": 4, "soft_pair_count": 6, - "serialized_chars": 4265, + "serialized_chars": 4266, "estimated_input_tokens": 1067, "estimated_eval_tokens": 2390 }, @@ -20153,55 +20082,58 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These are four extremely sparse closed issues with only generic titles and no useful body details. There isn\u2019t enough evidence to confidently treat them as duplicates, though the three \"huggingface\" issues are the closest match by title alone.", - "confidence": 0.41, - "canonical_issue_reason": "issue 43845 is the earliest of the identically titled 'huggingface' reports, so it is the most plausible placeholder if one issue must represent the cluster.", + "summary": "Three issues share the same Hugging Face title and look like duplicate placeholder reports; the \"Fett\" issue appears unrelated.", + "confidence": 0.66, + "canonical_issue_reason": "issue #43845 is the earliest of the near-identical Hugging Face issues and is the best stand-in for the duplicate set.", "canonical_pr_reason": null, - "best_issue_reason": "43845 is the least arbitrary representative because it is the earliest of the near-identical 'huggingface' issues, but the cluster is too under-specified for a strong canonical choice.", + "best_issue_reason": "issue #43845 is the cleanest representative of the repeated Hugging Face report cluster; the other matching issues can fold into it.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43835", "right": "issue:43845", "accept": false, - "reason": "Different titles ('Fett' vs 'huggingface') and no body/target/file evidence tying them to the same bug; soft similarity alone is not enough." + "reason": "\"Fett\" does not match the Hugging Face report and there is no evidence they describe the same issue." }, { "left": "issue:43835", "right": "issue:43846", "accept": false, - "reason": "Different titles and no concrete shared context beyond generic similarity, so there is no basis to merge them as the same issue." + "reason": "\"Fett\" looks unrelated to the Hugging Face issue cluster." }, { "left": "issue:43835", "right": "issue:43859", "accept": false, - "reason": "Different titles and no supporting details; cannot conclude the same underlying problem." + "reason": "No shared topic beyond generic similarity; this does not look like the same underlying issue." }, { "left": "issue:43845", "right": "issue:43846", - "accept": false, - "reason": "Identical generic titles, but there is no body content, explicit target, or other evidence that they describe the same bug rather than two separate reports." + "accept": true, + "reason": "Same title and very similar metadata/body size; these look like duplicate reports of the same Hugging Face issue." }, { "left": "issue:43845", "right": "issue:43859", - "accept": false, - "reason": "Only a shared generic title fragment is available; that is insufficient to mark them as duplicates." + "accept": true, + "reason": "All are the same Hugging Face report topic, and this pair appears to be duplicate issue filing." }, { "left": "issue:43846", "right": "issue:43859", - "accept": false, - "reason": "Only soft similarity from the title is present, with no concrete evidence of the same underlying issue." + "accept": true, + "reason": "Same repeated Hugging Face issue title and closely related report content; likely duplicates." } ] }, - "evaluator_result": null, - "evaluator_used": false, + "evaluator_result": { + "accept": true, + "feedback": "The summary is supported by the packet: 43845/43846/43859 share the same title and have relatively high soft-similarity, while 43835 is notably less similar and looks separate. The canonical-choice rationale is also grounded in creation order. Soft-edge verdicts are conservative enough overall." + }, + "evaluator_used": true, "retried": false, - "accepted_nontrivial_soft_edge": false, + "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null }, @@ -20235,88 +20167,88 @@ "issue:44572|issue:44937", "issue:44573|issue:44937" ], - "prepared_review_unit_hash": "a590427a08c95e7735686ce91c6f9900c54d3f310f2a7876948e4425a863249a", + "prepared_review_unit_hash": "c08bac667a3ecb1bbee2ab87c35add7f46ee6f6c5b4ce93827498346d48e5c92", "component_budget": { "node_count": 5, "item_count": 5, "soft_pair_count": 9, - "serialized_chars": 5791, - "estimated_input_tokens": 1448, - "estimated_eval_tokens": 3152 + "serialized_chars": 5794, + "estimated_input_tokens": 1449, + "estimated_eval_tokens": 3154 }, "budget": { "node_count": 5, "item_count": 5, "soft_pair_count": 9, - "serialized_chars": 5791, - "estimated_input_tokens": 1448, - "estimated_eval_tokens": 3152 + "serialized_chars": 5794, + "estimated_input_tokens": 1449, + "estimated_eval_tokens": 3154 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "All items look like low-signal spam or placeholder issues, with no concrete bug/change details. None of the soft pairs describe the same underlying problem in a way that would justify deduping them as true duplicates.", - "confidence": 0.98, - "canonical_issue_reason": "issue:44418 is the least obviously spammy and has the most generic, reusable title, but it is still too vague to represent a real canonical bug report.", + "summary": "This cluster looks like low-signal spam / unrelated placeholder issues, with no evidence of a shared underlying bug or change. The similar titles are not enough to treat them as duplicates.", + "confidence": 0.96, + "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "issue:44418 is the best available representative only by relative title quality; it is not substantively informative.", + "best_issue_reason": "Issue 44418 is the least obviously spammy and the only one that reads like a real user-facing report, but there is still no evidence it is a duplicate of the others.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:44371", "right": "issue:44572", "accept": false, - "reason": "Both are spam placeholders, but there is no evidence of the same underlying issue or change." + "reason": "Both are just '' placeholders with no body or targets; similarity is not enough to infer the same bug." }, { "left": "issue:44371", "right": "issue:44573", "accept": false, - "reason": "Both are spam placeholders with no concrete shared bug report or actionable content." + "reason": "Both are generic spam placeholders, but there is no concrete issue content tying them to the same underlying problem." }, { "left": "issue:44371", "right": "issue:44418", "accept": false, - "reason": "Titles are unrelated spam/generic text; not enough to indicate the same issue." + "reason": "One is spam and the other is a documentation suggestion; the titles are not evidence of the same issue." }, { "left": "issue:44371", "right": "issue:44937", "accept": false, - "reason": "No substantive overlap beyond low-signal spam-like content." + "reason": "A spam placeholder and a titled issue about 'Google LLC' are not the same underlying bug or change." }, { "left": "issue:44418", "right": "issue:44572", "accept": false, - "reason": "Generic documentation wording versus spam placeholder; no shared underlying bug/change." + "reason": "Documentation enhancement and spam placeholder are not a plausible duplicate pair." }, { "left": "issue:44418", "right": "issue:44573", "accept": false, - "reason": "Generic documentation wording versus spam placeholder; not the same concrete issue." + "reason": "No concrete overlap beyond weak text similarity; these do not look like the same issue." }, { "left": "issue:44572", "right": "issue:44573", "accept": false, - "reason": "Both are spam placeholders, but that alone is not evidence they report the same underlying problem." + "reason": "Both are spam placeholders, but there is no substantive evidence they report the same problem." }, { "left": "issue:44572", "right": "issue:44937", "accept": false, - "reason": "Spam placeholder versus unrelated promotional-style title; no concrete match." + "reason": "A spam placeholder and a titled issue about 'Google LLC' are unrelated." }, { "left": "issue:44573", "right": "issue:44937", "accept": false, - "reason": "Spam placeholder versus unrelated promotional-style title; not a duplicate pair." + "reason": "The titles and lack of body details do not support treating these as duplicates." } ] }, @@ -20346,7 +20278,7 @@ "soft_pairs": [ "issue:45145|issue:45151" ], - "prepared_review_unit_hash": "34ab5d2853ec55aca0dc75225a42370b9b4394d3dfe8dd64b231e73367be9add", + "prepared_review_unit_hash": "41cc2a898d4f3def52174adb836e789492e43e0df040d18800ee3d85791aad0e", "component_budget": { "node_count": 2, "item_count": 2, @@ -20368,24 +20300,24 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "The two closed issues appear to be duplicates: same title, very similar body length, and strong textual overlap with no distinguishing links or filenames. I would cluster them together and keep the earlier, slightly more referenced issue as canonical.", - "confidence": 0.99, - "canonical_issue_reason": "Issue #45145 is earlier and has at least one inbound reference, while #45151 shows no extra distinguishing signals; both appear to report the same energy-reduction writeup.", + "summary": "The two closed issues appear to be duplicate reports of the same energy-reduction proposal, with matching titles and very similar content.", + "confidence": 0.98, + "canonical_issue_reason": "Issue #45145 is the better canonical issue because it was created first and already has an inbound reference, so it is slightly more established as the primary thread.", "canonical_pr_reason": null, - "best_issue_reason": "#45145 is the best issue to represent the cluster because it is the earlier report and has more inbound reference activity, making it the more established duplicate target.", + "best_issue_reason": "Issue #45145 is the best issue representative for this cluster for the same reasons: earliest creation and a reference signal, while #45151 appears to be a near-identical duplicate.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:45145", "right": "issue:45151", "accept": true, - "reason": "Same exact title and highly similar issue bodies strongly indicate the same underlying report, with no evidence of distinct scope." + "reason": "Same title, same apparent topic, and highly similar body size/content indicate the same underlying issue rather than merely related reports." } ] }, "evaluator_result": { "accept": true, - "feedback": "The conclusion is grounded in the packet: the issues have the same title, similar body sizes, and a recorded soft-similarity signal, with no contrary evidence like distinct filenames or explicit divergent links. The canonical choice of the earlier issue is also conservative. One caution: the confidence is a bit high relative to the available evidence, but the clustering and soft-edge verdict are still reasonable." + "feedback": "The analyst output is grounded in the packet: the two issues share the exact title, have very similar body lengths, and the packet provides soft_similarity evidence. The canonical/best-issue reasoning is conservative and supported by the earlier creation time plus the inbound reference on #45145. No unsupported mergeability claims are made." }, "evaluator_used": true, "retried": false, @@ -20466,7 +20398,7 @@ "soft_pairs": [ "issue:45762|issue:45763" ], - "prepared_review_unit_hash": "afbc01cb4c66f43b4a2dc4dc97594e9e8932c3a3794e2b5df89caf7ce98205ee", + "prepared_review_unit_hash": "d4155d6c0f7c2028b26ea423ee47ae1700d44f1845642cd94b2bc8c4ac76ee83", "component_budget": { "node_count": 2, "item_count": 2, @@ -20488,18 +20420,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "The two issues are not duplicates: both are just person-name/date titles with no body, and they do not describe the same underlying bug or change.", + "summary": "The two closed issues are not duplicates; they are unrelated, title-only posts that appear to be personal names/dates rather than the same bug or request.", "confidence": 0.98, - "canonical_issue_reason": "No canonical issue can be selected; the issues appear unrelated and contain no shared problem description.", + "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "Neither issue is a better canonical representative because neither provides substantive issue context or matches the other.", + "best_issue_reason": null, "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:45762", "right": "issue:45763", "accept": false, - "reason": "The titles refer to different people and different dates, with no body or shared context indicating the same issue." + "reason": "No evidence of a shared bug, request, or code path; the titles are different names with different dates and there is no body or linkage tying them together." } ] }, @@ -20529,7 +20461,7 @@ "soft_pairs": [ "issue:45764|issue:45765" ], - "prepared_review_unit_hash": "b255d29728377e9fb9be382655ba15e6d26b0c5d88a5ac4cccb42d48029033fd", + "prepared_review_unit_hash": "c81581221d08c355e208dbce7ab99bdfb342c0a088cbdcdd915ab2972c665334", "component_budget": { "node_count": 2, "item_count": 2, @@ -20551,18 +20483,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These two closed issues do not appear to be duplicates or related bug reports; they are just short personal-name/date titles with no shared problem statement or code context.", - "confidence": 0.99, + "summary": "The two closed issues appear to be unrelated, bare-name entries with different people and dates; there is no indication of a shared bug, feature, or code change.", + "confidence": 0.98, "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "Neither issue is a suitable canonical representative for a duplicate cluster because there is no underlying technical issue to consolidate.", + "best_issue_reason": "Neither issue is a good canonical candidate because both are similarly content-free and do not describe an actionable product problem.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:45764", "right": "issue:45765", "accept": false, - "reason": "The titles reference different people and dates, with no shared bug, feature, or code-path context." + "reason": "These are just two different personal-name/date entries with no shared technical subject, so they do not look like duplicate reports of the same underlying issue." } ] }, @@ -20592,7 +20524,7 @@ "soft_pairs": [ "issue:45810|issue:46295" ], - "prepared_review_unit_hash": "3b6582f79786ffda528b3e6e362dd1d31027b1877831fcd6397b0ad9e7102efc", + "prepared_review_unit_hash": "4baa76b9aed350d90ff64f26f96ba08b8bfcb20cbb8cd3f42b194070f0d0c3d3", "component_budget": { "node_count": 2, "item_count": 2, @@ -20614,18 +20546,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These issues are not duplicates: one requests a new Qwen token-classification class for value modeling, while the other asks for FP8 training support. They involve different functionality and code paths.", + "summary": "The two issues are unrelated feature requests: one asks for Qwen3_5 token classification support, the other asks for FP8 training support. They should not be clustered as duplicates.", "confidence": 0.98, - "canonical_issue_reason": null, + "canonical_issue_reason": "Issue 45810 is the more concrete and actionable request, with a specific model class named in the title.", "canonical_pr_reason": null, - "best_issue_reason": "issue:45810 is the more concrete, narrowly scoped request, but it is not the same underlying change as issue:46295.", + "best_issue_reason": "Issue 45810 is the best representative because it is narrowly scoped and clearly describes the missing functionality.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:45810", "right": "issue:46295", "accept": false, - "reason": "Different feature requests in different areas: model class support vs FP8 training support. They do not describe the same bug or change." + "reason": "Different feature areas and code paths: model-class support versus FP8 training support. They do not describe the same underlying bug or change." } ] }, @@ -20665,12 +20597,12 @@ "issue:45871|issue:45873", "issue:45872|issue:45873" ], - "prepared_review_unit_hash": "ebdb37b4d4512c2040355edd3c64290b69244f217375db4f61713341514a1ae0", + "prepared_review_unit_hash": "ffcdecb03bf3e0c570233bb2220bbc9a49b883508a690d480a621d79264b82ac", "component_budget": { "node_count": 5, "item_count": 5, "soft_pair_count": 8, - "serialized_chars": 5728, + "serialized_chars": 5726, "estimated_input_tokens": 1432, "estimated_eval_tokens": 3120 }, @@ -20678,7 +20610,7 @@ "node_count": 5, "item_count": 5, "soft_pair_count": 8, - "serialized_chars": 5728, + "serialized_chars": 5726, "estimated_input_tokens": 1432, "estimated_eval_tokens": 3120 }, @@ -20687,66 +20619,66 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "This cluster contains two exact duplicate issue pairs: 45870/45871 and 45872/45873. Issue 45869 is a separate dataset-bundle request and is not the same underlying task as the example/service issues, so it should not be merged with them.", - "confidence": 0.95, - "canonical_issue_reason": "45870 is a strong canonical for the duplicated 'Example: Financial time series prediction with DataForge' request because it is one of an exact pair with 45871, and its topic is distinct from the separate dataset-bundle and scraping-service requests.", + "summary": "This is not one duplicate cluster but two separate exact-duplicate pairs: 45870/45871 and 45872/45873. Issue 45869 is a related dataset marketing/request item but not the same underlying ask as the others.", + "confidence": 0.97, + "canonical_issue_reason": "45872 is the best single representative if one must be chosen: it has an exact duplicate (45873) and is the most concrete, actionable DataForge request. The other exact-dupe pair (45870/45871) is a separate issue, and 45869 is a different dataset-bundle request.", "canonical_pr_reason": null, - "best_issue_reason": "45870 is the best single issue representative for the first exact-duplicate pair; it is text-identical to 45871 and clearer as the canonical duplicate target. 45872 is also a valid canonical for its own exact pair, but it is a different request.", + "best_issue_reason": "45872 is the strongest anchor for this set because it is an exact duplicate pair with a more specific feature-oriented title ('financial time series + scraping service').", "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 bundle request and the other is an example request; not the same underlying issue." + "reason": "Both mention financial data/time-series themes, but one is a dataset bundle offer and the other is a DataForge example request; not the same underlying issue." }, { "left": "issue:45869", "right": "issue:45871", "accept": false, - "reason": "Same as 45869 vs 45870: different artifact type and request scope, so not duplicates." + "reason": "Same as 45869/45870: related topic, but different artifact type and request scope." }, { "left": "issue:45870", "right": "issue:45871", "accept": true, - "reason": "Exact duplicate title/body pair for the same example request." + "reason": "Exact title match and same apparent example request; these look like duplicate issues." }, { "left": "issue:45870", "right": "issue:45872", "accept": false, - "reason": "Both mention DataForge and financial time series, but one is an example request and the other adds a scraping service; different underlying work." + "reason": "Both are DataForge/financial-time-series themed, but 45870 is an example while 45872 asks for a scraping service; not the same concrete change." }, { "left": "issue:45870", "right": "issue:45873", "accept": false, - "reason": "Different request: example prediction vs financial time series + scraping service." + "reason": "Same reasoning as 45870/45872; related subject, but different underlying request." }, { "left": "issue:45871", "right": "issue:45872", "accept": false, - "reason": "Different concrete request and no evidence they describe the same change or bug." + "reason": "Exact duplicate of 45870, but still a different request from the scraping-service issue." }, { "left": "issue:45871", "right": "issue:45873", "accept": false, - "reason": "Different concrete request and no evidence they describe the same change or bug." + "reason": "Different underlying request despite overlapping domain terminology." }, { "left": "issue:45872", "right": "issue:45873", "accept": true, - "reason": "Exact duplicate title/body pair for the same DataForge + scraping service request." + "reason": "Exact title match and same feature request; these are clear duplicates." } ] }, "evaluator_result": { "accept": true, - "feedback": "The summary is grounded in the packet: 45870/45871 and 45872/45873 are exact duplicate pairs based on identical titles and body lengths, while 45869 is a distinct dataset-bundle request. The soft-edge rejections stay conservative and do not overclaim equivalence across different request types." + "feedback": "The summary is grounded in the packet: 45870/45871 and 45872/45873 are exact title matches, while 45869 is clearly a different dataset-bundle request. The non-duplicate verdicts are appropriately conservative given only soft similarity evidence." }, "evaluator_used": true, "retried": false, @@ -20773,7 +20705,7 @@ "soft_pairs": [ "issue:46439|issue:46441" ], - "prepared_review_unit_hash": "d44b4cdfc242bd7db34b7d5746edfbedf1f1c38b93ab6a049d97b5584734513f", + "prepared_review_unit_hash": "e0011f7dd5e2206b7cb3b2e4207ae2ccadff8fbb842738e7fe711dfdcde82118", "component_budget": { "node_count": 2, "item_count": 2, @@ -20795,27 +20727,24 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "The two issues are in a related static-cache / linear-attention area, but the packet only shows soft similarity with moderate text overlap and no shared files or explicit targets. That is not enough to conclude they are the same bug, so the pair should be treated as possibly related rather than duplicates.", - "confidence": 0.57, - "canonical_issue_reason": "Issue 46439 is the broader report, but the evidence here is too thin to treat it as the same root cause as 46441.", + "summary": "These two issues are related to static-cache handling for linear attention, but the reported failures differ enough that they should not be merged as duplicates without more evidence.", + "confidence": 0.66, + "canonical_issue_reason": "Issue 46439 is the more general/root-cause-looking report: it describes static-cache early initialization corrupting linear-attention layers and producing shape errors, which could encompass narrower crash variants.", "canonical_pr_reason": null, - "best_issue_reason": "46439 is the better umbrella issue because it describes the more general failure mode, but it should not be used as proof that 46441 is a duplicate.", + "best_issue_reason": "Issue 46439 is the best representative because it names the suspected mechanism and the broader failure mode, making it more useful as the primary tracking issue.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:46439", "right": "issue:46441", "accept": false, - "reason": "Only soft_similarity supports the link; there are no shared files or targets, and the Jaccard overlap is moderate. The titles suggest a related static-cache / linear-attention area, but the packet does not ground a firm same-root-cause claim." + "reason": "Both mention static cache and linear attention, but one reports shape corruption during early initialization while the other reports a Qwen3.5 KeyError for 'linear_attention'; they may be related, but they do not clearly describe the same concrete bug." } ] }, - "evaluator_result": { - "accept": false, - "feedback": "The conclusion is too strong for the evidence in the packet. The pair only has `soft_similarity` support, with no shared files/targets and a moderate Jaccard score. The titles do suggest a related static-cache / linear-attention area, but the packet does not ground a firm same-root-cause claim. Please downgrade the verdict to something like \u201cpossibly related\u201d rather than accepting the pair as the same bug." - }, - "evaluator_used": true, - "retried": true, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, "accepted_nontrivial_soft_edge": false, "error_kind": null, "error_message": null @@ -20839,12 +20768,12 @@ "soft_pairs": [ "issue:46551|issue:46552" ], - "prepared_review_unit_hash": "fd86c59d778e837271fd1849b8a3db2d450e5d8ebd1ef3165f0815798d696eb4", + "prepared_review_unit_hash": "ae20f63216fbbe7cb3a90b3cb740641f9051c6eec93da73cd39884d1aa704961", "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 }, @@ -20852,7 +20781,7 @@ "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 1706, + "serialized_chars": 1707, "estimated_input_tokens": 427, "estimated_eval_tokens": 1110 }, @@ -20861,27 +20790,27 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "The two issues are near-identical duplicates describing the same Wav2Vec2CTCTokenizer v5 bug. Issue #46552 is the better canonical tracker because it is still open and has existing inbound references, while #46551 is already closed.", - "confidence": 0.99, - "canonical_issue_reason": "#46552 is the active open report with existing inbound references, so it is the better single issue to keep as canonical; #46551 is a closed duplicate of the same problem.", + "summary": "These two issues describe the same Wav2Vec2CTCTokenizer v5 regression verbatim. #46552 is the better canonical tracker because it is open and already has inbound references; #46551 appears to be an earlier overlapping report, but the packet does not explicitly confirm the closure reason.", + "confidence": 0.96, + "canonical_issue_reason": "#46552 is the better canonical issue because it is still open and has inbound references, while #46551 is the same bug report verbatim and appears to be the older overlapping record.", "canonical_pr_reason": null, - "best_issue_reason": "#46552 best represents the cluster because it remains open and can serve as the primary tracker for the shared tokenizer bug.", + "best_issue_reason": "#46552 is the best issue to keep as the primary tracker: it is open, already referenced, and matches #46551\u2019s title/body exactly.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:46551", "right": "issue:46552", "accept": true, - "reason": "Same title, same underlying Wav2Vec2CTCTokenizer v5 bug, and effectively identical report content; these are clear duplicates." + "reason": "Same exact title and same underlying tokenizer bug description; they are effectively the same issue, with no contrary evidence in the packet." } ] }, "evaluator_result": { - "accept": true, - "feedback": "Accept. The conclusion is grounded in the packet: the issues have identical titles and perfect soft similarity (jaccard 1.0), and choosing the open issue with inbound references as canonical is conservative. The soft-edge duplicate judgment is reasonable given the evidence." + "accept": false, + "feedback": "The pairing is well grounded as a strong duplicate-like match, but the writeup overstates the evidence by calling #46551 a \"closed duplicate.\" The packet only shows that #46551 is closed and #46552 is open with inbound references; it does not explicitly prove the closure reason. Keep the canonical recommendation, but phrase the duplicate/closure status more conservatively." }, "evaluator_used": true, - "retried": false, + "retried": true, "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null @@ -20905,7 +20834,7 @@ "soft_pairs": [ "issue:46657|issue:46692" ], - "prepared_review_unit_hash": "c13343db5dc24b11fd6584fc811defae9193712c3832d5fd5d17c6513b0c7a7d", + "prepared_review_unit_hash": "a5ac32533ba3db2c6aac740cfed3a70322dabd37c21ebf3e2df84cac217200a6", "component_budget": { "node_count": 2, "item_count": 2, @@ -20927,18 +20856,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These are both training-time issues in object detection, but they describe different failure modes: one is a tensor shape mismatch error, the other is an out-of-memory problem after several epochs in RT-DETR v2. They are related only at a broad subsystem level, not the same underlying bug.", - "confidence": 0.93, - "canonical_issue_reason": "#46657 is the better canonical issue because it reports a concrete, specific training error (tensor size mismatch) that is more clearly defined than the broader OOM symptom in #46692.", + "summary": "These are two separate training issues in the same general area, but they describe different failure modes: one is a tensor shape mismatch during object detection training, the other is an out-of-memory problem after several epochs with RT-DETR v2. I would not merge them.", + "confidence": 0.94, + "canonical_issue_reason": "Issue #46657 is the more concrete bug report: it names a specific failure mode (tensor size mismatch) and a broader object-detection training context, making it a better representative of the cluster if forced to choose one issue.", "canonical_pr_reason": null, - "best_issue_reason": "#46657 is the best representative of the cluster because it captures a distinct code-path bug with a clear runtime failure, whereas #46692 appears to be a separate resource/memory issue specific to RT-DETR v2.", + "best_issue_reason": "#46657 is the best issue candidate because it describes an actionable code-path problem rather than a resource-pressure symptom, and its title is more likely to map to a single underlying defect.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:46657", "right": "issue:46692", "accept": false, - "reason": "Do not merge: one issue is about tensor shape mismatch during object detection training, the other is about memory exhaustion after epochs. Same area, different underlying problem." + "reason": "They share a training-related theme, but the underlying problems differ: tensor dimension mismatch versus OOM during later epochs. These are not the same bug and would not plausibly be fixed by one change." } ] }, @@ -20968,12 +20897,12 @@ "soft_pairs": [ "issue:46822|issue:46823" ], - "prepared_review_unit_hash": "d4a687c715b35154053cf224b1f6168c57d745c5ea469cfac2ffd3e2af089b02", + "prepared_review_unit_hash": "6d78b03b27273ebd16d91ce5482824b1cdebe6c83a90e4df6536f20bbda4753b", "component_budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 1672, + "serialized_chars": 1671, "estimated_input_tokens": 418, "estimated_eval_tokens": 1092 }, @@ -20981,7 +20910,7 @@ "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 1672, + "serialized_chars": 1671, "estimated_input_tokens": 418, "estimated_eval_tokens": 1092 }, @@ -20990,18 +20919,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These are two distinct bug reports in different loading/dispatch paths: one about ZeRO-3 zero.Init failing to partition a composite multimodal submodule and causing OOM, the other about 4-bit + device_map='auto' + max_memory incorrectly offloading to CPU and raising ValueError. They do not look like duplicates.", - "confidence": 0.96, - "canonical_issue_reason": null, + "summary": "The two issues are both about model loading/device placement, but they describe different failure modes in different code paths, so they should not be clustered as duplicates.", + "confidence": 0.95, + "canonical_issue_reason": "Issue 46822 is the stronger anchor: it has more discussion and inbound references, and it describes a concrete ZeRO-3 zero.Init multi-GPU OOM scenario with a specific composite model submodule.", "canonical_pr_reason": null, - "best_issue_reason": null, + "best_issue_reason": "Issue 46822 is the best representative issue for the cluster because it is the more developed report and the symptom/cause are more specific and actionable.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:46822", "right": "issue:46823", "accept": false, - "reason": "Different code paths and symptoms: ZeRO-3 partitioning/OOM during multi-GPU load vs 4-bit auto device mapping/max_memory dispatching to CPU and failing with ValueError. No shared underlying bug is evident." + "reason": "Both involve loading models onto devices, but 46822 is a ZeRO-3 partitioning/OOM bug during zero.Init on a composite submodule, while 46823 is a 4-bit + device_map/max_memory routing bug that incorrectly dispatches to CPU and raises ValueError. Different mechanisms, symptoms, and likely fixes." } ] }, @@ -21031,7 +20960,7 @@ "soft_pairs": [ "pull_request:40546|pull_request:42781" ], - "prepared_review_unit_hash": "73bcd6d01e9f8532f5d725377365f6d03326d79cb434c451a2b8c58489266fe1", + "prepared_review_unit_hash": "cdd9e8ed7426e434ef423c6bb012b3959945c1c4e4d6dafb66c294041f567d27", "component_budget": { "node_count": 2, "item_count": 2, @@ -21053,18 +20982,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These are related VibeVoice implementation PRs, but they target different scope: the first adds the base VibeVoice model integration, while the second adds a realtime variant plus an acoustic tokenizer and extra generation/auto wiring. They overlap heavily in shared model files, but they do not look like the same underlying change.", + "summary": "Two open PRs target the VibeVoice family, but they represent different layers of work: the first is the initial VibeVoice integration, while the second adds VibeVoice Realtime and related model/processor changes.", "confidence": 0.91, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 40546 is the cleaner canonical candidate because it appears to introduce the base VibeVoice integration first, with broad model/docs/auto-registration work and substantial review activity.", + "canonical_pr_reason": "PR #40546 is the better canonical PR because it is the broader, foundational VibeVoice implementation and has substantially more review activity, making it the primary representative of this feature set.", "best_issue_reason": null, - "best_pr_reason": "PR 40546 is the best representative of the cluster because it is the foundational VibeVoice implementation; 42781 looks like a follow-on feature branch for realtime support rather than the same change.", + "best_pr_reason": "PR #40546 is the best overall representative for the cluster since it introduces the main VibeVoice support; #42781 is a later, more specialized follow-on for realtime functionality rather than the same mergeable change.", "soft_edge_verdicts": [ { "left": "pull_request:40546", "right": "pull_request:42781", "accept": false, - "reason": "Reject: despite shared VibeVoice files, 42781 adds realtime-specific functionality and an acoustic tokenizer, so it is a broader follow-up feature rather than the same concrete bug fix/change as 40546." + "reason": "Reject: despite overlapping VibeVoice files, #40546 is the base VibeVoice implementation and #42781 is a separate VibeVoice Realtime extension with additional feature-extractor/processing and generation changes. They are related, but not the same concrete code-path fix/change and would not plausibly be merged as one PR." } ] }, @@ -21094,12 +21023,12 @@ "soft_pairs": [ "pull_request:40755|pull_request:41224" ], - "prepared_review_unit_hash": "35eabbb70113a3738ad6cd409a22d24c3ad4b3f32bee1f121681a8624bc63eda", + "prepared_review_unit_hash": "c9d06f029a67398b6900ae3f9ca79a82b46b9389ade3a6fd44591ac7d500d1ef", "component_budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 4675, + "serialized_chars": 4676, "estimated_input_tokens": 1169, "estimated_eval_tokens": 2594 }, @@ -21107,7 +21036,7 @@ "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 4675, + "serialized_chars": 4676, "estimated_input_tokens": 1169, "estimated_eval_tokens": 2594 }, @@ -21116,18 +21045,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These two PRs are both model-specific feature additions, but they target unrelated architectures and code paths: TimesFM covariate forecasting vs. DINOv3 image-classification support.", + "summary": "The cluster contains two unrelated pull requests: one adds TimesFM covariate forecasting, the other adds DINOv3 image-classification support. They do not appear to fix the same underlying bug or implement the same change.", "confidence": 0.97, "canonical_issue_reason": null, "canonical_pr_reason": null, "best_issue_reason": null, - "best_pr_reason": "No single PR is a good global representative here; the changes are unrelated and do not point to the same underlying bug or fix.", + "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "pull_request:40755", "right": "pull_request:41224", "accept": false, - "reason": "Reject: they address different models and features (TimesFM forecasting with covariates vs. DINOv3 image classification), with no shared concrete bug or mergeable code-path fix." + "reason": "Different model families and different feature areas; TimesFM covariate forecasting is unrelated to DINOv3 image-classification support." } ] }, @@ -21258,202 +21187,208 @@ "pull_request:43823|pull_request:46911", "pull_request:43251|pull_request:43254" ], - "prepared_review_unit_hash": "a79bed600b59a15838a1a3abc757efa98b44d1eed33772c9ee02009e67ca1079", + "prepared_review_unit_hash": "474094e2db7289b57fa7fce08a3971ef3abdddc72fe56555d6e0ce9b5f84477b", "component_budget": { "node_count": 15, "item_count": 15, "soft_pair_count": 26, - "serialized_chars": 42253, - "estimated_input_tokens": 10564, - "estimated_eval_tokens": 21384 + "serialized_chars": 42243, + "estimated_input_tokens": 10561, + "estimated_eval_tokens": 21378 }, "budget": { "node_count": 15, "item_count": 15, "soft_pair_count": 26, - "serialized_chars": 42253, - "estimated_input_tokens": 10564, - "estimated_eval_tokens": 21384 + "serialized_chars": 42243, + "estimated_input_tokens": 10561, + "estimated_eval_tokens": 21378 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "This cluster is mostly a set of unrelated open PRs that only share broad Transformers-topic similarity. The only true duplicate-looking pair is the fixed_cross_entropy kwargs change (#43251 vs #43254); everything else is a different model, subsystem, or docs/test refactor.", - "confidence": 0.96, + "summary": "This cluster is mostly a grab bag of unrelated PRs. The only true duplicate-like pair is the two `fixed_cross_entropy` kwargs fixes; everything else targets different models, utilities, or tests and should stay separate.", + "confidence": 0.97, "canonical_issue_reason": null, - "canonical_pr_reason": "PR #43251 is the best canonical representative because it directly targets issue #43240 and cleanly implements the concrete loss-helper fix by passing supported kwargs through to nn.functional.cross_entropy.", + "canonical_pr_reason": "PR 43251 is the best canonical representative for the duplicate pair because it directly references the tracked bug (43240) and cleanly implements the `cross_entropy` kwargs passthrough in `fixed_cross_entropy`.", "best_issue_reason": null, - "best_pr_reason": "PR #43251 is the strongest standalone representative of the actual duplicate pair: it has the explicit issue link, fixes a concrete bug in fixed_cross_entropy, and is more complete than the alternative variant.", + "best_pr_reason": "PR 43251 is the strongest global representative of the cluster\u2019s only real duplicate candidate; it is narrowly scoped, bug-driven, and the closest match to the underlying fix.", "soft_edge_verdicts": [ { "left": "pull_request:41855", "right": "pull_request:43823", "accept": false, - "reason": "Different code paths: a Mistral tokenizer API addition vs a new model implementation. Same repo area, but not the same bug or change." + "reason": "Tokenizer missing-methods work for Mistral is unrelated to adding the MobileLLM model." }, { "left": "pull_request:41855", "right": "pull_request:43424", "accept": false, - "reason": "Tokenizer method work and Executorch dynamic-shape testing are unrelated changes with no shared underlying defect." + "reason": "Mistral tokenizer API additions and an executorch export test address different code paths and different symptoms." }, { "left": "pull_request:42668", "right": "pull_request:43823", "accept": false, - "reason": "Processor/auto-loading robustness fixes are unrelated to adding MobileLLM." + "reason": "Processor auto-loading robustness changes are unrelated to introducing MobileLLM." }, { "left": "pull_request:43251", "right": "pull_request:43823", "accept": false, - "reason": "Loss-helper kwargs passthrough and a new model addition are different features with no common code-path bug." + "reason": "A loss-helper kwargs fix is unrelated to MobileLLM model addition." }, { "left": "pull_request:43251", "right": "pull_request:44827", "accept": false, - "reason": "Cross-entropy kwargs plumbing is unrelated to Mistral4 test/model updates." + "reason": "`fixed_cross_entropy` kwargs support and Mistral4 test fixes are not the same underlying bug." }, { "left": "pull_request:43424", "right": "pull_request:46911", "accept": false, - "reason": "Executorch dynamic-shape export testing and Olmo3 RoPE changes are different concerns." + "reason": "Executorch dynamic-shape export testing and Olmo3 RoPE handling are different features in different subsystems." }, { "left": "pull_request:43424", "right": "pull_request:46707", "accept": false, - "reason": "Executorch export tests and FSDP2 migration are unrelated subsystems." + "reason": "A dynamic-shape export test is unrelated to adding native FSDP2 support." }, { "left": "pull_request:43424", "right": "pull_request:44827", "accept": false, - "reason": "Exportability tests and Mistral4 fixes do not share the same concrete bug." + "reason": "Executorch export testing and Mistral4 test/model fixes are separate concerns." }, { "left": "pull_request:43424", "right": "pull_request:43823", "accept": false, - "reason": "Dynamic-shape export testing is unrelated to adding MobileLLM." + "reason": "Executorch export tests and MobileLLM model scaffolding are unrelated." }, { "left": "pull_request:43823", "right": "pull_request:45655", "accept": false, - "reason": "New model addition and config-class resolution order are unrelated changes." + "reason": "MobileLLM model creation is unrelated to resolving `cls.config` annotation/attribute order." }, { "left": "pull_request:43823", "right": "pull_request:43995", "accept": false, - "reason": "MobileLLM scaffolding and Falcon output-interface refactoring are different code paths." + "reason": "MobileLLM addition and Falcon output-collection refactoring do not fix the same code path." }, { "left": "pull_request:43823", "right": "pull_request:45213", "accept": false, - "reason": "A model implementation and a model-creation skill/docs PR are not the same change." + "reason": "A model implementation PR is unrelated to the new-model skill/doc/tooling PR." + }, + { + "left": "pull_request:43823", + "right": "pull_request:46707", + "accept": false, + "reason": "MobileLLM model code and FSDP2 migration are unrelated changes." + }, + { + "left": "pull_request:43823", + "right": "pull_request:46911", + "accept": false, + "reason": "MobileLLM addition and Olmo3 RoPE-per-layer support are different model implementations." }, { "left": "pull_request:43995", "right": "pull_request:45213", "accept": false, - "reason": "Falcon refactoring and a repository skill/doc PR are unrelated." + "reason": "Falcon output-capture refactoring and the model-creation skill are not the same bug or change." }, { "left": "pull_request:43995", "right": "pull_request:44827", "accept": false, - "reason": "Falcon output-capture refactor and Mistral4 tests address different code paths." + "reason": "Falcon standardized output collection and Mistral4 test fixes are unrelated." }, { "left": "pull_request:44827", - "right": "pull_request:46911", + "right": "pull_request:45170", "accept": false, - "reason": "Mistral4 tests/model wiring and Olmo3 RoPE-per-layer changes are unrelated." + "reason": "Mistral4 test fixes and the `layrnorm` typo renaming in CLIP-like models are different changes." }, { "left": "pull_request:44827", - "right": "pull_request:46707", + "right": "pull_request:45213", "accept": false, - "reason": "Mistral4 fixes and native FSDP2 migration are different features." + "reason": "A model test-fix PR is unrelated to the model-creation skill repository changes." }, { "left": "pull_request:44827", - "right": "pull_request:45170", + "right": "pull_request:45655", "accept": false, - "reason": "Mistral4 tests and a CLIP-like weight-rename fix are not the same underlying bug." + "reason": "Mistral4 test updates and config-resolution order fixes are unrelated." }, { "left": "pull_request:44827", - "right": "pull_request:45213", + "right": "pull_request:46048", "accept": false, - "reason": "A model test/fix PR and the model-creation skill PR are clearly different changes." + "reason": "Mistral4 test fixes and TDT loss-kernel integration are separate code paths." }, { - "left": "pull_request:45170", - "right": "pull_request:46707", + "left": "pull_request:44827", + "right": "pull_request:46267", "accept": false, - "reason": "Weight-renaming for CLIP-like conversion and FSDP2 migration are unrelated." + "reason": "Mistral4 test fixes and chat-response parsing updates are unrelated." }, { - "left": "pull_request:46048", + "left": "pull_request:44827", "right": "pull_request:46707", "accept": false, - "reason": "TDT loss kernel integration and FSDP2 migration do not fix the same issue." + "reason": "Mistral4 test/model changes and native FSDP2 migration are not the same underlying issue." }, { "left": "pull_request:44827", - "right": "pull_request:46048", + "right": "pull_request:46911", "accept": false, - "reason": "Mistral4 test fixes and TDT loss kernel integration are different areas." + "reason": "Mistral4 test fixes and Olmo3 RoPE changes are unrelated model-specific work." }, { - "left": "pull_request:43823", + "left": "pull_request:45170", "right": "pull_request:46707", "accept": false, - "reason": "MobileLLM implementation and FSDP2 module migration are unrelated." + "reason": "A LayerNorm naming/weight-renaming fix in CLIP-like models is unrelated to FSDP2 migration." }, { - "left": "pull_request:44827", - "right": "pull_request:46911", + "left": "pull_request:46048", + "right": "pull_request:46707", "accept": false, - "reason": "Mistral4 test/model work and Olmo3 RoPE changes are different code paths." + "reason": "TDT loss-kernel integration and FSDP2 migration do not address the same bug or implementation change." }, { "left": "pull_request:44827", - "right": "pull_request:45655", + "right": "pull_request:45213", "accept": false, - "reason": "Mistral4 updates and config resolution-order fix are not the same bug." + "reason": "Mistral4 test/model work is unrelated to the model-creation skill PR." }, { "left": "pull_request:43823", - "right": "pull_request:46911", + "right": "pull_request:45213", "accept": false, - "reason": "MobileLLM implementation and Olmo3 RoPE changes are unrelated." + "reason": "MobileLLM model addition is unrelated to the model-creation skill and scaffolding changes." }, { "left": "pull_request:43251", "right": "pull_request:43254", "accept": true, - "reason": "Same file, same helper, same issue target, and the same concrete fix: pass supported kwargs through to nn.functional.cross_entropy in fixed_cross_entropy. These could plausibly be merged into one PR." - }, - { - "left": "pull_request:44827", - "right": "pull_request:46267", - "accept": false, - "reason": "Mistral4 fixes and serve chat-parsing updates are unrelated." + "reason": "Both PRs change `fixed_cross_entropy` in the same file to pass supported kwargs through to `nn.functional.cross_entropy`; they fix the same concrete loss-helper bug and could be merged into one PR." } ] }, "evaluator_result": { "accept": true, - "feedback": "The evaluation is grounded in the packet: the only clearly mergeable duplicate-looking pair is PR #43251 vs #43254, which both touch `src/transformers/loss/loss_utils.py`, share issue target #43240, and implement the same `fixed_cross_entropy` kwargs passthrough fix. The other pair verdicts are conservative and consistent with the diffs, and the summary does not overclaim broader duplication beyond broad Transformers-topic similarity." + "feedback": "The assessment is well grounded in the packet. The only accepted pair, 43251/43254, is a conservative match: both touch `fixed_cross_entropy` in the same file, point to the same issue target (43240), and are plausibly mergeable into one bugfix PR. The rest of the pairwise rejections are appropriately cautious and do not overstate similarity." }, "evaluator_used": true, "retried": false, @@ -21480,7 +21415,7 @@ "soft_pairs": [ "pull_request:42975|pull_request:42976" ], - "prepared_review_unit_hash": "bd6e659c7edcb92beabcfc68abc1799bd6a2e6bcd77c412211dfdd01a4e314b9", + "prepared_review_unit_hash": "26826399be55932318e38a3b55498429ddf195d36278f6a692bde16f602ca9b5", "component_budget": { "node_count": 2, "item_count": 2, @@ -21502,18 +21437,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Two open workflow-maintenance PRs overlap in GitHub Actions files, but they target different action upgrades and different change scopes, so they are not duplicates.", + "summary": "Two open PRs both upgrade GitHub Actions, but they target different action families and change sets: one is a broad Node 24 compatibility pass, the other is a Docker build action version bump. They are related but not duplicates.", "confidence": 0.92, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 42976 is the broader umbrella change: it upgrades several actions to newer releases across multiple workflows, whereas 42975 is a more specific Node 24 compatibility pass for checkout/cache/upload-artifact.", + "canonical_pr_reason": "PR #42975 is the better canonical representative because it has the clearest, broadest scope: upgrading multiple workflow actions for Node 24 compatibility across many CI workflows.", "best_issue_reason": null, - "best_pr_reason": "PR 42976 is the better representative of the cluster because it covers the wider set of workflow action version bumps and is less narrowly tied to one compatibility concern.", + "best_pr_reason": "PR #42975 is the strongest representative of the cluster because its title and file coverage summarize the overarching workflow-action upgrade theme more completely than the narrower Docker-action bump in #42976.", "soft_edge_verdicts": [ { "left": "pull_request:42975", "right": "pull_request:42976", "accept": false, - "reason": "Both are workflow dependency bumps, but they update different actions and address different maintenance goals: 42975 focuses on Node 24-compatible versions of checkout/cache/upload-artifact, while 42976 updates docker/build-push-action and other workflow actions. Same subsystem, not the same underlying change." + "reason": "Both are workflow dependency upgrades, but they affect different underlying actions and concrete code paths (checkout/cache/upload-artifact vs docker/build-push-action). They could be part of a larger maintenance sweep, but not the same fix or mergeable as one duplicate PR." } ] }, @@ -21554,12 +21489,12 @@ "pull_request:45490|pull_request:46562", "pull_request:46562|pull_request:46658" ], - "prepared_review_unit_hash": "6e029cdfbd562458d54d013d71800c00bf4cfb9b80b7c06b6826b43231fb9112", + "prepared_review_unit_hash": "d41c076ac823ac20841c19dac202ffd2266d4266abbeaf52bd1a3d72d4f228f2", "component_budget": { "node_count": 6, "item_count": 6, "soft_pair_count": 8, - "serialized_chars": 21197, + "serialized_chars": 21198, "estimated_input_tokens": 5300, "estimated_eval_tokens": 10856 }, @@ -21567,7 +21502,7 @@ "node_count": 6, "item_count": 6, "soft_pair_count": 8, - "serialized_chars": 21197, + "serialized_chars": 21198, "estimated_input_tokens": 5300, "estimated_eval_tokens": 10856 }, @@ -21576,60 +21511,60 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "This cluster is mostly unrelated model-add pull requests that overlap on generic integration files like docs, auto mappings, and repo checks. The only close pair is the two Step3p7 PRs, but one is the actual library integration and the other is a WIP conversion/debug-script PR, so they should not be merged as duplicates.", - "confidence": 0.89, + "summary": "Mostly a bundle of unrelated model-addition PRs. The only close-looking pair is Step3p7-related, but they still differ in scope: one integrates the model into Transformers, the other is a WIP scripts/mini-model conversion repo. No true duplicate pair stands out.", + "confidence": 0.9, "canonical_issue_reason": null, - "canonical_pr_reason": "PR #46562 is the most representative concrete library change in the cluster: a full Step 3.7 Flash model addition with code, docs, and tests. The other Step3p7 PR (#46658) is auxiliary tooling/assets rather than the same change.", + "canonical_pr_reason": "PR 46562 is the strongest representative of the only near-duplicate area: a full Step 3.7 Flash Transformers integration with docs, auto-mappings, and tests. The other Step3p7 PR is a separate WIP script/conversion effort, not the same change.", "best_issue_reason": null, - "best_pr_reason": "If one PR must stand in for the cluster, #46562 is the strongest candidate because it is a complete, high-signal model implementation PR. It has the clearest end-user impact and a coherent code path, unlike the others which are separate model additions or support scripts.", + "best_pr_reason": "PR 46562 is the best overall PR to anchor this cluster because it is the most complete, self-contained model integration and the only one that plausibly overlaps with another item in the set. The rest are different model additions.", "soft_edge_verdicts": [ { "left": "pull_request:42978", "right": "pull_request:43448", "accept": false, - "reason": "Both are model-add PRs, but they add different models (ViT NEPA vs Molmo) and only share generic scaffolding files." + "reason": "Different model families (ViT NEPA vs Molmo). Shared docs/auto-registration files are just boilerplate for model additions, not the same underlying change." }, { "left": "pull_request:42978", "right": "pull_request:43451", "accept": false, - "reason": "Different model families and different implementations; overlap is limited to common docs/auto-registration plumbing." + "reason": "Different model families (ViT NEPA vs Molmo2). Similar scaffolding changes do not make this the same bug or change." }, { "left": "pull_request:42978", "right": "pull_request:45490", "accept": false, - "reason": "ViT NEPA and CTSM are unrelated model additions with only generic repository integration overlap." + "reason": "Different model families (ViT NEPA vs CTSM). They touch shared auto/docs plumbing but implement unrelated additions." }, { "left": "pull_request:43448", "right": "pull_request:43451", "accept": false, - "reason": "Molmo and Molmo2 are related by name but are distinct model additions with separate code paths and assets." + "reason": "Molmo and Molmo2 are related by naming but are distinct model integrations with different files and code paths." }, { "left": "pull_request:43448", "right": "pull_request:45490", "accept": false, - "reason": "Molmo and CTSM are different model PRs; shared files are broad integration points, not the same change." + "reason": "Molmo vs CTSM are unrelated model additions; overlapping framework files are not enough to treat them as duplicates." }, { "left": "pull_request:43451", "right": "pull_request:45490", "accept": false, - "reason": "Molmo2 and CTSM target different models and do not look like one underlying bug or patch." + "reason": "Molmo2 vs CTSM are separate model integrations with different implementations and test coverage." }, { "left": "pull_request:45490", "right": "pull_request:46562", "accept": false, - "reason": "CTSM and Step3p7 are separate model additions; overlap is only in standard auto-mapping/docs infrastructure." + "reason": "CTSM and Step 3.7 Flash are unrelated model additions; the shared auto/doc changes are generic plumbing, not one underlying fix." }, { "left": "pull_request:46562", "right": "pull_request:46658", "accept": false, - "reason": "Although both concern Step3p7 Flash, #46562 is the actual model integration while #46658 is a WIP script/data-conversion repo for mini-model generation and debugging, not the same concrete code-path change." + "reason": "Both mention Step3p7, but 46562 adds the Transformers model while 46658 is a WIP mini-model/script conversion repo. They are different code paths and not plausibly merged as one PR." } ] }, @@ -21680,136 +21615,136 @@ "pull_request:46539|pull_request:46922", "pull_request:46886|pull_request:46922" ], - "prepared_review_unit_hash": "c280e1ba4680a5c25234b3868e1685b47a97b3af14559fc9ab038e2a602d67c7", + "prepared_review_unit_hash": "137c06b664034453c1b095ccb0d3aacf26c0c77f44f9c3c8d6b56f618668591d", "component_budget": { "node_count": 8, "item_count": 8, "soft_pair_count": 16, - "serialized_chars": 21724, - "estimated_input_tokens": 5431, - "estimated_eval_tokens": 11118 + "serialized_chars": 21660, + "estimated_input_tokens": 5415, + "estimated_eval_tokens": 11086 }, "budget": { "node_count": 8, "item_count": 8, "soft_pair_count": 16, - "serialized_chars": 21724, - "estimated_input_tokens": 5431, - "estimated_eval_tokens": 11118 + "serialized_chars": 21660, + "estimated_input_tokens": 5415, + "estimated_eval_tokens": 11086 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "This cluster is mostly a set of unrelated PRs across different subsystems. The notable overlap is the YOLOS pair: PR 46886 is a smaller, related fix for the same device-mismatch bug, and PR 46922 expands that same change with an additional YOLOS hardening tweak.", - "confidence": 0.96, + "summary": "Only one soft-edge pair looks duplicate-like: the two YOLOS device-mismatch fixes, where 46922 is a fuller version of 46886. The rest are different models, workflows, or unrelated code paths and should remain separate.", + "confidence": 0.97, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 46922 is the strongest canonical PR: it fixes the YOLOS device-mismatch in the shared forward path, targets the same issue as 46886, and also includes the broader YOLOS model hardening change.", + "canonical_pr_reason": "PR 46922 is the best canonical PR because it contains the more complete YOLOS device-placement fix, including the mid-position embedding device move and the no_split_modules adjustment, whereas 46886 is the narrower earlier variant.", "best_issue_reason": null, - "best_pr_reason": "PR 46922 is the best representative for the cluster because it subsumes the smaller YOLOS-only fix in 46886 while adding the related model-level adjustment; the rest of the PRs are distinct fixes in other areas.", + "best_pr_reason": "PR 46922 is the strongest representative of the only true duplicate-like pair in the cluster; it addresses the same YOLOS auto-device-map bug as 46886 and includes the broader fix surface.", "soft_edge_verdicts": [ { "left": "pull_request:43291", "right": "pull_request:43665", "accept": false, - "reason": "Different models and test-only changes; no same underlying bug or mergeable shared fix." + "reason": "Both are test-oriented, but they target different model suites and different failure modes (Whisper vs CLIP/ViT)." }, { "left": "pull_request:43291", "right": "pull_request:44815", "accept": false, - "reason": "Whisper tokenizer/test work versus FP8 loading/dequant work; unrelated code paths." + "reason": "Whisper tokenizer/test changes are unrelated to FP8 dequantization and model-loading logic." }, { "left": "pull_request:43291", "right": "pull_request:46380", "accept": false, - "reason": "Whisper decoding tests and generation watermarking validation are different issues." + "reason": "Whisper decoding/tests and generation-config validation are different subsystems with different bugs." }, { "left": "pull_request:43291", "right": "pull_request:46922", "accept": false, - "reason": "Whisper tokenizer fixes and YOLOS device placement fixes are unrelated." + "reason": "Whisper test/tokenizer work is unrelated to the YOLOS device-placement bug." }, { "left": "pull_request:43665", "right": "pull_request:46539", "accept": false, - "reason": "CLIP/ViT test failures are unrelated to Phimoe CI environment changes." + "reason": "Generic CLIP/ViT test edits do not match the Phimoe CI memory-sharing fix." }, { "left": "pull_request:43665", "right": "pull_request:46870", "accept": false, - "reason": "Model test fixes are unrelated to ASR pipeline model-type routing." + "reason": "Different areas: model test edits versus ASR pipeline classification changes." }, { "left": "pull_request:43665", "right": "pull_request:44815", "accept": false, - "reason": "Test assertion edits versus FP8 dequant/model-loading changes are not the same bug." + "reason": "These are not the same change or bug; one is model tests, the other is FP8/model-loading work." }, { "left": "pull_request:43665", "right": "pull_request:46380", "accept": false, - "reason": "Different subsystems and failure modes; not a duplicate change." + "reason": "Both touch broad model code/testing, but they address distinct issues and are not mergeable as one fix." }, { "left": "pull_request:43665", "right": "pull_request:46922", "accept": false, - "reason": "Different models and different code paths; no overlap in the concrete fix." + "reason": "CLIP/ViT test edits are unrelated to YOLOS device mismatch." }, { "left": "pull_request:44815", "right": "pull_request:46539", "accept": false, - "reason": "FP8 quantization/dequantization work is unrelated to CI memory/device-map testing." + "reason": "FP8 dequantization/model-loading changes are unrelated to CI runner memory limits and integration-test setup." }, { "left": "pull_request:44815", "right": "pull_request:46380", "accept": false, - "reason": "Distinct backend and validation changes; not the same underlying issue." + "reason": "FP8 kernel/loading fixes and generation watermarking validation are different code paths and bugs." }, { "left": "pull_request:44815", "right": "pull_request:46922", "accept": false, - "reason": "FP8 dequant fixes and YOLOS device placement fixes are unrelated." + "reason": "FP8 dequantization work is unrelated to YOLOS device placement." }, { "left": "pull_request:46380", "right": "pull_request:46539", "accept": false, - "reason": "Generation config validation and CI infrastructure fixes are different problems." + "reason": "Generation-config validation and Phimoe CI environment fixes are separate problems." }, { "left": "pull_request:46380", "right": "pull_request:46922", "accept": false, - "reason": "Generation watermarking validation and YOLOS device placement are different bugs." + "reason": "Watermarking/config validation changes are unrelated to YOLOS device mismatch." }, { "left": "pull_request:46539", "right": "pull_request:46922", "accept": false, - "reason": "CI memory-limit/device-map infrastructure is not the same change as the YOLOS model fix." + "reason": "CI instance-sharing fixes and YOLOS model device-placement fixes are not the same underlying issue." }, { "left": "pull_request:46886", "right": "pull_request:46922", "accept": true, - "reason": "Both address the same YOLOS device-mismatch bug in the same forward code path; 46922 is a broader variant that plausibly subsumes 46886." + "reason": "Both target the same YOLOS device-mismatch bug with device_map=\"auto\"; 46922 is the broader version that also fixes the forward-path embedding placement." } ] }, "evaluator_result": { "accept": false, - "feedback": "The cluster summary is slightly overstated: PR 46886 is not \u2018unrelated\u2019\u2014it shares the same YOLOS issue target and the same model file as 46922, and the packet explicitly marks that pair as a plausible same-bug overlap. The acceptance of 46886\u219446922 is reasonable, but the narrative should describe 46886 as a related/smaller variant of the same fix rather than lumping it in with unrelated PRs." + "feedback": "The summary is mostly grounded, but several soft-edge verdicts are not conservative and are internally inconsistent: e.g. `43665` vs `44815`, `46380`, and `46922` are marked `accept: true` even though the reasons explicitly say they are different subsystems / not duplicates. That overstates evidence. The only well-supported duplicate-like pair in the packet is `46886`/`46922`, which is plausibly mergeable as the same YOLOS bugfix." }, "evaluator_used": true, "retried": true, @@ -21839,7 +21774,7 @@ "pull_request:43488|pull_request:43532", "pull_request:43532|pull_request:46747" ], - "prepared_review_unit_hash": "30a2556ee88230029adfabf49429c4818b622e5a5e81e4ae0019033ab2d6d62c", + "prepared_review_unit_hash": "efa2f9436b91922ea3bbfae1e580a08be1193e208ff953122ddefdc78a09ae8f", "component_budget": { "node_count": 3, "item_count": 3, @@ -21861,30 +21796,30 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "All three items are separate '[don't merge]' test PRs touching different code paths (packaging/versioning, comment CI/workflow plumbing, and a workflow security gate). They are not the same underlying change, so none should be clustered as duplicates.", - "confidence": 0.96, + "summary": "These are three unrelated '[don't merge]' repository-bot / workflow test PRs with only superficial similarity in title and purpose. They touch different code paths: packaging/setup, comment/workflow automation, and security-gate workflow behavior. No duplicate pair is a strong same-change match.", + "confidence": 0.93, "canonical_issue_reason": null, "canonical_pr_reason": null, "best_issue_reason": null, - "best_pr_reason": null, + "best_pr_reason": "PR 43532 is the broadest representative of the cluster if one must be picked: it modifies multiple workflow files plus testing utilities and test code, whereas the others are narrower and target different areas.", "soft_edge_verdicts": [ { "left": "pull_request:43488", "right": "pull_request:46747", "accept": false, - "reason": "Different workflow areas and intents: one tweaks setup/versioning and test deps, the other changes the PR security gate trigger. No shared bug or mergeable fix." + "reason": "Different changes: 43488 is a setup/version and dependency tweak plus a test change, while 46747 only switches the PR CI security gate workflow trigger from pull_request_target to pull_request. Similar '[don't merge]' framing is not enough to make them the same underlying fix." }, { "left": "pull_request:43488", "right": "pull_request:43532", "accept": false, - "reason": "43488 is about packaging/version and pytest dependency setup; 43532 rewires multiple CI/workflow scripts and test utilities. They do not address the same concrete code-path problem." + "reason": "43488 is about packaging/versioning and adding pytest-random-order, whereas 43532 rewrites comment-driven CI/workflow automation and testing utilities. They do not fix the same concrete bug and would not plausibly merge into one PR." }, { "left": "pull_request:43532", "right": "pull_request:46747", "accept": false, - "reason": "Both touch CI workflows, but one is a broad self-comment CI refactor while the other only switches the security gate event. Different changes, not one underlying fix." + "reason": "Both touch workflows, but the actual code paths differ: 43532 updates self-comment CI, scheduled/slack workflows, testing helpers, and a vit test; 46747 only changes the security gate workflow trigger. This is too broad to treat as one duplicate change." } ] }, @@ -21914,7 +21849,7 @@ "soft_pairs": [ "pull_request:43612|pull_request:43613" ], - "prepared_review_unit_hash": "790400d940a50bd65fe8b8df6b034c14725202c785fc48f006173ffb562d50e9", + "prepared_review_unit_hash": "0c04026f2ad3c05ed44a2cf158e0232a2a95daead3a20cf6ce7d2f9a2628f927", "component_budget": { "node_count": 2, "item_count": 2, @@ -21936,18 +21871,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These are closely related feature PRs, but they implement different pipelines for different segmentation tasks and model stacks, so they should not be merged as duplicates.", - "confidence": 0.96, + "summary": "Two separate feature PRs: one adds a Promptable Concept Segmentation pipeline, the other adds a Promptable Visual Segmentation pipeline. They share documentation/registry plumbing, but they implement different pipelines and model integrations, so they are not duplicates.", + "confidence": 0.97, "canonical_issue_reason": null, - "canonical_pr_reason": "No canonical PR: 43612 adds Promptable Concept Segmentation, while 43613 adds Promptable Visual Segmentation; they share scaffolding but target different tasks and code paths.", + "canonical_pr_reason": "No single canonical PR: these are distinct feature additions with different pipeline names, model hooks, and tests, despite overlapping docs/metadata updates.", "best_issue_reason": null, - "best_pr_reason": "No single best PR for a duplicate cluster, because these are separate feature additions rather than the same underlying change.", + "best_pr_reason": "Neither PR is a global duplicate canonical; if a representative must be chosen, PR #43613 is slightly broader in model coverage, but it is still a separate feature from #43612.", "soft_edge_verdicts": [ { "left": "pull_request:43612", "right": "pull_request:43613", "accept": false, - "reason": "Reject: both add new pipeline/documentation scaffolding, but the concrete features differ. 43612 is for promptable concept segmentation (SAM3-focused), while 43613 is for promptable visual segmentation (SAM/EdgeTam/SAM2/SAM3 tracker). They are adjacent features, not the same bug or change, and should not be collapsed into one PR." + "reason": "Shared docs and registration files are incidental. One adds Promptable Concept Segmentation, the other adds Promptable Visual Segmentation; they target different tasks and code paths, so they should not be merged as the same change." } ] }, @@ -21977,7 +21912,7 @@ "soft_pairs": [ "pull_request:43651|pull_request:43663" ], - "prepared_review_unit_hash": "49015dbe288111c2628a20b8b413c12c8c8db653ea6a1f0ae731387b6fd2d612", + "prepared_review_unit_hash": "39f1a9a98253c83e682041f8254d7b47def04c8f42bb72b29eabdf731be9b5ec", "component_budget": { "node_count": 2, "item_count": 2, @@ -21999,18 +21934,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These two PRs both touch `trainer.py` but address different extension points in `Trainer`: one changes gradient-accumulation loss scaling, the other exposes signature-column filtering for custom trainers. They are not the same underlying bug/change, so the soft edge should be rejected.", - "confidence": 0.97, + "summary": "The two open PRs both touch `trainer.py`, but they implement different trainer extension hooks: one changes gradient-accumulation loss scaling, the other refactors signature-column filtering. They are not the same underlying change and should not be merged as duplicates.", + "confidence": 0.98, "canonical_issue_reason": null, - "canonical_pr_reason": "No clear canonical PR for this cluster because the PRs solve different problems in the same file rather than duplicating one another.", + "canonical_pr_reason": "No canonical PR: the patches address different code paths and different custom-trainer extensibility points.", "best_issue_reason": null, - "best_pr_reason": "Neither PR is a strong global representative of the other; if forced, PR #43651 is slightly more self-contained, but it still does not represent the unrelated column-filtering change in PR #43663.", + "best_pr_reason": "PR #43651 is the clearer self-contained behavior change, but it is still unrelated to PR #43663; it is not a duplicate candidate for the cluster.", "soft_edge_verdicts": [ { "left": "pull_request:43651", "right": "pull_request:43663", "accept": false, - "reason": "Different code paths and goals: #43651 changes gradient accumulation loss scaling, while #43663 adds an override hook for signature column filtering. Shared file alone is not enough to treat them as duplicates." + "reason": "Different functionality in the same file: #43651 changes gradient accumulation loss scaling, while #43663 adds a helper for model signature column filtering. They do not fix the same bug or implement the same change." } ] }, @@ -22040,12 +21975,12 @@ "soft_pairs": [ "pull_request:43656|pull_request:43842" ], - "prepared_review_unit_hash": "eb55fa5cb17605ff1f21f91c37518b1dd97fe140c8adc6cde9199243bbe433b2", + "prepared_review_unit_hash": "217d61badd7b399b69bbedfc98d80aaf3b96473363c652bf420d4e64509118ab", "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 }, @@ -22053,7 +21988,7 @@ "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 3663, + "serialized_chars": 3664, "estimated_input_tokens": 916, "estimated_eval_tokens": 2088 }, @@ -22062,18 +21997,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Both pull requests touch `src/transformers/cli/serve.py` and are about `TypeAdapter`-related CLI import/type-annotation failures, but they appear to fix different concrete failure modes: one is a broader annotations/import cleanup, the other specifically handles `pydantic` being absent. They are related, but not the same underlying change.", - "confidence": 0.82, + "summary": "These two open PRs touch the same file and both mention TypeAdapter-related NameErrors, but they address different root causes: one is a broader annotation/runtime typing cleanup in the CLI, the other is a narrow fix for optional pydantic imports. They are related, but not duplicates.", + "confidence": 0.91, "canonical_issue_reason": null, - "canonical_pr_reason": "PR #43656 is the better canonical representative because it has more discussion/review activity and appears to address the broader CLI `TypeAdapter` NameError path, while #43842 is a narrower follow-up for the no-pydantic case.", + "canonical_pr_reason": "PR 43656 is the broader CLI typing fix and likely the better canonical representative for this cluster; it addresses runtime annotation evaluation in serve.py rather than only the pydantic-import edge case.", "best_issue_reason": null, - "best_pr_reason": "PR #43656 is the strongest standalone candidate among the two: it has the higher activity, the explicit issue link, and a broader fix footprint in the same CLI module.", + "best_pr_reason": "PR 43656 is the strongest single representative because it appears to fix the underlying annotation NameError pattern more generally in the CLI code path, while PR 43842 is a narrower follow-up about TypeAdapter being unavailable when pydantic is missing.", "soft_edge_verdicts": [ { "left": "pull_request:43656", "right": "pull_request:43842", "accept": false, - "reason": "Do not merge as duplicates: #43656 is a broader type-annotation/import fix in `serve.py`, while #43842 specifically makes `TypeAdapter` safe when `pydantic` is not installed. Same file and same theme, but different concrete bugs and not one mergeable change." + "reason": "Reject: both are in serve.py and both mention NameError, but the diffs point to different bugs\u201443656 changes annotation evaluation with __future__ annotations and typing cleanup, while 43842 guards a pydantic TypeAdapter import behind TYPE_CHECKING. They are related but not the same underlying fix." } ] }, @@ -22266,12 +22201,12 @@ "pull_request:44044|pull_request:44154", "pull_request:44024|pull_request:44076" ], - "prepared_review_unit_hash": "890946ec2116d309130b56fb424b116c3e5ee71c455fdcde6520fa0a265175e2", + "prepared_review_unit_hash": "c687705031019804ae71c272450db66ee25afa8a0a1a6838cf2eeb99a4e8f792", "component_budget": { "node_count": 28, "item_count": 28, "soft_pair_count": 77, - "serialized_chars": 84508, + "serialized_chars": 84505, "estimated_input_tokens": 21127, "estimated_eval_tokens": 42510 }, @@ -22279,405 +22214,411 @@ "node_count": 28, "item_count": 28, "soft_pair_count": 72, - "serialized_chars": 76939, - "estimated_input_tokens": 19235, - "estimated_eval_tokens": 38726 + "serialized_chars": 76936, + "estimated_input_tokens": 19234, + "estimated_eval_tokens": 38724 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Mostly a broad family of standardized output-tracing refactors across different models, not one single duplicate change. I only accepted pairs that target the same concrete model/code path (GPT-Neo and GPT-J/CodeGen variants).", - "confidence": 0.87, + "summary": "This cluster is mostly a batch of model-specific refactors for standardized output tracing/capture_outputs. The only true duplicate-like overlaps are within the same concrete model file or a broader PR that subsumes the same file-level change; cross-model \u201csame issue target\u201d pairs should stay separate.", + "confidence": 0.92, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 44722 is the best canonical representative: it is the latest and most complete GPT-J/CodeGen output-tracing refactor, and its scope matches the same concrete code path as the other accepted GPT-J variants. The rest of the cluster is mostly separate model-by-model migrations.", + "canonical_pr_reason": "PR 44722 is the best canonical PR for the GPT-J/CodeGen duplicate subset: it is the latest and most complete variant, covers the shared GPT-J tracing refactor plus the CodeGen companion change, and aligns directly with the tracking issue.", "best_issue_reason": null, - "best_pr_reason": "44722 is the strongest single PR representative of the duplicated GPT-J/CodeGen change set because it covers the same files and refactor pattern in a more complete form than earlier iterations.", + "best_pr_reason": "PR 44722 is the strongest representative of the duplicate subset because it consolidates the shared GPT-J/CodeGen output-tracing change rather than a narrower or earlier variant.", "soft_edge_verdicts": [ { "left": "pull_request:43996", "right": "pull_request:44085", "accept": false, - "reason": "Different model families (CVT/FNet vs GPT-J); same issue target alone is not enough to make this a duplicate." + "reason": "Different model families (CVT/FNet vs GPT-J); same broad output-tracing theme is not enough." }, { "left": "pull_request:43996", "right": "pull_request:44044", "accept": false, - "reason": "Different models and code paths; both are output-tracing refactors, but not the same concrete change." + "reason": "Different models and code paths (CVT/FNet vs DeBERTa v2)." }, { "left": "pull_request:44066", "right": "pull_request:44085", "accept": true, - "reason": "Both are GPT-J/CodeGen output-tracing refactors in the same code path and could plausibly be merged into one PR." + "reason": "Both target the same GPT-J output-tracing refactor in the same file; 44085 looks like a narrower variant of the same change." }, { "left": "pull_request:44007", "right": "pull_request:44072", "accept": false, - "reason": "Different backbones (ResNet/RegNet/RT-DETR ResNet vs EfficientNet); same refactor theme, but not the same bug/change." + "reason": "Both are output-tracing refactors, but they are different model implementations (ResNet/RegNet/RT-DETR vs EfficientNet)." }, { "left": "pull_request:44072", "right": "pull_request:44722", "accept": false, - "reason": "EfficientNet and GPT-J are unrelated concrete code paths despite both being output-tracing cleanup." + "reason": "EfficientNet and GPT-J/CodeGen are unrelated code paths despite similar refactor style." }, { "left": "pull_request:44066", "right": "pull_request:44072", "accept": false, - "reason": "Different models (GPT-J/CodeGen vs EfficientNet); only the refactor style is similar." + "reason": "GPT-J/CodeGen vs EfficientNet; shared issue tracking does not make them duplicates." }, { "left": "pull_request:44013", "right": "pull_request:44044", "accept": false, - "reason": "MobileNetV2 and DeBERTa V2 are separate model implementations, so these are not duplicates." + "reason": "MobileNetV2 and DeBERTa v2 are unrelated implementations." }, { "left": "pull_request:44066", "right": "pull_request:44086", "accept": false, - "reason": "GPT-J/CodeGen vs MGP-STR: same broad tooling change, but different model code paths." + "reason": "GPT-J/CodeGen versus MGP-STR; different model internals and fixes." }, { "left": "pull_request:44018", "right": "pull_request:44068", "accept": true, - "reason": "Both target GPT-Neo output tracing in the same model/file family and appear to be the same underlying refactor." + "reason": "Same model/file family (GPT-Neo) and the diffs are the same output-tracing refactor shape." }, { "left": "pull_request:44066", "right": "pull_request:44071", "accept": false, - "reason": "GPT-J/CodeGen and MPT are unrelated implementations; not the same concrete fix." + "reason": "GPT-J/CodeGen vs MPT are different model code paths." }, { "left": "pull_request:44066", "right": "pull_request:44068", "accept": false, - "reason": "GPT-J/CodeGen vs GPT-Neo are different model paths, so this is not one duplicate PR." + "reason": "GPT-J/CodeGen versus GPT-Neo; similar refactor pattern but not the same change." }, { "left": "pull_request:44018", - "right": "pull_request:44019", + "right": "pull_request:44066", "accept": false, - "reason": "GPT-Neo and ResNet are unrelated models; both are standardized-output refactors only at a theme level." + "reason": "GPT-Neo and GPT-J/CodeGen are separate model-specific refactors." }, { "left": "pull_request:44019", "right": "pull_request:44722", "accept": false, - "reason": "ResNet and GPT-J/CodeGen are different concrete changes." + "reason": "ResNet output tracing and GPT-J/CodeGen are unrelated changes." }, { "left": "pull_request:44086", "right": "pull_request:44722", "accept": false, - "reason": "MGP-STR and GPT-J/CodeGen are different model implementations; not the same underlying change." + "reason": "MGP-STR and GPT-J/CodeGen are different models and different concrete fixes." }, { "left": "pull_request:44019", "right": "pull_request:44085", "accept": false, - "reason": "ResNet vs GPT-J/CodeGen are separate code paths." + "reason": "ResNet versus GPT-J; same issue tag is not enough." }, { "left": "pull_request:44018", - "right": "pull_request:44071", + "right": "pull_request:44019", "accept": false, - "reason": "GPT-Neo and MPT are different models; the shared decorator pattern is too broad to merge them." + "reason": "GPT-Neo and ResNet are different code paths." }, { "left": "pull_request:44019", "right": "pull_request:44071", "accept": false, - "reason": "ResNet and MPT are unrelated backbones." + "reason": "ResNet versus MPT are unrelated implementations." }, { "left": "pull_request:44068", "right": "pull_request:44722", "accept": false, - "reason": "GPT-Neo and GPT-J are adjacent but distinct model code paths, so these are not the same PR change." + "reason": "GPT-Neo and GPT-J/CodeGen are different model families." }, { "left": "pull_request:44068", "right": "pull_request:44116", "accept": false, - "reason": "GPT-Neo and Flaubert are unrelated models." + "reason": "GPT-Neo and Flaubert are unrelated model-specific refactors." }, { "left": "pull_request:44018", "right": "pull_request:44116", "accept": false, - "reason": "GPT-Neo vs Flaubert: different implementations, same general refactor theme only." + "reason": "GPT-Neo and Flaubert do not share the same concrete bug/change." }, { "left": "pull_request:44071", "right": "pull_request:44722", "accept": false, - "reason": "MPT and GPT-J/CodeGen are separate model code paths." + "reason": "MPT and GPT-J/CodeGen are different code paths." }, { "left": "pull_request:44018", "right": "pull_request:44085", "accept": false, - "reason": "GPT-Neo and GPT-J/CodeGen are different model families." + "reason": "GPT-Neo versus GPT-J; same pattern, different model fix." }, { "left": "pull_request:44018", "right": "pull_request:44086", "accept": false, - "reason": "GPT-Neo and MGP-STR are unrelated code paths." + "reason": "GPT-Neo and MGP-STR are unrelated." }, { "left": "pull_request:44068", "right": "pull_request:44086", "accept": false, - "reason": "GPT-Neo and MGP-STR are not the same concrete bug/change." + "reason": "GPT-Neo and MGP-STR are unrelated." }, { - "left": "pull_request:44059", - "right": "pull_request:44086", + "left": "pull_request:44071", + "right": "pull_request:44085", "accept": false, - "reason": "GPT2 and MGP-STR are different models; the similarity is only at the decorator refactor level." + "reason": "MPT and GPT-J are different model internals." }, { - "left": "pull_request:44098", - "right": "pull_request:44101", + "left": "pull_request:44085", + "right": "pull_request:44086", "accept": false, - "reason": "ViLT and Flaubert are distinct model implementations." + "reason": "GPT-J and MGP-STR are different changes." }, { "left": "pull_request:44059", - "right": "pull_request:44722", - "accept": false, - "reason": "GPT2 and GPT-J/CodeGen are related only by broad refactoring style, not the same code path." - }, - { - "left": "pull_request:44017", - "right": "pull_request:44161", + "right": "pull_request:44086", "accept": false, - "reason": "SegFormer and LongT5 are unrelated models." + "reason": "GPT-2 and MGP-STR are unrelated model-specific refactors." }, { - "left": "pull_request:44073", + "left": "pull_request:44098", "right": "pull_request:44101", "accept": false, - "reason": "VisualBERT and Flaubert are separate code paths." - }, - { - "left": "pull_request:44073", - "right": "pull_request:44161", - "accept": false, - "reason": "VisualBERT and LongT5 are different model implementations." + "reason": "ViLT versus Flaubert/XLM are different model families." }, { "left": "pull_request:44059", "right": "pull_request:44071", "accept": false, - "reason": "GPT2 and MPT are different concrete model changes." - }, - { - "left": "pull_request:44056", - "right": "pull_request:44101", - "accept": false, - "reason": "MPNet and Flaubert are unrelated model code paths." - }, - { - "left": "pull_request:44018", - "right": "pull_request:44059", - "accept": false, - "reason": "GPT-Neo and GPT2 are different models; shared output-tracing theme is too broad." - }, - { - "left": "pull_request:44059", - "right": "pull_request:44068", - "accept": false, - "reason": "GPT2 and GPT-Neo are distinct model paths." - }, - { - "left": "pull_request:44101", - "right": "pull_request:44161", - "accept": false, - "reason": "Flaubert and LongT5 are different model implementations." + "reason": "GPT-2 and MPT are separate implementations." }, { "left": "pull_request:43997", "right": "pull_request:44017", "accept": false, - "reason": "RegNet and SegFormer are different models." + "reason": "RegNet and Segformer are unrelated model code paths." }, { "left": "pull_request:43997", "right": "pull_request:44056", "accept": false, - "reason": "RegNet and MPNet are unrelated code paths." + "reason": "RegNet and MPNet are different model implementations." }, { "left": "pull_request:44071", "right": "pull_request:44076", "accept": false, - "reason": "MPT and ImageGPT are different model implementations." + "reason": "MPT and ImageGPT are unrelated changes." }, { "left": "pull_request:44074", "right": "pull_request:44101", "accept": false, - "reason": "TextNet and Flaubert are separate code paths." + "reason": "TextNet and Flaubert/XLM are different model-specific refactors." }, { "left": "pull_request:43997", "right": "pull_request:44018", "accept": false, - "reason": "RegNet and GPT-Neo are unrelated model families." + "reason": "RegNet and GPT-Neo are different code paths." }, { "left": "pull_request:44056", "right": "pull_request:44161", "accept": false, - "reason": "MPNet and LongT5 are different model changes." + "reason": "MPNet and LongT5 are unrelated model fixes." }, { "left": "pull_request:43997", "right": "pull_request:44161", "accept": false, - "reason": "RegNet and LongT5 are unrelated; same issue target does not make them duplicates." + "reason": "RegNet and LongT5 are different model families." }, { "left": "pull_request:44074", "right": "pull_request:44161", "accept": false, - "reason": "TextNet and LongT5 are distinct code paths." + "reason": "TextNet and LongT5 are unrelated." }, { "left": "pull_request:44018", "right": "pull_request:44161", "accept": false, - "reason": "GPT-Neo and LongT5 are different models." + "reason": "GPT-Neo and LongT5 are different model-specific changes." }, { "left": "pull_request:44010", "right": "pull_request:44071", "accept": false, - "reason": "SqueezeBert and MPT are unrelated models." + "reason": "SqueezeBert and MPT are unrelated implementations." }, { "left": "pull_request:44010", "right": "pull_request:44073", "accept": false, - "reason": "SqueezeBert and ViLT are distinct model implementations." + "reason": "SqueezeBert and VisualBert are different model code paths." }, { "left": "pull_request:44018", "right": "pull_request:44129", "accept": false, - "reason": "GPT-Neo and SpeechT5 are different code paths." + "reason": "GPT-Neo and SpeechT5 are unrelated model families." }, { "left": "pull_request:43997", "right": "pull_request:44074", "accept": false, - "reason": "RegNet and TextNet are unrelated model families." + "reason": "RegNet and TextNet are different model-specific refactors." + }, + { + "left": "pull_request:44056", + "right": "pull_request:44101", + "accept": false, + "reason": "MPNet and Flaubert/XLM are unrelated." + }, + { + "left": "pull_request:44071", + "right": "pull_request:44076", + "accept": false, + "reason": "MPT and ImageGPT are unrelated changes." + }, + { + "left": "pull_request:44018", + "right": "pull_request:44059", + "accept": false, + "reason": "GPT-Neo and GPT-2 are different model code paths." + }, + { + "left": "pull_request:44059", + "right": "pull_request:44068", + "accept": false, + "reason": "GPT-2 and GPT-Neo are separate refactors." + }, + { + "left": "pull_request:44068", + "right": "pull_request:44722", + "accept": false, + "reason": "GPT-Neo and GPT-J/CodeGen are not the same concrete change." + }, + { + "left": "pull_request:44071", + "right": "pull_request:44086", + "accept": false, + "reason": "MPT and MGP-STR are unrelated." + }, + { + "left": "pull_request:44017", + "right": "pull_request:44073", + "accept": false, + "reason": "Segformer and VisualBert are different model implementations." }, { "left": "pull_request:44010", "right": "pull_request:44018", "accept": false, - "reason": "SqueezeBert and GPT-Neo are different models." + "reason": "SqueezeBert and GPT-Neo are unrelated." }, { "left": "pull_request:43997", "right": "pull_request:44129", "accept": false, - "reason": "RegNet and SpeechT5 are not the same underlying change." + "reason": "RegNet and SpeechT5 are different model code paths." }, { "left": "pull_request:44071", "right": "pull_request:44129", "accept": false, - "reason": "MPT and SpeechT5 are unrelated code paths." + "reason": "MPT and SpeechT5 are unrelated changes." }, { "left": "pull_request:43996", "right": "pull_request:44161", "accept": false, - "reason": "CVT/FNet and LongT5 are different model implementations." + "reason": "CVT/FNet and LongT5 are unrelated model-specific refactors." }, { "left": "pull_request:43996", "right": "pull_request:44017", "accept": false, - "reason": "CVT/FNet and SegFormer are separate backbones." + "reason": "CVT/FNet and Segformer are different model families." }, { "left": "pull_request:44010", "right": "pull_request:44044", "accept": false, - "reason": "SqueezeBert and DeBERTa V2 are different model families." + "reason": "SqueezeBert and DeBERTa v2 are unrelated." }, { "left": "pull_request:44044", "right": "pull_request:44074", "accept": false, - "reason": "DeBERTa V2 and TextNet are unrelated concrete changes." + "reason": "DeBERTa v2 and TextNet are different code paths." }, { - "left": "pull_request:44066", - "right": "pull_request:44722", - "accept": true, - "reason": "Same GPT-J/CodeGen output-tracing refactor line; 44722 is a later variant of the same change." + "left": "pull_request:44024", + "right": "pull_request:44076", + "accept": false, + "reason": "FocalNet and ImageGPT are unrelated model implementations." }, { "left": "pull_request:44085", "right": "pull_request:44722", "accept": true, - "reason": "Both touch the GPT-J output-tracing refactor in the same model/file and could be merged as one PR." + "reason": "Same GPT-J output-tracing change; 44722 is the broader later consolidation that subsumes this file-level fix." }, { "left": "pull_request:44085", "right": "pull_request:44129", "accept": false, - "reason": "GPT-J/CodeGen and SpeechT5 are unrelated model code paths." - }, - { - "left": "pull_request:43996", - "right": "pull_request:44007", - "accept": false, - "reason": "CVT/FNet and ResNet/RegNet are different model implementations." - }, - { - "left": "pull_request:43996", - "right": "pull_request:44129", - "accept": false, - "reason": "CVT/FNet and SpeechT5 are unrelated concrete changes." + "reason": "GPT-J and SpeechT5 are unrelated model-specific refactors." }, { "left": "pull_request:44044", "right": "pull_request:44129", "accept": false, - "reason": "DeBERTa V2 and SpeechT5 are different model paths." + "reason": "DeBERTa v2 and SpeechT5 are different code paths." }, { "left": "pull_request:44044", "right": "pull_request:44154", "accept": false, - "reason": "DeBERTa V2 and ViTS are unrelated models; same refactor theme only." + "reason": "DeBERTa v2 and Vits are unrelated changes." }, { "left": "pull_request:44024", "right": "pull_request:44076", "accept": false, - "reason": "FocalNet and ImageGPT are different backbones; not the same duplicate change." + "reason": "FocalNet and ImageGPT are different model families." + }, + { + "left": "pull_request:44007", + "right": "pull_request:44019", + "accept": true, + "reason": "Both include the ResNet output-tracing refactor in the same concrete file; 44007 is the broader superset." + }, + { + "left": "pull_request:44101", + "right": "pull_request:44116", + "accept": true, + "reason": "Both cover the same Flaubert output-tracing change in the same file; 44101 additionally includes XLM." } ] }, "evaluator_result": { "accept": true, - "feedback": "The summary is grounded in the packet: it treats the cluster as mostly broad standardized output-tracing migrations and only groups PRs that clearly share the same concrete code path. The accepted soft-edge pairs are conservative and mergeable into single PRs for the same bugfix (GPT-Neo variants; GPT-J/CodeGen variants)." + "feedback": "The summary is grounded in the packet and stays conservative about cross-model overlaps. The accepted soft pairs are all supported by either the same concrete file or a clear broader PR that subsumes the narrower file-level change, so they look mergeable into a single bugfix PR. The rejected cross-model pairs are handled cautiously and are consistent with the evidence." }, "evaluator_used": true, "retried": false, @@ -22712,12 +22653,12 @@ "pull_request:43997|pull_request:44010", "pull_request:44010|pull_request:44129" ], - "prepared_review_unit_hash": "5f8a3ab61ab5500e14da5bfb03c29ca0841c499c30574bb547099c10c6dc110f", + "prepared_review_unit_hash": "06ad026ccb0114dbbf145150e503fdc2df6607ea7f4cd1319e30f78f11f50ad2", "component_budget": { "node_count": 28, "item_count": 28, "soft_pair_count": 77, - "serialized_chars": 84508, + "serialized_chars": 84505, "estimated_input_tokens": 21127, "estimated_eval_tokens": 42510 }, @@ -22734,42 +22675,42 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "These PRs are all model-specific refactors to standardized output tracing, but they target different architectures and code paths. They are thematically similar, not duplicate changes to the same bug or feature, so the soft edges should not be merged.", - "confidence": 0.93, + "summary": "All six PRs are model-specific refactors around standardized output tracing/capture, but they affect different architectures and code paths, so they are not duplicates of the same change.", + "confidence": 0.97, "canonical_issue_reason": null, - "canonical_pr_reason": "If a representative PR is needed, 44129 is the strongest candidate because it has an explicit issue target and a concrete, well-scoped SpeechT5 output-capture refactor. It is still not a duplicate of the other model-specific PRs.", + "canonical_pr_reason": "PR 44129 is the strongest representative because it is the only one tied to an explicit issue and makes a concrete SpeechT5 tracing refactor rather than a purely generic migration.", "best_issue_reason": null, - "best_pr_reason": "44129 is the best single representative of the theme because it is tied to an explicit issue and makes a concrete code-path change. The others are parallel migrations in different models, not one shared fix.", + "best_pr_reason": "PR 44129 has the clearest scope and traceability, with an explicit issue link and a concrete code-path change in SpeechT5.", "soft_edge_verdicts": [ { "left": "pull_request:44073", "right": "pull_request:44074", "accept": false, - "reason": "Same refactor theme, but different models and different files/code paths. Not the same underlying bug or change." + "reason": "Both are output-tracing refactors, but one is for VisualBert and the other for TextNet; different model code paths mean they are not the same change." }, { "left": "pull_request:44056", "right": "pull_request:44073", "accept": false, - "reason": "Both are output-tracing refactors, but for different architectures (MPNet vs VisualBert). Too broad to treat as duplicates." + "reason": "These touch different models (MPNet vs VisualBert) and different tracing implementations, so they are not mergeable as one duplicate fix." }, { "left": "pull_request:44056", "right": "pull_request:44074", "accept": false, - "reason": "Both are model-specific output-capture migrations, but they touch different model implementations and are not one mergeable change." + "reason": "MPNet and TextNet are separate model implementations; this is the same refactor theme, not the same underlying bug or code change." }, { "left": "pull_request:43997", "right": "pull_request:44010", "accept": false, - "reason": "RegNet and SqueezeBert are separate model implementations; similar decorator migration does not make them duplicate PRs." + "reason": "RegNet and SqueezeBert are unrelated model files with independent output-handling updates, so this is not a duplicate PR pair." }, { "left": "pull_request:44010", "right": "pull_request:44129", "accept": false, - "reason": "Different models and different refactor scopes. The shared standardized output tracing theme is not enough to merge them as one PR." + "reason": "SqueezeBert output collection and SpeechT5 output tracing/caching are different concrete code paths; shared tooling does not make them duplicates." } ] }, @@ -22849,274 +22790,274 @@ "pull_request:44028|pull_request:44030", "pull_request:44029|pull_request:44030" ], - "prepared_review_unit_hash": "e802bb08bd5887bcfd088c749ad235a330d6a6cc3d138da9c007a45b8ebcfd00", + "prepared_review_unit_hash": "5a20ed1d0ec6145ebdde55438fc49d476b5a425ec49115ef9e0ba92692bf18e5", "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": 40181, + "estimated_input_tokens": 10046, + "estimated_eval_tokens": 20348 }, "budget": { "node_count": 13, "item_count": 13, "soft_pair_count": 40, - "serialized_chars": 40180, - "estimated_input_tokens": 10045, - "estimated_eval_tokens": 20346 + "serialized_chars": 40181, + "estimated_input_tokens": 10046, + "estimated_eval_tokens": 20348 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "This cluster is a broad theme of output-tracing refactors across many unrelated model implementations, not a set of true duplicate changes. The PRs share wording and helper usage, but they target different code paths and would not plausibly be merged into one PR.", - "confidence": 0.93, + "summary": "These PRs are highly similar textually, but they are independent, model-specific output-tracing refactors across different architectures. They should not be deduplicated as the same change unless they touch the same concrete code path, which they do not.", + "confidence": 0.97, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 44025 is a good representative because it is a self-contained output-tracing refactor and touches two closely related files in the same model family; however, the cluster as a whole is not a duplicate set.", + "canonical_pr_reason": "If one PR must represent the cluster, 44025 is the best anchor: it is a self-contained refactor across two closely related depth-anything models and includes tests, making it the most complete example of the pattern.", "best_issue_reason": null, - "best_pr_reason": "PR 44025 is the most representative single PR for the cluster\u2019s theme, but only as a cluster exemplar, not as a duplicate canonical fix.", + "best_pr_reason": "44025 is the strongest representative of the family because it is broader than the single-file refactors and has accompanying test coverage, while still staying within one coherent model family.", "soft_edge_verdicts": [ { "left": "pull_request:43998", "right": "pull_request:44002", "accept": false, - "reason": "Different models and code paths (timm_backbone vs upernet); same refactor theme only." + "reason": "Different model families and code paths; same output-tracing refactor theme is too broad." }, { "left": "pull_request:43998", "right": "pull_request:44000", "accept": false, - "reason": "Unrelated model implementations; both mention output tracing but do not fix the same bug/change." + "reason": "Timm backbone vs vision-text dual encoder are unrelated implementations." }, { "left": "pull_request:43998", "right": "pull_request:44001", "accept": false, - "reason": "Different subsystems (vision backbone vs audio generation); no shared concrete code path." + "reason": "Different architectures; no shared concrete bug or mergeable code path." }, { "left": "pull_request:43998", "right": "pull_request:44003", "accept": false, - "reason": "Separate model families and files; too broad to be one duplicate PR." + "reason": "Timm backbone refactor is unrelated to Falcon/Mamba state/output handling." }, { "left": "pull_request:43998", "right": "pull_request:44029", "accept": false, - "reason": "Timm backbone output handling is unrelated to RWKV layer tracing." + "reason": "Different model code paths; only a broad tracing-refactor similarity." }, { "left": "pull_request:43999", "right": "pull_request:44001", "accept": false, - "reason": "MobileNetV1 and UnivNet are unrelated models; not the same underlying change." + "reason": "MobileNetV1 and UnivNet are separate models with separate output semantics." }, { "left": "pull_request:43999", "right": "pull_request:44002", "accept": false, - "reason": "Different model families and file targets; shared refactor pattern is insufficient." + "reason": "MobileNetV1 and UperNet are unrelated code paths." }, { "left": "pull_request:43999", "right": "pull_request:44025", "accept": false, - "reason": "MobileNetV1 vs Depth Anything are unrelated implementations and cannot be merged as one PR." + "reason": "Different model families; no same underlying bug or change." }, { "left": "pull_request:43999", "right": "pull_request:44026", "accept": false, - "reason": "Different model code paths; same helper usage does not imply duplicate fix." + "reason": "MobileNetV1 refactor is unrelated to vision encoder-decoder wrapper changes." }, { "left": "pull_request:43999", "right": "pull_request:44027", "accept": false, - "reason": "MobileNetV1 and Speech Encoder-Decoder are unrelated; no common concrete bug." + "reason": "Different architectures and concrete implementations." }, { "left": "pull_request:44000", "right": "pull_request:44001", "accept": false, - "reason": "Vision-text dual encoder vs UnivNet are unrelated; only the refactor wording matches." + "reason": "Vision-text dual encoder and UnivNet are unrelated model paths." }, { "left": "pull_request:44000", "right": "pull_request:44002", "accept": false, - "reason": "Different models and outputs; not the same code-path problem." + "reason": "Different models; same refactor motif is not enough to merge." }, { "left": "pull_request:44000", "right": "pull_request:44003", "accept": false, - "reason": "Vision-text dual encoder and Falcon/Mamba changes are separate model implementations." + "reason": "Vision-text dual encoder vs Falcon/Mamba output tracing are different changes." + }, + { + "left": "pull_request:44000", + "right": "pull_request:44004", + "accept": false, + "reason": "Different architectures and code paths; not the same fix." }, { "left": "pull_request:44000", "right": "pull_request:44025", "accept": false, - "reason": "No shared concrete behavior beyond output-tracing refactor style." + "reason": "Vision-text dual encoder is unrelated to depth-anything model tracing." }, { "left": "pull_request:44001", "right": "pull_request:44002", "accept": false, - "reason": "Unrelated audio and segmentation models; cannot be treated as duplicate fixes." + "reason": "UnivNet and UperNet are separate model families." }, { "left": "pull_request:44001", "right": "pull_request:44003", "accept": false, - "reason": "UnivNet and Mamba/Falcon-Mamba target different code paths and frameworks." + "reason": "Different concrete output/caching logic in different models." }, { "left": "pull_request:44001", "right": "pull_request:44025", "accept": false, - "reason": "Different model families and files; same broad theme only." + "reason": "UnivNet refactor is unrelated to depth-anything refactor." }, { "left": "pull_request:44002", - "right": "pull_request:44029", + "right": "pull_request:44003", "accept": false, - "reason": "UperNet segmentation vs RWKV language model tracing are unrelated changes." + "reason": "UperNet segmentation and Falcon/Mamba generation paths are unrelated." }, { "left": "pull_request:44002", - "right": "pull_request:44003", + "right": "pull_request:44004", "accept": false, - "reason": "Different model stacks and implementations; not plausibly one PR." + "reason": "Different models and output contracts; not mergeable as one fix." }, { "left": "pull_request:44003", "right": "pull_request:44028", "accept": false, - "reason": "Mamba/Falcon-Mamba refactor is unrelated to SuperPoint feature extraction output changes." + "reason": "Falcon/Mamba state tracing and SuperPoint feature extraction are unrelated." }, { "left": "pull_request:44003", "right": "pull_request:44004", "accept": false, - "reason": "Different model families and code paths; not the same concrete bug/change." + "reason": "Different concrete code paths: Mamba/Falcon state handling vs CodeGen cache handling." }, { "left": "pull_request:43998", "right": "pull_request:44004", "accept": false, - "reason": "Timm backbone output tracing and CodeGen internals are separate model changes." + "reason": "Timm backbone and CodeGen are unrelated model implementations." }, { "left": "pull_request:44001", "right": "pull_request:44004", "accept": false, - "reason": "UnivNet vs CodeGen are unrelated; same tracking theme only." + "reason": "UnivNet and CodeGen do not share the same bug or change." }, { "left": "pull_request:44002", "right": "pull_request:44004", "accept": false, - "reason": "Segmentation model refactor vs CodeGen tracing refactor are not the same fix." - }, - { - "left": "pull_request:44000", - "right": "pull_request:44004", - "accept": false, - "reason": "Vision-text dual encoder and CodeGen touch unrelated code paths." + "reason": "UperNet and CodeGen are different code paths." }, { "left": "pull_request:44025", "right": "pull_request:44026", "accept": false, - "reason": "Depth Anything and VisionEncoderDecoder are unrelated model implementations." + "reason": "Depth-anything models vs vision encoder-decoder wrappers are different concrete changes." }, { "left": "pull_request:44025", "right": "pull_request:44027", "accept": false, - "reason": "Depth Anything and Speech Encoder-Decoder are separate changes, not one duplicate bug." + "reason": "Depth-anything and speech encoder-decoder are unrelated model families." }, { "left": "pull_request:44025", "right": "pull_request:44028", "accept": false, - "reason": "Depth Anything refactor is unrelated to SuperPoint output tracing." + "reason": "Depth-anything refactor is unrelated to SuperPoint output tracing." }, { "left": "pull_request:44025", "right": "pull_request:44029", "accept": false, - "reason": "Different models and different code paths; cannot be merged as one PR." + "reason": "Depth-anything and RWKV are different architectures and output contracts." }, { "left": "pull_request:44025", "right": "pull_request:44030", "accept": false, - "reason": "Depth Anything and DPR are unrelated refactors despite similar helper usage." + "reason": "Depth-anything refactor is unrelated to DPR encoder output handling." }, { "left": "pull_request:44026", - "right": "pull_request:44029", + "right": "pull_request:44027", "accept": false, - "reason": "VisionEncoderDecoder and RWKV are unrelated subsystems." + "reason": "Vision encoder-decoder and speech encoder-decoder are separate wrapper implementations." }, { "left": "pull_request:44026", - "right": "pull_request:44027", + "right": "pull_request:44028", "accept": false, - "reason": "VisionEncoderDecoder vs SpeechEncoderDecoder are distinct model code paths." + "reason": "Different model families; not the same underlying bug or change." }, { "left": "pull_request:44026", - "right": "pull_request:44028", + "right": "pull_request:44029", "accept": false, - "reason": "Different model families; the output-tracing refactor is only superficially similar." + "reason": "Vision encoder-decoder wrapper and RWKV generation path are unrelated." }, { "left": "pull_request:44026", "right": "pull_request:44030", "accept": false, - "reason": "DPR and VisionEncoderDecoder have no shared concrete bug or change." + "reason": "Different models and code paths." }, { "left": "pull_request:44027", "right": "pull_request:44028", "accept": false, - "reason": "SpeechEncoderDecoder and SuperPoint are unrelated model implementations." + "reason": "Speech encoder-decoder and SuperPoint are unrelated implementations." }, { "left": "pull_request:44027", "right": "pull_request:44029", "accept": false, - "reason": "Speech encoder-decoder refactor is unrelated to RWKV tracing changes." + "reason": "Speech encoder-decoder wrapper changes do not match RWKV output tracing." }, { "left": "pull_request:44027", "right": "pull_request:44030", "accept": false, - "reason": "Different model families and file targets; not one duplicate PR." + "reason": "Different architectures and concrete fixes." }, { "left": "pull_request:44028", "right": "pull_request:44029", "accept": false, - "reason": "SuperPoint and RWKV are unrelated code paths." + "reason": "SuperPoint feature tracing and RWKV sequence modeling are unrelated." }, { "left": "pull_request:44028", "right": "pull_request:44030", "accept": false, - "reason": "SuperPoint output changes are unrelated to DPR model forwarding." + "reason": "Different model families and output structures." }, { "left": "pull_request:44029", "right": "pull_request:44030", "accept": false, - "reason": "RWKV and DPR are different model families with separate implementations." + "reason": "RWKV and DPR are separate code paths with different output behavior." } ] }, @@ -23146,7 +23087,7 @@ "soft_pairs": [ "pull_request:44662|pull_request:45401" ], - "prepared_review_unit_hash": "6f0565c1a146ffa5b6c357a043f795756014670e78ddbf1df75e1a93e974c293", + "prepared_review_unit_hash": "c9358f835f79bd06fb8233b38d0a62a9cbe91c1937b4fbe0c0460159464c1825", "component_budget": { "node_count": 2, "item_count": 2, @@ -23168,18 +23109,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These are two unrelated model-addition PRs that only overlap at the level of Transformers scaffolding for new models. They do not appear to fix the same code-path problem or belong in one merged change set.", - "confidence": 0.95, + "summary": "These are two unrelated model-addition PRs: one adds PenguinVL, the other adds Voxtral TTS. They share some generic integration files, but they target different models, different model codepaths, and only one links a specific issue. No duplicate relationship is present.", + "confidence": 0.98, "canonical_issue_reason": null, - "canonical_pr_reason": "PR #45401 is the better anchor for the cluster because it is explicitly tied to issue #45027 and presents a self-contained Voxtral TTS integration. PR #44662 is also a model-addition PR, but it is a separate PenguinVL implementation with no issue linkage.", + "canonical_pr_reason": "No canonical PR for deduplication: the PRs implement different models and should not be collapsed into one change.", "best_issue_reason": null, - "best_pr_reason": "PR #45401 is the more suitable representative PR: it has a clear product target, an explicit issue reference, and a narrower, well-scoped integration. PR #44662 is a distinct model implementation and should not be treated as a duplicate of it.", + "best_pr_reason": "PR #44662 is the better representative artifact only by maturity and review activity, but it is not a duplicate of #45401.", "soft_edge_verdicts": [ { "left": "pull_request:44662", "right": "pull_request:45401", "accept": false, - "reason": "They are different model integrations (PenguinVL vs Voxtral TTS). Shared docs/auto-registration/model files are generic new-model scaffolding, not evidence of the same underlying bug or change." + "reason": "Reject: PenguinVL and Voxtral TTS are distinct model implementations with different titles, filenames, and likely different codepaths. Overlap is limited to shared Transformers registration/docs plumbing, which is too broad to indicate the same underlying change." } ] }, @@ -23209,7 +23150,7 @@ "soft_pairs": [ "pull_request:44965|pull_request:46845" ], - "prepared_review_unit_hash": "649b47787a322cd102313d38046cd486d1543f091d34550537ca1452c0587a2b", + "prepared_review_unit_hash": "8866b115fbc0df34b9b0278ac2586f8ad455be36235585fcab6bf425c4750c64", "component_budget": { "node_count": 2, "item_count": 2, @@ -23231,18 +23172,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "The two PRs are not duplicates: #44965 rewrites a workflow to convert PRs to draft, while #46845 changes CI branch routing and includes a dummy model file edit. They address different code paths and different underlying purposes.", - "confidence": 0.94, + "summary": "These two pull requests are not duplicates: one is a workflow change to route PR CI to a smaller runner/branch target, while the other is an unrelated attempt to convert PRs to draft via a documentation workflow. They do not target the same code path or same underlying fix.", + "confidence": 0.97, "canonical_issue_reason": null, - "canonical_pr_reason": "No clear canonical PR: #44965 and #46845 are unrelated exploratory changes touching different workflows/code paths.", + "canonical_pr_reason": "PR 46845 is the more coherent and concrete change set, but it is not a duplicate of PR 44965; the two PRs address different automation paths and should remain separate.", "best_issue_reason": null, - "best_pr_reason": "No single best PR for a duplicate cluster; neither PR is a good representative of the other because they solve different problems.", + "best_pr_reason": "PR 46845 is the better representative of the cluster because it has a clearer, self-contained CI-related purpose and an explicit branch/workflow adjustment, whereas PR 44965 is a different draft-conversion workflow change.", "soft_edge_verdicts": [ { "left": "pull_request:44965", "right": "pull_request:46845", "accept": false, - "reason": "Different changes and intents: #44965 is about changing PRs to draft via a workflow, while #46845 adjusts CI branch selection and includes an unrelated placeholder code edit. They are not the same underlying fix and would not be merged as one PR." + "reason": "Different underlying automation changes: 44965 touches PR documentation/draft conversion workflow, while 46845 changes PR CI routing and adds unrelated code/test churn. They are not the same bug or mergeable as one PR." } ] }, @@ -23272,7 +23213,7 @@ "soft_pairs": [ "pull_request:44979|pull_request:45363" ], - "prepared_review_unit_hash": "3441ffa787956953558bb7b8370c216cc35d096fd7a2d09e335c36edd4d1d202", + "prepared_review_unit_hash": "55c4f258eca2e1b525148b1d9d7a30eb13b24b8e6db87fe21fcb35c42c094f24", "component_budget": { "node_count": 2, "item_count": 2, @@ -23294,18 +23235,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "The two PRs are related by the general idea of fusion, but they implement different features and code paths: one adds a generic module-fusion API, the other wires in n-to-1 kernel fusion through KernelConfig and related integrations. They should not be treated as duplicates.", - "confidence": 0.93, + "summary": "These two PRs are related to kernel/module fusion infrastructure, but they implement different layers of the stack and do not look like the same mergeable change. PR #44979 adds a standalone Module Fusion API and tests; PR #45363 adds n-to-1 kernel fusion wiring via KernelConfig across integrations/modeling utilities.", + "confidence": 0.94, "canonical_issue_reason": null, - "canonical_pr_reason": "PR #45363 is the more concrete, end-user-facing implementation of fusion and touches the actual model/integration code path, whereas #44979 is a broader API scaffold.", + "canonical_pr_reason": "PR #45363 is the better canonical representative for this cluster because it is the broader, end-user-facing fusion change and touches the runtime integration path rather than just introducing a lower-level API scaffold.", "best_issue_reason": null, - "best_pr_reason": "PR #45363 best represents the cluster because it specifies a concrete fusion mechanism (KernelConfig-based n-to-1 kernel fusion) rather than only introducing a generic abstraction.", + "best_pr_reason": "PR #45363 is the best overall fit: it has the clearest concrete behavior change (KernelConfig-driven fusion in model loading/execution) and is more likely to anchor follow-up discussion than the smaller API-only foundation PR.", "soft_edge_verdicts": [ { "left": "pull_request:44979", "right": "pull_request:45363", "accept": false, - "reason": "Different underlying changes: #44979 introduces a generic Module Fusion API, while #45363 implements KernelConfig-driven n-to-1 kernel fusion and integration wiring. They are related, but not the same fix or feature." + "reason": "Different code paths and scopes: #44979 introduces a general Module Fusion API in a new module, while #45363 implements n-to-1 kernel fusion through KernelConfig and integration hooks. Related, but not the same underlying change and not plausibly one merged PR." } ] }, @@ -23335,7 +23276,7 @@ "soft_pairs": [ "pull_request:45086|pull_request:45317" ], - "prepared_review_unit_hash": "bc174d5b46e0743ce30a5adb391be867b74f8b15edcfe26f48e3fdd12c29a558", + "prepared_review_unit_hash": "b56b14de3cfd1956a4b112c7dfaa72472769978f8a25a25342556eefe297eed7", "component_budget": { "node_count": 2, "item_count": 2, @@ -23357,18 +23298,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Both pull requests patch the same Mistral pre-tokenizer AttributeError in the same file and target the same issue; the second PR is a slightly fuller version with tests, so they look like duplicate fixes for one code-path bug.", - "confidence": 0.98, + "summary": "Both PRs patch the same AttributeError in `_patch_mistral_regex` by switching from `tokenizer.backend_tokenizer.pre_tokenizer` to `tokenizer.pre_tokenizer` and updating the sequence assignment accordingly; one PR just adds tests and a slightly broader diff around the same fix.", + "confidence": 0.99, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 45317 is the better canonical PR because it contains the same fix as 45086 plus test coverage, making it the more complete representative of the duplicate pair.", + "canonical_pr_reason": "PR #45317 is the better canonical PR because it contains the same core fix as #45086 plus test coverage, making it the more complete representation of the change.", "best_issue_reason": null, - "best_pr_reason": "PR 45317 is the best PR choice: same underlying fix, same issue target, and it adds tests while 45086 is the minimal code-only variant.", + "best_pr_reason": "PR #45317 is the best representative because it fixes the same concrete code-path problem as #45086 and includes validation via tests, so it is the strongest standalone record of the change.", "soft_edge_verdicts": [ { "left": "pull_request:45086", "right": "pull_request:45317", "accept": true, - "reason": "Same bug, same changed code path in tokenization_utils_tokenizers.py, same issue target 45081, and the second PR is a superset with tests; these could plausibly be merged into one PR." + "reason": "Accept: these PRs make the same exact code change in the same file/function to fix the same AttributeError, and the right PR simply adds tests around the same fix." } ] }, @@ -23398,7 +23339,7 @@ "soft_pairs": [ "pull_request:45115|pull_request:45569" ], - "prepared_review_unit_hash": "97e019f724c147f7ef3109f82a9c18c95052dbedd50251889a08d699a1f48706", + "prepared_review_unit_hash": "f0e6332fac5f2e9bbab1fc838db6bde1d843efec31d5488637067b433f01637e", "component_budget": { "node_count": 2, "item_count": 2, @@ -23420,18 +23361,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These PRs are related to NemotronH, but they address different changes: #45115 is a narrower refactor/conversion-mapping update for inheriting GraniteMoeHybrid, while #45569 is a broader model split introducing nemotron_h_dense/sparse plus docs/tests. They should not be treated as duplicates.", + "summary": "Both items are NemotronH-related PRs, but they address different changes: one refactors conversion/mapping compatibility, while the other splits NemotronH into dense/sparse model variants and updates docs/auto-mapping.", "confidence": 0.91, "canonical_issue_reason": null, - "canonical_pr_reason": "PR #45569 is the better cluster representative because it is the broader, more complete NemotronH family change and includes the new dense/sparse model files, docs, and tests. PR #45115 is a narrower refactor focused on conversion and inheritance.", + "canonical_pr_reason": "No clear canonical PR for duplication: #45115 is a conversion-mapping refactor, while #45569 is a model-architecture split and documentation/API update.", "best_issue_reason": null, - "best_pr_reason": "PR #45569 is the strongest standalone candidate: it makes the concrete product change (new NemotronHDense/NemotronHSparse support) and is more likely to be the primary landed implementation. #45115 is a smaller preparatory/refactor PR.", + "best_pr_reason": "#45569 is the stronger representative of the main product change because it introduces the new dense/sparse NemotronH model structure and corresponding integration work.", "soft_edge_verdicts": [ { "left": "pull_request:45115", "right": "pull_request:45569", "accept": false, - "reason": "Not the same underlying fix/change. #45115 is a refactor and weight-conversion mapping update for NemotronH inheritance, while #45569 introduces separate dense/sparse model variants and documentation. They overlap in shared core files but are not mergeable as one duplicate PR." + "reason": "They share the NemotronH subsystem, but not the same underlying fix. #45115 only adds conversion-key renames for checkpoint loading, whereas #45569 reworks the model/package layout into dense and sparse variants plus docs/tests. These could not plausibly be merged as one PR without mixing distinct work." } ] }, @@ -23461,40 +23402,40 @@ "soft_pairs": [ "pull_request:45332|pull_request:45333" ], - "prepared_review_unit_hash": "7ddeb2c02f5d9db1e6c564f529f4cadadfdf3d9f2f9b0f2758abf5a3a9f42b60", + "prepared_review_unit_hash": "c6bc5fe1c029dde9faaf12eb3df4567fbeeb2378b17d5727a4995521a961a653", "component_budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 5952, - "estimated_input_tokens": 1488, - "estimated_eval_tokens": 3232 + "serialized_chars": 5953, + "estimated_input_tokens": 1489, + "estimated_eval_tokens": 3234 }, "budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 5952, - "estimated_input_tokens": 1488, - "estimated_eval_tokens": 3232 + "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 closely related but not the same change: PR 45333 adds per-layer heterogeneous configuration support, while PR 45332 extends that into broader heterogeneous model/runtime support. They overlap on shared config/docs/test scaffolding, but the code paths and scope are different enough that I would not collapse them as duplicates.", - "confidence": 0.87, + "summary": "These are related feature PRs for heterogeneous model support, but they are not duplicates: #45333 is a config-only slice, while #45332 includes config plus broader modeling/runtime integration. I would keep #45332 as the representative PR for the cluster and reject the soft-duplicate link.", + "confidence": 0.91, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 45332 is the broader umbrella change and appears to subsume the configuration work while adding the model/runtime pieces.", + "canonical_pr_reason": "#45332 is the broader, more complete implementation: it covers per-layer config plus modeling/runtime changes across cache, masking, mxfp4, and heterogeneous modeling utilities, so it best represents the underlying feature work.", "best_issue_reason": null, - "best_pr_reason": "PR 45332 is the better representative for the cluster because it covers the wider heterogeneous-model feature set, not just configuration plumbing.", + "best_pr_reason": "#45332 is the best PR to anchor the cluster because it subsumes the config work and extends into the core code paths needed for actual heterogeneous model support.", "soft_edge_verdicts": [ { "left": "pull_request:45332", "right": "pull_request:45333", "accept": false, - "reason": "Related but not duplicate: 45333 is configuration-only, while 45332 adds broader heterogeneous modeling/runtime support and touches additional code paths. They could be sequenced, but they are not the same concrete fix/change." + "reason": "They share the same feature area and some files, but #45333 is only per-layer configuration while #45332 also adds modeling/runtime support. That makes #45333 a subset, not the same concrete PR-level change." } ] }, @@ -23524,7 +23465,7 @@ "soft_pairs": [ "pull_request:45415|pull_request:45425" ], - "prepared_review_unit_hash": "137257cc27a2595eb559ca953e0a90fe2a94c1527c3c8506ca1d6018cd20d3ac", + "prepared_review_unit_hash": "bb1ce2ec1b2af41f1f415c3778ed90eac49c748e6290c7c41765f4065b02ff99", "component_budget": { "node_count": 2, "item_count": 2, @@ -23546,18 +23487,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Both PRs are type-checking/typing-related, but they address different changes: #45415 is a broad mypy/typing cleanup across many `src/transformers/*` files, while #45425 adds specific typing aliases/protocols to support `modeling_utils` and PEFT integration. They are not the same underlying fix, so this should not be merged as a duplicate pair.", - "confidence": 0.93, + "summary": "These are both typing-focused PRs in transformers, but they address different scopes: #45415 is a broad type-checking sweep across many files, while #45425 adds/adjusts specific typing protocols and annotations around modeling_utils and related helpers. Shared typing infrastructure is not enough to make them the same change.", + "confidence": 0.86, "canonical_issue_reason": null, - "canonical_pr_reason": "PR #45415 is the broader representative of the cluster because it targets widespread type-checking cleanup across many modules and config, whereas #45425 is a narrower typing-support change.", + "canonical_pr_reason": "#45425 is the more focused, concrete typing change set; #45415 is a larger repo-wide type-checking cleanup and not the same underlying change.", "best_issue_reason": null, - "best_pr_reason": "PR #45415 is the best single representative if one must be chosen, since it spans the largest portion of the observed typing work and is more central to the cluster theme.", + "best_pr_reason": "#45425 is the better representative PR for this cluster because it makes a narrower, self-contained typing fix rather than a broad linting/type-check sweep.", "soft_edge_verdicts": [ { "left": "pull_request:45415", "right": "pull_request:45425", "accept": false, - "reason": "Different concrete changes: #45415 is broad type-checking enablement and ignore tuning across many files, while #45425 adds specific typing protocols/aliases for modeling_utils/PEFT. Shared files and subsystem are not enough to treat them as the same fix." + "reason": "Both are typing-related, but they fix different problems at different scopes. #45415 is a wide sweep to add type checking across many modules and silence type-check issues, while #45425 introduces specific typing protocols/aliases for modeling_utils and related integrations. They do not look mergeable as one PR." } ] }, @@ -23587,7 +23528,7 @@ "soft_pairs": [ "pull_request:45548|pull_request:45662" ], - "prepared_review_unit_hash": "1efed35ad68af6364a43123f6957355f9a120c836742e2cc05f74e0c372f9681", + "prepared_review_unit_hash": "50adef346135bb2fb9bd94d598c43001bf14ee390cb6ed197af01f39c891c4c2", "component_budget": { "node_count": 2, "item_count": 2, @@ -23609,18 +23550,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Both PRs touch expert-parallel loading, but they fix different bugs on different code paths: one is about config serialization and accelerate/ZeRO-3 loading, the other about FSDP2 sharded parameter assignment and rank-0 broadcast overwrites. They are related but not duplicates.", - "confidence": 0.93, + "summary": "Two PRs touch expert-parallel loading paths, but they fix different failure modes: one is about preserving EP-related config and launch-time loading for DeepSpeed ZeRO-3, while the other prevents EP experts from being overwritten during FSDP2/rank-0 broadcast. They are related by subsystem, not the same underlying change.", + "confidence": 0.94, "canonical_issue_reason": null, - "canonical_pr_reason": "If a single representative is needed, PR #45662 is the better canonical PR because it targets a more concrete runtime corruption bug in the EP loading path and includes the main code-path fix rather than a mostly compatibility/serialization adjustment.", + "canonical_pr_reason": "PR #45662 is the better canonical PR for this cluster because it targets a concrete EP loading corruption bug in the weight-loading path and spans the relevant post-shard / distributed tensor handling. It is the more substantive and behaviorally specific fix.", "best_issue_reason": null, - "best_pr_reason": "PR #45662 is the stronger standalone fix: it addresses a specific data-overwrite failure mode during EP/FSDP2 loading, has a clear symptom, and changes the relevant loading machinery directly.", + "best_pr_reason": "PR #45662 is the best representative of the cluster because it addresses a clearer end-to-end correctness problem in the model-loading code path, whereas PR #45548 is a narrower config/launch integration fix.", "soft_edge_verdicts": [ { "left": "pull_request:45548", "right": "pull_request:45662", "accept": false, - "reason": "They share the expert-parallel theme, but #45548 fixes distributed_config serialization and accelerate/DeepSpeed ZeRO-3 loading, while #45662 fixes FSDP2 expert parameters being overwritten by rank-0 broadcast. Different underlying bugs and not safely mergeable into one PR." + "reason": "Both involve expert parallelism, but they fix different bugs and code paths: #45548 preserves distributed config and adds an EP flag for accelerate/ZeRO-3 loading, while #45662 fixes experts being overwritten after shard wrapping in FSDP2. Not mergeable as one PR." } ] }, @@ -23650,40 +23591,40 @@ "soft_pairs": [ "pull_request:45639|pull_request:45645" ], - "prepared_review_unit_hash": "f9fd4b929b6e2376392cfcf56c68f6a53a007b62e0b743db68d0495df2651edf", + "prepared_review_unit_hash": "c4eff0bb07013c6542e403b08445f3e8da8df8b54921f871c4c9ebb30fbde358", "component_budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 4672, - "estimated_input_tokens": 1168, - "estimated_eval_tokens": 2592 + "serialized_chars": 4673, + "estimated_input_tokens": 1169, + "estimated_eval_tokens": 2594 }, "budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 4672, - "estimated_input_tokens": 1168, - "estimated_eval_tokens": 2592 + "serialized_chars": 4673, + "estimated_input_tokens": 1169, + "estimated_eval_tokens": 2594 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Both PRs target the same xdist-related captured_info logging problem in test utilities; one is a narrower fix, the other bundles the generation fix with CI log collection changes, so they are duplicate/same-change candidates.", - "confidence": 0.92, + "summary": "Both PRs target the same xdist-related captured_info logging problem in transformers testing utilities; PR 45645 is the broader, more complete version because it includes the core xdist-safe file handling plus the CI workflow change to aggregate multiple captured_info files.", + "confidence": 0.88, "canonical_issue_reason": null, - "canonical_pr_reason": "PR #45645 is the better canonical PR because it covers the full end-to-end fix: xdist-safe file creation in testing_utils plus workflow changes to collect all captured_info artifacts in CI.", + "canonical_pr_reason": "PR 45645 is the better canonical PR because it covers the underlying fix in testing_utils and also updates the CI workflow to read all xdist-split captured_info artifacts, making the end-to-end solution more complete.", "best_issue_reason": null, - "best_pr_reason": "PR #45645 is the strongest representative of the underlying bug/fix since it handles both producing separate artifacts per worker and reading them safely in CI, making it the more complete merge target.", + "best_pr_reason": "PR 45645 is the strongest representative of the change set: it fixes the xdist-safe output path and preserves CI debug log collection across multiple worker files, whereas 45639 appears to cover only the code-side portion.", "soft_edge_verdicts": [ { "left": "pull_request:45639", "right": "pull_request:45645", "accept": true, - "reason": "Both PRs fix the same concrete captured_info xdist collision path in the test logging flow; #45645 is a superset of #45639's core change and could plausibly subsume it into one PR." + "reason": "Same underlying bug: captured_info logging breaks under xdist. 45645 looks like a superset of 45639 by including the same testing_utils fix plus the workflow-side aggregation of split log files, so they could plausibly be merged into one PR." } ] }, @@ -23713,7 +23654,7 @@ "soft_pairs": [ "pull_request:45720|pull_request:46259" ], - "prepared_review_unit_hash": "506211fc586356f6783d1e9c5de670dab9b692c54567f4627100da2ac2c2260f", + "prepared_review_unit_hash": "5ba90fed84929d18bddad0ac449ae14298bfeed26c270e5403b55b031364e63f", "component_budget": { "node_count": 2, "item_count": 2, @@ -23735,18 +23676,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These two PRs are both lint-rule implementation changes in Transformers, but they address different problems: one standardizes forward signatures/return handling (TRF003) while the other refactors weight initialization logic (TRF018). They overlap in some model filenames, but not in the underlying change.", - "confidence": 0.93, + "summary": "The two PRs are both linter-rule changes, but they implement different rules on different code paths (TRF003 vs TRF018). They should not be treated as duplicates.", + "confidence": 0.96, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 45720 is the cleaner representative: it has a focused, concrete scope around forward return handling and signature cleanup across model classes.", + "canonical_pr_reason": "PR #46259 is the better cluster representative because it adds the more substantial modeling-rule implementation across many model files, while #45720 is a narrower partial TRF003 change.", "best_issue_reason": null, - "best_pr_reason": "PR 45720 is the best PR in this cluster because its change is narrower and more clearly tied to one specific code-path rule implementation.", + "best_pr_reason": "PR #46259 is the strongest standalone candidate: it has a broader, clearer scope and a more complete rule implementation, making it more suitable as the representative PR for this cluster.", "soft_edge_verdicts": [ { "left": "pull_request:45720", "right": "pull_request:46259", "accept": false, - "reason": "Reject: PR 45720 implements TRF003 by changing forward return handling/signatures, while PR 46259 implements TRF018 by changing _init_weights initialization behavior. They touch some of the same model files but fix different code paths and would not plausibly be merged as one PR." + "reason": "Reject: these are different lint rules with different intended behavior. #45720 implements part of TRF003, while #46259 adds TRF018; they do not appear to fix the same concrete bug or change." } ] }, @@ -23845,12 +23786,12 @@ "pull_request:45932|pull_request:46891", "pull_request:46100|pull_request:46891" ], - "prepared_review_unit_hash": "2d28991448b9b0a139a8aed023118262813037421fd71be46f3ffd30a15cf6b8", + "prepared_review_unit_hash": "74cbde91c9343bcab1d2ba59e55a9ab7b4ca9a8704db2853d15f74f99cd10ea3", "component_budget": { "node_count": 3, "item_count": 3, "soft_pair_count": 3, - "serialized_chars": 7882, + "serialized_chars": 7883, "estimated_input_tokens": 1971, "estimated_eval_tokens": 4198 }, @@ -23858,7 +23799,7 @@ "node_count": 3, "item_count": 3, "soft_pair_count": 3, - "serialized_chars": 7882, + "serialized_chars": 7883, "estimated_input_tokens": 1971, "estimated_eval_tokens": 4198 }, @@ -23867,30 +23808,30 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These PRs are only loosely similar as CI/test-related work, but they do not fix the same underlying problem. The cluster appears to be a false positive: each PR targets a different change set and would not plausibly be merged into one PR.", - "confidence": 0.94, + "summary": "These are three unrelated open test/CI-oriented PRs that only overlap superficially in labels and workflow/testing areas. They do not appear to fix the same underlying problem or belong in one duplicate cluster.", + "confidence": 0.92, "canonical_issue_reason": null, - "canonical_pr_reason": "No true duplicate PR exists here; if one must be selected as the best representative, PR #46891 is the most focused and self-contained CI-related change.", + "canonical_pr_reason": "PR #45932 is the broadest and most central of the three, with the largest surface area and explicit CI/workflow changes, but it is not a true duplicate of the others.", "best_issue_reason": null, - "best_pr_reason": "PR #46891 is the cleanest representative of the cluster because it has a single coherent intent (PEFT integration CI) and a narrow diff compared with the broader, more mixed test/debug changes in the other PRs.", + "best_pr_reason": "#45932 is the best representative PR in this set because it is the most substantial and most centrally about CI/test plumbing, whereas #46100 is a test migration and #46891 is a separate PEFT integration CI experiment.", "soft_edge_verdicts": [ { "left": "pull_request:45932", "right": "pull_request:46100", "accept": false, - "reason": "Both are test/CI-adjacent, but the diffs are unrelated: #45932 mixes Grafana/OTel workflow changes with model code edits, while #46100 is a test migration adding kernel/DDP test files. They are not the same bug/change." + "reason": "Superficial overlap only: both are test-related, but #45932 changes model code and CI secrets, while #46100 adds kernel/model tests and a DDP YAML. Different concrete changes and no shared bug path." }, { "left": "pull_request:45932", "right": "pull_request:46891", "accept": false, - "reason": "They both touch the CI workflow file, but #45932 is a broad debug/test PR with model config changes, while #46891 is specifically about PEFT integration CI. Shared filename alone is not enough to treat them as duplicates." + "reason": "Both touch the CI caller workflow, but #45932 is a grafana/testing PR with model mapping changes, while #46891 is a PEFT integration test/CI branch update. Same subsystem, different change intent and code path." }, { "left": "pull_request:46100", "right": "pull_request:46891", "accept": false, - "reason": "#46100 is a test migration and new model test suite, whereas #46891 changes PEFT integration and CI wiring. Different code paths and goals; they would not merge cleanly as one PR." + "reason": "Only shared theme is testing. #46100 is a broad test migration across vision models; #46891 is specifically about PEFT integration CI and related integration code. Not the same underlying change." } ] }, @@ -23920,12 +23861,12 @@ "soft_pairs": [ "pull_request:46126|pull_request:46269" ], - "prepared_review_unit_hash": "204d0e74cd1b370d2e7e6cdd90dd9ea53ebbd9ea2e4d55cd3c48b77a1e7f6206", + "prepared_review_unit_hash": "db55ce1e3880d9c4c7b2baae9c0df521c6a1beb5b018d0aa9f6e2ec1cf1f92fc", "component_budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 5526, + "serialized_chars": 5525, "estimated_input_tokens": 1382, "estimated_eval_tokens": 3020 }, @@ -23933,7 +23874,7 @@ "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 5526, + "serialized_chars": 5525, "estimated_input_tokens": 1382, "estimated_eval_tokens": 3020 }, @@ -23942,18 +23883,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Two unrelated pull requests in the distributed stack: one focused on EP/Trainer integration, the other a broad distributed branch base touching many subsystems. They are not the same change and should not be merged as duplicates.", - "confidence": 0.91, + "summary": "These are not duplicates: PR 46126 is a focused EP/Trainer integration change, while PR 46269 is a broad distributed-branch base that happens to touch overlapping distributed files but represents a much larger, different change set.", + "confidence": 0.92, "canonical_issue_reason": null, - "canonical_pr_reason": "PR #46126 is the more focused, self-contained change: EP + Trainer integration with targeted edits in tensor parallel, utils, Qwen3 MoE config, and trainer code.", + "canonical_pr_reason": "PR 46126 is the more canonical artifact for this cluster because it describes a specific, coherent feature/change with a clear scope; PR 46269 is a broad base branch with many unrelated distributed updates and CI/config churn.", "best_issue_reason": null, - "best_pr_reason": "PR #46126 is the best representative for this cluster because it has a clear single feature goal and bounded scope, unlike #46269 which is an umbrella branch base spanning many unrelated distributed updates.", + "best_pr_reason": "PR 46126 is the best representative PR here: it targets one concrete code path and could be reviewed/merged on its own, unlike PR 46269 which is too sprawling to serve as the canonical duplicate target.", "soft_edge_verdicts": [ { "left": "pull_request:46126", "right": "pull_request:46269", "accept": false, - "reason": "Shared filenames are not enough here: #46126 is a specific EP routing/trainer feature, while #46269 is a broad distributed branch base with many unrelated changes. They do not look like the same concrete bug or change." + "reason": "Reject: the overlap is only at the level of shared distributed filenames. PR 46126 implements EP + Trainer integration, whereas PR 46269 is a wide distributed branch base touching CI, FSDP, modeling, and multiple integration paths; they are not the same underlying fix/change and would not plausibly be merged as one PR." } ] }, @@ -24019,208 +23960,220 @@ "pull_request:46504|pull_request:46563", "pull_request:46504|pull_request:46716" ], - "prepared_review_unit_hash": "be03ab53034eb26b9465300f6aed8630b48da508c2672aaab92643deaec0bb57", + "prepared_review_unit_hash": "8f152e3bcec53dc722aa5106a0a0c4c6688499307cc2b0c643fdef4a16010dcb", "component_budget": { "node_count": 10, "item_count": 10, "soft_pair_count": 29, - "serialized_chars": 25928, - "estimated_input_tokens": 6482, - "estimated_eval_tokens": 13220 + "serialized_chars": 25931, + "estimated_input_tokens": 6483, + "estimated_eval_tokens": 13222 }, "budget": { "node_count": 10, "item_count": 10, "soft_pair_count": 29, - "serialized_chars": 25928, - "estimated_input_tokens": 6482, - "estimated_eval_tokens": 13220 + "serialized_chars": 25931, + "estimated_input_tokens": 6483, + "estimated_eval_tokens": 13222 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "This cluster is a loose similarity set of unrelated workflow-maintenance PRs. The only tight subcluster is several sequential edits to `.github/workflows/pr-ci-security-gate.yml`, but they address different hardening changes rather than the same bug or change.", + "summary": "These are mostly unrelated workflow-maintenance PRs that only look similar at the title level. Even the repeated pr-ci-security-gate.yml changes address different concerns, so there is no safe duplicate merge here.", "confidence": 0.96, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 46554 is the most central anchor in the only closely related subcluster: it is the earliest pr-ci-security-gate workflow update and shares the exact filename with 46558/46563/46716.", + "canonical_pr_reason": "No canonical duplicate PR stands out because the cluster is heterogeneous. If one representative is needed, PR 46554 is the most substantive and central security-gate workflow change, but it is not a duplicate of the others.", "best_issue_reason": null, - "best_pr_reason": "PR 46554 is the best representative PR by connectivity and filename overlap, though it is not a true duplicate of the others.", + "best_pr_reason": "PR 46554 is the best single representative because it makes the core pr-ci-security-gate workflow logic change and has the strongest overlap with the other security-gate variants, though the group is not a true duplicate set.", "soft_edge_verdicts": [ { "left": "pull_request:46138", "right": "pull_request:46554", "accept": false, - "reason": "Different workflows and unrelated changes (`self-comment-ci` vs `pr-ci-security-gate`)." + "reason": "Different workflow files and different changes: self-comment CI stdin handling vs security-gate checks." }, { "left": "pull_request:46138", "right": "pull_request:46558", "accept": false, - "reason": "Different workflows; no shared code path or same underlying fix." + "reason": "Different workflow files; one updates comment parsing, the other changes PR CI gate permissions/approval flow." }, { "left": "pull_request:46138", "right": "pull_request:46563", "accept": false, - "reason": "Different workflows and distinct changes, not one mergeable fix." + "reason": "Unrelated workflows: self-comment CI command plumbing vs adding safety notes to the security gate reusable workflow." }, { "left": "pull_request:46138", "right": "pull_request:46716", "accept": false, - "reason": "Different workflows; the PRs solve unrelated CI issues." + "reason": "Different workflow files and unrelated changes; not the same underlying bug or fix." }, { "left": "pull_request:46138", "right": "pull_request:46504", "accept": false, - "reason": "Separate workflow files and different shell-quoting changes; not the same change." + "reason": "Different workflow files and different behavior; only a broad CI-workflow similarity." }, { "left": "pull_request:46432", "right": "pull_request:46503", "accept": false, - "reason": "`ssh-runner.yml` vs `benchmark_v2.yml`; unrelated workflow updates." + "reason": "SSH runner hardening vs benchmark trigger logic; same subsystem is too broad to merge." }, { "left": "pull_request:46432", "right": "pull_request:46501", "accept": false, - "reason": "Different workflows and different behavior changes." + "reason": "Unrelated workflow changes in different files: SSH runner hardening vs anti-slop permissions layout." }, { "left": "pull_request:46432", "right": "pull_request:46502", "accept": false, - "reason": "Different workflows; no same concrete bug or code path." + "reason": "Different workflows and problems; SSH runner secret handling is not the same as reviewer assignment checkout behavior." }, { "left": "pull_request:46432", "right": "pull_request:46554", "accept": false, - "reason": "`ssh-runner.yml` hardening is unrelated to `pr-ci-security-gate.yml` logic." + "reason": "Different files and distinct fixes; SSH runner environment cleanup is unrelated to security-gate logic." }, { "left": "pull_request:46432", "right": "pull_request:46558", "accept": false, - "reason": "Different workflows and unrelated fixes." + "reason": "SSH runner hardening is unrelated to PR CI security-gate permissions/approval changes." }, { "left": "pull_request:46501", "right": "pull_request:46503", "accept": false, - "reason": "`anti-slop.yml` vs `benchmark_v2.yml`; unrelated workflow changes." + "reason": "Different workflow files and different triggers/permissions; similar title only." }, { "left": "pull_request:46501", "right": "pull_request:46502", "accept": false, - "reason": "Different workflows; no shared bug or code path." + "reason": "Anti-slop permissions and reviewer checkout are separate workflow concerns, not one bug." }, { "left": "pull_request:46501", "right": "pull_request:46504", "accept": false, - "reason": "Different workflow files and unrelated edits." + "reason": "Different workflow files and distinct behavior changes; not mergeable as one fix." }, { "left": "pull_request:46501", "right": "pull_request:46554", "accept": false, - "reason": "`anti-slop.yml` and `pr-ci-security-gate.yml` are different workflows with different purposes." + "reason": "Different workflow files; anti-slop permissions vs security-gate code-path changes." }, { "left": "pull_request:46502", "right": "pull_request:46503", "accept": false, - "reason": "Different workflows and unrelated behavior changes." + "reason": "Reviewer checkout ref and benchmark trigger conditions are unrelated." }, { "left": "pull_request:46502", "right": "pull_request:46504", "accept": false, - "reason": "`assign-reviewers.yml` vs `build-ci-docker-images.yml`; unrelated changes." + "reason": "Different workflow files and unrelated changes; only superficial CI similarity." }, { "left": "pull_request:46502", "right": "pull_request:46554", "accept": false, - "reason": "Different workflows; no common underlying bug." + "reason": "Assign-reviewers checkout behavior is unrelated to security-gate logic." }, { "left": "pull_request:46503", "right": "pull_request:46504", "accept": false, - "reason": "`benchmark_v2.yml` and `build-ci-docker-images.yml` address different CI paths." + "reason": "Benchmark dispatch trigger and build-ci docker commit parsing are distinct workflow fixes." }, { "left": "pull_request:46503", "right": "pull_request:46554", "accept": false, - "reason": "Different workflows and unrelated fixes." + "reason": "Benchmark runner trigger logic is unrelated to PR CI security gate changes." }, { "left": "pull_request:46504", "right": "pull_request:46554", "accept": false, - "reason": "`build-ci-docker-images.yml` and `pr-ci-security-gate.yml` are different CI concerns." + "reason": "Build-ci tag selection from commit messages is unrelated to security-gate workflow hardening." }, { "left": "pull_request:46504", "right": "pull_request:46558", "accept": false, - "reason": "Different workflow files and distinct changes." + "reason": "Different workflow files and fixes; no shared underlying defect." }, { "left": "pull_request:46554", "right": "pull_request:46558", "accept": false, - "reason": "Same workflow file, but different hardening changes (`checkout/diff checks` vs `permissions`/job gating)." + "reason": "Same file, but different concrete changes: one adds checkout and file-change checks, the other adjusts permissions and approval gating." }, { "left": "pull_request:46554", "right": "pull_request:46563", "accept": false, - "reason": "Same file, but one changes the reusable workflow note and the other changes gate logic; not the same fix." + "reason": "Same workflow file, but one is security logic and the other is explanatory comments only." }, { "left": "pull_request:46554", "right": "pull_request:46716", "accept": false, - "reason": "Same workflow file, but `secrets: {}` is a different security tweak from the earlier gate changes." + "reason": "Different security-gate edits: one adds checks, the other changes secrets inheritance; not the same change." }, { "left": "pull_request:46558", "right": "pull_request:46563", "accept": false, - "reason": "Same file, but the edits are different security adjustments and not one concrete change." + "reason": "Permissions/approval logic vs comment-only update; not the same underlying fix." }, { "left": "pull_request:46558", "right": "pull_request:46716", "accept": false, - "reason": "Same workflow file, but `permissions`/`if` changes are distinct from `secrets: {}`." + "reason": "Different changes in the same workflow file: approval/permissions vs secrets handling." }, { "left": "pull_request:46563", "right": "pull_request:46716", "accept": false, - "reason": "Same file, but different aspects of the reusable workflow call; not duplicate fixes." + "reason": "Comment-only documentation change vs functional secrets scoping change." }, { "left": "pull_request:46504", "right": "pull_request:46563", "accept": false, - "reason": "Different workflow files and unrelated updates." + "reason": "Different workflow files and unrelated changes." }, { "left": "pull_request:46504", "right": "pull_request:46716", "accept": false, - "reason": "Different workflow files; no shared concrete bug or change." + "reason": "Different workflow files and unrelated fixes; not duplicate PRs." + }, + { + "left": "pull_request:46558", + "right": "pull_request:46563", + "accept": false, + "reason": "Different functional area in the same file; not the same concrete problem." + }, + { + "left": "pull_request:46558", + "right": "pull_request:46716", + "accept": false, + "reason": "Permissions/approval gating vs secrets scoping are distinct changes." } ] }, @@ -24250,12 +24203,12 @@ "soft_pairs": [ "pull_request:46152|pull_request:46207" ], - "prepared_review_unit_hash": "1668bb7bb3f908b3b282782212ef3c8789a92fb8c3ea5f7bb39930bc518f6ab2", + "prepared_review_unit_hash": "8924788f6a3cf7debd8f6938790d69f7ffcd4ae233d2678bcc306843be5a83f4", "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 }, @@ -24263,7 +24216,7 @@ "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 4276, + "serialized_chars": 4275, "estimated_input_tokens": 1069, "estimated_eval_tokens": 2394 }, @@ -24272,18 +24225,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These two open PRs share a broad theme of replacing file-based source access, but they target different code paths and different problems: one changes import-time resource resolution in `import_utils.py`, the other changes attention-implementation detection in `modeling_utils.py`. They are not the same underlying bug/fix, so they should not be clustered as duplicates.", - "confidence": 0.92, + "summary": "Both PRs modernize source-file inspection to avoid direct open() calls, but they address different code paths in different modules and are not duplicates.", + "confidence": 0.91, "canonical_issue_reason": null, - "canonical_pr_reason": null, + "canonical_pr_reason": "PR #46152 is the broader representative of the shared theme: it introduces a reusable source-path resolution helper in import_utils to replace file-open based self-inspection. PR #46207 is a separate, narrower fix in modeling_utils.", "best_issue_reason": null, - "best_pr_reason": "PR 46207 is the stronger standalone fix: it targets a specific heuristic in `_can_set_attn_implementation`, includes a focused test file, and has a clearly defined behavioral change.", + "best_pr_reason": "PR #46152 is the best cluster representative because it captures the more general refactor pattern and the core change in the shared code-reading approach, while PR #46207 is a specific follow-up in a different subsystem.", "soft_edge_verdicts": [ { "left": "pull_request:46152", "right": "pull_request:46207", "accept": false, - "reason": "Same broad idea (avoid `open()` for source inspection), but different modules, different functions, and different failure modes. They do not appear to fix the same concrete bug or to be mergeable into one PR." + "reason": "Same high-level goal, but different concrete functions and files with different implementation paths; they do not look mergeable into one PR or like the same underlying bug." } ] }, @@ -24376,12 +24329,12 @@ "soft_pairs": [ "pull_request:46281|pull_request:46890" ], - "prepared_review_unit_hash": "2aa5eacadff400bc00e77411ff37b2ea76ce59f9da4d6f45ed1f03dbbf294195", + "prepared_review_unit_hash": "983f6eb82adc9cde687cb700fc7cc81aac57f8d404c3e3451bf20c605bd6dbeb", "component_budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 4270, + "serialized_chars": 4271, "estimated_input_tokens": 1068, "estimated_eval_tokens": 2392 }, @@ -24389,7 +24342,7 @@ "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 4270, + "serialized_chars": 4271, "estimated_input_tokens": 1068, "estimated_eval_tokens": 2392 }, @@ -24398,18 +24351,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Two open PRs both harden local checkpoint loading against path traversal, but they fix different code paths: sharded checkpoint shard filenames from the index vs the explicit `transformers_weights` filename from config. Similar security theme, not the same underlying change.", + "summary": "Both items are security-focused PRs that harden path handling against traversal, but they target different untrusted inputs and different code paths: sharded checkpoint weight_map vs. transformers_weights config. Similar theme, not the same underlying fix.", "confidence": 0.92, "canonical_issue_reason": null, - "canonical_pr_reason": "PR #46281 is the more concrete, self-contained fix: it adds containment checks directly in the sharded-checkpoint loader and includes a focused test path. The other PR targets a separate config-driven filename path.", + "canonical_pr_reason": "PR #46281 is the more concrete and self-contained fix: it addresses a specific loader path traversal in sharded-checkpoint resolution from the index weight_map, with a clear containment check and targeted test coverage.", "best_issue_reason": null, - "best_pr_reason": "PR #46890 is also a solid security fix, but it addresses a different resolver path. It is not a better duplicate canonical than #46281; it is simply another related hardening patch.", + "best_pr_reason": "PR #46281 is the best representative of the cluster because it has the clearest, most explicit bug surface and a direct fix for a specific code path; #46890 is related security hardening but for a distinct input path.", "soft_edge_verdicts": [ { "left": "pull_request:46281", "right": "pull_request:46890", "accept": false, - "reason": "Both are path-traversal mitigations, but they operate on different inputs and different loading code paths. They are related security fixes, not the same concrete bug, and would not naturally collapse into one PR." + "reason": "Both are path-traversal mitigations, but they fix different sources of untrusted filenames in different loaders. They do not appear to be the same bug or a single change that could plausibly be merged into one PR." } ] }, @@ -24439,12 +24392,12 @@ "soft_pairs": [ "pull_request:46310|pull_request:46546" ], - "prepared_review_unit_hash": "d1af469877bbdc19aba1f89d4662e3be134d03b4846c309693dfda0602678a46", + "prepared_review_unit_hash": "16bbd7e8b1dac8d6148a7dec9a38a54f938fda6e5d43dc199242550340da243e", "component_budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 4850, + "serialized_chars": 4851, "estimated_input_tokens": 1213, "estimated_eval_tokens": 2682 }, @@ -24452,7 +24405,7 @@ "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 4850, + "serialized_chars": 4851, "estimated_input_tokens": 1213, "estimated_eval_tokens": 2682 }, @@ -24461,18 +24414,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Both PRs fix the same off-by-one error in the streaming sliding-window mask right boundary (`-dist <= right_window_size` vs `<`). PR #46546 is the stronger representative because it covers the same fix in Moonshine and adds tests, though it also includes a parallel Gemma4 adjustment.", - "confidence": 0.95, + "summary": "Both PRs fix the same off-by-one in the right side of the sliding-window mask (`-dist <= right_window_size` instead of `<`); PR 46546 is the more complete change because it also updates Gemma4 and adds tests.", + "confidence": 0.93, "canonical_issue_reason": null, - "canonical_pr_reason": "PR #46546 is the best canonical PR because it addresses the same Moonshine streaming right-window boundary bug as #46310 and includes test coverage, making it the more complete representative of the fix.", + "canonical_pr_reason": "PR 46546 is the best canonical PR because it contains the full fix across all affected implementations and includes test coverage, while PR 46310 is a narrower partial fix.", "best_issue_reason": null, - "best_pr_reason": "PR #46546 is the best overall PR in this cluster: same underlying boundary bug, broader coverage across both Moonshine streaming variants, and tests to validate the behavior.", + "best_pr_reason": "PR 46546 is the strongest representative of the change: broader coverage, same core code fix, and explicit tests make it the more complete/merge-ready patch.", "soft_edge_verdicts": [ { "left": "pull_request:46310", "right": "pull_request:46546", "accept": true, - "reason": "Same concrete bug: an off-by-one in the streaming right-window mask boundary for Moonshine. The diffs change the same condition and target the same issue, so they are duplicates/mergeable as one fix." + "reason": "Same underlying boundary bug in the sliding-window right mask, with identical fix in the shared Moonshine streaming code path; 46546 is a superset rather than a different change." } ] }, @@ -24511,12 +24464,12 @@ "pull_request:46425|pull_request:46850", "pull_request:46850|pull_request:46917" ], - "prepared_review_unit_hash": "8570e39809da7d77741095cd2c8df7fe1e669bc200f10cb811937cfa2e22da97", + "prepared_review_unit_hash": "3c005043df5839de116ea965e0c35267380c4722ace45042be3c834038934941", "component_budget": { "node_count": 6, "item_count": 6, "soft_pair_count": 6, - "serialized_chars": 16839, + "serialized_chars": 16838, "estimated_input_tokens": 4210, "estimated_eval_tokens": 8676 }, @@ -24524,7 +24477,7 @@ "node_count": 6, "item_count": 6, "soft_pair_count": 6, - "serialized_chars": 16839, + "serialized_chars": 16838, "estimated_input_tokens": 4210, "estimated_eval_tokens": 8676 }, @@ -24533,48 +24486,48 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "This cluster is heterogeneous: multimodal continuous batching, two new model integrations, a Gemma masking bugfix, and a deprecation cleanup. None look like the same underlying change.", - "confidence": 0.98, + "summary": "These PRs are all distinct: three separate model additions, one Gemma sliding-window masking bugfix, one continuous-batching multimodal feature, and one deprecation cleanup. The soft links are broad-topic similarity only, not duplicate code changes.", + "confidence": 0.97, "canonical_issue_reason": null, - "canonical_pr_reason": null, + "canonical_pr_reason": "PR 46425 is the best anchor for the cluster because it is the most central and broadly model-integration oriented PR, but it still addresses a different model family than the others.", "best_issue_reason": null, - "best_pr_reason": null, + "best_pr_reason": "PR 46425 is the best overall representative of the cluster\u2019s loose similarity pattern, since it is the hub with the most soft connections and the broadest feature-addition scope.", "soft_edge_verdicts": [ { "left": "pull_request:46376", "right": "pull_request:46417", "accept": false, - "reason": "Continuous batching multimodal support and HunYuan VL model integration are different feature areas with different code paths." + "reason": "Different features for different model families: continuous batching multimodal support vs. HunYuan VL model support." }, { "left": "pull_request:46376", "right": "pull_request:46425", "accept": false, - "reason": "Multimodal continuous batching changes are unrelated to adding RADIO model support." + "reason": "Unrelated PRs: continuous batching multimodal support is a runtime feature change; RADIO is a model implementation/docs addition." }, { "left": "pull_request:46425", "right": "pull_request:46509", "accept": false, - "reason": "RADIO and Nemotron Omni are different model additions; they do not appear to fix or implement the same change." + "reason": "Both add model support, but for different architectures and code paths (RADIO vs. Nemotron Omni), so they are not the same change." }, { "left": "pull_request:46425", "right": "pull_request:46917", "accept": false, - "reason": "New model support is unrelated to removing deprecated image/training args APIs." + "reason": "RADIO model support and deprecated-args cleanup are unrelated changes touching different subsystems." }, { "left": "pull_request:46425", "right": "pull_request:46850", "accept": false, - "reason": "RADIO model support and the Gemma bidirectional masking fix affect different models and code paths." + "reason": "Model addition vs. Gemma masking bugfix; they do not fix the same concrete problem or change the same code path." }, { "left": "pull_request:46850", "right": "pull_request:46917", "accept": false, - "reason": "Gemma masking logic and deprecated-args cleanup are unrelated changes." + "reason": "Gemma attention-mask boundary fix and deprecated training/image-processor args cleanup are unrelated." } ] }, @@ -24604,7 +24557,7 @@ "soft_pairs": [ "pull_request:46574|pull_request:46580" ], - "prepared_review_unit_hash": "ba31b287190b92a5b3f3bad397173be41e9de548c77def29c798a7c63747c5d6", + "prepared_review_unit_hash": "916d065f7e80368b21f4ac9eb2733ff15fc216c2250c5735967cb41a42c7268c", "component_budget": { "node_count": 2, "item_count": 2, @@ -24626,18 +24579,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "The two PRs are unrelated: one fixes CodeLlama decode behavior stripping real leading spaces, the other fixes DeepSeek-Coder/DeepSeek-LLM checkpoints loading the wrong tokenizer class. They do not look like the same underlying bug or mergeable change.", - "confidence": 0.98, + "summary": "Two unrelated open PRs: one fixes CodeLlama decode stripping of real leading spaces, the other fixes DeepSeek-Coder/DeepSeek-LLM auto-tokenizer class selection. They should not be treated as duplicates.", + "confidence": 0.99, "canonical_issue_reason": null, - "canonical_pr_reason": "No canonical PR: the PRs target different models and different failure modes, so they are not duplicates.", + "canonical_pr_reason": "No canonical PR: the PRs fix different bugs in different tokenizer pathways (CodeLlama decode behavior vs. auto-tokenizer checkpoint routing).", "best_issue_reason": null, - "best_pr_reason": "No single best PR for the cluster; 46574 is a self-contained tokenizer decoding fix, while 46580 is a separate tokenizer-class routing fix for different checkpoints.", + "best_pr_reason": "46580 is the broader tokenizer-routing fix, but it is still unrelated to 46574 and should not be merged with it as a duplicate.", "soft_edge_verdicts": [ { "left": "pull_request:46574", "right": "pull_request:46580", "accept": false, - "reason": "Different model families, different code paths, and different bugs (decode stripping vs wrong tokenizer-class selection); not the same underlying change." + "reason": "Reject: 46574 changes CodeLlama decoder setup to preserve leading spaces on decode, while 46580 changes auto-tokenizer checkpoint-prefix logic for DeepSeek models. Different model families, files, and code paths." } ] }, @@ -24667,7 +24620,7 @@ "soft_pairs": [ "pull_request:46603|pull_request:46604" ], - "prepared_review_unit_hash": "0de44052700f005d1335fd5456ff3143988f02bf5ba70acbdff9a87e2e3afb38", + "prepared_review_unit_hash": "566a339282b9105a15a9cb0ba86dc86ac6750ee494b66b7c69a371fff3c21513", "component_budget": { "node_count": 2, "item_count": 2, @@ -24689,18 +24642,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These two open Mistral PRs touch the same area and files, but they solve different problems: one refactors/moves the existing Mistral converter into an integrations package, while the other adds tekken tokenizer detection/conversion/save support. They are related, but not duplicates and should not be merged into one PR.", - "confidence": 0.95, + "summary": "Two open Mistral PRs share files but address different changes: one refactors/moves the Mistral converter into an integrations package, while the other adds tekken tokenizer detection/conversion/save utilities. They are related maintenance around the same subsystem, but not the same concrete change.", + "confidence": 0.97, "canonical_issue_reason": null, - "canonical_pr_reason": null, + "canonical_pr_reason": "PR 46604 is the broader functional change focused on tokenizer detection/conversion/save support; PR 46603 is mainly a code organization refactor.", "best_issue_reason": null, - "best_pr_reason": "PR #46604 is the more substantive, user-facing Mistral change; it adds tekken tokenizer support rather than just relocating existing converter code.", + "best_pr_reason": "PR 46604 is the best representative of the cluster because it introduces the main user-facing tokenizer capability change, whereas 46603 is an internal move/refactor.", "soft_edge_verdicts": [ { "left": "pull_request:46603", "right": "pull_request:46604", "accept": false, - "reason": "Shared Mistral files are not enough: #46603 is a code move/refactor for MistralConverter, while #46604 adds tekken tokenizer detection, conversion, and save utilities. Different concrete changes and not the same bug/fix." + "reason": "Different underlying changes: 46603 relocates MistralConverter into the integrations package, while 46604 implements tekken tokenizer detection/conversion/save utilities. Shared files/subsystem are not enough to merge them as duplicates." } ] }, @@ -24732,7 +24685,7 @@ "pull_request:46741|pull_request:46773", "pull_request:46741|pull_request:46781" ], - "prepared_review_unit_hash": "00ffd92a0cbca1f26535ab8e027699c331bd072b18c49591db2ed9fa8cd17286", + "prepared_review_unit_hash": "f601a5a2ccc1087797963cafe9e386982baa67b5ba775adcb1c9c7da68eb7a55", "component_budget": { "node_count": 3, "item_count": 3, @@ -24754,24 +24707,24 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These PRs are broadly related to model/cache behavior, but they address different concrete bugs and should not be merged as duplicates. The most central representative is the multi-model Mamba2 chunked-prefill/speculative-decoding fix.", - "confidence": 0.89, + "summary": "These are three distinct PRs around linear-attention/cache behavior, but they fix different bugs: one for chunked-prefill/speculative decoding state reuse, one for padding-mask application at batch size 1, and one for QuantizedCache handling on hybrid models. None are true duplicates of each other.", + "confidence": 0.95, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 46741 is the broadest and most central item: it fixes a concrete generation/cache-path bug across several Mamba2-like model implementations and has the most discussion/activity, making it the best representative of the cluster.", + "canonical_pr_reason": "PR #46741 is the best canonical PR because it is the broadest concrete fix in the shared model-cache area and the title/diff clearly center on a single bug family: incorrect Mamba2 chunked-prefill/speculative-decoding behavior across several affected models.", "best_issue_reason": null, - "best_pr_reason": "PR 46741 is the strongest standalone representative because it targets a specific runtime correctness issue, spans the widest relevant set of model files, and appears to be the main cross-model fix rather than a narrow edge-case patch.", + "best_pr_reason": "PR #46741 is the strongest representative of the cluster since it addresses a specific forward/cache-path bug with clear model-level impact and the most substantial shared code changes.", "soft_edge_verdicts": [ { "left": "pull_request:46741", "right": "pull_request:46773", "accept": false, - "reason": "Both touch linear-attention/model files, but 46741 fixes chunked-prefill/speculative decoding cache reuse, while 46773 fixes padding mask application when batch size is 1. Different code paths and failure modes." + "reason": "Both touch similar linear-attention model files, but the bugs are different: #46741 changes cache/state reuse for chunked-prefill/speculative decoding, while #46773 fixes padding-mask application being skipped for batch size 1. Different code paths and failure modes." }, { "left": "pull_request:46741", "right": "pull_request:46781", "accept": false, - "reason": "46741 is a model-forward/cache-flow fix in several model implementations; 46781 fixes QuantizedCache construction for hybrid SSM+attention models in cache_utils. Different underlying bugs and not mergeable into one PR." + "reason": "#46741 is a model forward/cache bug fix across several architectures, whereas #46781 fixes QuantizedCache construction for hybrid SSM+attention models in cache_utils. They are related to caches broadly but not the same underlying change." } ] }, @@ -24804,12 +24757,12 @@ "pull_request:46895|pull_request:46916", "pull_request:46898|pull_request:46916" ], - "prepared_review_unit_hash": "e0c9cde16c54493b4df1e3c593f9ca2c5d3239cd13eb117cf530fa7a6d8e4aab", + "prepared_review_unit_hash": "1b4af156e3127d63c0efe87c8cbbe60194522d25ba57d9e4c093528e2c16cb0d", "component_budget": { "node_count": 3, "item_count": 3, "soft_pair_count": 3, - "serialized_chars": 7714, + "serialized_chars": 7713, "estimated_input_tokens": 1929, "estimated_eval_tokens": 4114 }, @@ -24817,7 +24770,7 @@ "node_count": 3, "item_count": 3, "soft_pair_count": 3, - "serialized_chars": 7714, + "serialized_chars": 7713, "estimated_input_tokens": 1929, "estimated_eval_tokens": 4114 }, @@ -24826,30 +24779,30 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "All three PRs address the same training-loss double-shift bug pattern in seq2seq ASR/VLM forwards: labels are shifted for decoder inputs, then the loss path must avoid shifting again. PR 46916 is the broadest representative because it covers both CohereASR and Florence2, effectively subsuming the model-specific fixes in 46895 and 46898.", - "confidence": 0.88, + "summary": "Two of the PRs are model-specific variants of the same training-loss double-shift fix, while the third is a bundled PR that contains both fixes. The only clear non-duplicate pair is CohereASR vs Florence2.", + "confidence": 0.76, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 46916 is the best canonical representative because it implements the same double-shift loss fix across both affected model families, while the other two are narrower model-specific subsets of that same change.", + "canonical_pr_reason": "PR #46916 is the broadest representative: it contains the same concrete loss/shifting fix for CohereASR and Florence2, so it best captures the cluster\u2019s combined change.", "best_issue_reason": null, - "best_pr_reason": "46916 is the most complete and reusable patch for the shared bug pattern, making it the strongest single PR to anchor the cluster.", + "best_pr_reason": "PR #46916 is the best single representative because it subsumes the narrower CohereASR-only and Florence2-only fixes and points to both affected code paths.", "soft_edge_verdicts": [ { "left": "pull_request:46895", "right": "pull_request:46898", - "accept": true, - "reason": "Both PRs fix the same concrete bug class: training loss was being computed after labels were already shifted to build decoder inputs. They affect different model implementations, but the underlying change is the same and could plausibly live in one combined PR." + "accept": false, + "reason": "Different models and different issue targets; they share the symptom pattern but not the same concrete code path or change." }, { "left": "pull_request:46895", "right": "pull_request:46916", "accept": true, - "reason": "46916 includes the CohereASR double-shift fix from 46895, and adds the same style of fix for Florence2. This is a true overlap on the same code-path problem, with 46916 being a superset." + "reason": "46916 includes the same CohereASR training-loss double-shift fix in the same files; it is a superset PR, but this edge still matches the same underlying CohereASR bug/change." }, { "left": "pull_request:46898", "right": "pull_request:46916", "accept": true, - "reason": "46916 includes the Florence2 double-shift fix from 46898, and also adds the CohereASR counterpart. Same underlying bug, with 46916 subsuming the narrower PR." + "reason": "46916 includes the same Florence2 training-loss double-shift fix in the same files; despite bundling CohereASR too, this is the same underlying Florence2 bug/change." } ] }, @@ -24871,48 +24824,114 @@ "source": "llm", "cache_hit": false, "model": "gpt-5.4-mini?service_tier=flex", - "cluster_id": "cluster-46907-2", + "cluster_id": "cluster-46907-5", "nodes": [ "pull_request:46907", - "pull_request:46908" + "pull_request:46908", + "pull_request:46926", + "pull_request:46927", + "pull_request:46928" ], "soft_pairs": [ - "pull_request:46907|pull_request:46908" + "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": "d821ab6b5916835b08e631de8728bc8195978e6c90b1ccd40df366aa38bdee7d", + "prepared_review_unit_hash": "c668f32b49e267942925b11311db62ebd69e17449d695d494e85c96d8fa14584", "component_budget": { - "node_count": 2, - "item_count": 2, - "soft_pair_count": 1, - "serialized_chars": 4383, - "estimated_input_tokens": 1096, - "estimated_eval_tokens": 2448 + "node_count": 5, + "item_count": 5, + "soft_pair_count": 10, + "serialized_chars": 11925, + "estimated_input_tokens": 2982, + "estimated_eval_tokens": 6220 }, "budget": { - "node_count": 2, - "item_count": 2, - "soft_pair_count": 1, - "serialized_chars": 4383, - "estimated_input_tokens": 1096, - "estimated_eval_tokens": 2448 + "node_count": 5, + "item_count": 5, + "soft_pair_count": 10, + "serialized_chars": 11925, + "estimated_input_tokens": 2982, + "estimated_eval_tokens": 6220 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Two unrelated pull requests: one fixes EdgeTam\u2019s model forward path to ignore a processor-only kwarg, and the other patches DeepSeek-V3.2 integration tests to provide an offload folder during `from_pretrained`. They do not share the same underlying bug or code path.", - "confidence": 0.98, + "summary": "These PRs are all model-specific fixes with different root causes: one real code-path bug in EdgeTam, two offload/load-test workarounds for different models, one dtype-cast fix in GLM46V, and one tokenizer registry update for BLIP-2. They are not duplicates of the same underlying change.", + "confidence": 0.95, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 46907 is the better canonical representative because it contains the actual runtime code fix in the model implementation; PR 46908 is a test-only accommodation for a separate loading issue.", + "canonical_pr_reason": "PR 46907 is the strongest representative because it fixes an actual runtime bug in the model forward path, while the others are unrelated model-specific test or registration fixes.", "best_issue_reason": null, - "best_pr_reason": "PR 46907 is the stronger exemplar of a concrete product bug fix, while PR 46908 is narrowly scoped to test setup for a different model.", + "best_pr_reason": "PR 46907 is the best single PR to represent the cluster since it contains a concrete code fix rather than a test-only workaround and is clearly distinct from the other PRs.", "soft_edge_verdicts": [ + { + "left": "pull_request:46907", + "right": "pull_request:46927", + "accept": false, + "reason": "Different bugs in different models: EdgeTam drops processor-only kwargs, while GLM46V casts input_ids to long." + }, + { + "left": "pull_request:46907", + "right": "pull_request:46926", + "accept": false, + "reason": "EdgeTam forward-path fix vs MiniMax-M3-VL test offload_folder workaround; not the same underlying issue." + }, { "left": "pull_request:46907", "right": "pull_request:46908", "accept": false, - "reason": "Different models, different failures, and different fixes: EdgeTam drops `original_sizes` before backbone forwarding, while DeepSeek-V3.2 tests add `offload_folder` for `device_map=\"auto\"`. They are not the same underlying bug and would not plausibly merge into one PR." + "reason": "Different models and fixes: EdgeTam code-path cleanup vs DeepSeek-V3.2 test offload_folder workaround." + }, + { + "left": "pull_request:46907", + "right": "pull_request:46928", + "accept": false, + "reason": "EdgeTam runtime fix vs BLIP-2 tokenizer auto-registration update; unrelated changes." + }, + { + "left": "pull_request:46908", + "right": "pull_request:46926", + "accept": false, + "reason": "Both are load-related test changes, but they target different models and different load failures, so they are not the same bug." + }, + { + "left": "pull_request:46908", + "right": "pull_request:46927", + "accept": false, + "reason": "DeepSeek test offload workaround and GLM46V input dtype fix are unrelated." + }, + { + "left": "pull_request:46926", + "right": "pull_request:46927", + "accept": false, + "reason": "MiniMax offload_folder test workaround vs GLM46V forward input cast; different code paths and failures." + }, + { + "left": "pull_request:46926", + "right": "pull_request:46928", + "accept": false, + "reason": "MiniMax load workaround and BLIP-2 tokenizer registry change are not the same underlying change." + }, + { + "left": "pull_request:46927", + "right": "pull_request:46928", + "accept": false, + "reason": "GLM46V input_ids casting fix vs BLIP-2 tokenizer registration update; unrelated model fixes." + }, + { + "left": "pull_request:46908", + "right": "pull_request:46928", + "accept": false, + "reason": "DeepSeek load_error workaround vs BLIP-2 tokenizer registry update; different models and different bugs." } ] }, diff --git a/analysis/current/manifest.json b/analysis/current/manifest.json index 0eab05d600bb5b0aad15080cc2bc87c518b3b6cb..5990750f80d9a49d0b8927a755e1df97a59f12d0 100644 --- a/analysis/current/manifest.json +++ b/analysis/current/manifest.json @@ -1,8 +1,8 @@ { - "analysis_id": "hybrid-model-20260627t000016z", + "analysis_id": "hybrid-model-20260627t060016z", "archived_artifacts": { - "hybrid": "snapshots/20260627T000016Z/analysis-runs/hybrid-model-20260627t000016z/analysis-report-hybrid.json", - "hybrid_reviews": "snapshots/20260627T000016Z/analysis-runs/hybrid-model-20260627t000016z/analysis-report-hybrid.llm-reviews.json" + "hybrid": "snapshots/20260627T060016Z/analysis-runs/hybrid-model-20260627t060016z/analysis-report-hybrid.json", + "hybrid_reviews": "snapshots/20260627T060016Z/analysis-runs/hybrid-model-20260627t060016z/analysis-report-hybrid.llm-reviews.json" }, "artifacts": { "hybrid": "analysis/current/analysis-report-hybrid.json", @@ -10,9 +10,9 @@ }, "channel": "canonical", "model": null, - "published_at": "2026-06-27T00:09:46Z", + "published_at": "2026-06-27T06:09:30Z", "repo": "huggingface/transformers", "schema_version": 1, - "snapshot_id": "20260627T000016Z", + "snapshot_id": "20260627T060016Z", "variant": "hybrid" } diff --git a/snapshots/20260627T060016Z/analysis-runs/hybrid-model-20260627t060016z/analysis-report-hybrid.json b/snapshots/20260627T060016Z/analysis-runs/hybrid-model-20260627t060016z/analysis-report-hybrid.json new file mode 100644 index 0000000000000000000000000000000000000000..077f43bc950eec8e8d92d54cd29fb1c8c89dd3e7 --- /dev/null +++ b/snapshots/20260627T060016Z/analysis-runs/hybrid-model-20260627t060016z/analysis-report-hybrid.json @@ -0,0 +1,2495 @@ +{ + "schema_version": "1.0", + "repo": "huggingface/transformers", + "snapshot_id": "20260627T060016Z", + "generated_at": "2026-06-27T06:09:28Z", + "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-46895-4", + "summary": "Cluster of 1 issues and 3 PRs centered on issue #46897.", + "status": "open", + "confidence": 0.8, + "canonical_issue_number": 46897, + "canonical_pr_number": 46895, + "issue_numbers": [ + 46897 + ], + "pr_numbers": [ + 46895, + 46898, + 46916 + ], + "evidence_types": [ + "closing_reference", + "shared_issue_target", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 46895, + "right_pr_number": 46898, + "code_similarity": 0.121, + "size_similarity": 0.483, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.164, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 46895, + "right_pr_number": 46916, + "code_similarity": 0.341, + "size_similarity": 0.644, + "file_overlap": 0.4, + "area_overlap": 0.194, + "patch_similarity": 0.162, + "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-44002-2", + "summary": "Cluster of 2 related pull requests linked by soft_similarity.", + "status": "open", + "confidence": 0.5, + "canonical_issue_number": null, + "canonical_pr_number": 44002, + "issue_numbers": [], + "pr_numbers": [ + 44002, + 44029 + ], + "evidence_types": [ + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 44002, + "right_pr_number": 44029, + "code_similarity": 0.135, + "size_similarity": 0.43, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.329, + "shared_filenames": [], + "shared_file_areas": [] + } + ] + }, + { + "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": [], + "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-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-44002-2", + "canonical_pr_number": 44002, + "duplicate_pr_numbers": [ + 44029 + ], + "target_issue_number": null, + "reason": "PRs in cluster-44002-2 are treated as duplicates because they share 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-46895-4", + "canonical_pr_number": 46895, + "duplicate_pr_numbers": [ + 46898, + 46916 + ], + "target_issue_number": 46897, + "reason": "PRs in cluster-46895-4 are treated as duplicates because they converge on issue #46897 with closing_reference, shared_issue_target, soft_similarity evidence." + } + ], + "best_issue": { + "cluster_id": "cluster-46895-4", + "issue_number": 46897, + "reason": "Issue #46897 is the strongest global issue candidate because it is open, belongs to a cluster with 4 artifacts, and carries 0 discussion comments plus 9 inbound references.", + "score": 149.0 + }, + "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/20260627T060016Z/analysis-runs/hybrid-model-20260627t060016z/analysis-report-hybrid.llm-reviews.json b/snapshots/20260627T060016Z/analysis-runs/hybrid-model-20260627t060016z/analysis-report-hybrid.llm-reviews.json new file mode 100644 index 0000000000000000000000000000000000000000..ce9680039f87f860885d47014df122c63b564335 --- /dev/null +++ b/snapshots/20260627T060016Z/analysis-runs/hybrid-model-20260627t060016z/analysis-report-hybrid.llm-reviews.json @@ -0,0 +1,24946 @@ +{ + "schema_version": "1.0", + "repo": "huggingface/transformers", + "snapshot_id": "20260627T060016Z", + "generated_at": "2026-06-27T06:09:28Z", + "model": "gpt-5.4-mini?service_tier=flex", + "reviews": [ + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 1, + "review_unit_count": 60, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-500", + "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:44560", + "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:44560|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:43541|issue:43716", + "issue:43828|issue:44292", + "issue:43716|issue:46468", + "issue:44292|issue:44918", + "issue:45507|issue:45685", + "issue:43716|issue:44292", + "issue:44918|issue:45161", + "issue:43716|issue:45685", + "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": "f0d510929ef519664e5f4f6f3903a8a98b16db6eda79a20cf63f2b02096456ee", + "component_budget": { + "node_count": 500, + "item_count": 500, + "soft_pair_count": 1839, + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 63, + "serialized_chars": 46573, + "estimated_input_tokens": 11644, + "estimated_eval_tokens": 23544 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a mix of unrelated bug reports. Only two soft pairs look like true duplicates: the ZeRO-3 BertModel index-out-of-bounds reports, and the torch 2.4.1 compatibility reports.", + "confidence": 0.93, + "canonical_issue_reason": "Issue 43596 is the clearest canonical for the Bert/ZeRO-3 duplicate pair: it describes the concrete index-out-of-bounds failure on a standard BertModel, and 43638 is just the non-pretrained variant of the same bug.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 43596 is the best single issue in the set to anchor triage because it has the strongest true duplicate match and a concrete, reusable failure signature.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44322", + "right": "issue:45464", + "accept": false, + "reason": "Different bugs: Qwen3.5 config attribute missing vs streaming chat/completions failure." + }, + { + "left": "issue:45237", + "right": "issue:45685", + "accept": false, + "reason": "Unrelated failures: AMD GPU support for GPT-OSS vs an MPS histogram op error." + }, + { + "left": "issue:43716", + "right": "issue:45237", + "accept": false, + "reason": "Different model/runtime issues: Mistral-3 image dtype mismatch vs GPT-OSS on AMD GPUs." + }, + { + "left": "issue:43828", + "right": "issue:45237", + "accept": false, + "reason": "Different bugs: autocast dtype mismatch in Phi-tiny-MoE vs GPT-OSS AMD GPU failure." + }, + { + "left": "issue:45237", + "right": "issue:46468", + "accept": false, + "reason": "Different models and symptoms; no shared concrete code path is evident." + }, + { + "left": "issue:45237", + "right": "issue:45507", + "accept": false, + "reason": "Different issues: GPT-OSS AMD GPU support vs GraniteMoEHybrid invalid method call." + }, + { + "left": "issue:30277", + "right": "issue:35545", + "accept": false, + "reason": "Different failures: DeepSpeed ZeRO-3 collective mismatch vs ONNX export error." + }, + { + "left": "issue:35545", + "right": "issue:42915", + "accept": false, + "reason": "Different problems: ONNX export of ModernBERT vs Qwen3Moe FP8 failure." + }, + { + "left": "issue:35545", + "right": "issue:41720", + "accept": false, + "reason": "Different code paths: export/tracing bug vs Qwen3 auto device-map cuda assert." + }, + { + "left": "issue:45901", + "right": "issue:46015", + "accept": false, + "reason": "Different topics: table QA crash vs outdated RAG docs." + }, + { + "left": "issue:44509", + "right": "issue:44512", + "accept": false, + "reason": "Both are v5 docs cleanup, but they remove different obsolete references in different documentation areas." + }, + { + "left": "issue:35545", + "right": "issue:36010", + "accept": false, + "reason": "Different bugs: ONNX export error vs GenerationMixin import regression." + }, + { + "left": "issue:44918", + "right": "issue:46165", + "accept": false, + "reason": "Different training failures: Qwen3.5 embedding unpacking vs resume-training with trust_remote_code." + }, + { + "left": "issue:46165", + "right": "issue:46869", + "accept": false, + "reason": "Unrelated model/runtime problems; no common underlying bug is clear." + }, + { + "left": "issue:44805", + "right": "issue:45823", + "accept": false, + "reason": "Different symptoms and subsystems: mask shape/indexing mismatch vs Gemma4 device-map PLE mismatch." + }, + { + "left": "issue:43638", + "right": "issue:44805", + "accept": false, + "reason": "Both mention tensor/index errors, but the failures are different and not the same code path." + }, + { + "left": "issue:43638", + "right": "issue:46165", + "accept": false, + "reason": "Different issues: Bert ZeRO-3 index error vs resume-training with local model and remote code." + }, + { + "left": "issue:43596", + "right": "issue:44805", + "accept": false, + "reason": "Different failures: Bert ZeRO-3 loading bug vs mask/tensor indexing mismatch." + }, + { + "left": "issue:44805", + "right": "issue:46294", + "accept": false, + "reason": "Different modalities and errors: mask shape mismatch vs video token/feature mismatch." + }, + { + "left": "issue:44805", + "right": "issue:45161", + "accept": false, + "reason": "Different problems: tensor mask indexing error vs GPT-OSS MoE tensor-parallel issue." + }, + { + "left": "issue:46165", + "right": "issue:46468", + "accept": false, + "reason": "Different bugs with different models and call paths." + }, + { + "left": "issue:44560", + "right": "issue:46165", + "accept": false, + "reason": "Different issues: Qwen3-vl video StopIteration vs training resume with remote code." + }, + { + "left": "issue:34634", + "right": "issue:35545", + "accept": false, + "reason": "Different components: BarkProcessor voice_preset vs ModernBERT ONNX export." + }, + { + "left": "issue:30277", + "right": "issue:43638", + "accept": false, + "reason": "Both involve DeepSpeed, but one is a collective mismatch and the other is a Bert ZeRO-3 index error." + }, + { + "left": "issue:43596", + "right": "issue:43638", + "accept": true, + "reason": "Same underlying bug: both are DeepSpeed ZeRO-3 BertModel loads failing with index 0 out of bounds; only the pretrained status differs." + }, + { + "left": "issue:43039", + "right": "issue:45161", + "accept": false, + "reason": "Different failures: Liger Kernel cross-entropy still being called vs GPT-OSS MoE tensor-parallel behavior." + }, + { + "left": "issue:43975", + "right": "issue:44457", + "accept": false, + "reason": "Different bugs: DeepSeek-Coder detokenization vs LoRA merge/save/reload output mismatch." + }, + { + "left": "issue:44292", + "right": "issue:44457", + "accept": false, + "reason": "Different problems: Qwen-3-8B-NVFP4 runtime error vs LoRA merge/reload inconsistency." + }, + { + "left": "issue:44841", + "right": "issue:46661", + "accept": false, + "reason": "Different issues: processor failure for Voxtral vs remote-code config_class AttributeError." + }, + { + "left": "issue:45859", + "right": "issue:45902", + "accept": false, + "reason": "Same class name, but different bugs: CPU-offload forward compatibility vs uninitialized inv_freq after meta materialization." + }, + { + "left": "issue:45685", + "right": "issue:46869", + "accept": false, + "reason": "Different hardware/runtime problems: MPS Int histogram op vs YOLO multi-GPU device-map auto failure." + }, + { + "left": "issue:43638", + "right": "issue:45161", + "accept": false, + "reason": "Different bugs: Bert ZeRO-3 index error vs GPT-OSS MoE tensor-parallel issue." + }, + { + "left": "issue:34928", + "right": "issue:36331", + "accept": false, + "reason": "Different training failures: activation checkpointing/FSDP tensor recomputation vs CustomTrainer compute_loss signature mismatch." + }, + { + "left": "issue:43596", + "right": "issue:45161", + "accept": false, + "reason": "Different bugs and models; no shared concrete failure mode." + }, + { + "left": "issue:43716", + "right": "issue:45507", + "accept": false, + "reason": "Different model-specific issues: Mistral-3 dtype mismatch vs GraniteMoEHybrid invalid method." + }, + { + "left": "issue:44292", + "right": "issue:45237", + "accept": false, + "reason": "Different model/runtime failures; not the same underlying bug." + }, + { + "left": "issue:45874", + "right": "issue:46294", + "accept": false, + "reason": "Different features: reverse embedding lookup on inputs_embeds vs video chat-template feature/token mismatch." + }, + { + "left": "issue:39290", + "right": "issue:45081", + "accept": false, + "reason": "Different breakages: Gemma3 vLLM config attribute error vs mistral regex patch tokenizer backend error." + }, + { + "left": "issue:45800", + "right": "issue:46515", + "accept": true, + "reason": "Same compatibility issue: both report torch 2.4.1 incompatibility, with no meaningful distinction in the failure itself." + }, + { + "left": "issue:45823", + "right": "issue:45874", + "accept": false, + "reason": "Different Gemma4 issues: PLE device mismatch vs expensive reverse embedding lookup." + }, + { + "left": "issue:43425", + "right": "issue:44292", + "accept": false, + "reason": "Different problems: torch 2.10 compatibility vs Qwen-3-8B-NVFP4 runtime error." + }, + { + "left": "issue:34928", + "right": "issue:41720", + "accept": false, + "reason": "Different issues: FSDP/activation-checkpointing size mismatch vs Qwen3 auto device-map cuda assert." + }, + { + "left": "issue:43828", + "right": "issue:45685", + "accept": false, + "reason": "Different bugs: autocast dtype mismatch vs MPS histogram op not implemented for Int." + }, + { + "left": "issue:43828", + "right": "issue:44292", + "accept": false, + "reason": "Different failures and models; no evidence they share a root cause." + }, + { + "left": "issue:43716", + "right": "issue:46468", + "accept": false, + "reason": "Different model families and symptoms." + }, + { + "left": "issue:44292", + "right": "issue:44918", + "accept": false, + "reason": "Different code paths: Qwen NVFP4 runtime error vs TRL SFT trainer input embedding unpacking." + }, + { + "left": "issue:45507", + "right": "issue:45685", + "accept": false, + "reason": "Different bugs: GraniteMoEHybrid invalid method call vs MPS histogram op unsupported for Int." + }, + { + "left": "issue:43716", + "right": "issue:44292", + "accept": false, + "reason": "Different model/runtime failures; not the same bug." + }, + { + "left": "issue:44918", + "right": "issue:45161", + "accept": false, + "reason": "Different issues: TRL SFT embedding unpacking vs GPT-OSS MoE tensor-parallel behavior." + }, + { + "left": "issue:43716", + "right": "issue:45685", + "accept": false, + "reason": "Different subsystems and failure modes." + }, + { + "left": "issue:43039", + "right": "issue:44918", + "accept": false, + "reason": "Different issues: Liger Kernel cross-entropy invocation vs Qwen3.5 TRL embedding unpacking." + }, + { + "left": "issue:41762", + "right": "issue:43638", + "accept": false, + "reason": "Both are ZeRO-3 index errors, but they affect different models and loading paths; too broad to treat as the same bug." + }, + { + "left": "issue:43827", + "right": "issue:44512", + "accept": false, + "reason": "Different docs removals: pipeline task references vs transformers run command references." + }, + { + "left": "issue:43828", + "right": "issue:45685", + "accept": false, + "reason": "Different bugs: autocast dtype mismatch vs MPS histogram op unsupported for Int." + }, + { + "left": "issue:43039", + "right": "issue:44805", + "accept": false, + "reason": "Different failures: Liger Kernel cross_entropy call vs mask shape/indexing mismatch." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays conservative. The two accepted duplicate pairs are well supported by the issue titles and descriptions, and the rejected soft pairs are phrased cautiously without overclaiming shared root cause. The canonical Bert ZeRO-3 issue choice is also defensible." + }, + "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": 60, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-500", + "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:43901|issue:44512", + "issue:44589|issue:46304", + "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:43404|issue:43866", + "issue:45507|issue:46869", + "issue:41093|issue:41762", + "issue:45399|issue:46661", + "issue:43572|issue:45070", + "issue:45800|issue:45823" + ], + "prepared_review_unit_hash": "d55e3f913db7038362b35e55a5d54edd731dafe62cbcf19ba693a4368b81a49a", + "component_budget": { + "node_count": 500, + "item_count": 500, + "soft_pair_count": 1839, + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 42, + "serialized_chars": 40040, + "estimated_input_tokens": 10010, + "estimated_eval_tokens": 20276 + }, + "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 bugs, with one clear duplicate pair around the v5 AutoTokenizer token-ID regression.", + "confidence": 0.76, + "canonical_issue_reason": "issue:45920 is the broadest and most representative report of the shared tokenizer regression, and it subsumes the Granite-only repro in issue:45812.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:45920 best represents the actual duplicate-worthy topic in the cluster: a v5 AutoTokenizer regression affecting multiple model families.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43866", + "right": "issue:46402", + "accept": false, + "reason": "Different models and failures: Ovis2 checkpoint corruption vs Sam3Model text-encoder weights not loaded." + }, + { + "left": "issue:45800", + "right": "issue:46304", + "accept": false, + "reason": "Both mention compatibility, but one is a generic torch/transformers version mismatch while the other is a USE_DISTRIBUTED=0 breakage." + }, + { + "left": "issue:43296", + "right": "issue:46304", + "accept": false, + "reason": "PaddleOCR-VL loading in vLLM and USE_DISTRIBUTED=0 are unrelated code paths and failure modes." + }, + { + "left": "issue:41762", + "right": "issue:43596", + "accept": false, + "reason": "Both hit a zero3-related IndexError, but the repros are on different model-loading paths and there is not enough evidence they are the same bug." + }, + { + "left": "issue:39290", + "right": "issue:41720", + "accept": false, + "reason": "Gemma3TextConfig missing sliding_window_pattern and Qwen3 auto device-map cuda asserts are different regressions." + }, + { + "left": "issue:43901", + "right": "issue:44512", + "accept": false, + "reason": "Separate documentation issues in different areas, not the same underlying bug or change." + }, + { + "left": "issue:44589", + "right": "issue:46304", + "accept": false, + "reason": "Float8 storage lookup failure is unrelated to the distributed-support regression in transformers v5.9.0." + }, + { + "left": "issue:45237", + "right": "issue:45901", + "accept": false, + "reason": "Different subsystems and symptoms: AMD GPU support for GPT-OSS vs table-question-answering crash." + }, + { + "left": "issue:43975", + "right": "issue:46015", + "accept": false, + "reason": "A DeepSeek detokenization regression and outdated RAG docs are unrelated." + }, + { + "left": "issue:43039", + "right": "issue:44560", + "accept": false, + "reason": "Liger Kernel cross-entropy dispatch and Qwen3.5-vl video StopIteration are different bugs." + }, + { + "left": "issue:45823", + "right": "issue:46294", + "accept": false, + "reason": "Different models and mismatches: Gemma4 PLE device placement vs Qwen3.5 video/chat-template token mismatch." + }, + { + "left": "issue:44841", + "right": "issue:45823", + "accept": false, + "reason": "Voxtral processor failure and Gemma4 device mismatch are separate model-specific problems." + }, + { + "left": "issue:43866", + "right": "issue:46772", + "accept": false, + "reason": "Ovis2 checkpoint corruption is unrelated to DiffusionGemma NVFP4 compatibility." + }, + { + "left": "issue:44863", + "right": "issue:46015", + "accept": false, + "reason": "NemotronH checkpoint loading and RAG docs being outdated are unrelated." + }, + { + "left": "issue:45020", + "right": "issue:46304", + "accept": false, + "reason": "Remote-code model breakage and USE_DISTRIBUTED=0 support regression are not the same underlying issue." + }, + { + "left": "issue:45812", + "right": "issue:45920", + "accept": true, + "reason": "Both report the same v5 AutoTokenizer wrong-token-ID regression; 45920 is the broader multi-model report and 45812 is a narrower Granite repro." + }, + { + "left": "issue:43866", + "right": "issue:44863", + "accept": false, + "reason": "Different model-loading failures with no shared concrete code path." + }, + { + "left": "issue:43872", + "right": "issue:46304", + "accept": false, + "reason": "bitsandbytes Int8Params API incompatibility is unrelated to the distributed-build regression." + }, + { + "left": "issue:44292", + "right": "issue:46015", + "accept": false, + "reason": "Qwen-3-8B-NVFP4 runtime failure and outdated RAG examples are unrelated." + }, + { + "left": "issue:39290", + "right": "issue:43531", + "accept": false, + "reason": "Both mention sliding_window, but one is a Gemma3 config attribute issue and the other is a Qwen3-MoE model-specific sliding-window bug." + }, + { + "left": "issue:41093", + "right": "issue:43638", + "accept": false, + "reason": "The mask-shape mismatch and the zero3 BertModel index error are different failure signatures and likely different bugs." + }, + { + "left": "issue:43792", + "right": "issue:45901", + "accept": false, + "reason": "Whisper-large-v2 runtime failure and table-question-answering crash are unrelated." + }, + { + "left": "issue:43854", + "right": "issue:43866", + "accept": false, + "reason": "GLM-4.7-Flash loading in tests and Ovis2 checkpoint corruption are distinct issues." + }, + { + "left": "issue:43425", + "right": "issue:45070", + "accept": false, + "reason": "Both are version-related, but they affect different compatibility surfaces and are not the same regression." + }, + { + "left": "issue:43975", + "right": "issue:45901", + "accept": false, + "reason": "Detokenization regression and table QA crash are unrelated." + }, + { + "left": "issue:43792", + "right": "issue:45070", + "accept": false, + "reason": "Whisper runtime failure is unrelated to the PretrainedConfig pydantic breakage." + }, + { + "left": "issue:43596", + "right": "issue:45800", + "accept": false, + "reason": "Zero3 BertModel IndexError and torch/transformers incompatibility are not the same bug." + }, + { + "left": "issue:45362", + "right": "issue:45464", + "accept": false, + "reason": "Both touch Qwen3.5 chat flows, but one is a chat-template crash and the other is a streaming API failure; too different to merge as duplicates." + }, + { + "left": "issue:45464", + "right": "issue:45685", + "accept": false, + "reason": "Chat streaming failure and an MPS histogram op issue are unrelated." + }, + { + "left": "issue:36010", + "right": "issue:39290", + "accept": false, + "reason": "GenerationMixin import error and Gemma3 sliding_window attribute error are unrelated." + }, + { + "left": "issue:43541", + "right": "issue:45685", + "accept": false, + "reason": "Mixtral torch-dynamo grouped_mm tracing and the MPS histogram issue are separate bugs." + }, + { + "left": "issue:43541", + "right": "issue:43828", + "accept": false, + "reason": "Both involve model execution, but one is a grouped_mm tracing failure and the other is an autocast dtype mismatch." + }, + { + "left": "issue:44841", + "right": "issue:45084", + "accept": false, + "reason": "Voxtral processor failure and a compile error about non-template nodes are unrelated." + }, + { + "left": "issue:43404", + "right": "issue:43866", + "accept": false, + "reason": "Untied lm_head weights in Mistral3 and Ovis2 checkpoint corruption are different issues." + }, + { + "left": "issue:45507", + "right": "issue:46869", + "accept": false, + "reason": "GraniteMoEHybrid invalid method and YOLO multi-GPU auto device-map failure are unrelated." + }, + { + "left": "issue:41093", + "right": "issue:41762", + "accept": false, + "reason": "Different error shapes and symptoms: mask-size mismatch vs zero3 index-out-of-bounds during Gemma3 loading." + }, + { + "left": "issue:45399", + "right": "issue:46661", + "accept": false, + "reason": "Flash-attn fallback gating and remote-code model_class/config_class handling are unrelated." + }, + { + "left": "issue:43572", + "right": "issue:45070", + "accept": false, + "reason": "StableLm pad_token_idx regression and a PretrainedConfig pydantic breakage are different compatibility bugs." + }, + { + "left": "issue:45800", + "right": "issue:45823", + "accept": false, + "reason": "Torch/transformers version incompatibility and Gemma4 device-mapping mismatch are not the same concrete bug." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is well grounded in the packet: issue:45920 and issue:45812 are the only clearly duplicate-worthy pair, and the rest of the cluster appears heterogeneous. The soft-edge verdicts are conservative overall and do not overclaim evidence. No PR-pair mergeability concerns apply here." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 3, + "review_unit_count": 60, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-500", + "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:45632", + "issue:45698", + "issue:45759", + "issue:45800", + "issue:45902", + "issue:46165", + "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:43596|issue:46165", + "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" + ], + "prepared_review_unit_hash": "6a538354b3f80a6e75e4adc1734b0412bce501176282a4e361eb87891925c4fb", + "component_budget": { + "node_count": 500, + "item_count": 500, + "soft_pair_count": 1839, + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 31, + "serialized_chars": 36887, + "estimated_input_tokens": 9222, + "estimated_eval_tokens": 18700 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly heterogeneous; most soft edges are false positives. The only plausible duplicate groups are around trust_remote_code custom-module loading, Qwen3.5 video-input token accounting, and input validation/error-message wording.", + "confidence": 0.67, + "canonical_issue_reason": "No single issue cleanly represents the whole cluster, but issue 45698 is the strongest anchor among the near-duplicates because it captures the broad custom-module misloading symptom that also covers the leaf-directory collision case.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 45698 is the most actionable and generally worded issue among the candidates, and it plausibly subsumes the closely related 45632 report.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43296", + "right": "issue:43366", + "accept": false, + "reason": "Different model/setup problems: PaddleOCR-VL load failure vs GGUF gpt-oss architecture support." + }, + { + "left": "issue:44387", + "right": "issue:45005", + "accept": false, + "reason": "One is int4 quantization memory/OOM behavior; the other is tied-weights handling for translation models." + }, + { + "left": "issue:44805", + "right": "issue:44918", + "accept": false, + "reason": "Both are runtime errors, but the shapes and code paths differ: mask indexing vs TRL input embedding unpacking." + }, + { + "left": "issue:43541", + "right": "issue:45161", + "accept": false, + "reason": "Both mention MoE/Mixtral/GPT-OSS, but one is torch dynamo grouped_mm tracing and the other is tensor-parallel support." + }, + { + "left": "issue:43828", + "right": "issue:45464", + "accept": false, + "reason": "Different multimodal/dtype failures with different models and failure modes." + }, + { + "left": "issue:43366", + "right": "issue:45464", + "accept": false, + "reason": "Unrelated: GGUF gpt-oss architecture support vs Qwen3.5 streaming inference failure." + }, + { + "left": "issue:44291", + "right": "issue:45070", + "accept": false, + "reason": "Both are v5 regressions, but one is init_empty_weights API breakage and the other is pydantic config-field behavior." + }, + { + "left": "issue:43531", + "right": "issue:45070", + "accept": false, + "reason": "Qwen3-MoE sliding_window behavior is unrelated to PretrainedConfig/pydantic issues." + }, + { + "left": "issue:41093", + "right": "issue:43596", + "accept": false, + "reason": "Similar error type only; the mask-size mismatch and zero3/BertModel index error are different bugs." + }, + { + "left": "issue:45020", + "right": "issue:45800", + "accept": false, + "reason": "Remote-code breakage and torch/transformers version incompatibility are not the same concrete issue." + }, + { + "left": "issue:43716", + "right": "issue:46294", + "accept": false, + "reason": "Both involve multimodal inputs, but one is image-preprocessor dtype mismatch and the other is video-template token accounting." + }, + { + "left": "issue:43054", + "right": "issue:43994", + "accept": false, + "reason": "Both are about Siglip2 quality, but one is text-embedding degradation and the other is incorrect overall outputs/loading behavior." + }, + { + "left": "issue:44479", + "right": "issue:46294", + "accept": true, + "reason": "Both describe the same Qwen3.5 video-input regression where chat-template/video-token accounting mismatches the provided video features." + }, + { + "left": "issue:46468", + "right": "issue:46869", + "accept": false, + "reason": "Different domains: Nemotron example breakage vs YOLO multi-GPU device-map auto failure." + }, + { + "left": "issue:45759", + "right": "issue:45902", + "accept": false, + "reason": "Composite-config unwrapping and uninitialized rotary-embedding memory are separate bugs." + }, + { + "left": "issue:43596", + "right": "issue:46165", + "accept": false, + "reason": "Zero3/BertModel index error and trust_remote_code resume issues are unrelated." + }, + { + "left": "issue:46640", + "right": "issue:46661", + "accept": false, + "reason": "Flash Attention 2 usage and remote-code config_class attribute errors are different code paths." + }, + { + "left": "issue:43716", + "right": "issue:44918", + "accept": false, + "reason": "Both are multimodal-related, but image dtype mismatch is not the same as the TRL embedding-unpacking failure." + }, + { + "left": "issue:39186", + "right": "issue:44918", + "accept": false, + "reason": "FSDP 2-D weight enforcement is unrelated to the Qwen3.5/TRl tensor-shape issue." + }, + { + "left": "issue:40990", + "right": "issue:44960", + "accept": false, + "reason": "High perplexity on gpt-oss and GLM5 are unrelated model-quality reports." + }, + { + "left": "issue:43645", + "right": "issue:45399", + "accept": false, + "reason": "Custom-model notebook initialization and flash-attn2 fallback logic are unrelated." + }, + { + "left": "issue:45632", + "right": "issue:45698", + "accept": true, + "reason": "Both point to the same trust_remote_code custom-module loading/caching bug; 45632 is the specific leaf-directory collision variant of 45698's wrong-module symptom." + }, + { + "left": "issue:43404", + "right": "issue:45005", + "accept": false, + "reason": "Tied lm_head weights in Mistral3 and translation-model tied-weight issues are different model families and failures." + }, + { + "left": "issue:42915", + "right": "issue:45399", + "accept": false, + "reason": "FineGrainedFP8Config failure and flash-attn2 fallback gating are unrelated." + }, + { + "left": "issue:43873", + "right": "issue:45005", + "accept": false, + "reason": "Quantization/offloading behavior is not the same bug as tied-weight handling." + }, + { + "left": "issue:44530", + "right": "issue:45005", + "accept": false, + "reason": "PagedAttentionCache linear_attention crash and tied weights are unrelated." + }, + { + "left": "issue:44360", + "right": "issue:44512", + "accept": false, + "reason": "A DSA indexer note and a docs typo are not the same underlying change." + }, + { + "left": "issue:44492", + "right": "issue:44512", + "accept": false, + "reason": "Cache-strategy typo and removed run-command docs are unrelated documentation issues." + }, + { + "left": "issue:44863", + "right": "issue:45005", + "accept": false, + "reason": "NemotronH checkpoint loading and translation-model tied weights are different bugs." + }, + { + "left": "issue:45183", + "right": "issue:46463", + "accept": true, + "reason": "Both are the same validation-family bug: incorrect XOR/input validation between input_ids and inputs_embeds leading to misleading error messages, with 46463 as a T5-specific manifestation." + }, + { + "left": "issue:46489", + "right": "issue:46491", + "accept": false, + "reason": "Both are tokenizer regressions, but they affect different models and different round-trip behaviors." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is conservative and stays within what the packet supports. The accepted soft edges are plausibly related by the titles, and the rejected pairs are not over-merged. The only potentially close call is 45632/45698, but the reasoning is still cautious enough to pass." + }, + "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": 4, + "review_unit_count": 60, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-500", + "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: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", + "issue:45987", + "issue:46489", + "issue:46858" + ], + "soft_pairs": [ + "issue:43975|issue:44292", + "issue:44918|issue:45507", + "issue:43950|issue:44534", + "issue:39692|issue:43054", + "issue:43644|issue:43950", + "issue:36331|issue:41762", + "issue:43872|issue:44589", + "issue:43828|issue:44928", + "issue:45987|issue:46858", + "issue:45920|issue:46489", + "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:43643|issue:45020", + "issue:41762|issue:42915", + "issue:44991|issue:45399", + "issue:43824|issue:43872", + "issue:43782|issue:44960", + "issue:43881|issue:43883", + "issue:45685|issue:45823" + ], + "prepared_review_unit_hash": "9e1943f5cfb7c064917e63f085e3a01ff3815c954071dc7854a976de8481c9a4", + "component_budget": { + "node_count": 500, + "item_count": 500, + "soft_pair_count": 1839, + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 31, + "serialized_chars": 37030, + "estimated_input_tokens": 9258, + "estimated_eval_tokens": 18772 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster mostly contains unrelated issue reports, but a few soft pairs do look like the same underlying bug: non-persistent buffer corruption in v5, StaticCache.get_seq_length returning a tensor instead of int, tokenizer regression in v5+, and the SigLIP2 implementation discrepancy. The rest are different models or code paths.", + "confidence": 0.78, + "canonical_issue_reason": "issue:43950 is the clearest root-cause regression report: it names the transformers 5.x non-persistent-buffer corruption directly and cleanly subsumes narrower manifestations like issue:44534 and likely issue:45902.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:43950 is the best canonical anchor because it is broad, technically specific, and framed as the regression itself rather than a single model-specific symptom.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43975", + "right": "issue:44292", + "accept": false, + "reason": "Different failure modes and models: DeepSeek-Coder detokenization vs Qwen NVFP4 loading/runtime error." + }, + { + "left": "issue:44918", + "right": "issue:45507", + "accept": false, + "reason": "Different models and code paths; one is Qwen3.5 input-embedding unpacking with TRL, the other GraniteMoEHybrid method invocation." + }, + { + "left": "issue:43950", + "right": "issue:44534", + "accept": true, + "reason": "Same underlying bug: transformers v5 non-persistent buffer handling/corruption, with 44534 phrased as the same regression." + }, + { + "left": "issue:39692", + "right": "issue:43054", + "accept": false, + "reason": "One is a docs/example mismatch and quantization failure; the other is a SigLIP2 quality/performance complaint." + }, + { + "left": "issue:43644", + "right": "issue:43950", + "accept": false, + "reason": "Unrelated regressions: missing config fields from trust_remote_code vs non-persistent buffer corruption." + }, + { + "left": "issue:36331", + "right": "issue:41762", + "accept": false, + "reason": "Different subsystems: Trainer loss signature change vs DeepSpeed ZeRO-3 model loading IndexError." + }, + { + "left": "issue:43872", + "right": "issue:44589", + "accept": false, + "reason": "Both are quantization/backend issues, but they describe different concrete incompatibilities and code paths." + }, + { + "left": "issue:43828", + "right": "issue:44928", + "accept": false, + "reason": "Different models and symptoms; dtype mismatch in Phi-tiny-MoE is not the same bug as Qwen3.5 RLHF gradient explosion." + }, + { + "left": "issue:45987", + "right": "issue:46858", + "accept": true, + "reason": "Same bug family: get_seq_length returning a tensor/device tensor instead of an int, causing downstream static-cache arithmetic failures." + }, + { + "left": "issue:45920", + "right": "issue:46489", + "accept": true, + "reason": "Same tokenizer regression in transformers v5+: both report incorrect token IDs/output for affected model families." + }, + { + "left": "issue:43054", + "right": "issue:45701", + "accept": false, + "reason": "SigLIP2 embedding quality complaint vs a broad tokenizer-change report; not the same concrete issue." + }, + { + "left": "issue:38175", + "right": "issue:43054", + "accept": false, + "reason": "Both mention SigLIP2, but one is zero probabilities and the other is relative embedding degradation; no clear shared root cause." + }, + { + "left": "issue:43950", + "right": "issue:45902", + "accept": true, + "reason": "Same non-persistent-buffer regression, with 45902 describing a specific manifestation after meta\u2192to_empty(cuda) materialization." + }, + { + "left": "issue:41553", + "right": "issue:43883", + "accept": false, + "reason": "Bad AutoTokenizer error message for Voxtral is unrelated to Molmo's missing tied-weights attribute." + }, + { + "left": "issue:43054", + "right": "issue:43493", + "accept": true, + "reason": "Likely the same SigLIP2 implementation discrepancy: HF vs JAX mismatch plausibly explains the degraded text embeddings." + }, + { + "left": "issue:44910", + "right": "issue:45859", + "accept": false, + "reason": "Both involve Qwen3.5/Qwen3_5Moe, but one is FlashAttention position_id handling and the other is CPU-offload incompatibility in rotary embedding." + }, + { + "left": "issue:42947", + "right": "issue:44387", + "accept": false, + "reason": "Different performance problems: LoRA gradient checkpointing vs int4 quantization memory bloat/OOM." + }, + { + "left": "issue:45663", + "right": "issue:45902", + "accept": false, + "reason": "Different model paths and failures: Gemma4 FSDP2 attention KeyError vs non-persistent buffer materialization." + }, + { + "left": "issue:41553", + "right": "issue:43881", + "accept": false, + "reason": "Different model-loading failures; no shared underlying bug is evident." + }, + { + "left": "issue:44403", + "right": "issue:44451", + "accept": false, + "reason": "Generic load-time noise vs a specific model loading failure for ScandiBERT." + }, + { + "left": "issue:43975", + "right": "issue:45520", + "accept": false, + "reason": "Tokenizer detokenization bug vs Python 3.13 flash_attn import mapping issue." + }, + { + "left": "issue:44403", + "right": "issue:44488", + "accept": false, + "reason": "Different artifacts and failures; both are loading-related but not the same bug." + }, + { + "left": "issue:24643", + "right": "issue:28530", + "accept": false, + "reason": "Training with DeepSpeed weight-shape error vs early-stopping metric selection; unrelated." + }, + { + "left": "issue:42947", + "right": "issue:43856", + "accept": false, + "reason": "Different memory-related training issues with different causes and model setups." + }, + { + "left": "issue:43643", + "right": "issue:45020", + "accept": false, + "reason": "Related theme of remote_code, but one is missing config fields and the other is a broader compatibility regression; not the same concrete bug." + }, + { + "left": "issue:41762", + "right": "issue:42915", + "accept": false, + "reason": "Gemma3 ZeRO-3 loading IndexError vs Qwen3MoE FineGrainedFP8 failure; different code paths." + }, + { + "left": "issue:44991", + "right": "issue:45399", + "accept": false, + "reason": "Tokenizer loading failure and flash-attn fallback blocking are unrelated." + }, + { + "left": "issue:43824", + "right": "issue:43872", + "accept": false, + "reason": "Importing a model class vs bitsandbytes Int8Params constructor incompatibility." + }, + { + "left": "issue:43782", + "right": "issue:44960", + "accept": false, + "reason": "Qwen3VL weight_only load error vs a vague GLM5 issue; no shared root cause." + }, + { + "left": "issue:43881", + "right": "issue:43883", + "accept": false, + "reason": "Both are model-loading issues, but they affect different models and failure mechanisms." + }, + { + "left": "issue:45685", + "right": "issue:45823", + "accept": false, + "reason": "MPS histogram/int limitation vs Gemma4 PLE device mismatch are different backend bugs." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "Accept. The canonical anchor choice and the soft-pair judgments are conservative and grounded in the packet: the accepted pairs are all clear same-bug families by title/body framing (non-persistent buffers, StaticCache.get_seq_length, tokenizer regression, SigLIP2 discrepancy), and the rejected pairs are appropriately kept separate by model/code-path differences. I don\u2019t see an obvious overstatement." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 5, + "review_unit_count": 60, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-500", + "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:43827|issue:44509", + "issue:45488|issue:46489", + "issue:46894|issue:46897", + "issue:43872|issue:46869", + "issue:43404|issue:43950", + "issue:45759|issue:46661", + "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": "100af846bdef61f1ee147fce57474a9894ecc6623c94204688c10dec265418af", + "component_budget": { + "node_count": 500, + "item_count": 500, + "soft_pair_count": 1839, + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 + }, + "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": "The packet is mostly heterogeneous. The docs regression pair and the mask-shape IndexError pair look like true duplicates; the Whisper/fbank pair is only a weak thematic match and should not be accepted on the evidence shown.", + "confidence": 0.81, + "canonical_issue_reason": "No single issue represents the whole packet well. The clearest duplicate subclusters are 43827/44509 for the stale v5 docs references and 41093/44805 for the same mask-shape IndexError; 43827 is the cleanest docs representative.", + "canonical_pr_reason": null, + "best_issue_reason": "43827 is the strongest overall issue representative because it cleanly states the v5 docs regression and has a clear duplicate in 44509.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:45081", + "right": "issue:45399", + "accept": false, + "reason": "Tokenizer regex patching and flash-attn2 fallback checks are different bugs with different code paths." + }, + { + "left": "issue:45020", + "right": "issue:45823", + "accept": false, + "reason": "Remote-code loading breakage and a device-map forward mismatch are not the same concrete failure." + }, + { + "left": "issue:44661", + "right": "issue:45020", + "accept": false, + "reason": "add-new-model-like/tokenizer mapping logic is unrelated to the broader remote_code regression." + }, + { + "left": "issue:42491", + "right": "issue:44368", + "accept": false, + "reason": "LoRA compatibility across versions is not the same bug as a missing gradient-checkpointing flag warning." + }, + { + "left": "issue:43828", + "right": "issue:45823", + "accept": false, + "reason": "Dtype mismatch under autocast and device mismatch with auto device_map are different failure modes." + }, + { + "left": "issue:45020", + "right": "issue:45356", + "accept": false, + "reason": "Generic remote_code breakage and a tokenizer codec regression do not share one concrete bug path." + }, + { + "left": "issue:45715", + "right": "issue:46315", + "accept": false, + "reason": "These are both tokenizer performance issues, but they affect opposite conversion directions and likely different code paths." + }, + { + "left": "issue:43827", + "right": "issue:44509", + "accept": true, + "reason": "Same stale v5 docs regression: both report documentation still referencing removed pipeline tasks." + }, + { + "left": "issue:45488", + "right": "issue:46489", + "accept": false, + "reason": "Both involve tokenizers, but the affected models and bugs are different and not shown to be the same underlying issue." + }, + { + "left": "issue:46894", + "right": "issue:46897", + "accept": false, + "reason": "The titles suggest a similar training-loss pattern, but different models and insufficient evidence prevent treating them as the same bug." + }, + { + "left": "issue:43872", + "right": "issue:46869", + "accept": false, + "reason": "bitsandbytes constructor incompatibility is unrelated to the YOLO multi-GPU device_map failure." + }, + { + "left": "issue:43404", + "right": "issue:43950", + "accept": false, + "reason": "lm_head tying in a specific multimodal model is unrelated to non-persistent buffer corruption in from_pretrained()." + }, + { + "left": "issue:45759", + "right": "issue:46661", + "accept": false, + "reason": "Different model-construction failures: text_config unwrapping versus remote-code config_class access." + }, + { + "left": "issue:43872", + "right": "issue:44291", + "accept": false, + "reason": "Both mention _is_hf_initialized, but the concrete failures differ: bitsandbytes Int8Params versus init_empty_weights." + }, + { + "left": "issue:45020", + "right": "issue:46661", + "accept": false, + "reason": "Broad remote-code breakage and missing config_class on model_class are distinct loading failures." + }, + { + "left": "issue:43792", + "right": "issue:44220", + "accept": false, + "reason": "Both mention Whisper/fbank, but the evidence here does not establish the same concrete buggy code path, so this is only uncertain." + }, + { + "left": "issue:43257", + "right": "issue:45305", + "accept": false, + "reason": "Both touch DeepSpeed/Qwen training, but one is missing MoE weight conversion and the other is gradient averaging with GAS." + }, + { + "left": "issue:41553", + "right": "issue:44991", + "accept": false, + "reason": "Bad tokenizer error messaging for Voxtral is unrelated to the est-roberta tokenizer load failure." + }, + { + "left": "issue:36331", + "right": "issue:41093", + "accept": false, + "reason": "Unexpected compute_loss kwarg and mask/tensor shape mismatch are different trainer/runtime bugs." + }, + { + "left": "issue:42947", + "right": "issue:45923", + "accept": false, + "reason": "Gradient checkpointing being ineffective with PEFT LoRA is not the same as a missing supports_gradient_checkpointing flag." + }, + { + "left": "issue:43854", + "right": "issue:44863", + "accept": false, + "reason": "Different model-loading problems for different model families; no shared underlying bug is evident." + }, + { + "left": "issue:43940", + "right": "issue:45313", + "accept": false, + "reason": "Both are ZeRO-3 loading failures on Qwen variants, but the symptoms and affected components differ enough to keep them separate." + }, + { + "left": "issue:41553", + "right": "issue:44488", + "accept": false, + "reason": "Different tokenizer-loading failures: a Voxtral error message issue versus a model loading failure for cjvt/sleng-bert." + }, + { + "left": "issue:44928", + "right": "issue:45588", + "accept": false, + "reason": "RLHF NaN/SDPA fallback issue is unrelated to flash_attention.py crashing on sink-less models." + }, + { + "left": "issue:41093", + "right": "issue:44805", + "accept": true, + "reason": "Same mask-shape IndexError; 44805 is a later, more detailed report of the same bug." + }, + { + "left": "issue:44479", + "right": "issue:44560", + "accept": false, + "reason": "Both are video regressions in Qwen-family models, but the affected models and failure signatures differ." + }, + { + "left": "issue:24643", + "right": "issue:30064", + "accept": false, + "reason": "DeepSpeed training weight-shape error is unrelated to void segmentation map processing." + }, + { + "left": "issue:44530", + "right": "issue:44863", + "accept": false, + "reason": "PagedAttentionCache linear_attention crash and NemotronH checkpoint loading are different model/backend failures." + }, + { + "left": "issue:43872", + "right": "issue:44291", + "accept": false, + "reason": "Reiterating the shared keyword is not enough; the constructor/type mismatch contexts differ and should stay separate." + }, + { + "left": "issue:45020", + "right": "issue:46661", + "accept": false, + "reason": "Remote-code breakage and model_class.config_class access errors are distinct problems." + } + ] + }, + "evaluator_result": { + "accept": false, + "feedback": "Mostly grounded, but one soft-edge acceptance is too strong for the packet: 43792/44220 only share a Whisper/fbank theme from titles, and the evidence shown here does not establish the same concrete bug path. The docs pair (43827/44509) and mask-shape pair (41093/44805) look fine, but the Whisper pair should be downgraded to uncertain rather than accepted." + }, + "evaluator_used": true, + "retried": true, + "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": 60, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-500", + "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:44910|issue:46858", + "issue:43716|issue:44560", + "issue:43645|issue:43646", + "issue:44560|issue:45685", + "issue:43550|issue:44910", + "issue:42617|issue:43366", + "issue:44683|issue:46515", + "issue:44560|issue:44918", + "issue:36010|issue:42915", + "issue:45812|issue:46489", + "issue:45823|issue:46515", + "issue:44661|issue:45823", + "issue:38175|issue:43994", + "issue:44560|issue:44805", + "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": "6e2d7f93d566bf7e242bc3d9f595168bf62ba865e2a93038586c65a87e34cf98", + "component_budget": { + "node_count": 500, + "item_count": 500, + "soft_pair_count": 1839, + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 35, + "serialized_chars": 38346, + "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 transformer bug reports, with only a few true duplicate pairs: the v5 non-persistent-buffer regression, the custom-model initialization regression, a SigLIP2 output bug pair, and a Gemma4 4-bit dtype/casting pair. Most other soft links are broad-similarity false positives across different models or code paths.", + "confidence": 0.86, + "canonical_issue_reason": "43644 is the clearest canonical anchor because it is a tightly scoped v5 regression with a direct duplicate in 44534 and cleanly represents one real duplicate subgroup in an otherwise mixed cluster.", + "canonical_pr_reason": null, + "best_issue_reason": "43644 is the best issue representative: concrete, reproducible, and backed by an exact duplicate (44534), making it the strongest anchor for this cluster.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43366", + "right": "issue:43828", + "accept": false, + "reason": "Different bugs and different model families: GGUF architecture support vs Phi autocast dtype mismatch." + }, + { + "left": "issue:43596", + "right": "issue:46515", + "accept": false, + "reason": "Zero3/BertModel index error is unrelated to a general torch version incompatibility report." + }, + { + "left": "issue:45325", + "right": "issue:45902", + "accept": false, + "reason": "Both involve model internals, but one is a Qwen2.5-VL rope scaling bug and the other is Qwen3.5-MoE uninitialized memory after meta materialization." + }, + { + "left": "issue:43366", + "right": "issue:43531", + "accept": false, + "reason": "GGUF gpt-oss support and Qwen3-MoE sliding_window are unrelated issues." + }, + { + "left": "issue:43638", + "right": "issue:46515", + "accept": false, + "reason": "Same broad area of loading/runtime failures, but the concrete bugs are different: Bert/zero3 index error vs torch 2.4.1 incompatibility." + }, + { + "left": "issue:45663", + "right": "issue:45823", + "accept": false, + "reason": "Both are Gemma4, but one is an FSDP2 shared_kv_states training bug and the other is a device_map forward mismatch." + }, + { + "left": "issue:44155", + "right": "issue:45584", + "accept": false, + "reason": "Both are audio-related, but one is batched inference leakage in AudioFlamingo3 and the other is Whisper empty-transcription generation failure." + }, + { + "left": "issue:43366", + "right": "issue:44589", + "accept": false, + "reason": "GGUF architecture support is unrelated to the Float8 storage-object loading error." + }, + { + "left": "issue:43824", + "right": "issue:43957", + "accept": false, + "reason": "Importing Qwen2_5_VLForConditionalGeneration and meta-device loading failures are different code paths and different symptoms." + }, + { + "left": "issue:43950", + "right": "issue:45834", + "accept": false, + "reason": "Non-persistent buffer corruption and Kosmos2.5 OCR index errors are unrelated." + }, + { + "left": "issue:42915", + "right": "issue:43957", + "accept": false, + "reason": "Qwen3-MoE FineGrainedFP8Config failure is unrelated to generic meta-device loading breakage." + }, + { + "left": "issue:44910", + "right": "issue:46858", + "accept": false, + "reason": "Both involve attention/cache mechanics, but the concrete failures are different: packed 3D position_ids on Qwen3.5 vs device tensor seq_length arithmetic on non-CUDA backends." + }, + { + "left": "issue:43716", + "right": "issue:44560", + "accept": false, + "reason": "Mistral-3 image dtype mismatch and Qwen3-vl video StopIteration are distinct multimodal bugs." + }, + { + "left": "issue:43645", + "right": "issue:43646", + "accept": true, + "reason": "These are the same underlying Transformers v5 custom model initialization regression; the notebook mention is just a reproduction context." + }, + { + "left": "issue:44560", + "right": "issue:45685", + "accept": false, + "reason": "Qwen3-vl video StopIteration is unrelated to the MPS histogram implementation error for MoE." + }, + { + "left": "issue:43550", + "right": "issue:44910", + "accept": false, + "reason": "Different failure modes: torch.compile with SDPA on Bamba vs FlashAttention illegal memory access on Qwen3.5." + }, + { + "left": "issue:42617", + "right": "issue:43366", + "accept": false, + "reason": "3d_parallel.py runtime failure and GGUF gpt-oss support are unrelated." + }, + { + "left": "issue:44683", + "right": "issue:46515", + "accept": false, + "reason": "Compiled flex_attention on torch >=2.9 and torch 2.4.1 incompatibility are different issues." + }, + { + "left": "issue:44560", + "right": "issue:44918", + "accept": false, + "reason": "Different Qwen3.x multimodal/training failures with no shared concrete code path." + }, + { + "left": "issue:36010", + "right": "issue:42915", + "accept": false, + "reason": "GenerationMixin import error and Qwen3-MoE FP8 failure are unrelated." + }, + { + "left": "issue:45812", + "right": "issue:46489", + "accept": false, + "reason": "Both are tokenizer regressions, but they affect different model families and different tokenizer code paths." + }, + { + "left": "issue:45823", + "right": "issue:46515", + "accept": false, + "reason": "Gemma4 device_map mismatch and generic torch incompatibility are not the same bug." + }, + { + "left": "issue:44661", + "right": "issue:45823", + "accept": false, + "reason": "Tokenizer-mapping failure and Gemma4 PLE device mismatch are unrelated." + }, + { + "left": "issue:38175", + "right": "issue:43994", + "accept": true, + "reason": "Same SigLIP2 model, same symptom class of bad/incorrect outputs; these look like the same underlying model/processor bug." + }, + { + "left": "issue:44560", + "right": "issue:44805", + "accept": false, + "reason": "Qwen3-vl video StopIteration and a generic mask/tensor shape IndexError are different failures." + }, + { + "left": "issue:45672", + "right": "issue:46899", + "accept": true, + "reason": "Both describe Gemma4 4-bit inference breakage caused by unsafe dtype handling in multimodal/audio-vision embedders." + }, + { + "left": "issue:44877", + "right": "issue:45030", + "accept": false, + "reason": "Config validation regressions on different model families are too broad to merge as one duplicate." + }, + { + "left": "issue:43632", + "right": "issue:45834", + "accept": false, + "reason": "_is_hf_initialized flag breakage is unrelated to Kosmos2.5 OCR indexing." + }, + { + "left": "issue:43638", + "right": "issue:44661", + "accept": false, + "reason": "Zero3/Bert index error and add-new-model-like/tokenizer mapping failure are unrelated." + }, + { + "left": "issue:43596", + "right": "issue:44661", + "accept": false, + "reason": "Same as above: Bert zero3 initialization bug vs tokenizer mapping bug." + }, + { + "left": "issue:42491", + "right": "issue:43720", + "accept": false, + "reason": "LoRA hf4.x/hf5.x compatibility and BitNet packed-weight loading are different problems." + }, + { + "left": "issue:44661", + "right": "issue:44805", + "accept": false, + "reason": "Tokenizer mapping failure and generic mask-shape IndexError are unrelated." + }, + { + "left": "issue:43317", + "right": "issue:44756", + "accept": false, + "reason": "device_map auto dequantized-model offload failure and Strix Halo mmap OOM are different loading/resource issues." + }, + { + "left": "issue:38175", + "right": "issue:39692", + "accept": false, + "reason": "Both involve SigLIP2, but 39692 is a docs/example-specific model/processor mismatch plus quantization failure, not the same concrete runtime bug as 38175." + }, + { + "left": "issue:43644", + "right": "issue:44534", + "accept": true, + "reason": "Exact same non-persistent-buffer regression in Transformers v5; 44534 is a later duplicate of 43644." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The cluster summary is grounded in the packet and stays conservative: it treats most links as false positives, and the accepted soft edges are supported by closely aligned titles (especially the near-identical v5 buffer-regression pair and the custom-model init pair). The canonical 43644 anchor is well supported by its exact duplicate 44534." + }, + "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": 60, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-500", + "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:43856|issue:44387", + "issue:43854|issue:43931", + "issue:43708|issue:44075", + "issue:42491|issue:45335", + "issue:44960|issue:46050", + "issue:46897|issue:46901", + "issue:45488|issue:45812", + "issue:45276|issue:45335", + "issue:45362|issue:45406", + "issue:44164|issue:44291", + "issue:44910|issue:45588", + "issue:44402|issue:44488", + "issue:39692|issue:44960", + "issue:43425|issue:43957", + "issue:41720|issue:44155" + ], + "prepared_review_unit_hash": "2c1e5e71bea78e2ac069db7d107676d421e06d1a52e1391d59f9800899bbf0e3", + "component_budget": { + "node_count": 500, + "item_count": 500, + "soft_pair_count": 1839, + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 30, + "serialized_chars": 36099, + "estimated_input_tokens": 9025, + "estimated_eval_tokens": 18306 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a loose collection of Transformers 5.x/model-specific regressions rather than one unified bug. I only see one strong duplicate pair: the `resize_token_embeddings` / output-embedding regression across Gemma4 and T5Gemma; the rest are too model-, API-, or failure-mode-specific to merge safely.", + "confidence": 0.58, + "canonical_issue_reason": "`issue:45276` is the clearest representative of the duplicate `resize_token_embeddings` regression: it names the concrete API, the affected embedding surfaces, and the expected propagation failure.", + "canonical_pr_reason": null, + "best_issue_reason": "`issue:45276` is the strongest issue to anchor the cluster because it is specific, actionable, and broad enough to explain the matched T5Gemma report as the same root problem.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:42491", + "right": "issue:45276", + "accept": false, + "reason": "Different bugs: LoRA/qwen3_moe compatibility versus a `resize_token_embeddings` embedding-propagation regression." + }, + { + "left": "issue:41720", + "right": "issue:42915", + "accept": false, + "reason": "Both mention Qwen3-related failures, but one is auto device mapping/cuda assert and the other is fine-grained FP8 config." + }, + { + "left": "issue:42915", + "right": "issue:44756", + "accept": false, + "reason": "Distinct failures: Qwen3Moe FP8 config versus mmap/OOM on Strix Halo." + }, + { + "left": "issue:44451", + "right": "issue:44488", + "accept": false, + "reason": "Both are model-loading complaints, but they target different checkpoints and no shared root cause is evident." + }, + { + "left": "issue:44960", + "right": "issue:44977", + "accept": false, + "reason": "Unrelated model/runtime failures: GLM5 versus Qwen3.5 flash-attention generation." + }, + { + "left": "issue:44863", + "right": "issue:46846", + "accept": false, + "reason": "Different code paths: NemotronH checkpoint loading versus Qwen3.5 tensor-parallel planning." + }, + { + "left": "issue:46459", + "right": "issue:46512", + "accept": false, + "reason": "Both involve load-time state/dtype handling, but one is composite-checkpoint dtype propagation and the other is `from_config` stale config metadata." + }, + { + "left": "issue:45313", + "right": "issue:45357", + "accept": false, + "reason": "ZeRO-3 weight loading and save_pretrained visual-key regression are separate bugs." + }, + { + "left": "issue:44488", + "right": "issue:44991", + "accept": false, + "reason": "Both are tokenizer-load regressions, but they affect different models and likely different failure points." + }, + { + "left": "issue:45507", + "right": "issue:46468", + "accept": false, + "reason": "Different model implementations and symptoms: invalid method call in GraniteMoEHybrid versus a broken Nemotron example." + }, + { + "left": "issue:43644", + "right": "issue:43645", + "accept": false, + "reason": "Transformers 5.0 regressions, but one is buffer initialization and the other is Jupyter custom-model construction." + }, + { + "left": "issue:42491", + "right": "issue:45292", + "accept": false, + "reason": "LoRA/qwen3_moe incompatibility is not the same bug as `resize_token_embeddings` not updating output embeddings." + }, + { + "left": "issue:39692", + "right": "issue:43994", + "accept": false, + "reason": "Different SigLIP/SigLIP2 model-loading and inference problems; not the same concrete defect." + }, + { + "left": "issue:43257", + "right": "issue:43854", + "accept": false, + "reason": "Qwen3 MoE deepspeed conversion and GLM-4.7-Flash unit-test loading are unrelated." + }, + { + "left": "issue:43856", + "right": "issue:43873", + "accept": false, + "reason": "Both are quantization/memory complaints, but one is training-memory usage and the other is offloading behavior." + }, + { + "left": "issue:43856", + "right": "issue:44387", + "accept": false, + "reason": "Same broad area, but the specific problem differs: training memory usage versus int4 quantization reserved-memory OOM." + }, + { + "left": "issue:43854", + "right": "issue:43931", + "accept": false, + "reason": "Different models and failure modes: GLM load issue versus Qwen3-VL weight-shape mismatch." + }, + { + "left": "issue:43708", + "right": "issue:44075", + "accept": false, + "reason": "Trainer resume/max-steps logic and SGD argument handling are unrelated." + }, + { + "left": "issue:42491", + "right": "issue:45335", + "accept": false, + "reason": "LoRA/qwen3_moe incompatibility is not the same concrete issue as T5Gemma embedding resizing." + }, + { + "left": "issue:44960", + "right": "issue:46050", + "accept": false, + "reason": "GLM5 loading/generation versus quantized-model save/load corruption are different bugs." + }, + { + "left": "issue:46897", + "right": "issue:46901", + "accept": false, + "reason": "They share a \"double-shift\" theme, but the affected models and exact training paths differ, so this is too weak to merge." + }, + { + "left": "issue:45488", + "right": "issue:45812", + "accept": false, + "reason": "Both are tokenizer regressions, but one is a pre-tokenizer override bug and the other is wrong token IDs for Granite models." + }, + { + "left": "issue:45276", + "right": "issue:45335", + "accept": true, + "reason": "Both report the same `resize_token_embeddings` propagation bug: model-specific output/decoder embeddings are not updated after resizing." + }, + { + "left": "issue:45362", + "right": "issue:45406", + "accept": false, + "reason": "Qwen3.5 chat crash and Gemma4Processor serve crash are unrelated failures." + }, + { + "left": "issue:44164", + "right": "issue:44291", + "accept": false, + "reason": "Save/from_pretrained extra_state handling and init_empty_weights argument compatibility are different loading paths." + }, + { + "left": "issue:44910", + "right": "issue:45588", + "accept": false, + "reason": "Both are flash-attention crashes, but one is illegal memory access from 3D position IDs and the other is an AttributeError on `s_aux=None`." + }, + { + "left": "issue:44402", + "right": "issue:44488", + "accept": false, + "reason": "Tokenizer vocab-size mismatch and model-loading failure are not the same bug." + }, + { + "left": "issue:39692", + "right": "issue:44960", + "accept": false, + "reason": "Completely different model families and symptoms: SigLIP2 example errors versus GLM5." + }, + { + "left": "issue:43425", + "right": "issue:43957", + "accept": false, + "reason": "Torch version incompatibility and meta-device model-loading regression are separate issues." + }, + { + "left": "issue:41720", + "right": "issue:44155", + "accept": false, + "reason": "Qwen3 device-mapping CUDA assert and AudioFlamingo3 batched inference leakage are unrelated." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays conservative: it correctly treats most pairs as model-specific or failure-mode-specific, and the only accepted duplicate pair (issue:45276 with issue:45335) is well supported by the titles. No overstatement of evidence stood out." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 8, + "review_unit_count": 60, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-500", + "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:44079|issue:46829", + "issue:43404|issue:43454", + "issue:44291|issue:44756", + "issue:43873|issue:44387", + "issue:42947|issue:45704", + "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:43854|issue:45216", + "issue:42796|issue:46326", + "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": "4b15f3806416f168c260cbeeed709a0526076d87446ba77e255c614a01a60181", + "component_budget": { + "node_count": 500, + "item_count": 500, + "soft_pair_count": 1839, + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 + }, + "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": "This cluster is mostly noisy and heterogeneous. The only clear duplicate family is the resize_token_embeddings regression across Gemma/t5gemma variants; most other pairs are merely superficially similar and should stay separate.", + "confidence": 0.71, + "canonical_issue_reason": "issue:45292 is the best anchor for the only real duplicate family here: it states the generic resize_token_embeddings failure, while issue:45276 and issue:45335 are narrower model-specific variants of the same regression.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:45292 is the most representative and reusable issue title in the cluster: it describes the shared bug at the right abstraction level without overfitting to one model name.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:45663", + "right": "issue:45759", + "accept": false, + "reason": "Different subsystems and failure modes: Gemma4 FSDP2 shared_kv_states rebuilding vs Qwen multimodal config unwrapping." + }, + { + "left": "issue:44079", + "right": "issue:46829", + "accept": false, + "reason": "Both involve output containers, but one is a ModelOutput key assignment bug and the other is a VideoPrism hidden_states field mix-up." + }, + { + "left": "issue:43404", + "right": "issue:43454", + "accept": false, + "reason": "Both mention lm_head tying, but they are different model implementations and not clearly the same code-path bug." + }, + { + "left": "issue:44291", + "right": "issue:44756", + "accept": false, + "reason": "init_empty_weights argument handling and mmap/OOM on Strix Halo are unrelated issues." + }, + { + "left": "issue:43873", + "right": "issue:44387", + "accept": false, + "reason": "Both touch quantization memory behavior, but one is offloading and the other is int4 reserved-memory/OOM; too broad to merge." + }, + { + "left": "issue:42947", + "right": "issue:45704", + "accept": false, + "reason": "Gradient checkpointing/LoRA effectiveness and T5 apex RMSNorm memory leak are unrelated." + }, + { + "left": "issue:44292", + "right": "issue:44912", + "accept": false, + "reason": "Different quantization formats and symptoms; one is NVFP4 loading, the other MXFP4 fallback to bf16." + }, + { + "left": "issue:36010", + "right": "issue:39404", + "accept": false, + "reason": "ImportError for GenerationMixin and Whisper pipeline return_language are unrelated." + }, + { + "left": "issue:45663", + "right": "issue:45859", + "accept": false, + "reason": "FSDP2 shared_kv_states KeyError and CPU-offload incompatibility in Qwen rotary embedding are different bugs." + }, + { + "left": "issue:44387", + "right": "issue:44756", + "accept": false, + "reason": "Both mention OOM, but the causes are different: int4 quantization memory growth vs disabling mmap on Strix Halo." + }, + { + "left": "issue:42947", + "right": "issue:43381", + "accept": false, + "reason": "LoRA gradient checkpointing ineffectiveness is not the same as checkpointing being disallowed in eval mode." + }, + { + "left": "issue:44155", + "right": "issue:45823", + "accept": false, + "reason": "AudioFlamingo3 batch leakage and Gemma4 device mismatch are unrelated model-specific forward bugs." + }, + { + "left": "issue:43957", + "right": "issue:44756", + "accept": false, + "reason": "meta-device loading regressions and Strix Halo mmap OOM are not the same issue." + }, + { + "left": "issue:43335", + "right": "issue:45663", + "accept": false, + "reason": "SwitchTransformers sparse-layer construction and Gemma4 FSDP2 state rebuilding are unrelated." + }, + { + "left": "issue:45276", + "right": "issue:45292", + "accept": true, + "reason": "Same underlying resize_token_embeddings regression; 45292 is the generic version and 45276 is a model-specific manifestation." + }, + { + "left": "issue:44863", + "right": "issue:46468", + "accept": false, + "reason": "NemotronH checkpoint loading and the official Nemotron example being broken are related only at the model family level, not clearly the same bug." + }, + { + "left": "issue:44479", + "right": "issue:44877", + "accept": false, + "reason": "Video input regression for Qwen VL models and strict config blocking granite_speech loading are unrelated." + }, + { + "left": "issue:43381", + "right": "issue:44928", + "accept": false, + "reason": "Checkpointing-in-eval and RLHF gradient explosion from 3D position_ids are different problems." + }, + { + "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, but one is read-only file permissions during save_pretrained and the other is loading the wrong custom module." + }, + { + "left": "issue:45588", + "right": "issue:45663", + "accept": false, + "reason": "Flash attention s_aux=None crash and Gemma4 shared_kv_states rebuilding are unrelated." + }, + { + "left": "issue:43854", + "right": "issue:45216", + "accept": false, + "reason": "GLM-4.7-Flash unit-test loading and Qwen3.5 checkpoint save_pretrained regression are different model-specific issues." + }, + { + "left": "issue:42796", + "right": "issue:46326", + "accept": false, + "reason": "Both are tokenizer-related, but BPE marker handling and continue_final_message behavior are distinct bugs." + }, + { + "left": "issue:45292", + "right": "issue:45335", + "accept": true, + "reason": "Same resize_token_embeddings bug family; both report embeddings not being updated after resize in different model wrappers." + }, + { + "left": "issue:46894", + "right": "issue:46901", + "accept": false, + "reason": "Both describe double-shift loss errors, but they are in different model implementations and not clearly one shared code-path bug." + }, + { + "left": "issue:43828", + "right": "issue:43957", + "accept": false, + "reason": "Autocast dtype mismatch and meta-device loading failures are unrelated." + }, + { + "left": "issue:44402", + "right": "issue:44451", + "accept": false, + "reason": "Tokenizer vocab-size mismatch and ScandiBERT loading failure are not the same bug." + }, + { + "left": "issue:43646", + "right": "issue:43708", + "accept": false, + "reason": "Custom model initialization in notebooks and Trainer resume max_steps miscalculation are unrelated." + }, + { + "left": "issue:43645", + "right": "issue:44164", + "accept": false, + "reason": "Custom model init in Jupyter and extra_state save/load failures are different defects." + }, + { + "left": "issue:44589", + "right": "issue:44756", + "accept": false, + "reason": "Float8 storage lookup error and mmap-related OOM are unrelated." + }, + { + "left": "issue:43425", + "right": "issue:43828", + "accept": false, + "reason": "Torch 2.10 compatibility and Phi-tiny-MoE autocast dtype mismatch are not the same issue." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "Grounded and conservative overall. The resize_token_embeddings family is the only clearly supported duplicate cluster in the packet, and the other pairwise rejections stay appropriately cautious rather than over-claiming equivalence." + }, + "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": 60, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-500", + "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:46315|issue:46491", + "issue:45812|issue:46315", + "issue:39404|issue:45701", + "issue:39404|issue:46665", + "issue:44530|issue:45084", + "issue:43611|issue:43645", + "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:44265|issue:44898", + "issue:36032|issue:43452", + "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": "dc9f7882e8112fc4bf12345cf5e8b69d51a6dc7ee971808871054e820c56df2b", + "component_budget": { + "node_count": 500, + "item_count": 500, + "soft_pair_count": 1839, + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 33, + "serialized_chars": 37732, + "estimated_input_tokens": 9433, + "estimated_eval_tokens": 19122 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "A loose v5-regression cluster dominated by tokenizer/model-loading failures, with a few unrelated export/training/memory bugs mixed in. I would keep it mostly split; I don\u2019t see any soft pair that is clearly the same underlying bug.", + "confidence": 0.72, + "canonical_issue_reason": "issue:43611 is the broadest and most central v5 loading-regression report in the set; it best serves as an umbrella anchor, even though the cluster itself is heterogeneous.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:43611 is the best representative because it describes a generic Transformers 5 loading breakage rather than a single model-specific symptom.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:46315", + "right": "issue:46491", + "accept": false, + "reason": "Both are tokenizer regressions, but one is a slowdown in convert_tokens_to_ids and the other is a CodeLlama encode/decode space bug; different code paths and fixes." + }, + { + "left": "issue:45812", + "right": "issue:46315", + "accept": false, + "reason": "Granite token ID corruption and slow slow-tokenizer conversion are distinct bugs; same subsystem, not the same underlying defect." + }, + { + "left": "issue:39404", + "right": "issue:45701", + "accept": false, + "reason": "Whisper pipeline return_language regression is unrelated to a general tokenization behavior change report." + }, + { + "left": "issue:39404", + "right": "issue:46665", + "accept": false, + "reason": "Pipeline return_language failure and from_pretrained-in-classes inconsistency are different API behaviors and different fixes." + }, + { + "left": "issue:44530", + "right": "issue:45084", + "accept": false, + "reason": "PagedAttentionCache invalid group type and template-node compilation failure are separate internals with no clear shared root cause." + }, + { + "left": "issue:43611", + "right": "issue:43645", + "accept": false, + "reason": "base_model_prefix loading breakage and Jupyter custom-model initialization issues are both v5 regressions, but not the same code-path problem." + }, + { + "left": "issue:44683", + "right": "issue:45823", + "accept": false, + "reason": "Compiled flex_attention on torch>=2.9 and Gemma4 device mismatch are unrelated compatibility bugs." + }, + { + "left": "issue:43425", + "right": "issue:46869", + "accept": false, + "reason": "Torch 2.10 incompatibility is not the same as multi-GPU device_map auto failure on YOLO models." + }, + { + "left": "issue:43425", + "right": "issue:43716", + "accept": false, + "reason": "Torch version incompatibility and Mistral-3 image-preprocessor dtype mismatch are different issues." + }, + { + "left": "issue:43425", + "right": "issue:45823", + "accept": false, + "reason": "General torch incompatibility does not match the Gemma4 PLE device-mapping bug." + }, + { + "left": "issue:44589", + "right": "issue:45084", + "accept": false, + "reason": "Missing Float8 storage and non-template-node compilation errors are unrelated failures." + }, + { + "left": "issue:44479", + "right": "issue:44805", + "accept": false, + "reason": "Qwen video-input regression and a mask-shape IndexError are not the same underlying bug." + }, + { + "left": "issue:43906", + "right": "issue:44530", + "accept": false, + "reason": "An isolated reproduction of a prior issue and a Qwen3.5 PagedAttentionCache crash are different problems." + }, + { + "left": "issue:43901", + "right": "issue:44509", + "accept": false, + "reason": "These are both docs regressions, but they concern different removed/changed pipeline APIs." + }, + { + "left": "issue:44671", + "right": "issue:44991", + "accept": false, + "reason": "CamemBERT masked-LM prediction errors and est-roberta tokenizer loading failures are unrelated." + }, + { + "left": "issue:43994", + "right": "issue:44079", + "accept": false, + "reason": "SigLIP2 nonsensical outputs and ModelOutput key assignment with None are different bugs." + }, + { + "left": "issue:44265", + "right": "issue:44898", + "accept": false, + "reason": "torch.export failure with torch_compilable_check and Perceiver non-default resolution failure are distinct export/model issues." + }, + { + "left": "issue:36032", + "right": "issue:43452", + "accept": false, + "reason": "T5 tokenizer method-name conflict and gguf_file load failures are different tokenizer/loading bugs." + }, + { + "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": "sinq save/load breakage and continue_final_message on mistral_common are separate features and code paths." + }, + { + "left": "issue:43906", + "right": "issue:46326", + "accept": false, + "reason": "The reproduction of an older issue is not the same as the mistral_common continue_final_message bug." + }, + { + "left": "issue:44991", + "right": "issue:45081", + "accept": false, + "reason": "Tokenizer load failure for est-roberta and a Mistral regex patch backend_tokenizer error are different tokenizer issues." + }, + { + "left": "issue:43299", + "right": "issue:43931", + "accept": false, + "reason": "Both involve Qwen3-VL loading, but one is a dev0 MoE loading regression and the other is a specific weight-shape mismatch; too little evidence of one shared bug." + }, + { + "left": "issue:45823", + "right": "issue:46899", + "accept": false, + "reason": "Gemma4 device-mismatch during forward and Gemma4 4-bit multimodal embedder dtype-cast breakage are separate runtime bugs." + }, + { + "left": "issue:45479", + "right": "issue:46894", + "accept": false, + "reason": "Single-label classification zero-loss behavior is not the same as a training-loss double-shift issue." + }, + { + "left": "issue:43856", + "right": "issue:44756", + "accept": false, + "reason": "Qwen3 MoE memory usage and Strix Halo mmap OOM are different resource issues." + }, + { + "left": "issue:43596", + "right": "issue:44683", + "accept": false, + "reason": "Deepspeed ZeRO3/BertModel index error and torch>=2.9 compiled flex_attention failure are unrelated." + }, + { + "left": "issue:45479", + "right": "issue:46897", + "accept": false, + "reason": "Degenerate classification loss and Florence2 double-shift training loss are not the same defect." + }, + { + "left": "issue:44164", + "right": "issue:46821", + "accept": false, + "reason": "extra_state save/load handling and TimesFM window_size attribute access are different from-pretrained/config issues." + }, + { + "left": "issue:44488", + "right": "issue:44987", + "accept": false, + "reason": "Loading cjvt/sleng-bert and loading physical-intelligence/fast on transformers>=5.1.0 are model-specific failures with no clear shared root cause." + }, + { + "left": "issue:44488", + "right": "issue:45081", + "accept": false, + "reason": "Generic model loading failure and Mistral regex patch backend_tokenizer crash are different problems." + }, + { + "left": "issue:44488", + "right": "issue:45081", + "accept": false, + "reason": "Same pair as above; the symptoms and code paths do not align." + }, + { + "left": "issue:43611", + "right": "issue:43644", + "accept": false, + "reason": "Model-loading with base_model_prefix and non-persistent buffer initialization are separate v5 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": 60, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-500", + "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:44534", + "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" + ], + "soft_pairs": [ + "issue:44462|issue:46489", + "issue:42796|issue:43404", + "issue:46285|issue:46402", + "issue:36032|issue:42915", + "issue:32090|issue:39404", + "issue:43716|issue:45084", + "issue:44737|issue:45859", + "issue:44661|issue:44683", + "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:34567|issue:40444", + "issue:42175|issue:46178", + "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:44987|issue:44991", + "issue:43317|issue:43482", + "issue:45084|issue:45507", + "issue:43262|issue:43377", + "issue:43746|issue:46326", + "issue:43883|issue:44991", + "issue:43646|issue:44534" + ], + "prepared_review_unit_hash": "af9882f1d3323fb9213d212c18ff5b929da7aee986ea066b9d82de2d3a47dbf4", + "component_budget": { + "node_count": 500, + "item_count": 500, + "soft_pair_count": 1839, + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 30, + "serialized_chars": 36693, + "estimated_input_tokens": 9174, + "estimated_eval_tokens": 18604 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Most pairs are near-matches only by subsystem or release theme, not the same concrete bug. The strongest duplicates are the audio feature-extraction reports and the GGUF loading reports; the rest should stay separate.", + "confidence": 0.71, + "canonical_issue_reason": "Issue 43452 is the broadest, most concrete loading regression here: it describes a generic GGUF/from_pretrained break that can subsume specific model reports more cleanly than the narrower, model-specific issues.", + "canonical_pr_reason": null, + "best_issue_reason": "43452 is the best canonical issue candidate because it states the failure mode at the right abstraction level (GGUF loading via from_pretrained) while still naming the impacted APIs clearly.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44462", + "right": "issue:46489", + "accept": false, + "reason": "Both are tokenizer-loading regressions, but 46489 is a model-specific DeepSeek-Coder v1 bug tied to a gap in PR #44801, not clearly the same root cause as \"ignores tokenizer.json\"." + }, + { + "left": "issue:42796", + "right": "issue:43404", + "accept": false, + "reason": "Tokenizer BPE-marker handling and Mistral3 lm_head tying are unrelated code paths and symptoms." + }, + { + "left": "issue:46285", + "right": "issue:46402", + "accept": false, + "reason": "One is a CLIP model structure inconsistency, the other is missing weight loading in Sam3Model; different model families and bugs." + }, + { + "left": "issue:36032", + "right": "issue:42915", + "accept": false, + "reason": "T5 tokenizer special-token name conflict and Qwen3Moe FP8 failure are different failures with no shared concrete code path." + }, + { + "left": "issue:32090", + "right": "issue:39404", + "accept": false, + "reason": "Trainer GPU broadcast NoneType error and Whisper return_language pipeline regression are unrelated." + }, + { + "left": "issue:43716", + "right": "issue:45084", + "accept": false, + "reason": "Image preprocessor dtype mismatch is not the same bug as template compilation failing on non-template nodes." + }, + { + "left": "issue:44737", + "right": "issue:45859", + "accept": false, + "reason": "Both involve device placement, but XLNet CPU arange and Qwen3_Moe CPU offload are distinct implementation bugs." + }, + { + "left": "issue:44661", + "right": "issue:44683", + "accept": false, + "reason": "Tokenizer mapping/model-addition logic and compiled flex_attention on torch 2.9 are unrelated." + }, + { + "left": "issue:43643", + "right": "issue:44530", + "accept": false, + "reason": "Missing fields from trust_remote_code config return and PagedAttentionCache linear_attention failure are different issues." + }, + { + "left": "issue:44451", + "right": "issue:44987", + "accept": false, + "reason": "Both are model-loading failures, but the affected models and likely failure points differ; too broad to merge." + }, + { + "left": "issue:40990", + "right": "issue:43421", + "accept": false, + "reason": "High perplexity on a model benchmark and runtime post-processor updates after special-token edits are not the same bug." + }, + { + "left": "issue:42915", + "right": "issue:44530", + "accept": false, + "reason": "FP8 config failure for Qwen3Moe and PagedAttentionCache group-type crash are separate model/runtime problems." + }, + { + "left": "issue:43541", + "right": "issue:45084", + "accept": false, + "reason": "Torch dynamo grouped_mm tracing failure and template compilation errors are different code paths." + }, + { + "left": "issue:43644", + "right": "issue:43646", + "accept": false, + "reason": "Both are v5 regressions around custom model initialization, but one is specifically about non-persistent buffers being junk-filled while the other is broader; not enough to treat as the same bug." + }, + { + "left": "issue:43582", + "right": "issue:43632", + "accept": false, + "reason": "Apple Silicon warmup TypeError and _is_hf_initialized flag breakage are unrelated regressions." + }, + { + "left": "issue:34567", + "right": "issue:40444", + "accept": false, + "reason": "TrainerState token counting and multi-image iterable dataset finetuning failures are not the same issue." + }, + { + "left": "issue:42175", + "right": "issue:46178", + "accept": false, + "reason": "Backend packaging omission for pip extras and an unsupported tensorflow_text backend mapping error are distinct problems." + }, + { + "left": "issue:43883", + "right": "issue:44488", + "accept": false, + "reason": "Missing tied-weights metadata on Molmo and a specific tokenizer load failure for cjvt/sleng-bert are unrelated." + }, + { + "left": "issue:40444", + "right": "issue:46326", + "accept": false, + "reason": "Multi-image Qwen2.5-VL finetuning and Mistral-common continue_final_message behavior are different features and failures." + }, + { + "left": "issue:44206", + "right": "issue:44220", + "accept": true, + "reason": "These look like the same audio feature-extraction bug: both point at _torch_extract_fbank_features() / feature extractor crashes, with the second being the helper-function manifestation of the first." + }, + { + "left": "issue:44811", + "right": "issue:45584", + "accept": false, + "reason": "Both are Whisper-related, but one is batch_decode skip_special_tokens handling and the other is empty-transcription failure after align_special_tokens; different behavior." + }, + { + "left": "issue:43452", + "right": "issue:43482", + "accept": true, + "reason": "Both are GGUF loading regressions in transformers v5: 43452 describes the generic from_pretrained break, and 43482 is a specific Qwen2.5-GGUF manifestation of that same loading path failure." + }, + { + "left": "issue:36032", + "right": "issue:43645", + "accept": false, + "reason": "T5 tokenizer method conflict and custom model init failures in notebooks are different regressions." + }, + { + "left": "issue:44987", + "right": "issue:44991", + "accept": false, + "reason": "Both are version-dependent tokenizer/model load failures, but they affect different models and do not clearly share the same root cause." + }, + { + "left": "issue:43317", + "right": "issue:43482", + "accept": false, + "reason": "GPU/CPU offload dequantized-model loading and GGUF loading on v5 are different failure modes." + }, + { + "left": "issue:45084", + "right": "issue:45507", + "accept": false, + "reason": "Template compilation errors and GraniteMoEHybrid invalid method calls are unrelated." + }, + { + "left": "issue:43262", + "right": "issue:43377", + "accept": false, + "reason": "Sampling-rate defaulting in audio chat templates and MIMI batched-vs-single padding mask mismatch are different bugs." + }, + { + "left": "issue:43746", + "right": "issue:46326", + "accept": false, + "reason": "PEFT adapter loading for GraniteSpeech and mistral_common continue_final_message behavior are unrelated." + }, + { + "left": "issue:43883", + "right": "issue:44991", + "accept": false, + "reason": "Molmo tied-weights metadata and tokenizer loading for EMBEDDIA/est-roberta are not the same problem." + }, + { + "left": "issue:43646", + "right": "issue:44534", + "accept": false, + "reason": "Both are Transformers v5 regressions, but one is custom model initialization and the other is non-persistent buffers being filled with junk; related theme, not the same concrete bug." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays conservative overall. The analyst mostly rejects weak subsystem-level similarities, and the two accepted pairs are plausibly the same regressions based on the issue titles alone: the audio feature-extraction pair shares the same helper/path, and the GGUF pair is a generic vs. specific manifestation of the same loading failure. No PR-pair mergeability issue applies here." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 11, + "review_unit_count": 60, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-500", + "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: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:45084", + "issue:45216", + "issue:45325", + "issue:45357", + "issue:45987", + "issue:46315", + "issue:46821", + "issue:46869", + "issue:46899" + ], + "soft_pairs": [ + "issue:45084|issue:46869", + "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:44008|issue:46899", + "issue:44987|issue:45081", + "issue:43994|issue:44610", + "issue:44451|issue:45081", + "issue:28530|issue:34567", + "issue:42915|issue:44291", + "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:44991|issue:46315", + "issue:43531|issue:44877", + "issue:43425|issue:44403", + "issue:43296|issue:43531", + "issue:44220|issue:44683", + "issue:43611|issue:44534" + ], + "prepared_review_unit_hash": "d37f949bf763e90852afa16d45be39485f2ccb3984715c31c86b2efba4f29a4d", + "component_budget": { + "node_count": 500, + "item_count": 500, + "soft_pair_count": 1839, + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 30, + "serialized_chars": 36744, + "estimated_input_tokens": 9186, + "estimated_eval_tokens": 18628 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a loose collection of Transformers regressions; only two pairs look like true duplicates. No PRs are present.", + "confidence": 0.44, + "canonical_issue_reason": "issue:45216 is the clearest duplicate anchor because issue:45357 looks like the same Qwen3.5 save_pretrained / visual-encoder-key regression continuing into a later release.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:43611 is the broadest and most central issue in the set, covering a core Transformers v5 model-loading regression theme rather than a single model-specific symptom.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:45084", + "right": "issue:46869", + "accept": false, + "reason": "Different subsystems and failure modes: template compilation vs YOLO multi-GPU device mapping." + }, + { + "left": "issue:43792", + "right": "issue:46869", + "accept": false, + "reason": "Unrelated models and bugs; Whisper loading/can't run is not the same as YOLO device-map failures." + }, + { + "left": "issue:44403", + "right": "issue:45081", + "accept": false, + "reason": "Both touch loading behavior, but one is generic loading noise and the other is a tokenizer regex crash; not the same bug." + }, + { + "left": "issue:43645", + "right": "issue:44534", + "accept": false, + "reason": "Both are v5 regressions, but one is custom-model init in notebooks and the other is non-persistent buffer corruption." + }, + { + "left": "issue:44849", + "right": "issue:45325", + "accept": false, + "reason": "Same family, but different concrete symptoms: hidden-states output bug vs RoPE/position-id scaling bug." + }, + { + "left": "issue:43377", + "right": "issue:43819", + "accept": false, + "reason": "Both are audio-model issues, but they involve different models and different code paths." + }, + { + "left": "issue:43404", + "right": "issue:44466", + "accept": false, + "reason": "Both mention tied lm_head weights, but one is missing tying and the other is serialization varying by device; not the same concrete defect." + }, + { + "left": "issue:44464", + "right": "issue:45987", + "accept": false, + "reason": "Compiled generation inconsistency and cache sequence-length return type are different issues." + }, + { + "left": "issue:44008", + "right": "issue:46899", + "accept": false, + "reason": "Different multimodal models and distinct bugs: variable shadowing vs unsafe dtype casting." + }, + { + "left": "issue:44987", + "right": "issue:45081", + "accept": false, + "reason": "Different tokenizer/loading failures; no clear shared code-path or defect." + }, + { + "left": "issue:43994", + "right": "issue:44610", + "accept": false, + "reason": "Different model families and symptoms: bad SigLIP outputs vs OmDet-Turbo image-size mismatch." + }, + { + "left": "issue:44451", + "right": "issue:45081", + "accept": false, + "reason": "Unrelated loading failure vs Mistral tokenizer regex crash." + }, + { + "left": "issue:28530", + "right": "issue:34567", + "accept": false, + "reason": "Trainer early-stopping config issue and token-count tracking bug are unrelated." + }, + { + "left": "issue:42915", + "right": "issue:44291", + "accept": false, + "reason": "Different problems: quantization/config handling vs init_empty_weights argument compatibility." + }, + { + "left": "issue:36032", + "right": "issue:43742", + "accept": false, + "reason": "Tokenizer method-name collision and MobileLLM key error are not the same bug." + }, + { + "left": "issue:44062", + "right": "issue:44987", + "accept": false, + "reason": "Both are tokenizer-related, but one is an AddedToken argument conflict and the other is a model/tokenizer loading failure." + }, + { + "left": "issue:44265", + "right": "issue:44610", + "accept": false, + "reason": "torch.export/compilable-check failure is unrelated to an image processor size mismatch." + }, + { + "left": "issue:42915", + "right": "issue:44589", + "accept": false, + "reason": "Both involve version compatibility, but the concrete failures differ completely." + }, + { + "left": "issue:43089", + "right": "issue:45987", + "accept": false, + "reason": "Generation performance overhead is not the same as cache length contract behavior." + }, + { + "left": "issue:43322", + "right": "issue:43540", + "accept": false, + "reason": "Both are multimodal loading/runtime issues, but they affect different models and different failure points." + }, + { + "left": "issue:44479", + "right": "issue:45325", + "accept": true, + "reason": "These look like the same Qwen VL RoPE/temporal-position regression in v5.3.0, with one report focusing on video inputs and the other on still-image position_ids." + }, + { + "left": "issue:43632", + "right": "issue:46821", + "accept": false, + "reason": "Different regressions: HF initialization flag breakage vs TimesFM window_size attribute error." + }, + { + "left": "issue:36032", + "right": "issue:41720", + "accept": false, + "reason": "Tokenizer loading conflict and Qwen3 auto device-map CUDA assert are unrelated." + }, + { + "left": "issue:45216", + "right": "issue:45357", + "accept": true, + "reason": "Both report the same Qwen3.5 save_pretrained / checkpoint serialization regression, with 45357 sounding like a later manifestation of the same bug." + }, + { + "left": "issue:44991", + "right": "issue:46315", + "accept": false, + "reason": "Tokenizer loading failure and tokenizer conversion slowdown are different issues." + }, + { + "left": "issue:43531", + "right": "issue:44877", + "accept": false, + "reason": "Different model/config problems with no shared concrete bug." + }, + { + "left": "issue:43296", + "right": "issue:43531", + "accept": false, + "reason": "vLLM PaddleOCR-VL load failure and Qwen3-MoE sliding_window behavior are unrelated." + }, + { + "left": "issue:44220", + "right": "issue:44683", + "accept": false, + "reason": "Audio feature extraction and compiled flex_attention failures are unrelated." + }, + { + "left": "issue:43611", + "right": "issue:44534", + "accept": false, + "reason": "Both are v5 regressions, but one is base_model_prefix loading and the other is junk-filled non-persistent buffers; not the same underlying defect." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet: there are no PRs, and the claim that only two issue pairs look duplicate is conservative. The accepted soft pairs are supported by closely related titles, and the rejected pairs avoid overclaiming sameness. The best-issue and canonical-issue rationales are subjective but still consistent with 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": 12, + "review_unit_count": 60, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-500", + "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:43404", + "issue:43421", + "issue:43482", + "issue:43525", + "issue:43531", + "issue:43582", + "issue:43632", + "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:43716|issue:46899", + "issue:40444|issue:45325", + "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:43404|issue:46899", + "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:43632|issue:44534", + "issue:43531|issue:44291", + "issue:42175|issue:45542", + "issue:44462|issue:45812", + "issue:41720|issue:45081", + "issue:43317|issue:43856", + "issue:44079|issue:44534", + "issue:44509|issue:46909", + "issue:43296|issue:46821", + "issue:43377|issue:45491" + ], + "prepared_review_unit_hash": "99911d182514ad6d720d6d268e78aaf4eed313eca8c95de75c7da0bc5e36b3bc", + "component_budget": { + "node_count": 500, + "item_count": 500, + "soft_pair_count": 1839, + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 29, + "serialized_chars": 35904, + "estimated_input_tokens": 8976, + "estimated_eval_tokens": 18208 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is very heterogeneous: most items are separate model-specific regressions, tokenizer quirks, packaging issues, or docs bugs. I do not see any true duplicate PR/issue pair among the soft edges; at best they share broad themes like v5 regressions or tokenizer behavior.", + "confidence": 0.69, + "canonical_issue_reason": "Issue 44534 is the broadest core regression in the set and the best representative of the cluster\u2019s overall v5-era breakage theme. It is still not a duplicate of the others, just the most canonical umbrella-style report.", + "canonical_pr_reason": null, + "best_issue_reason": "44534 has the widest scope and is closest to an underlying platform-level regression, whereas most other issues are narrower model- or subsystem-specific reports.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43716", + "right": "issue:46899", + "accept": false, + "reason": "Both are multimodal dtype problems, but they affect different models and code paths; not the same bug." + }, + { + "left": "issue:40444", + "right": "issue:45325", + "accept": false, + "reason": "Both are Qwen2.5-VL, but one is a finetuning/data-pipeline failure and the other is a temporal position-id scaling bug." + }, + { + "left": "issue:45812", + "right": "issue:46491", + "accept": false, + "reason": "Both are tokenizer regressions, but Granite token IDs and CodeLlama leading-space round-tripping are different behaviors." + }, + { + "left": "issue:44403", + "right": "issue:44991", + "accept": false, + "reason": "Loading-related, but one is log noise and the other is a concrete tokenizer loading failure for a specific model." + }, + { + "left": "issue:43381", + "right": "issue:45923", + "accept": false, + "reason": "Both mention gradient checkpointing, but one is an eval-mode restriction and the other is a missing config flag on a trust_remote_code variant." + }, + { + "left": "issue:43632", + "right": "issue:44756", + "accept": false, + "reason": "Completely different problems: an internal initialization flag regression vs mmap-related OOM avoidance." + }, + { + "left": "issue:42796", + "right": "issue:43746", + "accept": false, + "reason": "Tokenizer BPE marker handling is unrelated to PEFT/local checkpoint loading." + }, + { + "left": "issue:43792", + "right": "issue:44589", + "accept": false, + "reason": "No shared code path: Whisper runtime failure versus a float8 storage deserialization/type error." + }, + { + "left": "issue:33666", + "right": "issue:40444", + "accept": false, + "reason": "Both are Qwen vision-training issues, but they describe different failures and likely different fixes." + }, + { + "left": "issue:43404", + "right": "issue:46899", + "accept": false, + "reason": "Different multimodal bugs: weight tying in Mistral3 vs unsafe dtype casting in Gemma4 embedders." + }, + { + "left": "issue:40990", + "right": "issue:44811", + "accept": false, + "reason": "Unrelated: perplexity evaluation issue vs processor.batch_decode special-token handling." + }, + { + "left": "issue:44291", + "right": "issue:44589", + "accept": false, + "reason": "Both are v5 compatibility failures, but one is init_empty_weights argument handling and the other is float8 storage lookup." + }, + { + "left": "issue:42796", + "right": "issue:46710", + "accept": false, + "reason": "Tokenizer marker handling and accuracy regression are different bug classes with no shared concrete code path." + }, + { + "left": "issue:43482", + "right": "issue:43525", + "accept": false, + "reason": "Both are model-loading problems, but they affect different models and different missing/unsupported config fields." + }, + { + "left": "issue:35707", + "right": "issue:40444", + "accept": false, + "reason": "Progressive generation with inputs_embeds/past_key_values is unrelated to Qwen2.5-VL finetuning with multiple images." + }, + { + "left": "issue:44079", + "right": "issue:45663", + "accept": false, + "reason": "ModelOutput key assignment and Gemma4 FSDP2 shared_kv_states are unrelated." + }, + { + "left": "issue:41553", + "right": "issue:43582", + "accept": false, + "reason": "Bad tokenizer error message and AppleSilicon TypeError are unrelated." + }, + { + "left": "issue:43482", + "right": "issue:45538", + "accept": false, + "reason": "GGUF loading failure and CLIPTokenizer max length are different tokenizer/model-loading issues." + }, + { + "left": "issue:43421", + "right": "issue:44568", + "accept": false, + "reason": "Both involve special-token behavior, but one is runtime post-processor updates and the other is add_special_tokens not adding BOS/EOS for a specific tokenizer." + }, + { + "left": "issue:43632", + "right": "issue:44534", + "accept": false, + "reason": "Different regression types: initialization flag breakage versus non-persistent buffer corruption." + }, + { + "left": "issue:43531", + "right": "issue:44291", + "accept": false, + "reason": "Qwen3-MoE sliding-window behavior is unrelated to init_empty_weights/_is_hf_initialized." + }, + { + "left": "issue:42175", + "right": "issue:45542", + "accept": false, + "reason": "Both touch TensorFlow packaging, but one is missing TF in the torch extra and the other is tensorboard-installed-without-TF backend handling." + }, + { + "left": "issue:44462", + "right": "issue:45812", + "accept": false, + "reason": "Both are tokenizer regressions, but one is ignoring tokenizer.json and the other is wrong token IDs for Granite models." + }, + { + "left": "issue:41720", + "right": "issue:45081", + "accept": false, + "reason": "Auto device mapping cuda assert and Mistral regex patch crash are unrelated." + }, + { + "left": "issue:43317", + "right": "issue:43856", + "accept": false, + "reason": "Device-map/offload load failure and Qwen3 MoE memory inefficiency are different problems." + }, + { + "left": "issue:44079", + "right": "issue:44534", + "accept": false, + "reason": "ModelOutput key handling is unrelated to junk-filled non-persistent buffers." + }, + { + "left": "issue:44509", + "right": "issue:46909", + "accept": false, + "reason": "Both are docs issues, but they are different documents and different defects." + }, + { + "left": "issue:43296", + "right": "issue:46821", + "accept": false, + "reason": "PaddleOCR-VL/vLLM load failure and TimesFM window_size AttributeError are unrelated." + }, + { + "left": "issue:43377", + "right": "issue:45491", + "accept": false, + "reason": "Both mention batching/padding, but they are different models and different failure modes (encoder mismatch vs NaN embeddings)." + } + ] + }, + "evaluator_result": 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": 13, + "review_unit_count": 60, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-500", + "nodes": [ + "issue:34634", + "issue:35707", + "issue:39692", + "issue:40444", + "issue:40990", + "issue:41720", + "issue:42175", + "issue:42796", + "issue:42915", + "issue:43262", + "issue:43296", + "issue:43454", + "issue:43482", + "issue:43493", + "issue:43550", + "issue:43606", + "issue:43611", + "issue:43644", + "issue:43646", + "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: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:43611|issue:43646", + "issue:43818|issue:46402", + "issue:44829|issue:45479", + "issue:43644|issue:44079", + "issue:42175|issue:43296", + "issue:43818|issue:45357", + "issue:44265|issue:45072", + "issue:43493|issue:43994", + "issue:40444|issue:45381", + "issue:45325|issue:45834", + "issue:45325|issue:45381", + "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:35707|issue:41720", + "issue:43828|issue:46821", + "issue:34634|issue:35707", + "issue:39692|issue:40444", + "issue:43454|issue:43550", + "issue:43606|issue:43819", + "issue:44186|issue:44265", + "issue:41720|issue:45823" + ], + "prepared_review_unit_hash": "a0fb84dccc83f43d5e18157864a399b6692f95b229f3fad85e0e9deb2c2e7562", + "component_budget": { + "node_count": 500, + "item_count": 500, + "soft_pair_count": 1839, + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 30, + "serialized_chars": 36225, + "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 mostly a collection of unrelated Transformers issues across different models/subsystems. Only the two SigLIP2 reports look like a plausible duplicate pair.", + "confidence": 0.74, + "canonical_issue_reason": "issue:43493 is the better canonical target for the SigLIP2 duplicate pair because it states the underlying HF-vs-JAX implementation discrepancy, which likely explains the downstream bad outputs reported in issue:43994.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:43873 is the strongest overall standalone issue in the cluster: it has the broadest scope, the most discussion/inbound references, and a concrete reproducible failure mode around quantized offloading.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43611", + "right": "issue:43646", + "accept": false, + "reason": "Different failure modes: loading with `base_model_prefix` vs. custom model initialization under Transformers 5." + }, + { + "left": "issue:43818", + "right": "issue:46402", + "accept": false, + "reason": "Both involve model internals, but one is a Video-LLaVA tower/temporal-attention bug and the other is Sam3 text-encoder weight loading." + }, + { + "left": "issue:44829", + "right": "issue:45479", + "accept": false, + "reason": "Both describe degenerate training, but one is tied to `flash_attention_3` and the other is a `num_labels=1` single-label classification loss issue." + }, + { + "left": "issue:43644", + "right": "issue:44079", + "accept": false, + "reason": "Non-persistent buffer serialization junk and `ModelOutput` key assignment are different code paths." + }, + { + "left": "issue:42175", + "right": "issue:43296", + "accept": false, + "reason": "Packaging/backend dependency issue versus a model-loading failure for PaddleOCR-VL." + }, + { + "left": "issue:43818", + "right": "issue:45357", + "accept": false, + "reason": "Video tower architecture bug vs. incorrect `save_pretrained` visual-encoder key handling." + }, + { + "left": "issue:44265", + "right": "issue:45072", + "accept": false, + "reason": "`torch.export`/`torch_compilable_check` failure is unrelated to a bfloat16 dtype mismatch during inference." + }, + { + "left": "issue:43493", + "right": "issue:43994", + "accept": true, + "reason": "Same SigLIP2 output-correctness bug family: one reports HF-vs-JAX discrepancy, the other reports nonsensical outputs via AutoModel/pipeline, both pointing to the same model implementation path." + }, + { + "left": "issue:40444", + "right": "issue:45381", + "accept": false, + "reason": "Multiple-images finetuning with IterableDataset is a different bug from Qwen2.5-VL video position-id handling." + }, + { + "left": "issue:45325", + "right": "issue:45834", + "accept": false, + "reason": "Qwen2.5-VL rope/position-id scaling bug and Kosmos2.5 OCR index error are unrelated." + }, + { + "left": "issue:45325", + "right": "issue:45381", + "accept": false, + "reason": "Both are Qwen2.5-VL multimodal regressions, but one is still-image temporal position scaling and the other is video `vision_position_ids`; not the same concrete bug." + }, + { + "left": "issue:43454", + "right": "issue:44466", + "accept": false, + "reason": "One is a tied-weight generation bug in AyaVision, the other is device-dependent serialization inconsistency for `lm_head.weight`." + }, + { + "left": "issue:42915", + "right": "issue:43482", + "accept": false, + "reason": "FineGrainedFP8Config failure and GGUF loading failure are separate model/version issues." + }, + { + "left": "issue:44291", + "right": "issue:44877", + "accept": false, + "reason": "Unexpected `_is_hf_initialized` argument and strict config rejection are different loading regressions." + }, + { + "left": "issue:43701", + "right": "issue:45325", + "accept": false, + "reason": "Checkpoint key mismatch is unrelated to Qwen2.5-VL rope-index scaling." + }, + { + "left": "issue:43873", + "right": "issue:45701", + "accept": false, + "reason": "Quantized offloading behavior and version-dependent tokenization changes do not share the same code path." + }, + { + "left": "issue:42796", + "right": "issue:45381", + "accept": false, + "reason": "Tokenizer BPE-marker handling is not the same as Qwen2.5-VL video position-id regression." + }, + { + "left": "issue:43606", + "right": "issue:44265", + "accept": false, + "reason": "CPU offload device mismatch and `torch.export` failure on `torch_compilable_check` are distinct problems." + }, + { + "left": "issue:40990", + "right": "issue:45584", + "accept": false, + "reason": "Perplexity regression on GPT-OSS and Whisper empty-transcription failure are unrelated model behaviors." + }, + { + "left": "issue:43262", + "right": "issue:43746", + "accept": false, + "reason": "Audio chat-template sampling-rate defaulting and PEFT checkpoint loading are different subsystems." + }, + { + "left": "issue:44479", + "right": "issue:45823", + "accept": false, + "reason": "Qwen video-input regression and Gemma4 PLE device mismatch are unrelated model-specific bugs." + }, + { + "left": "issue:43881", + "right": "issue:44488", + "accept": false, + "reason": "GLM-4V loading failure and sleng-bert loading failure are separate model issues." + }, + { + "left": "issue:35707", + "right": "issue:41720", + "accept": false, + "reason": "Progressive generation with `inputs_embeds/past_key_values` and Qwen3 auto device-mapping CUDA asserts are different failures." + }, + { + "left": "issue:43828", + "right": "issue:46821", + "accept": false, + "reason": "Autocast dtype mismatch and TimesFM `window_size` AttributeError are unrelated." + }, + { + "left": "issue:34634", + "right": "issue:35707", + "accept": false, + "reason": "BarkProcessor voice preset behavior and generation with `inputs_embeds/past_key_values` are unrelated." + }, + { + "left": "issue:39692", + "right": "issue:40444", + "accept": false, + "reason": "SigLIP2 doc-example mistakes and Qwen2.5-VL multi-image finetuning failure are different issues." + }, + { + "left": "issue:43454", + "right": "issue:43550", + "accept": false, + "reason": "AyaVision tied-weight loading/generation bug and Bamba SDPA compile bug do not match." + }, + { + "left": "issue:43606", + "right": "issue:43819", + "accept": false, + "reason": "CPU offload device mismatch is unrelated to DAC `from_latents` vs forward-pass mismatch." + }, + { + "left": "issue:44186", + "right": "issue:44265", + "accept": false, + "reason": "Tokenizer crashes on NER/padding and `torch.export` failures are different code paths." + }, + { + "left": "issue:41720", + "right": "issue:45823", + "accept": false, + "reason": "Qwen3 auto device-mapping CUDA assert and Gemma4 PLE forward mismatch are unrelated." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is well grounded in the packet. The choice of issue:43873 as the strongest standalone issue is supported by its higher discussion and inbound-reference counts, and the SigLIP2 pairing of issue:43493 with issue:43994 is a conservative, evidence-based duplicate call. The soft-edge verdicts are appropriately cautious and do not overstate similarity." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 14, + "review_unit_count": 60, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-500", + "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:44610", + "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:43377|issue:44610", + "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:43749|issue:44534", + "issue:43317|issue:43873", + "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:43065|issue:46402", + "issue:44877|issue:44991", + "issue:40990|issue:45701", + "issue:44186|issue:45072", + "issue:26413|issue:28530", + "issue:41720|issue:42617", + "issue:43550|issue:43720" + ], + "prepared_review_unit_hash": "d909e08fba73060cf519b69c3cc4f531d24287d4bf57e3ca343cc453cfcd59e0", + "component_budget": { + "node_count": 500, + "item_count": 500, + "soft_pair_count": 1839, + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 31, + "serialized_chars": 37008, + "estimated_input_tokens": 9252, + "estimated_eval_tokens": 18760 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a heterogeneous mix of Transformers bug reports across tokenization, model loading/offload, quantization, config, and training behavior. The pairwise similarities are mostly superficial, so none of the soft pairs look like true duplicates.", + "confidence": 0.93, + "canonical_issue_reason": "No single canonical issue stands out: the cluster spans unrelated bugs and subsystems, and the apparent overlaps are mostly broad theme similarity rather than the same underlying defect.", + "canonical_pr_reason": null, + "best_issue_reason": "No global best issue is suitable as a representative duplicate target because the items are too diverse to collapse into one concrete bug report.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43377", + "right": "issue:44610", + "accept": false, + "reason": "Both involve batched input handling, but one is a MIMI padding-mask discrepancy and the other is an OmDet image size mismatch; different code paths and fixes." + }, + { + "left": "issue:46305", + "right": "issue:46315", + "accept": false, + "reason": "Sliding-window off-by-one in streaming masks and slow `convert_tokens_to_ids` are unrelated bugs." + }, + { + "left": "issue:43257", + "right": "issue:45923", + "accept": false, + "reason": "Qwen3 MoE weight conversion/loading and a missing gradient-checkpointing flag are different problems." + }, + { + "left": "issue:44568", + "right": "issue:45920", + "accept": false, + "reason": "Both are tokenizer regressions, but one is missing BOS/EOS insertion for a specific tokenizer while the other is wrong token IDs across many models; not the same bug." + }, + { + "left": "issue:42796", + "right": "issue:46209", + "accept": false, + "reason": "Tokenizer BPE marker handling and FP8 inference failure on B200 are unrelated." + }, + { + "left": "issue:43454", + "right": "issue:45072", + "accept": false, + "reason": "AyaVision weight tying and dtype mismatches in inference are separate model/runtime issues." + }, + { + "left": "issue:45823", + "right": "issue:46821", + "accept": false, + "reason": "Device-map forward mismatch for Gemma4 PLE and a TimesFM `window_size` AttributeError are different failures." + }, + { + "left": "issue:43377", + "right": "issue:44898", + "accept": false, + "reason": "Both are batching-related, but one is missing padding-mask support and the other is a Perceiver resolution/interpolation bug; not the same defect." + }, + { + "left": "issue:43454", + "right": "issue:43720", + "accept": false, + "reason": "AyaVision tie-weights failure and BitNet packed-weight unpacking during accelerate loading are unrelated." + }, + { + "left": "issue:40990", + "right": "issue:42947", + "accept": false, + "reason": "High perplexity on a specific model and ineffective gradient checkpointing with LoRA are different training/inference issues." + }, + { + "left": "issue:42617", + "right": "issue:42915", + "accept": false, + "reason": "3d_parallel.py execution failure and Qwen3Moe fine-grained FP8 failure are unrelated." + }, + { + "left": "issue:43749", + "right": "issue:44534", + "accept": false, + "reason": "Both touch loading/state handling, but FSDP CPU RAM efficient loading and non-persistent buffer junk are different concrete bugs." + }, + { + "left": "issue:43317", + "right": "issue:43873", + "accept": false, + "reason": "Both mention offload/quantization, but one is dequantized-model loading with device_map auto and the other is a broader quantization offloading issue; too different to merge." + }, + { + "left": "issue:43572", + "right": "issue:45759", + "accept": false, + "reason": "Missing `pad_token_idx` in config and failure to unwrap `text_config` are different config-loading bugs." + }, + { + "left": "issue:43526", + "right": "issue:43883", + "accept": false, + "reason": "BEiT label reduction and missing `all_tied_weights_keys` are unrelated." + }, + { + "left": "issue:43322", + "right": "issue:46050", + "accept": false, + "reason": "LLaVA Next segmentation fault and sinq save/load breakage are unrelated loading bugs." + }, + { + "left": "issue:41720", + "right": "issue:43322", + "accept": false, + "reason": "Qwen3 auto device-mapping CUDA assert and LLaVA Next segfault are different model/runtime failures." + }, + { + "left": "issue:43881", + "right": "issue:44991", + "accept": false, + "reason": "glm-4v-9b loading failure and tokenizer loading regression are unrelated." + }, + { + "left": "issue:36331", + "right": "issue:42617", + "accept": false, + "reason": "`compute_loss` signature breakage and `3d_parallel.py` execution failure are unrelated trainer/script issues." + }, + { + "left": "issue:44387", + "right": "issue:45704", + "accept": false, + "reason": "Both are memory-related, but one is int4 quantization memory growth and the other is an apex RMSNorm leak; different causes and fixes." + }, + { + "left": "issue:41720", + "right": "issue:44977", + "accept": false, + "reason": "Qwen3 auto device-map CUDA assert and Qwen3.5 flash-attention generation issues are not the same bug." + }, + { + "left": "issue:36331", + "right": "issue:42915", + "accept": false, + "reason": "`compute_loss` keyword mismatch and Qwen3Moe FP8 failure are unrelated." + }, + { + "left": "issue:44466", + "right": "issue:45357", + "accept": false, + "reason": "Both are serialization regressions, but one concerns tied `lm_head.weight` behavior and the other incorrect visual encoder keys; different artifacts." + }, + { + "left": "issue:43450", + "right": "issue:45701", + "accept": false, + "reason": "Batched video shape and version-dependent tokenization are unrelated." + }, + { + "left": "issue:43065", + "right": "issue:46402", + "accept": false, + "reason": "Sam3PixelDecoder dummy Conv2d and Sam3 text encoder weights not loading are different subcomponent issues." + }, + { + "left": "issue:44877", + "right": "issue:44991", + "accept": false, + "reason": "Strict config blocking `granite_speech` and tokenizer loading regression are unrelated." + }, + { + "left": "issue:40990", + "right": "issue:45701", + "accept": false, + "reason": "Extremely high perplexity and tokenization changes across versions are not the same issue." + }, + { + "left": "issue:44186", + "right": "issue:45072", + "accept": false, + "reason": "LayoutLMv2 tokenizer crashes and dtype mismatches in inference are unrelated." + }, + { + "left": "issue:26413", + "right": "issue:28530", + "accept": false, + "reason": "Checkpoint resume epoch-iterator failure and early-stopping metric lookup are different trainer behaviors." + }, + { + "left": "issue:41720", + "right": "issue:42617", + "accept": false, + "reason": "CUDA assert on Qwen3 device mapping and running 3d_parallel.py are unrelated." + }, + { + "left": "issue:43550", + "right": "issue:43720", + "accept": false, + "reason": "Bamba torch.compile SDPA failure and BitNet packed-weight unpacking 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": 15, + "review_unit_count": 60, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-500", + "nodes": [ + "issue:28530", + "issue:33666", + "issue:36010", + "issue:36331", + "issue:41720", + "issue:42617", + "issue:42915", + "issue:43295", + "issue:43316", + "issue:43317", + "issue:43404", + "issue:43425", + "issue:43526", + "issue:43531", + "issue:43550", + "issue:43572", + "issue:43577", + "issue:43606", + "issue:43646", + "issue:43708", + "issue:43873", + "issue:43881", + "issue:44008", + "issue:44186", + "issue:44361", + "issue:44423", + "issue:44464", + "issue:44610", + "issue:44671", + "issue:44734", + "issue:44743", + "issue:44756", + "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:43873|issue:44756", + "issue:46612|issue:46858", + "issue:43577|issue:43708", + "issue:43295|issue:45701", + "issue:43550|issue:43606", + "issue:45685|issue:46821", + "issue:44186|issue:44610", + "issue:43646|issue:46665", + "issue:43404|issue:45357", + "issue:44734|issue:46858", + "issue:44610|issue:44898", + "issue:42915|issue:43317", + "issue:36010|issue:42617", + "issue:44423|issue:45406", + "issue:44008|issue:45823", + "issue:46821|issue:46869", + "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" + ], + "prepared_review_unit_hash": "ccf1e3ba3f196e5fc3e8a476e854487221cd15c0c472a4446629d5c194e4533e", + "component_budget": { + "node_count": 500, + "item_count": 500, + "soft_pair_count": 1839, + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 28, + "serialized_chars": 36122, + "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: it mixes unrelated Transformers regressions across loading, tokenization, generation, quantization, offloading, and serve code paths. None of the soft pairs look like the same concrete bug/change, so all candidate edges are rejected. The broad custom-model initialization regression is the best representative issue in the set.", + "confidence": 0.74, + "canonical_issue_reason": "Issue 43646 is the broadest, most central regression report here: a Transformers 5.0 breakage around custom model initialization, which best fits the cluster\u2019s overall version/loading-regression theme.", + "canonical_pr_reason": null, + "best_issue_reason": "43646 is the best single representative because it is concrete, high-level, and closest to several custom-loading/version-regression reports, even though the cluster is not a tight duplicate set.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43873", + "right": "issue:44756", + "accept": false, + "reason": "Different bugs: quantization/offload behavior versus mmap/OOM on Strix Halo; no shared code path." + }, + { + "left": "issue:46612", + "right": "issue:46858", + "accept": false, + "reason": "Both involve cache/generation, but one is beam-search reorder skipping in specific models and the other is static-cache slice arithmetic on non-CUDA backends." + }, + { + "left": "issue:43577", + "right": "issue:43708", + "accept": false, + "reason": "Unrelated: dtype propagation in Blip2 loading versus Trainer max_steps calculation on checkpoint resume." + }, + { + "left": "issue:43295", + "right": "issue:45701", + "accept": false, + "reason": "Both are version-related regressions, but one is processor/tokenizer API breakage and the other is tokenization output changes; different underlying issues." + }, + { + "left": "issue:43550", + "right": "issue:43606", + "accept": false, + "reason": "Different models and failures: torch.compile with SDPA on Bamba versus CPU-offload device mismatch on bark-small." + }, + { + "left": "issue:45685", + "right": "issue:46821", + "accept": false, + "reason": "MoE MPS histogram support error is unrelated to TimesFM window_size AttributeError." + }, + { + "left": "issue:44186", + "right": "issue:44610", + "accept": false, + "reason": "Different model-specific preprocessing bugs: LayoutLMv2 tokenizer padding/NER crash versus OmDet-Turbo image-size mismatch." + }, + { + "left": "issue:43646", + "right": "issue:46665", + "accept": false, + "reason": "Related broadly to loading/custom code, but one is a Transformers 5.0 initialization regression and the other is from_pretrained behavior inside Python classes." + }, + { + "left": "issue:43404", + "right": "issue:45357", + "accept": false, + "reason": "Different models and failure modes: untied lm_head in Mistral3 versus wrong visual encoder keys saved for Qwen3.5." + }, + { + "left": "issue:44734", + "right": "issue:46858", + "accept": false, + "reason": "Both touch cache/generation, but one crashes in serve KV-cache continuation and the other is a static-cache device-tensor arithmetic bug." + }, + { + "left": "issue:44610", + "right": "issue:44898", + "accept": false, + "reason": "Both concern input sizing, but they affect different models and mechanisms; not the same bug." + }, + { + "left": "issue:42915", + "right": "issue:43317", + "accept": false, + "reason": "Both involve quantization/offload, but one is FineGrainedFP8 on Qwen3MoE and the other is dequantized loading with device_map auto." + }, + { + "left": "issue:36010", + "right": "issue:42617", + "accept": false, + "reason": "Completely unrelated: import error for GenerationMixin versus inability to run a 3d_parallel script." + }, + { + "left": "issue:44423", + "right": "issue:45406", + "accept": false, + "reason": "Both are serve multimodal crashes, but the exceptions and root causes differ: string `.to()` misuse versus missing `_tokenizer` on Gemma4Processor." + }, + { + "left": "issue:44008", + "right": "issue:45823", + "accept": false, + "reason": "Different Gemma model families and distinct forward-time failures: variable-name collision versus device mismatch with device_map auto." + }, + { + "left": "issue:45632", + "right": "issue:45684", + "accept": false, + "reason": "Different filesystem/save_pretrained problems: trust_remote_code cache collisions versus read-only permission propagation." + }, + { + "left": "issue:44464", + "right": "issue:46858", + "accept": false, + "reason": "Both involve compiled generation, but one is inconsistent outputs with chunked generation and the other is a cache-length tensor arithmetic failure." + }, + { + "left": "issue:41720", + "right": "issue:45161", + "accept": false, + "reason": "Different model-parallelism issues: cudaErrorAssert on Qwen3 auto device mapping versus TP not working on GPT-OSS MoE." + }, + { + "left": "issue:28530", + "right": "issue:33666", + "accept": false, + "reason": "Unrelated: early stopping metric resolution versus Qwen2-VL multi-GPU training." + }, + { + "left": "issue:44671", + "right": "issue:44743", + "accept": false, + "reason": "Different bugs in different models: CamemBERT MLM predictions versus Qwen3.5 recurrent state reset with cache." + }, + { + "left": "issue:43531", + "right": "issue:43572", + "accept": false, + "reason": "Unrelated model-specific issues: sliding window behavior in Qwen3-MoE versus missing pad_token_idx in StableLmConfig." + }, + { + "left": "issue:42915", + "right": "issue:46291", + "accept": false, + "reason": "Different problems: Qwen3MoE FP8 failure versus pipeline import failing due to kernels version/revision requirements." + }, + { + "left": "issue:36331", + "right": "issue:41720", + "accept": false, + "reason": "Trainer compute_loss signature mismatch is unrelated to Qwen3 device-map CUDA assert." + }, + { + "left": "issue:44186", + "right": "issue:44361", + "accept": false, + "reason": "Both are tokenizer/task crashes, but they affect different tokenizers and different error conditions." + }, + { + "left": "issue:43526", + "right": "issue:43881", + "accept": false, + "reason": "Different issues: BeitImageProcessorFast reduce_labels bug versus glm-4v-9b loading failure." + }, + { + "left": "issue:43316", + "right": "issue:44913", + "accept": false, + "reason": "Both are config-related, but one is an API discrepancy in Gemma3TextConfig and the other is a GPTNeoX rotary_pct reload bug." + }, + { + "left": "issue:43425", + "right": "issue:46821", + "accept": false, + "reason": "No meaningful overlap: Torch 2.10 incompatibility versus TimesFM window_size AttributeError." + } + ] + }, + "evaluator_result": 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": 60, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-500", + "nodes": [ + "issue:28530", + "issue:29159", + "issue:30064", + "issue:33357", + "issue:34567", + "issue:34634", + "issue:35707", + "issue:36032", + "issue:41720", + "issue:42490", + "issue:42915", + "issue:43322", + "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:44991", + "issue:45059", + "issue:45072", + "issue:45357", + "issue:45406", + "issue:45464", + "issue:45701", + "issue:45704", + "issue:45759", + "issue:45859", + "issue:46402", + "issue:46665", + "issue:46821", + "issue:46858" + ], + "soft_pairs": [ + "issue:42915|issue:43322", + "issue:28530|issue:30064", + "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:44492|issue:44509", + "issue:43606|issue:43720", + "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:45759|issue:45859", + "issue:44451|issue:44991", + "issue:45357|issue:46402", + "issue:43531|issue:44589", + "issue:44077|issue:44625", + "issue:43644|issue:46665", + "issue:44062|issue:44451", + "issue:44898|issue:45072" + ], + "prepared_review_unit_hash": "cf41b9722903f31118a16702984d6cc895a3ce6d69d24f5eb13a3b03f7816897", + "component_budget": { + "node_count": 500, + "item_count": 500, + "soft_pair_count": 1839, + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 30, + "serialized_chars": 36754, + "estimated_input_tokens": 9189, + "estimated_eval_tokens": 18634 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a loose semantic grouping of unrelated Transformers issues. The only clearly duplicate-like theme is the v5 non-persistent-buffer regression, but none of the provided soft-edge pairs are strong enough to merge as true duplicates/changes.", + "confidence": 0.33, + "canonical_issue_reason": "Issue 43644 is the clearest concrete regression report in the cluster and is the most canonical anchor for the v5 non-persistent-buffer bug theme; the nearby 44534 report looks like a rephrase of the same problem.", + "canonical_pr_reason": null, + "best_issue_reason": "43644 is the best representative because it describes a specific, reproducible framework regression with a clear failure mode, while most other issues here are model-specific or unrelated.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:42915", + "right": "issue:43322", + "accept": false, + "reason": "Different models and failures: FP8 config loading vs Llava Next segfault." + }, + { + "left": "issue:28530", + "right": "issue:30064", + "accept": false, + "reason": "Early stopping/metric selection is unrelated to void segmentation map processing." + }, + { + "left": "issue:33357", + "right": "issue:35707", + "accept": false, + "reason": "MacOS CLIP bus error is a different code path from generation with inputs_embeds/past_key_values." + }, + { + "left": "issue:29159", + "right": "issue:36032", + "accept": false, + "reason": "Both involve tokenizers, but one is slow/fast inconsistency and the other is a T5 tokenizer class-loading conflict." + }, + { + "left": "issue:43540", + "right": "issue:43782", + "accept": false, + "reason": "Different Qwen3 multimodal bugs: video input handling vs weight_only loading." + }, + { + "left": "issue:43550", + "right": "issue:43653", + "accept": false, + "reason": "Unrelated subsystems: torch.compile/SDPA vs BigBird tokenizer special-token registration." + }, + { + "left": "issue:44387", + "right": "issue:45701", + "accept": false, + "reason": "Quantization OOM and tokenization changes are different issues with no shared bug path." + }, + { + "left": "issue:43873", + "right": "issue:45704", + "accept": false, + "reason": "Offloading/quantization behavior is unrelated to T5's RMSNorm memory leak." + }, + { + "left": "issue:30064", + "right": "issue:34634", + "accept": false, + "reason": "Image segmentation-map processing and BarkProcessor voice_preset are unrelated." + }, + { + "left": "issue:44492", + "right": "issue:44509", + "accept": false, + "reason": "A typo in cache strategies is not the same as stale docs mentioning removed pipeline tasks." + }, + { + "left": "issue:43606", + "right": "issue:43720", + "accept": false, + "reason": "Different failure modes and models: CPU offload device mismatch vs packed-weight unpacking during accelerate loading." + }, + { + "left": "issue:44534", + "right": "issue:46665", + "accept": false, + "reason": "Non-persistent buffer corruption is unrelated to from_pretrained behavior inside Python classes." + }, + { + "left": "issue:43531", + "right": "issue:45759", + "accept": false, + "reason": "Qwen3-MoE sliding_window bug and Qwen multimodal config unwrapping are different problems." + }, + { + "left": "issue:41720", + "right": "issue:43746", + "accept": false, + "reason": "Auto device-mapping CUDA assert is unrelated to GraniteSpeech PEFT local checkpoint loading." + }, + { + "left": "issue:42490", + "right": "issue:43329", + "accept": false, + "reason": "Trailing-slash import warning and undefined video-branch variables are different bugs." + }, + { + "left": "issue:43388", + "right": "issue:45059", + "accept": false, + "reason": "Metric gathering tuple-label truncation is unrelated to SAM3 text/box behavior." + }, + { + "left": "issue:45406", + "right": "issue:45464", + "accept": false, + "reason": "Gemma4Processor serve crash and Qwen3.5 streaming API failure are distinct issues." + }, + { + "left": "issue:43606", + "right": "issue:43653", + "accept": false, + "reason": "CPU offload device mismatch is not the same as tokenizer special-token registration." + }, + { + "left": "issue:34567", + "right": "issue:35707", + "accept": false, + "reason": "TrainerState token counting and progressive generation are unrelated." + }, + { + "left": "issue:43531", + "right": "issue:46821", + "accept": false, + "reason": "Different models and bugs: Qwen3-MoE sliding_window vs TimesFM window_size attribute." + }, + { + "left": "issue:34634", + "right": "issue:41720", + "accept": false, + "reason": "BarkProcessor voice_preset and Qwen3 auto device mapping are unrelated." + }, + { + "left": "issue:43540", + "right": "issue:46858", + "accept": false, + "reason": "Video-input processing and static-cache seq-length tensor arithmetic are different code paths." + }, + { + "left": "issue:45759", + "right": "issue:45859", + "accept": false, + "reason": "Both are Qwen3.5-related, but one is config unwrapping and the other is CPU-offload rotary embedding compatibility." + }, + { + "left": "issue:44451", + "right": "issue:44991", + "accept": false, + "reason": "Model-specific tokenizer load failures are not enough to treat these as the same bug." + }, + { + "left": "issue:45357", + "right": "issue:46402", + "accept": false, + "reason": "Saving incorrect visual encoder keys and missing text-encoder weights on load are different directions and symptoms." + }, + { + "left": "issue:43531", + "right": "issue:44589", + "accept": false, + "reason": "Sliding-window behavior and missing Float8 storage are unrelated." + }, + { + "left": "issue:44077", + "right": "issue:44625", + "accept": false, + "reason": "patchtsmixer post_init policy and Qwen3.5 num_labels propagation are unrelated." + }, + { + "left": "issue:43644", + "right": "issue:46665", + "accept": false, + "reason": "Junk-filled non-persistent buffers are unrelated to from_pretrained behavior in user-defined classes." + }, + { + "left": "issue:44062", + "right": "issue:44451", + "accept": false, + "reason": "AddedToken keyword conflicts and ScandiBERT loading failure are not the same concrete bug." + }, + { + "left": "issue:44898", + "right": "issue:45072", + "accept": false, + "reason": "Perceiver interpolation failure and bfloat16 dtype 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": 17, + "review_unit_count": 60, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-500", + "nodes": [ + "issue:30064", + "issue:36010", + "issue:39692", + "issue:41720", + "issue:42886", + "issue:43381", + "issue:43421", + "issue:43493", + "issue:43653", + "issue:43720", + "issue:43746", + "issue:43782", + "issue:43819", + "issue:43844", + "issue:43873", + "issue:43927", + "issue:43994", + "issue:44186", + "issue:44368", + "issue:44514", + "issue:44545", + "issue:44610", + "issue:44625", + "issue:44671", + "issue:44843", + "issue:44918", + "issue:44936", + "issue:44987", + "issue:44995", + "issue:45042", + "issue:45081", + "issue:45356", + "issue:45357", + "issue:45381", + "issue:45479", + "issue:45704", + "issue:45759", + "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:44987|issue:45042", + "issue:44625|issue:45479", + "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:43746|issue:43873", + "issue:43421|issue:43653", + "issue:43994|issue:44671", + "issue:39692|issue:43493", + "issue:46032|issue:46463", + "issue:43421|issue:44186", + "issue:43782|issue:45381", + "issue:45081|issue:45356", + "issue:44995|issue:46858", + "issue:30064|issue:36010", + "issue:45356|issue:45907", + "issue:39692|issue:43873", + "issue:43720|issue:43819", + "issue:45987|issue:46829", + "issue:46829|issue:46832", + "issue:43653|issue:43720" + ], + "prepared_review_unit_hash": "49330532c7e9d13990193e435fe55c4141b47fe9c45321075041108012557ee3", + "component_budget": { + "node_count": 500, + "item_count": 500, + "soft_pair_count": 1839, + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 31, + "serialized_chars": 37549, + "estimated_input_tokens": 9388, + "estimated_eval_tokens": 19032 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a set of unrelated issues that were linked by loose lexical similarity. The only clear duplicate pair is the two Qwen2_5_VLProcessor batched padding=False crash reports; everything else should stay separate.", + "confidence": 0.96, + "canonical_issue_reason": "issue:44545 is the cleanest representative of the one true duplicate pair: it states the batched padding=False crash clearly and matches issue:44514 on the same underlying bug.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:44545 is the strongest canonical issue candidate because it is the clearer, more concise report of the duplicated Qwen2_5_VLProcessor.apply_chat_template failure.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44987", + "right": "issue:45042", + "accept": false, + "reason": "Different failures in different subsystems: loading a specific model vs PIL backend image processors requiring torchvision." + }, + { + "left": "issue:44625", + "right": "issue:45479", + "accept": false, + "reason": "One is Qwen3.5 config propagation into text_config; the other is a generic sequence-classification loss bug across models." + }, + { + "left": "issue:43381", + "right": "issue:43844", + "accept": false, + "reason": "Eval-mode gradient checkpointing restriction is unrelated to a ZeRO-3 gradient explosion regression." + }, + { + "left": "issue:44368", + "right": "issue:45357", + "accept": false, + "reason": "A warning about tie_word_embeddings in LoRA fine-tuning is not the same bug as incorrect visual encoder keys being saved." + }, + { + "left": "issue:39692", + "right": "issue:44610", + "accept": false, + "reason": "Both mention vision models, but one is a SigLIP2 docs/model mismatch and the other is an OmDet-Turbo image-size mismatch." + }, + { + "left": "issue:46655", + "right": "issue:46693", + "accept": false, + "reason": "Video padding calculation in a processor is unrelated to a FlashAttention2 long-context performance regression." + }, + { + "left": "issue:43421", + "right": "issue:43927", + "accept": false, + "reason": "Runtime special-token post-processing and save/load of custom token IDs are different tokenizer bugs." + }, + { + "left": "issue:41720", + "right": "issue:45704", + "accept": false, + "reason": "Qwen3 device-mapping CUDA asserts and T5 apex RMSNorm memory leak are unrelated." + }, + { + "left": "issue:43746", + "right": "issue:46050", + "accept": false, + "reason": "PEFT/local checkpoint loading failure is a different code path from quantized save_pretrained/from_pretrained corruption." + }, + { + "left": "issue:43381", + "right": "issue:45704", + "accept": false, + "reason": "Eval-mode checkpointing behavior is unrelated to the T5 apex memory leak." + }, + { + "left": "issue:42886", + "right": "issue:44843", + "accept": false, + "reason": "Both involve offline mode, but one is cache loading and the other is an unconditional model_info() call in _patch_mistral_regex." + }, + { + "left": "issue:45704", + "right": "issue:45923", + "accept": false, + "reason": "T5 apex memory leak and missing supports_gradient_checkpointing on a trust_remote_code model are separate issues." + }, + { + "left": "issue:44918", + "right": "issue:44936", + "accept": false, + "reason": "Qwen3.5 embedding unpacking in TRL SFT is not the same as trainer.evaluate() failing after trainer.train()." + }, + { + "left": "issue:44514", + "right": "issue:44545", + "accept": true, + "reason": "Same bug report twice: Qwen2_5_VLProcessor.apply_chat_template crashes on batched input when padding=False." + }, + { + "left": "issue:45759", + "right": "issue:45923", + "accept": false, + "reason": "Unwrapping text_config in composite multimodal configs is unrelated to the gradient-checkpointing flag on trust_remote_code variants." + }, + { + "left": "issue:43746", + "right": "issue:43873", + "accept": false, + "reason": "Local checkpoint loading for GraniteSpeech adapters is unrelated to quantization offloading behavior." + }, + { + "left": "issue:43421", + "right": "issue:43653", + "accept": false, + "reason": "Post-processor updates after special-token mutation are different from BigBirdTokenizer not registering mask as special." + }, + { + "left": "issue:43994", + "right": "issue:44671", + "accept": false, + "reason": "SigLIP2 pipeline/model output issues and CamemBERT masked LM regressions are unrelated." + }, + { + "left": "issue:39692", + "right": "issue:43493", + "accept": false, + "reason": "Both are SigLIP2-related, but one is a broken documentation example and the other is an implementation discrepancy against JAX." + }, + { + "left": "issue:46032", + "right": "issue:46463", + "accept": false, + "reason": "A silent use_cache correctness bug is not the same as an incorrect error message in T5 decoder input validation." + }, + { + "left": "issue:43421", + "right": "issue:44186", + "accept": false, + "reason": "Tokenizer post-processor updates and LayoutLMv2 NER/batching crashes are different bugs." + }, + { + "left": "issue:43782", + "right": "issue:45381", + "accept": false, + "reason": "Qwen3VL weight_only loading error and Qwen2.5-VL video position-id bug are separate multimodal issues." + }, + { + "left": "issue:45081", + "right": "issue:45356", + "accept": false, + "reason": "Both touch tokenizer regex handling, but one is a crash in _patch_mistral_regex and the other is a Kimi-K2.5 codec/regression report." + }, + { + "left": "issue:44995", + "right": "issue:46858", + "accept": false, + "reason": "A stale indexer cache on second forward pass is unrelated to static-cache seq_length returning a device tensor." + }, + { + "left": "issue:30064", + "right": "issue:36010", + "accept": false, + "reason": "Void segmentation map processing and a GenerationMixin import error have no meaningful overlap." + }, + { + "left": "issue:45356", + "right": "issue:45907", + "accept": false, + "reason": "Tokenizer codec/regex regression and list_repo_files revision forwarding are unrelated." + }, + { + "left": "issue:39692", + "right": "issue:43873", + "accept": false, + "reason": "SigLIP2 example/model issues and quantization offloading behavior are different bugs." + }, + { + "left": "issue:43720", + "right": "issue:43819", + "accept": false, + "reason": "BitNet packed-weight loading and DAC from_latents/STE mismatch are unrelated code paths." + }, + { + "left": "issue:45987", + "right": "issue:46829", + "accept": false, + "reason": "StaticCache seq_length contract bug and VideoPrism output-field mismatch are unrelated." + }, + { + "left": "issue:46829", + "right": "issue:46832", + "accept": false, + "reason": "VideoPrism hidden_states misassignment is different from RT-DETR feature-level forwarding errors." + }, + { + "left": "issue:43653", + "right": "issue:43720", + "accept": false, + "reason": "BigBirdTokenizer special-token registration and BitNet accelerate-loading issues are unrelated." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "Grounded and conservative overall. The only clearly duplicate pair in the packet is issue:44514/issue:44545, which have near-identical titles and matching descriptions. The rest of the soft-edge rejections are appropriately cautious and tied to materially different failure modes 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": 18, + "review_unit_count": 60, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-500", + "nodes": [ + "issue:30064", + "issue:34634", + "issue:36010", + "issue:36331", + "issue:39692", + "issue:41720", + "issue:43295", + "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:44361", + "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:46294", + "issue:46665", + "issue:46693", + "issue:46858" + ], + "soft_pairs": [ + "issue:43295|issue:45834", + "issue:43550|issue:44361", + "issue:43550|issue:43927", + "issue:44062|issue:44488", + "issue:43596|issue:44220", + "issue:45381|issue:46693", + "issue:43540|issue:46294", + "issue:44008|issue:46858", + "issue:43531|issue:43643", + "issue:43611|issue:46665", + "issue:43653|issue:43819", + "issue:44220|issue:44661", + "issue:44186|issue:44898", + "issue:43618|issue:43697", + "issue:41720|issue:44991", + "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": "59d160e69839dc23ae5a4ec954f7ca06ad2280edd4875bf5099e422609e66620", + "component_budget": { + "node_count": 500, + "item_count": 500, + "soft_pair_count": 1839, + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 29, + "serialized_chars": 36588, + "estimated_input_tokens": 9147, + "estimated_eval_tokens": 18550 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is heterogeneous: most items are distinct model-, tokenizer-, config-, or performance-specific issues rather than duplicates. None of the soft edge pairs looks like the same underlying bug/change. If one issue must serve as the anchor, #43611 is the broadest regression in the set.", + "confidence": 0.86, + "canonical_issue_reason": "Issue #43611 is the most broadly applicable regression here: it affects model loading across Transformers v5 via `base_model_prefix`, making it a better cluster anchor than the more model-specific reports.", + "canonical_pr_reason": null, + "best_issue_reason": "#43611 is the best representative issue because it describes a cross-cutting runtime regression with wide user impact, whereas the others are narrower or isolated to specific models/code paths.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43295", + "right": "issue:45834", + "accept": false, + "reason": "Different bugs: one is a processor/tokenizer regression, the other is a Kosmos2.5 long-OCR index error." + }, + { + "left": "issue:43550", + "right": "issue:44361", + "accept": false, + "reason": "Different subsystems and failure modes: Bamba torch.compile/SDPA vs MLukeTokenizer task handling." + }, + { + "left": "issue:43550", + "right": "issue:43927", + "accept": false, + "reason": "Unrelated issues: Bamba compile failure vs DiaConfig losing custom token IDs on save/load." + }, + { + "left": "issue:44062", + "right": "issue:44488", + "accept": false, + "reason": "Both are tokenizer-related, but one is an AddedToken constructor error and the other is a model-loading failure for a specific tokenizer." + }, + { + "left": "issue:43596", + "right": "issue:44220", + "accept": false, + "reason": "Different code paths: deepspeed ZeRO-3/BertModel init vs `_torch_extract_fbank_features()`." + }, + { + "left": "issue:45381", + "right": "issue:46693", + "accept": false, + "reason": "Qwen2.5-VL video position IDs bug is not the same as the FA2 long-context performance regression." + }, + { + "left": "issue:43540", + "right": "issue:46294", + "accept": false, + "reason": "Both involve Qwen video flows, but they are different models and different mismatches/errors in the pipeline." + }, + { + "left": "issue:44008", + "right": "issue:46858", + "accept": false, + "reason": "Gemma3n forward variable collision is unrelated to the static-cache slice arithmetic bug." + }, + { + "left": "issue:43531", + "right": "issue:43643", + "accept": false, + "reason": "Sliding-window behavior in Qwen3-MoE is unrelated to `trust_remote_code=True` returning incomplete configs." + }, + { + "left": "issue:43611", + "right": "issue:46665", + "accept": false, + "reason": "Both concern loading behavior, but the specific regression is different: base_model_prefix loading vs from_pretrained inside Python classes." + }, + { + "left": "issue:43653", + "right": "issue:43819", + "accept": false, + "reason": "Tokenizer special-token registration bug is unrelated to DAC.from_latents not matching forward without STE." + }, + { + "left": "issue:44220", + "right": "issue:44661", + "accept": false, + "reason": "Different features entirely: fbank extraction vs add-new-model-like failing on TOKENIZER_MAPPING_NAMES." + }, + { + "left": "issue:44186", + "right": "issue:44898", + "accept": false, + "reason": "LayoutLMv2 tokenizer padding/NER crash is not the same bug as Perceiver image classification at non-default resolution." + }, + { + "left": "issue:43618", + "right": "issue:43697", + "accept": false, + "reason": "CLIPOutput attentions assignment and RTDetrV2 output drift are separate model bugs." + }, + { + "left": "issue:41720", + "right": "issue:44991", + "accept": false, + "reason": "Qwen3 auto device-map CUDA assert and est-roberta tokenizer loading failure are unrelated." + }, + { + "left": "issue:30064", + "right": "issue:36331", + "accept": false, + "reason": "Image processor segmentation-map failure and `compute_loss` signature mismatch are completely different issues." + }, + { + "left": "issue:43643", + "right": "issue:44877", + "accept": false, + "reason": "Both are config-loading problems, but one is missing fields from `trust_remote_code`, the other is strict config rejection for granite_speech." + }, + { + "left": "issue:43550", + "right": "issue:43819", + "accept": false, + "reason": "Torch.compile SDPA failure in Bamba is unrelated to DAC latent decoding mismatch." + }, + { + "left": "issue:43335", + "right": "issue:43756", + "accept": false, + "reason": "SwitchTransformers sparse-layer creation and Smollm3 RoPE-layer dropping are different model-configuration bugs." + }, + { + "left": "issue:44843", + "right": "issue:45081", + "accept": false, + "reason": "Same helper name, but different root causes: offline `model_info()` call vs missing `backend_tokenizer` attribute." + }, + { + "left": "issue:44821", + "right": "issue:45084", + "accept": false, + "reason": "AutoImageProcessor URL loading and template-node compilation are unrelated failures." + }, + { + "left": "issue:44186", + "right": "issue:44442", + "accept": false, + "reason": "LayoutLMv2 tokenizer crashes and AutoTokenizer failing on FastSpeech2ConformerTokenizer are distinct tokenizer bugs." + }, + { + "left": "issue:39692", + "right": "issue:45834", + "accept": false, + "reason": "Documentation/example errors for SigLIP2 are unrelated to Kosmos2.5 OCR long-input indexing." + }, + { + "left": "issue:44448", + "right": "issue:44671", + "accept": false, + "reason": "Different models and causes: Pegasus v4/v5 output drift vs CamemBERT masked-LM prediction regression." + }, + { + "left": "issue:43540", + "right": "issue:44008", + "accept": false, + "reason": "Qwen3OmniMoe video processing error and Gemma3n audio mask variable collision are unrelated multimodal bugs." + }, + { + "left": "issue:43644", + "right": "issue:43749", + "accept": false, + "reason": "Non-persistent buffer initialization junk and FSDP CPU RAM efficient loading are different v5 regressions." + }, + { + "left": "issue:34634", + "right": "issue:36010", + "accept": false, + "reason": "BarkProcessor voice preset handling and GenerationMixin import failure do not describe the same defect." + }, + { + "left": "issue:45183", + "right": "issue:45479", + "accept": false, + "reason": "Input validation error messages across models are unrelated to the degenerate zero-loss classification bug." + }, + { + "left": "issue:45584", + "right": "issue:45834", + "accept": false, + "reason": "Whisper empty-transcription generation failure is not the same as Kosmos2.5 long OCR input handling." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 19, + "review_unit_count": 60, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-500", + "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:43950", + "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:45987", + "issue:46032", + "issue:46291", + "issue:46294", + "issue:46463", + "issue:46512", + "issue:46665", + "issue:46846" + ], + "soft_pairs": [ + "issue:43606|issue:45072", + "issue:43322|issue:45685", + "issue:43550|issue:45072", + "issue:43388|issue:46032", + "issue:44479|issue:45834", + "issue:43825|issue:43994", + "issue:43782|issue:44977", + "issue:43582|issue:46512", + "issue:43577|issue:43818", + "issue:44991|issue:46291", + "issue:43421|issue:44811", + "issue:44561|issue:45020", + "issue:43716|issue:44008", + "issue:43950|issue:45987", + "issue:43931|issue:46846", + "issue:46294|issue:46463", + "issue:43825|issue:43827", + "issue:28218|issue:29159", + "issue:44805|issue:44936", + "issue:44610|issue:45072", + "issue:43540|issue:45381", + "issue:43720|issue:45072", + "issue:43749|issue:44291", + "issue:43577|issue:43582", + "issue:43645|issue:46665", + "issue:44671|issue:45701", + "issue:42175|issue:45081" + ], + "prepared_review_unit_hash": "667680724238723077c7d844a2931b7d5e3aaf8ae47df7cb973d6057fd5353c7", + "component_budget": { + "node_count": 500, + "item_count": 500, + "soft_pair_count": 1839, + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 27, + "serialized_chars": 36220, + "estimated_input_tokens": 9055, + "estimated_eval_tokens": 18366 + }, + "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 bugs, with one clear duplicate pair around tokenizer added-token spacing/slow-vs-fast inconsistency. I would only merge the tokenizer pair; the rest are different models, code paths, or failure modes.", + "confidence": 0.84, + "canonical_issue_reason": "issue:28218 is the most concrete description of the shared tokenizer regression (extra space after an added token), and issue:29159 reads like a broader manifestation of the same problem.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:28218 is the best issue anchor because it is specific, actionable, and captures the actual symptom more directly than the broader tokenizer-consistency report.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43606", + "right": "issue:45072", + "accept": false, + "reason": "Both mention model/backend incompatibility, but one is a CPU-offload device mismatch and the other is a bfloat16 dtype mismatch on different models." + }, + { + "left": "issue:43322", + "right": "issue:45685", + "accept": false, + "reason": "Different failures: Llava Next loading segfault versus an MPS histogram kernel not implemented for Int." + }, + { + "left": "issue:43550", + "right": "issue:45072", + "accept": false, + "reason": "torch.compile+SDPA compile failure is a different code-path problem from dtype mismatch inference errors." + }, + { + "left": "issue:43388", + "right": "issue:46032", + "accept": false, + "reason": "Metric gathering drops labels in the last batch; the other is incorrect cache behavior for seq_len>1. Not the same bug." + }, + { + "left": "issue:44479", + "right": "issue:45834", + "accept": false, + "reason": "Both are multimodal input shape issues, but they affect different models and different preprocessing paths." + }, + { + "left": "issue:43825", + "right": "issue:43994", + "accept": false, + "reason": "One is an error-message/translation-support issue; the other is a bad model output regression with AutoModel/pipeline." + }, + { + "left": "issue:43782", + "right": "issue:44977", + "accept": false, + "reason": "Weight-only loading failure and flash-attention generation failure are distinct problems." + }, + { + "left": "issue:43582", + "right": "issue:46512", + "accept": false, + "reason": "Apple Silicon warmup TypeError is unrelated to stale config dtype after from_config(dtype=...)." + }, + { + "left": "issue:43577", + "right": "issue:43818", + "accept": false, + "reason": "BLIP2 dtype propagation and Video-LLaVA temporal-attention/weight-sharing are different model-specific bugs." + }, + { + "left": "issue:44991", + "right": "issue:46291", + "accept": false, + "reason": "Tokenizer loading failure and kernels dependency/version import failure are unrelated." + }, + { + "left": "issue:43421", + "right": "issue:44811", + "accept": false, + "reason": "Post-processor not updating after special-token changes is different from batch_decode ignoring skip_special_tokens." + }, + { + "left": "issue:44561", + "right": "issue:45020", + "accept": false, + "reason": "Both involve recent-version remote_code breakage, but the evidence is too broad to treat them as the same concrete issue." + }, + { + "left": "issue:43716", + "right": "issue:44008", + "accept": false, + "reason": "Image preprocessor/model dtype mismatch is unrelated to a variable-name collision in Gemma3nModel.forward()." + }, + { + "left": "issue:43950", + "right": "issue:45987", + "accept": false, + "reason": "Silent corruption of non-persistent buffers is a different regression from StaticCache.get_seq_length returning the wrong type." + }, + { + "left": "issue:43931", + "right": "issue:46846", + "accept": false, + "reason": "Qwen3-VL weight-shape loading errors and Qwen3.5 TP-plan OOM/shape issues are not the same bug." + }, + { + "left": "issue:46294", + "right": "issue:46463", + "accept": false, + "reason": "Qwen3.5 video chat-template mismatch and T5 decoder-input validation messaging are unrelated." + }, + { + "left": "issue:43825", + "right": "issue:43827", + "accept": false, + "reason": "One concerns a runtime pipeline error message; the other is documentation still referencing pipeline() after removals." + }, + { + "left": "issue:28218", + "right": "issue:29159", + "accept": true, + "reason": "These look like the same tokenizer regression: extra spacing after added tokens fits the broader slow/fast tokenizer inconsistency report." + }, + { + "left": "issue:44805", + "right": "issue:44936", + "accept": false, + "reason": "Mask/index shape mismatch at inference is unrelated to trainer.evaluate() failing after trainer.train()." + }, + { + "left": "issue:44610", + "right": "issue:45072", + "accept": false, + "reason": "Processor image-size mismatch and dtype mismatch inference failures are different issues." + }, + { + "left": "issue:43540", + "right": "issue:45381", + "accept": false, + "reason": "Both are video-input bugs, but they affect different Qwen model families and different failure modes." + }, + { + "left": "issue:43720", + "right": "issue:45072", + "accept": false, + "reason": "Packed-weight loading during accelerate is unrelated to dtype mismatches in inference." + }, + { + "left": "issue:43749", + "right": "issue:44291", + "accept": false, + "reason": "FSDP CPU RAM-efficient loading is a different loading path from init_empty_weights rejecting an unexpected argument." + }, + { + "left": "issue:43577", + "right": "issue:43582", + "accept": false, + "reason": "These are unrelated dtype/model-loading and Apple Silicon allocator problems." + }, + { + "left": "issue:43645", + "right": "issue:46665", + "accept": false, + "reason": "Jupyter notebook custom-model initialization and from_pretrained-inside-classes are separate contexts with different failure causes." + }, + { + "left": "issue:44671", + "right": "issue:45701", + "accept": false, + "reason": "Masked LM prediction regression and tokenizer behavior changes are too different to merge as one bug." + }, + { + "left": "issue:42175", + "right": "issue:45081", + "accept": false, + "reason": "Package/backend install issues are unrelated to a Mistral tokenizer regex patch crash." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays mostly conservative on the soft pairs. The only accepted pair, issue:28218 with issue:29159, is a plausible tokenizer duplicate based on the titles and cluster context. The other verdicts avoid overreach and correctly treat different models, code paths, or failure modes as separate." + }, + "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": 20, + "review_unit_count": 60, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-500", + "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:46697|issue:46728", + "issue:43653|issue:44912", + "issue:43262|issue:43819", + "issue:43782|issue:45325", + "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:44060|issue:44368", + "issue:45084|issue:45538", + "issue:43452|issue:44991", + "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:43638|issue:44936", + "issue:42175|issue:46291", + "issue:42490|issue:43257", + "issue:43720|issue:44610" + ], + "prepared_review_unit_hash": "4b884a015de29ab89f0312a8a215e01ccef9cb2ab4407b0f7154239fde6a38f1", + "component_budget": { + "node_count": 500, + "item_count": 500, + "soft_pair_count": 1839, + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 28, + "serialized_chars": 36524, + "estimated_input_tokens": 9131, + "estimated_eval_tokens": 18518 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "All suggested soft matches are false positives: the paired issues mostly share only broad themes like loading, trainers, or image processors, but they describe different models, different code paths, or different failure modes. None of these pairs look like the same underlying bug/change.", + "confidence": 0.95, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:34634", + "right": "issue:36331", + "accept": false, + "reason": "Unrelated subsystems: BarkProcessor voice presets vs a Trainer compute_loss signature regression." + }, + { + "left": "issue:45003", + "right": "issue:46285", + "accept": false, + "reason": "Both touch modeling code, but one is unsafe sys.modules access and the other is CLIP text-model structure mismatch; different bugs." + }, + { + "left": "issue:46697", + "right": "issue:46728", + "accept": false, + "reason": "Both are image-processor counting/padding issues, but for different models and different mechanisms." + }, + { + "left": "issue:43653", + "right": "issue:44912", + "accept": false, + "reason": "Tokenizer special-token registration vs MXFP4 quantization loading; no shared bug." + }, + { + "left": "issue:43262", + "right": "issue:43819", + "accept": false, + "reason": "Audio-related, but one is chat-template sampling-rate defaulting and the other is DAC latent reconstruction/STE behavior." + }, + { + "left": "issue:43782", + "right": "issue:45325", + "accept": false, + "reason": "Both involve Qwen VL models, but one is a from_pretrained weight_only load error and the other is a rope index scaling bug." + }, + { + "left": "issue:45081", + "right": "issue:45406", + "accept": false, + "reason": "Both are processor/tokenizer loading problems, but they fail on different missing attributes and different model families." + }, + { + "left": "issue:45372", + "right": "issue:46326", + "accept": false, + "reason": "One is an import dependency break in mistral_common; the other is continue_final_message behavior for a different tokenizer path." + }, + { + "left": "issue:43582", + "right": "issue:44857", + "accept": false, + "reason": "A platform-specific allocator TypeError and a CUDA AMP loss crash are not the same code-path problem." + }, + { + "left": "issue:44589", + "right": "issue:45042", + "accept": false, + "reason": "Float8 storage lookup failure vs PIL backend torchvision dependency regression; unrelated." + }, + { + "left": "issue:43550", + "right": "issue:44912", + "accept": false, + "reason": "Different model/runtime bugs: torch.compile with SDPA vs MXFP4 loading fallback." + }, + { + "left": "issue:43742", + "right": "issue:45020", + "accept": false, + "reason": "A specific MobileLLM load key error is not the same as the broad remote_code regression." + }, + { + "left": "issue:43329", + "right": "issue:45600", + "accept": false, + "reason": "Video multimodal token counting bug vs auto_mappings referencing removed configs; different affected code." + }, + { + "left": "issue:44060", + "right": "issue:44368", + "accept": false, + "reason": "Both mention tied-weights warnings, but they concern different Qwen variants and different causes." + }, + { + "left": "issue:45084", + "right": "issue:45538", + "accept": false, + "reason": "Template compilation error and CLIPTokenizer max_length issue are unrelated." + }, + { + "left": "issue:43452", + "right": "issue:44991", + "accept": false, + "reason": "Both are from_pretrained loading failures, but one is gguf_file handling and the other is a tokenizer regression for a specific model." + }, + { + "left": "issue:43526", + "right": "issue:43697", + "accept": false, + "reason": "One is a BeitImageProcessorFast label-reduction bug; the other is a model output mismatch in RTDetrV2." + }, + { + "left": "issue:43257", + "right": "issue:43329", + "accept": false, + "reason": "Qwen3 MoE loading with accelerate/deepspeed is unrelated to multimodal token counting in video branches." + }, + { + "left": "issue:43526", + "right": "issue:44554", + "accept": false, + "reason": "Different domains and failure modes: 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": "Both are API breakages, but import path breakage and Trainer signature change are distinct regressions." + }, + { + "left": "issue:43819", + "right": "issue:43927", + "accept": false, + "reason": "DAC latent decoding mismatch vs DiaConfig token-ID persistence; different components." + }, + { + "left": "issue:36010", + "right": "issue:43645", + "accept": false, + "reason": "GenerationMixin import failure is unrelated to custom-model initialization in notebooks." + }, + { + "left": "issue:43531", + "right": "issue:45042", + "accept": false, + "reason": "Qwen3-MoE sliding_window bug and PIL/torchvision dependency regression are unrelated." + }, + { + "left": "issue:43638", + "right": "issue:44936", + "accept": false, + "reason": "Training/evaluation failures, but one is a deepspeed zero3 index error and the other is evaluate-after-train behavior; no clear same root cause." + }, + { + "left": "issue:42175", + "right": "issue:46291", + "accept": false, + "reason": "Different packaging/import failures: missing TensorFlow backend in install extras vs kernels version/revision requirement." + }, + { + "left": "issue:42490", + "right": "issue:43257", + "accept": false, + "reason": "Local-path trailing-slash import warning and Qwen3 MoE accelerate/deepspeed loading are separate loading bugs." + }, + { + "left": "issue:43720", + "right": "issue:44610", + "accept": false, + "reason": "Packed-weight unpacking during accelerate loading and OmDet-Turbo processor 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": 60, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-500", + "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:43454", + "issue:43525", + "issue:43540", + "issue:43577", + "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:45042", + "issue:45356", + "issue:45459", + "issue:45759", + "issue:45803", + "issue:45859", + "issue:46291", + "issue:46421", + "issue:46424", + "issue:46701" + ], + "soft_pairs": [ + "issue:39692|issue:44912", + "issue:43746|issue:44811", + "issue:43854|issue:44977", + "issue:43643|issue:45759", + "issue:38175|issue:44610", + "issue:43257|issue:43696", + "issue:43262|issue:44811", + "issue:44387|issue:44617", + "issue:43257|issue:43931", + "issue:45459|issue:45803", + "issue:43329|issue:43450", + "issue:46421|issue:46424", + "issue:43329|issue:44479", + "issue:42915|issue:43742", + "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:43540|issue:46701", + "issue:28282|issue:29870", + "issue:43454|issue:43577", + "issue:43525|issue:46291", + "issue:44568|issue:45356" + ], + "prepared_review_unit_hash": "e2e421971c3dbc471809f2450382084f156e860c7eae227c3ab9466f61a5b4b4", + "component_budget": { + "node_count": 500, + "item_count": 500, + "soft_pair_count": 1839, + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 27, + "serialized_chars": 36014, + "estimated_input_tokens": 9004, + "estimated_eval_tokens": 18264 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a grab bag of mostly unrelated closed issues across loading, tokenization, multimodal/video, quantization, cache, and CPU-offload regressions. None of the soft pairs look like true duplicates; they share only broad subsystem keywords, not the same concrete bug path.", + "confidence": 0.78, + "canonical_issue_reason": "issue:43749 is the strongest anchor: it has the most discussion/inbound references and describes a concrete, high-impact loading regression with a clearly reproducible scope.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:43749 is the best representative issue for the cluster because it is well-scoped, impactful, and more central than the other candidates; it is also easier to use as a triage anchor than the narrower model-specific reports.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:39692", + "right": "issue:44912", + "accept": false, + "reason": "Different models and failures: SigLIP2 doc-example errors vs gpt-oss-20b MXFP4 quantization fallback/OOM behavior." + }, + { + "left": "issue:43746", + "right": "issue:44811", + "accept": false, + "reason": "Different code paths: PEFT local-checkpoint loading for GraniteSpeech vs Whisper processor batch_decode skip_special_tokens handling." + }, + { + "left": "issue:43854", + "right": "issue:44977", + "accept": false, + "reason": "Different model families and bugs: GLM-4.7-Flash unit-test load failure vs Qwen3.5 flash-attention generation regression." + }, + { + "left": "issue:43643", + "right": "issue:45759", + "accept": false, + "reason": "Both involve config loading, but one is missing fields with trust_remote_code and the other is text_config unwrapping for composite multimodal configs; not the same bug." + }, + { + "left": "issue:38175", + "right": "issue:44610", + "accept": false, + "reason": "Different models and symptoms: siglip2 zero probabilities vs OmDet-Turbo processor/model input-size mismatch." + }, + { + "left": "issue:43257", + "right": "issue:43696", + "accept": false, + "reason": "Qwen3 MOE weight conversion under accelerate+deepspeed is unrelated to GPT-oss-20b CUDA OOM." + }, + { + "left": "issue:43262", + "right": "issue:44811", + "accept": false, + "reason": "Audio template sampling-rate defaulting and Whisper decode token filtering are separate processor behaviors." + }, + { + "left": "issue:44387", + "right": "issue:44617", + "accept": false, + "reason": "Quantization memory regression in Transformers 5.x is not the same as Sam3Video model-specific OOM." + }, + { + "left": "issue:43257", + "right": "issue:43931", + "accept": false, + "reason": "Different Qwen-related failures: MOE weight conversion vs Qwen3-VL shape mismatch on load." + }, + { + "left": "issue:45459", + "right": "issue:45803", + "accept": false, + "reason": "Both mention hidden errors, but they are different exception-handling bugs in different components and code paths." + }, + { + "left": "issue:43329", + "right": "issue:43450", + "accept": false, + "reason": "Undefined helper/variable usage in multimodal token counting is not the same as batched video processor shape regression." + }, + { + "left": "issue:46421", + "right": "issue:46424", + "accept": false, + "reason": "Both are cache-related, but one is lazy-init recompilation and the other is max_cache_length applicability." + }, + { + "left": "issue:43329", + "right": "issue:44479", + "accept": false, + "reason": "The first is an undefined-name bug in token counting; the second is a broader video-input regression across Qwen VL variants." + }, + { + "left": "issue:42915", + "right": "issue:43742", + "accept": false, + "reason": "Different failures: FineGrainedFP8Config handling for Qwen3Moe vs MobileLLM-125M load key error." + }, + { + "left": "issue:43749", + "right": "issue:44756", + "accept": false, + "reason": "Both are loading/memory issues, but one is FSDP CPU RAM efficient loading and the other is mmap behavior on Strix Halo." + }, + { + "left": "issue:34567", + "right": "issue:40990", + "accept": false, + "reason": "TrainerState token counters not updating is unrelated to GPT-oss-20b perplexity on WikiText-2." + }, + { + "left": "issue:43650", + "right": "issue:43906", + "accept": false, + "reason": "No shared bug: placeholder issue text versus an isolated reproduction of a different upstream issue." + }, + { + "left": "issue:44291", + "right": "issue:45042", + "accept": false, + "reason": "Model init_empty_weights argument handling is unrelated to PIL image processor torchvision dependency regression." + }, + { + "left": "issue:42915", + "right": "issue:44821", + "accept": false, + "reason": "GPT-oss-20b loading/quantization issues do not match AutoImageProcessor URL-loading failure." + }, + { + "left": "issue:29159", + "right": "issue:34567", + "accept": false, + "reason": "Tokenizer slow/fast inconsistencies are unrelated to TrainerState token-count tracking." + }, + { + "left": "issue:43606", + "right": "issue:45859", + "accept": false, + "reason": "Both mention CPU offload, but they affect different models and likely different offload paths." + }, + { + "left": "issue:43927", + "right": "issue:44912", + "accept": false, + "reason": "DiaConfig save/load token-ID loss is unrelated to GPT-oss-20b MXFP4 quantization fallback." + }, + { + "left": "issue:43540", + "right": "issue:46701", + "accept": false, + "reason": "Different modalities and failures: Qwen3OmniMoe video-input ValueError vs DiffusionGemma pipeline output issue." + }, + { + "left": "issue:28282", + "right": "issue:29870", + "accept": false, + "reason": "PyTorch-missing ImportError and Bart generate() decoder_attention_mask UnboundLocalError are unrelated." + }, + { + "left": "issue:43454", + "right": "issue:43577", + "accept": false, + "reason": "AyaVision weight-tying causing garbage text and Blip2 dtype propagation are different model-loading bugs." + }, + { + "left": "issue:43525", + "right": "issue:46291", + "accept": false, + "reason": "Missing pad_token_id in Llama4Config is unrelated to pipeline import failing on kernels version/revision requirements." + }, + { + "left": "issue:44568", + "right": "issue:45356", + "accept": false, + "reason": "Both are tokenizer regressions, but they concern different tokenizer implementations, models, and failure modes." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 22, + "review_unit_count": 60, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-500", + "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:43540", + "issue:43577", + "issue:43582", + "issue:43697", + "issue:43742", + "issue:43749", + "issue:43824", + "issue:44062", + "issue:44291", + "issue:44479", + "issue:44488", + "issue:44560", + "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:44987|issue:45538", + "issue:28282|issue:34689", + "issue:45538|issue:46869", + "issue:44857|issue:46832", + "issue:43697|issue:45357", + "issue:43425|issue:46291", + "issue:33357|issue:34567", + "issue:43404|issue:45127", + "issue:42915|issue:45538", + "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:43577|issue:46512", + "issue:42915|issue:43749", + "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:43540|issue:44560", + "issue:44792|issue:44964" + ], + "prepared_review_unit_hash": "1102b6c8863cab3a7efd83da51ccf3c9883996c7d3cb5a6ce50845665e7c6267", + "component_budget": { + "node_count": 500, + "item_count": 500, + "soft_pair_count": 1839, + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 28, + "serialized_chars": 36020, + "estimated_input_tokens": 9005, + "estimated_eval_tokens": 18266 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is mostly a set of unrelated issue candidates that only share broad Transformers theme words. I found no strong duplicate PRs, and only one possible soft duplicate is still too weak to accept under strict triage.", + "confidence": 0.91, + "canonical_issue_reason": "issue:45488 is the clearest, most actionable report: it is a concrete regression with a precise title, active discussion, and inbound references, making it a good cluster representative.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:45488 is the strongest overall issue to keep as the canonical example because it describes a specific user-facing regression with enough detail to triage and verify.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44987", + "right": "issue:45538", + "accept": false, + "reason": "Different problems: one is a model-loading failure for a specific repo, the other is a CLIPTokenizer max-length bug." + }, + { + "left": "issue:28282", + "right": "issue:34689", + "accept": false, + "reason": "Generic missing-PyTorch import error vs a specific Llama 3.2 vision model-loading regression; not the same bug." + }, + { + "left": "issue:45538", + "right": "issue:46869", + "accept": false, + "reason": "Tokenizer configuration bug vs multi-GPU device-map failure; unrelated failure modes." + }, + { + "left": "issue:44857", + "right": "issue:46832", + "accept": false, + "reason": "Both are detection-model issues, but one is AMP/CUDA crash in loss and the other is a forward shape/type error; different code paths." + }, + { + "left": "issue:43697", + "right": "issue:45357", + "accept": false, + "reason": "RT-DETR output regression vs Qwen3.5 save_pretrained key mismatch; different subsystems and symptoms." + }, + { + "left": "issue:43425", + "right": "issue:46291", + "accept": false, + "reason": "Torch compatibility complaint vs kernels package revision/version requirement; unrelated dependency issues." + }, + { + "left": "issue:33357", + "right": "issue:34567", + "accept": false, + "reason": "MacOS bus error in CLIP model loading vs TrainerState token counter not updating; no shared underlying bug." + }, + { + "left": "issue:43404", + "right": "issue:45127", + "accept": false, + "reason": "Both mention tied weights, but one is a missing tie in a specific model class and the other is LoRA merge corruption with extended vocab; not the same fix." + }, + { + "left": "issue:42915", + "right": "issue:45538", + "accept": false, + "reason": "FineGrainedFP8Config failure in Qwen3Moe vs CLIPTokenizer max_length bug; unrelated." + }, + { + "left": "issue:45137", + "right": "issue:45405", + "accept": false, + "reason": "DeepSpeed ZeRO3 deque crash vs PEFT version bump/release mismatch; completely different issues." + }, + { + "left": "issue:42175", + "right": "issue:43824", + "accept": false, + "reason": "Packaging/backend install problem vs missing Qwen2.5-VL symbol import; different dependency layers." + }, + { + "left": "issue:28282", + "right": "issue:29127", + "accept": false, + "reason": "Missing PyTorch import error vs LayoutLMv3 error-message clarity issue; not duplicates." + }, + { + "left": "issue:44062", + "right": "issue:44991", + "accept": false, + "reason": "These are both tokenizer-loading failures, but the evidence is too broad to confirm the same root cause from the titles alone." + }, + { + "left": "issue:44479", + "right": "issue:46693", + "accept": false, + "reason": "Video-input regression for Qwen models vs FA2 long-context performance regression; different symptoms and likely different code paths." + }, + { + "left": "issue:43742", + "right": "issue:44291", + "accept": false, + "reason": "MobileLLM key error vs unexpected _is_hf_initialized argument during init_empty_weights; both are load-time errors but not the same bug." + }, + { + "left": "issue:44936", + "right": "issue:46821", + "accept": false, + "reason": "Trainer evaluate-after-train failure vs TimesFM window_size attribute error; unrelated components." + }, + { + "left": "issue:43577", + "right": "issue:46512", + "accept": false, + "reason": "Model/QFormer dtype staying float32 vs stale config.dtype after from_config; related theme but different bugs and fixes." + }, + { + "left": "issue:42915", + "right": "issue:43749", + "accept": false, + "reason": "Qwen3Moe FP8 config failure vs FSDP CPU RAM efficient loading breakage; different features." + }, + { + "left": "issue:44849", + "right": "issue:45701", + "accept": false, + "reason": "Qwen3.5 output_hidden_states bug vs version-dependent tokenization changes; unrelated." + }, + { + "left": "issue:43452", + "right": "issue:43582", + "accept": false, + "reason": "gguf_file loader breakage vs Apple Silicon allocator TypeError; different areas." + }, + { + "left": "issue:27957", + "right": "issue:28282", + "accept": false, + "reason": "Flash Attention 2 issue in XLMRoberta vs missing-PyTorch import error; unrelated." + }, + { + "left": "issue:24643", + "right": "issue:29159", + "accept": false, + "reason": "DeepSpeed training RuntimeError vs tokenizer slow/fast inconsistency; no shared code-path problem." + }, + { + "left": "issue:44987", + "right": "issue:45907", + "accept": false, + "reason": "Model-loading regression vs list_repo_files revision-forwarding bug; not duplicates." + }, + { + "left": "issue:41720", + "right": "issue:43540", + "accept": false, + "reason": "Auto device-mapping CUDA assert vs video-input processing ValueError in Qwen3OmniMoe; distinct failures." + }, + { + "left": "issue:42175", + "right": "issue:44488", + "accept": false, + "reason": "Backend installation packaging issue vs a model-loading failure for a specific repository; different problems." + }, + { + "left": "issue:44568", + "right": "issue:45488", + "accept": false, + "reason": "Both are tokenizer regressions, but one is BOS/EOS insertion behavior and the other is ByteLevel pre-tokenizer being overridden; not the same bug." + }, + { + "left": "issue:44792", + "right": "issue:44964", + "accept": false, + "reason": "Janus image-generation test failure vs Phi-4 multimodal model-loading issue; different models and failure points." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 23, + "review_unit_count": 60, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-500", + "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:44568", + "issue:44589", + "issue:44625", + "issue:44737", + "issue:44821", + "issue:44877", + "issue:44898", + "issue:45042", + "issue:45072", + "issue:45081", + "issue:45325", + "issue:45357", + "issue:45381", + "issue:45405", + "issue:45600", + "issue:45672", + "issue:45674", + "issue:45823", + "issue:45920", + "issue:46399", + "issue:46463", + "issue:46468", + "issue:46491", + "issue:46693", + "issue:46701", + "issue:46821", + "issue:46829" + ], + "soft_pairs": [ + "issue:44479|issue:45381", + "issue:38175|issue:44898", + "issue:44291|issue:44821", + "issue:43540|issue:44479", + "issue:34567|issue:36331", + "issue:43257|issue:43756", + "issue:44062|issue:45081", + "issue:45042|issue:45081", + "issue:43582|issue:46821", + "issue:43742|issue:44589", + "issue:45672|issue:45674", + "issue:43756|issue:46399", + "issue:43756|issue:45325", + "issue:44568|issue:46491", + "issue:45405|issue:45600", + "issue:43716|issue:46701", + "issue:43577|issue:45072", + "issue:43531|issue:44821", + "issue:43906|issue:46468", + "issue:24643|issue:29870", + "issue:44737|issue:46463", + "issue:43906|issue:46693", + "issue:45823|issue:46463", + "issue:43452|issue:44488", + "issue:43697|issue:46829", + "issue:45357|issue:45920", + "issue:44625|issue:44877" + ], + "prepared_review_unit_hash": "54d75f45b823f8e8382284e7b026704fb9f63638c384c64c6d9a5d9a768355a5", + "component_budget": { + "node_count": 500, + "item_count": 500, + "soft_pair_count": 1839, + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 27, + "serialized_chars": 35594, + "estimated_input_tokens": 8899, + "estimated_eval_tokens": 18054 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Most pairs are clearly unrelated. The only plausible duplicate edge is the Qwen video-input regression pair (44479 \u2194 45381), which appears to describe the same underlying position-id bug across related Qwen VL variants.", + "confidence": 0.74, + "canonical_issue_reason": "issue:44479 is the best canonical issue because it is the broadest, most central report of the Qwen video-input regression and already spans multiple affected Qwen variants.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:44479 is the strongest representative issue in this cluster: it describes a concrete regression with multiple affected models and is the most likely target for consolidation.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44479", + "right": "issue:45381", + "accept": true, + "reason": "Both report the same Qwen VL video-input regression around wrong position ids / temporal scaling in v5.3.0; 45381 looks like a narrower manifestation of the broader bug in 44479." + }, + { + "left": "issue:38175", + "right": "issue:44898", + "accept": false, + "reason": "Different models and different failure modes: unexpected zero probabilities in SigLIP2 vs Perceiver non-default-resolution failure." + }, + { + "left": "issue:44291", + "right": "issue:44821", + "accept": false, + "reason": "Unrelated load-time errors: unexpected _is_hf_initialized argument vs AutoImageProcessor URL loading." + }, + { + "left": "issue:43540", + "right": "issue:44479", + "accept": false, + "reason": "Both involve video inputs, but they affect different Qwen model families and do not look like the same underlying bug." + }, + { + "left": "issue:34567", + "right": "issue:36331", + "accept": false, + "reason": "TrainerState token-count tracking and CustomTrainer compute_loss signature incompatibility are separate training issues." + }, + { + "left": "issue:43257", + "right": "issue:43756", + "accept": false, + "reason": "Different subsystems: DeepSpeed/Accelerate MoE weight conversion vs SmolLM3 RoPE layer count regression." + }, + { + "left": "issue:44062", + "right": "issue:45081", + "accept": false, + "reason": "Tokenizer AddedToken construction error is unrelated to the Mistral regex patch AttributeError." + }, + { + "left": "issue:45042", + "right": "issue:45081", + "accept": false, + "reason": "Image processor torchvision dependency regression is unrelated to tokenizer regex patching." + }, + { + "left": "issue:43582", + "right": "issue:46821", + "accept": false, + "reason": "Apple Silicon allocator TypeError and TimesFM window_size AttributeError are unrelated." + }, + { + "left": "issue:43742", + "right": "issue:44589", + "accept": false, + "reason": "MobileLLM key error and Float8 storage lookup failure are different loading problems." + }, + { + "left": "issue:45672", + "right": "issue:45674", + "accept": false, + "reason": "Both mention 4-bit inference, but one is a torch.finfo uint8 audio-module error and the other is a BitsAndBytesConfig skip-module exclusion bug." + }, + { + "left": "issue:43756", + "right": "issue:46399", + "accept": false, + "reason": "SmolLM3 RoPE-layer issue and Qwen3.5 kernel-registration crash are unrelated." + }, + { + "left": "issue:43756", + "right": "issue:45325", + "accept": false, + "reason": "Both mention RoPE, but the affected models and concrete failures are different." + }, + { + "left": "issue:44568", + "right": "issue:46491", + "accept": false, + "reason": "Tokenizer special-token insertion regression and CodeLlama round-trip spacing regression are different tokenizer bugs." + }, + { + "left": "issue:45405", + "right": "issue:45600", + "accept": false, + "reason": "Released-version requirement mismatch and removed-config reference in auto_mappings are unrelated CI/release issues." + }, + { + "left": "issue:43716", + "right": "issue:46701", + "accept": false, + "reason": "Different models and pipelines: Mistral-3 dtype mismatch vs DiffusionGemma image-text-to-text pipeline failure." + }, + { + "left": "issue:43577", + "right": "issue:45072", + "accept": false, + "reason": "Both are dtype-related, but they concern different models/code paths and do not look mergeable as one fix." + }, + { + "left": "issue:43531", + "right": "issue:44821", + "accept": false, + "reason": "Qwen3-MoE sliding_window behavior is unrelated to AutoImageProcessor URL loading." + }, + { + "left": "issue:43906", + "right": "issue:46468", + "accept": false, + "reason": "An isolated reproduction of a prior issue is not the same as the Nemotron example failure." + }, + { + "left": "issue:24643", + "right": "issue:29870", + "accept": false, + "reason": "DeepSpeed weight-dimension error and BART generate decoder mask initialization error are unrelated." + }, + { + "left": "issue:44737", + "right": "issue:46463", + "accept": false, + "reason": "XLNet CPU-side arange bug and T5 decoder-input validation messaging are unrelated." + }, + { + "left": "issue:43906", + "right": "issue:46693", + "accept": false, + "reason": "Prior-issue reproduction and FA2 long-context performance regression are unrelated." + }, + { + "left": "issue:45823", + "right": "issue:46463", + "accept": false, + "reason": "Gemma4 device mismatch and T5 decoder validation message are unrelated." + }, + { + "left": "issue:43452", + "right": "issue:44488", + "accept": false, + "reason": "gguf_file loading breakage and sleng-bert load failure are different loader problems." + }, + { + "left": "issue:43697", + "right": "issue:46829", + "accept": false, + "reason": "RTDetrV2 output drift and VideoPrism hidden_states field mix-up are different model-output bugs." + }, + { + "left": "issue:45357", + "right": "issue:45920", + "accept": false, + "reason": "Qwen3.5 save_pretrained key regression and tokenizer token-id regression affect different code paths." + }, + { + "left": "issue:44625", + "right": "issue:44877", + "accept": false, + "reason": "Qwen3.5 num_labels propagation and granite_speech strict-config loading are separate config issues." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The cluster summary is mostly grounded in the packet, and the only accepted soft edge (44479 \u2194 45381) is a conservative duplicate call based on strong title overlap within the same Qwen video-input regression family. I did not see evidence of overstated mergeability for any other pair." + }, + "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": 60, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-500", + "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:45356", + "issue:45372", + "issue:45375", + "issue:45479", + "issue:45542", + "issue:45588", + "issue:45663", + "issue:46050", + "issue:46121", + "issue:46170", + "issue:46178", + "issue:46463", + "issue:46693", + "issue:46701", + "issue:46829", + "issue:46846", + "issue:46899" + ], + "soft_pairs": [ + "issue:45663|issue:46846", + "issue:43540|issue:46693", + "issue:44913|issue:46121", + "issue:44737|issue:45588", + "issue:28282|issue:32090", + "issue:46463|issue:46899", + "issue:43697|issue:45216", + "issue:43526|issue:45479", + "issue:43720|issue:44898", + "issue:43643|issue:45375", + "issue:43761|issue:46829", + "issue:45200|issue:46463", + "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:43688|issue:43749", + "issue:43819|issue:44811", + "issue:38175|issue:43295", + "issue:43994|issue:44898", + "issue:43525|issue:46701", + "issue:45372|issue:46170", + "issue:45542|issue:46178", + "issue:43295|issue:45356" + ], + "prepared_review_unit_hash": "81b8c520fc3e77676de82ba99032f07922ccf9722ff4b9cb0e5d595da142b8ae", + "component_budget": { + "node_count": 500, + "item_count": 500, + "soft_pair_count": 1839, + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 26, + "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": "This cluster is mostly a grab-bag of unrelated bug reports. The soft pairs share superficial vocabulary (loading, tokenizer, config, vision, etc.) but not the same concrete code path, so none look safe to merge as duplicates.", + "confidence": 0.94, + "canonical_issue_reason": "Issue 43749 is the broadest and most discussed loading/regression report here, so it is the closest thing to a cluster anchor, but the set is not a true duplicate cluster.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 43749 has the strongest activity and broad applicability among the items, making it the best representative issue in this mixed cluster.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:45663", + "right": "issue:46846", + "accept": false, + "reason": "Different models and failures: Gemma-4 FSDP2 KeyError vs Qwen3.5 TP-plan OOM/shape error. Only both are distributed-training themed." + }, + { + "left": "issue:43540", + "right": "issue:46693", + "accept": false, + "reason": "One is a video-input validation error in Qwen3OmniMoe; the other is a long-context FlashAttention performance regression. Different bug classes." + }, + { + "left": "issue:44913", + "right": "issue:46121", + "accept": false, + "reason": "Both touch RoPE-related code, but one is config reload persistence and the other is JSON-loaded list handling in validation. Not the same defect." + }, + { + "left": "issue:44737", + "right": "issue:45588", + "accept": false, + "reason": "XLNet CPU placement bug and flash-attention None crash are unrelated code paths and symptoms." + }, + { + "left": "issue:28282", + "right": "issue:32090", + "accept": false, + "reason": "ImportError for missing PyTorch and Trainer GPU broadcast TypeError are separate environment/runtime issues." + }, + { + "left": "issue:46463", + "right": "issue:46899", + "accept": false, + "reason": "T5 decoder-input error-message issue vs Gemma-4 4-bit quantization dtype cast bug. Different models and root causes." + }, + { + "left": "issue:43697", + "right": "issue:45216", + "accept": false, + "reason": "Both are regressions, but one is RTDetrV2 output mismatch and the other is Qwen3.5 checkpoint saving corruption. Not the same code path." + }, + { + "left": "issue:43526", + "right": "issue:45479", + "accept": false, + "reason": "BeitImageProcessorFast label reduction bug is unrelated to the sequence-classification zero-loss problem." + }, + { + "left": "issue:43720", + "right": "issue:44898", + "accept": false, + "reason": "BitNet packed-weight loading and Perceiver interpolation failure are different architectures and loading/inference paths." + }, + { + "left": "issue:43643", + "right": "issue:45375", + "accept": false, + "reason": "Both mention missing fields, but one is generic AutoConfig with trust_remote_code and the other is a specific Qwen3_5MoeVisionConfig strict-field drop." + }, + { + "left": "issue:43761", + "right": "issue:46829", + "accept": false, + "reason": "Both involve output fields, but CLIPVisionModel hidden_states regression and VideoPrism output mis-assignment are separate model-specific bugs." + }, + { + "left": "issue:45200", + "right": "issue:46463", + "accept": false, + "reason": "Gemma 4 mm_token_type_ids defaulting and T5 decoder-input validation messaging are unrelated." + }, + { + "left": "issue:44466", + "right": "issue:45005", + "accept": false, + "reason": "Both relate loosely to tied weights, but one is device-dependent serialization and the other is translation-model tied-weight handling. Too broad to merge." + }, + { + "left": "issue:44361", + "right": "issue:44442", + "accept": false, + "reason": "Different tokenizers and failures: MLukeTokenizer task AttributeError vs AutoTokenizer loading FastSpeech2ConformerTokenizer." + }, + { + "left": "issue:44493", + "right": "issue:46050", + "accept": false, + "reason": "UNEXPECTED Key with position id and sinq quantized save/load breakage are distinct loading/serialization problems." + }, + { + "left": "issue:43653", + "right": "issue:44361", + "accept": false, + "reason": "BigBirdTokenizer special-token decode bug is unrelated to MLukeTokenizer task AttributeError." + }, + { + "left": "issue:41762", + "right": "issue:44936", + "accept": false, + "reason": "Gemma3 ZeRO-3 load-time IndexError and trainer.evaluate() after train() are not the same failure mode or subsystem." + }, + { + "left": "issue:43452", + "right": "issue:44843", + "accept": false, + "reason": "Both affect from_pretrained, but gguf_file support breakage and offline model_info calls are different concrete bugs." + }, + { + "left": "issue:43688", + "right": "issue:43749", + "accept": false, + "reason": "Aux-loss normalization in OLMoE/GPT Oss is unrelated to FSDP CPU RAM-efficient loading." + }, + { + "left": "issue:43819", + "right": "issue:44811", + "accept": false, + "reason": "DAC latent reconstruction mismatch and Whisper skip_special_tokens decode behavior are different model APIs and bugs." + }, + { + "left": "issue:38175", + "right": "issue:43295", + "accept": false, + "reason": "Siglip2 zero-probabilities and processor.tokenizer/image forwarding regression are unrelated issues." + }, + { + "left": "issue:43994", + "right": "issue:44898", + "accept": false, + "reason": "Both are vision-related, but Siglip2 nonsensical outputs and Perceiver interpolation failure are different models and root causes." + }, + { + "left": "issue:43525", + "right": "issue:46701", + "accept": false, + "reason": "Llama4Config pad_token_id missing and DiffusionGemma image-text-to-text pipeline breakage are unrelated." + }, + { + "left": "issue:45372", + "right": "issue:46170", + "accept": false, + "reason": "Gemma4 processor import failure via mistral_common and Windows ROCm torch.distributed dependency issues are separate dependency problems." + }, + { + "left": "issue:45542", + "right": "issue:46178", + "accept": false, + "reason": "Both concern backend availability, but tensorboard-without-TF vs tensorflow_text backend lookup are not the same bug." + }, + { + "left": "issue:43295", + "right": "issue:45356", + "accept": false, + "reason": "Both are tokenizer regressions, but the reported behaviors and implicated code paths are different." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 25, + "review_unit_count": 60, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-500", + "nodes": [ + "issue:29159", + "issue:32090", + "issue:33357", + "issue:33666", + "issue:36010", + "issue:41720", + "issue:42898", + "issue:43452", + "issue:43504", + "issue:43531", + "issue:43550", + "issue:43577", + "issue:43643", + "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:44554", + "issue:44561", + "issue:44568", + "issue:44625", + "issue:44704", + "issue:44734", + "issue:44743", + "issue:44821", + "issue:44843", + "issue:44977", + "issue:44991", + "issue:45440", + "issue:45491", + "issue:45701", + "issue:45704", + "issue:46002", + "issue:46050", + "issue:46402", + "issue:46491", + "issue:46612", + "issue:46693", + "issue:46821" + ], + "soft_pairs": [ + "issue:43531|issue:43742", + "issue:42898|issue:46491", + "issue:29159|issue:33357", + "issue:43653|issue:44568", + "issue:43646|issue:43749", + "issue:43452|issue:43577", + "issue:43756|issue:46402", + "issue:43688|issue:45704", + "issue:44843|issue:44991", + "issue:43653|issue:43927", + "issue:44464|issue:44743", + "issue:43688|issue:45701", + "issue:44484|issue:44704", + "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:44734|issue:46612", + "issue:44466|issue:46050", + "issue:43504|issue:43720", + "issue:43504|issue:43550", + "issue:44977|issue:46693", + "issue:44464|issue:44977", + "issue:41720|issue:44464", + "issue:43643|issue:44561", + "issue:43577|issue:44554" + ], + "prepared_review_unit_hash": "dcec38aa04d8b0b0ac8eab02d339ccfa4559e434202d7c224374ba11a4f039b9", + "component_budget": { + "node_count": 500, + "item_count": 500, + "soft_pair_count": 1839, + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 28, + "serialized_chars": 35538, + "estimated_input_tokens": 8885, + "estimated_eval_tokens": 18026 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This is a loose, heterogeneous set of Transformers issues: tokenization regressions, model-loading failures, generation/cache bugs, and a few unrelated performance/config problems. None of the soft pairs look like the same concrete bug or change.", + "confidence": 0.91, + "canonical_issue_reason": "If a representative issue is needed, issue 44464 is the best anchor: it is a concrete, reproducible generation regression and sits closest to several other generation/cache-related soft matches in this set.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 44464 is the most usable representative because it is specific, actionable, and broadly similar in theme to the other generation-related items without being overly model-specific.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43531", + "right": "issue:43742", + "accept": false, + "reason": "Different bug classes and models: sliding_window/recurrent-state behavior in Qwen3-MoE vs a key error while loading MobileLLM-125M." + }, + { + "left": "issue:42898", + "right": "issue:46491", + "accept": false, + "reason": "Both are tokenizer regressions, but 42898 is a broad v5 clean_up_tokenization_spaces change while 46491 is a CodeLlama leading-space round-trip bug; different code path and symptom." + }, + { + "left": "issue:29159", + "right": "issue:33357", + "accept": false, + "reason": "Slow-vs-fast tokenizer inconsistency is unrelated to the CLIP MacOS bus error." + }, + { + "left": "issue:43653", + "right": "issue:44568", + "accept": false, + "reason": "Both concern special tokens, but one is BigBird mask-token registration and the other is mdeberta BOS/EOS insertion; not the same underlying bug." + }, + { + "left": "issue:43646", + "right": "issue:43749", + "accept": false, + "reason": "AutoConfig trust_remote_code missing fields is unrelated to FSDP_CPU_RAM_EFFICIENT_LOADING being broken." + }, + { + "left": "issue:43452", + "right": "issue:43577", + "accept": false, + "reason": "gguf AutoTokenizer/AutoModel loading failure and BLIP2 dtype propagation are distinct loading issues with different root causes." + }, + { + "left": "issue:43756", + "right": "issue:46402", + "accept": false, + "reason": "Missing RoPE layers in Smollm3 and Sam3 text-encoder weights not loading are different model-specific load bugs." + }, + { + "left": "issue:43688", + "right": "issue:45704", + "accept": false, + "reason": "Auxiliary-loss normalization and a T5 apex.FusedRMSNorm memory leak are unrelated bugs." + }, + { + "left": "issue:44843", + "right": "issue:44991", + "accept": false, + "reason": "Both are tokenizer-loading regressions, but one is an offline-mode model_info() call in _patch_mistral_regex and the other is a specific tokenizer load failure for EMBEDDIA/est-roberta." + }, + { + "left": "issue:43653", + "right": "issue:43927", + "accept": false, + "reason": "Both involve token handling, but 43653 is a missing special-token registration and 43927 is custom token IDs being lost across save/load." + }, + { + "left": "issue:44464", + "right": "issue:44743", + "accept": false, + "reason": "Both are generation/cache issues, but 44464 is compiled-forward chunked generation inconsistency while 44743 is recurrent state resetting with cache in modular_qwen3_5." + }, + { + "left": "issue:43688", + "right": "issue:45701", + "accept": false, + "reason": "Aux-loss normalization is unrelated to a general version-dependent tokenization change." + }, + { + "left": "issue:44484", + "right": "issue:44704", + "accept": false, + "reason": "save_pretrained shard-size behavior and AutoProcessor kwarg forwarding are different API issues." + }, + { + "left": "issue:45491", + "right": "issue:46002", + "accept": false, + "reason": "Both mention NaNs, but Gemma3 mixed-length sliding-window attention on GPU is a different failure mode from NaNs in classification heads after checkout/init." + }, + { + "left": "issue:44479", + "right": "issue:44625", + "accept": false, + "reason": "A video-input regression in Qwen VL models is unrelated to 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." + }, + { + "left": "issue:43688", + "right": "issue:45440", + "accept": false, + "reason": "Aux-loss normalization and native DeepseekV3MoE divergence are different correctness issues in different code paths." + }, + { + "left": "issue:33666", + "right": "issue:36010", + "accept": false, + "reason": "Qwen2-VL multi-GPU training and the GenerationMixin import error are unrelated." + }, + { + "left": "issue:32090", + "right": "issue:33666", + "accept": false, + "reason": "Trainer GPU broadcast of NoneType and Qwen2-VL multi-GPU training are not the same bug." + }, + { + "left": "issue:44734", + "right": "issue:46612", + "accept": false, + "reason": "Both involve generation state, but 44734 is a KV-cache continuation indexing crash in serve while 46612 is beam-search cache reorder being skipped for specific models." + }, + { + "left": "issue:44466", + "right": "issue:46050", + "accept": false, + "reason": "Inconsistent lm_head.weight serialization and broken quantized save/load for sinq are both persistence bugs, but they are different mechanisms." + }, + { + "left": "issue:43504", + "right": "issue:43720", + "accept": false, + "reason": "Beit legacy-field loading and BitNet packed-weight unpacking during accelerate loading are distinct load-time failures." + }, + { + "left": "issue:43504", + "right": "issue:43550", + "accept": false, + "reason": "A Beit pretrained-preset load issue is unrelated to Bamba torch.compile+SDPA runtime failure." + }, + { + "left": "issue:44977", + "right": "issue:46693", + "accept": false, + "reason": "Both mention Qwen3.5 and FA2, but 44977 is a functional generation regression while 46693 is a performance regression after 5.2.0." + }, + { + "left": "issue:41720", + "right": "issue:44464", + "accept": false, + "reason": "cudaErrorAssert on auto device mapping for Qwen3 is unrelated to compiled-forward chunked generation inconsistency." + }, + { + "left": "issue:43643", + "right": "issue:44561", + "accept": false, + "reason": "Both touch trust_remote_code, but 43643 is missing fields from AutoConfig and 44561 is a breakage from removing is_torch_fx_available." + }, + { + "left": "issue:43577", + "right": "issue:44554", + "accept": false, + "reason": "BLIP2 dtype propagation and an MPS attention correctness issue with mismatched head dims 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": 26, + "review_unit_count": 60, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-500", + "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:43606", + "issue:43611", + "issue:43646", + "issue:43688", + "issue:43701", + "issue:43720", + "issue:43782", + "issue:43825", + "issue:43844", + "issue:43927", + "issue:43994", + "issue:44361", + "issue:44451", + "issue:44466", + "issue:44514", + "issue:44554", + "issue:44568", + "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:46489", + "issue:46693", + "issue:46710", + "issue:46846" + ], + "soft_pairs": [ + "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:44964|issue:45405", + "issue:44514|issue:45290", + "issue:43493|issue:43825", + "issue:44554|issue:46365", + "issue:43525|issue:43572", + "issue:44977|issue:46209", + "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:33357|issue:33666", + "issue:44849|issue:46209", + "issue:44568|issue:46489", + "issue:43278|issue:43701" + ], + "prepared_review_unit_hash": "f4f783095bedc81f5aa4b360f33463f145381a9e0599c8176c83f5b9d1524f4d", + "component_budget": { + "node_count": 500, + "item_count": 500, + "soft_pair_count": 1839, + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 30, + "serialized_chars": 36794, + "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 highly heterogeneous: it mixes unrelated tokenizer, processor, model-loading, training, and performance bugs across many model families. None of the soft-edge pairs look like the same underlying fix target, so I would not merge any of them as duplicates.", + "confidence": 0.96, + "canonical_issue_reason": "No single canonical duplicate stands out; the cluster is too broad and the issues do not share one concrete bug. If forced to pick a representative, issue 43611 is the broadest v5-loading regression, but it is still not a true cluster anchor.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 43611 is the most general, user-facing regression among the set and best represents the broad theme of v5 breakage, though it should not be treated as a duplicate hub for the whole cluster.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43611", + "right": "issue:46846", + "accept": false, + "reason": "Different failures: base_model_prefix loading regression vs missing linear_attn TP plan entries." + }, + { + "left": "issue:43720", + "right": "issue:43927", + "accept": false, + "reason": "BitNet packed-weight loading bug vs DiaConfig token-id persistence bug; unrelated code paths." + }, + { + "left": "issue:43504", + "right": "issue:44610", + "accept": false, + "reason": "BEiT legacy-field load failure vs OmDet-Turbo processor image-size mismatch." + }, + { + "left": "issue:43606", + "right": "issue:43927", + "accept": false, + "reason": "CPU offload device mismatch in Bark vs config token-id loss in Dia; different bugs." + }, + { + "left": "issue:33666", + "right": "issue:36331", + "accept": false, + "reason": "Multi-GPU Qwen2-VL training issue vs CustomTrainer signature mismatch; not the same problem." + }, + { + "left": "issue:43278", + "right": "issue:43688", + "accept": false, + "reason": "Embedding dtype regression vs auxiliary-loss normalization bug; separate model behaviors." + }, + { + "left": "issue:43782", + "right": "issue:46693", + "accept": false, + "reason": "Qwen3VL weight_only loading error vs FA2 long-context performance regression." + }, + { + "left": "issue:43262", + "right": "issue:43295", + "accept": false, + "reason": "Different processor regressions: sampling-rate default vs processor.tokenizer/images compatibility." + }, + { + "left": "issue:43720", + "right": "issue:44361", + "accept": false, + "reason": "Packed-weight loading bug vs MLukeTokenizer task AttributeError; unrelated." + }, + { + "left": "issue:44964", + "right": "issue:45405", + "accept": false, + "reason": "Phi-4 multimodal loading failure vs unreleased PEFT version bump; not duplicates." + }, + { + "left": "issue:44514", + "right": "issue:45290", + "accept": false, + "reason": "Both involve apply_chat_template, but the crashing inputs and failure modes differ." + }, + { + "left": "issue:43493", + "right": "issue:43825", + "accept": false, + "reason": "SigLIP2 implementation discrepancy vs pipeline error-message regression; different layers." + }, + { + "left": "issue:44554", + "right": "issue:46365", + "accept": false, + "reason": "MPS attention correctness bug vs Trainer import conflict with datasets; unrelated." + }, + { + "left": "issue:43525", + "right": "issue:43572", + "accept": false, + "reason": "Both are missing pad-token fields, but they are model-specific config issues with different attributes and fixes." + }, + { + "left": "issue:44977", + "right": "issue:46209", + "accept": false, + "reason": "Qwen3.5 flash-attention generation bug vs FP8 inference failure on B200; different hardware/code paths." + }, + { + "left": "issue:44625", + "right": "issue:45584", + "accept": false, + "reason": "num_labels propagation bug vs Whisper empty-transcription generation bug." + }, + { + "left": "issue:43452", + "right": "issue:44451", + "accept": false, + "reason": "gguf_file loading regression vs ScandiBERT load failure; not the same root cause." + }, + { + "left": "issue:28218", + "right": "issue:33357", + "accept": false, + "reason": "Tokenizer spacing bug vs MacOS bus error in CLIP model; unrelated." + }, + { + "left": "issue:45823", + "right": "issue:46050", + "accept": false, + "reason": "Gemma4 device-map mismatch vs sinq save/load quantization breakage." + }, + { + "left": "issue:42796", + "right": "issue:44977", + "accept": false, + "reason": "Tokenizer BPE-marker handling vs Qwen3.5 flash-attention generation; different failures." + }, + { + "left": "issue:43994", + "right": "issue:46192", + "accept": false, + "reason": "SigLIP2 output quality issue vs return_tensors being ignored; unrelated API behavior." + }, + { + "left": "issue:43493", + "right": "issue:43844", + "accept": false, + "reason": "SigLIP2 implementation discrepancy vs ZeRO-3 gradient increase; not the same bug." + }, + { + "left": "issue:39692", + "right": "issue:43295", + "accept": false, + "reason": "Documentation example errors vs processor.tokenizer regression; different artifact types and causes." + }, + { + "left": "issue:43278", + "right": "issue:43844", + "accept": false, + "reason": "dtype regression in evaluation vs gradient growth under ZeRO-3; separate training issues." + }, + { + "left": "issue:43646", + "right": "issue:44466", + "accept": false, + "reason": "Custom model initialization breakage vs device-dependent lm_head serialization inconsistency." + }, + { + "left": "issue:46209", + "right": "issue:46710", + "accept": false, + "reason": "FP8 inference failure on B200 vs accuracy regression after upgrade; different symptoms and fixes." + }, + { + "left": "issue:33357", + "right": "issue:33666", + "accept": false, + "reason": "MacOS CLIP bus error vs Qwen2-VL multi-GPU training; no shared code path." + }, + { + "left": "issue:44849", + "right": "issue:46209", + "accept": false, + "reason": "Qwen3.5 hidden_states bug vs FP8 hardware-specific inference failure." + }, + { + "left": "issue:44568", + "right": "issue:46489", + "accept": false, + "reason": "Special-token insertion bug in mdeberta-v3-base vs DeepSeek-Coder v1 tokenizer regression; same subsystem but not the same concrete bug." + }, + { + "left": "issue:43278", + "right": "issue:43701", + "accept": false, + "reason": "Embedding dtype regression vs checkpoint resume key mismatch; 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": 27, + "review_unit_count": 60, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-500", + "nodes": [ + "issue:29942", + "issue:30333", + "issue:43257", + "issue:43299", + "issue:43334", + "issue:43504", + "issue:43572", + "issue:43575", + "issue:43606", + "issue:43653", + "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:44869", + "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:46701", + "issue:46710", + "issue:46752", + "issue:46821", + "issue:46832", + "issue:46846", + "issue:46858" + ], + "soft_pairs": [ + "issue:44792|issue:45405", + "issue:44488|issue:44843", + "issue:43927|issue:44361", + "issue:44568|issue:45812", + "issue:43824|issue:44589", + "issue:43575|issue:45405", + "issue:43504|issue:43606", + "issue:44849|issue:45357", + "issue:43257|issue:43299", + "issue:45245|issue:46710", + "issue:46399|issue:46846", + "issue:46821|issue:46832", + "issue:43606|issue:44112", + "issue:44493|issue:46285", + "issue:44589|issue:45520", + "issue:43334|issue:43572", + "issue:43334|issue:45759", + "issue:45072|issue:46832", + "issue:43504|issue:43653", + "issue:43572|issue:44291", + "issue:44464|issue:46424", + "issue:43937|issue:46701", + "issue:46832|issue:46858", + "issue:43844|issue:44964", + "issue:45200|issue:45229", + "issue:43873|issue:46050", + "issue:44869|issue:46752", + "issue:29942|issue:30333" + ], + "prepared_review_unit_hash": "44437a3e15d8804bd818974deeabe902a6cda31379185866587aca655a77df7c", + "component_budget": { + "node_count": 500, + "item_count": 500, + "soft_pair_count": 1839, + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 28, + "serialized_chars": 36307, + "estimated_input_tokens": 9077, + "estimated_eval_tokens": 18410 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is not a duplicate set: the items span unrelated model-loading, tokenizer, cache/offload, CI, and training issues. The suggested soft matches are mostly superficial keyword overlap (e.g., \"missing\", \"load\", \"OOM\", \"pad_token\") rather than the same concrete bug.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44792", + "right": "issue:45405", + "accept": false, + "reason": "Different bugs: a janus generation test failure vs an unrelated dependency version pin issue." + }, + { + "left": "issue:44488", + "right": "issue:44843", + "accept": false, + "reason": "Both mention loading/tokenizer behavior, but one is a model loading problem and the other is an offline hub call bug in regex patching." + }, + { + "left": "issue:43927", + "right": "issue:44361", + "accept": false, + "reason": "Different tokenizer/model failures with different symptoms and code paths; not the same underlying bug." + }, + { + "left": "issue:44568", + "right": "issue:45812", + "accept": false, + "reason": "Both involve tokenizer behavior, but one is missing BOS/EOS insertion for a specific model while the other is a broad Granite token-ID regression." + }, + { + "left": "issue:43824", + "right": "issue:44589", + "accept": false, + "reason": "One is an import error for a Qwen VL class; the other is a storage backend/type mismatch. No shared failure path." + }, + { + "left": "issue:43575", + "right": "issue:45405", + "accept": false, + "reason": "Qwen2-57B tensor-parallel OOM is unrelated to a PEFT minimum-version bump." + }, + { + "left": "issue:43504", + "right": "issue:43606", + "accept": false, + "reason": "Legacy field loading bug in BEiT is unrelated to Bark CPU offload device-mismatch failures." + }, + { + "left": "issue:44849", + "right": "issue:45357", + "accept": false, + "reason": "Different Qwen variants and different failures: output_hidden_states behavior vs incorrect visual encoder keys on save_pretrained." + }, + { + "left": "issue:43257", + "right": "issue:43299", + "accept": false, + "reason": "Both are Qwen3 MOE-related, but one is an accelerate+deepspeed conversion path and the other is a transformers dev loading regression for Qwen3VL." + }, + { + "left": "issue:45245", + "right": "issue:46710", + "accept": false, + "reason": "Category-count RuntimeError and DeepSeek accuracy regression are distinct problems with no common code path." + }, + { + "left": "issue:46399", + "right": "issue:46846", + "accept": false, + "reason": "Both are Qwen3.5-related, but one is a missing decorator for a kernel registration and the other is a missing TP plan entry causing OOM/shape errors." + }, + { + "left": "issue:46821", + "right": "issue:46832", + "accept": false, + "reason": "TimesFM window_size attribute error and RT-DETR feature-level mismatch are unrelated model-specific bugs." + }, + { + "left": "issue:43606", + "right": "issue:44112", + "accept": false, + "reason": "Both are CI failures, but they stem from different tests and different device-handling issues." + }, + { + "left": "issue:44493", + "right": "issue:46285", + "accept": false, + "reason": "Position-id key warnings across many models vs CLIP structure inconsistency; different regressions and fixes." + }, + { + "left": "issue:44589", + "right": "issue:45520", + "accept": false, + "reason": "Float8 storage lookup failure is unrelated to a flash_attn import-utils KeyError on Python 3.13." + }, + { + "left": "issue:43334", + "right": "issue:43572", + "accept": false, + "reason": "Qwen3-VL missing pad_token_id and StableLmConfig pad_token_idx are separate model/config bugs." + }, + { + "left": "issue:43334", + "right": "issue:45759", + "accept": false, + "reason": "Qwen3-VL pad_token_id AttributeError is unrelated to AutoModelForCausalLM.from_config not unwrapping text_config." + }, + { + "left": "issue:45072", + "right": "issue:46832", + "accept": false, + "reason": "Dtype mismatch CI failures vs RT-DETR feature-level mismatch; different subsystems and symptoms." + }, + { + "left": "issue:43504", + "right": "issue:43653", + "accept": false, + "reason": "BEiT legacy-field loading bug is unrelated to BigBirdTokenizer special-token registration/empty decode behavior." + }, + { + "left": "issue:43572", + "right": "issue:44291", + "accept": false, + "reason": "Missing pad_token_idx in StableLmConfig and unexpected _is_hf_initialized arg in init_empty_weights are different version-regression issues." + }, + { + "left": "issue:44464", + "right": "issue:46424", + "accept": false, + "reason": "Chunked generation inconsistency with compiled forward is unrelated to cache length handling for quantized cache only." + }, + { + "left": "issue:43937", + "right": "issue:46701", + "accept": false, + "reason": "GenerationConfig validation failure for GLM-5 is unrelated to DiffusionGemma image-text pipeline failure." + }, + { + "left": "issue:46832", + "right": "issue:46858", + "accept": false, + "reason": "RT-DETR backbone/output mismatch and static-cache seq-length tensor bug are separate runtime errors." + }, + { + "left": "issue:43844", + "right": "issue:44964", + "accept": false, + "reason": "Gradient explosion with ZeRO-3 and a Phi-4 multimodal loading failure are unrelated problems." + }, + { + "left": "issue:45200", + "right": "issue:45229", + "accept": false, + "reason": "Gemma 4 fine-tuning token defaults vs multi-GPU inference OOM are not the same underlying bug." + }, + { + "left": "issue:43873", + "right": "issue:46050", + "accept": false, + "reason": "Quantization offload behavior and sinq save/load breakage are different quantization issues with different code paths." + }, + { + "left": "issue:44869", + "right": "issue:46752", + "accept": false, + "reason": "Whisper timestamp decoding crash on trailing replacement char is unrelated to apply_chat_template on an empty list." + }, + { + "left": "issue:29942", + "right": "issue:30333", + "accept": false, + "reason": "Flash Attention 2 test failures and MLflow job-reporting issues are entirely 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": 60, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-500", + "nodes": [ + "issue:15354", + "issue:28218", + "issue:29159", + "issue:29870", + "issue:29942", + "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: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:46821" + ], + "soft_pairs": [ + "issue:45229|issue:45823", + "issue:43994|issue:44493", + "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:44423|issue:44734", + "issue:15354|issue:38175", + "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", + "issue:45200|issue:45823", + "issue:29870|issue:29942" + ], + "prepared_review_unit_hash": "df15d95f8ee061fd431cf0696a05a2350e9ab3a00fbccdb43c89b23e038dd6dc", + "component_budget": { + "node_count": 500, + "item_count": 500, + "soft_pair_count": 1839, + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 28, + "serialized_chars": 35560, + "estimated_input_tokens": 8890, + "estimated_eval_tokens": 18036 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a loose mix of unrelated Transformers issues across tokenizers, generation, serving, multimodal processors, quantization, and model loading. None of the soft-similarity pairs look like the same underlying bug/change, so I would not merge any of them as duplicates.", + "confidence": 0.86, + "canonical_issue_reason": "No true duplicate hub stands out from the titles alone; if one issue must anchor the cluster, issue:43873 is the broadest active report and has the strongest inbound reference signal, so it is the closest thing to a cluster representative.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:43873 is the best single issue representative because it is broad, currently open, and appears to be the most central runtime/memory-management report in the set.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:45229", + "right": "issue:45823", + "accept": false, + "reason": "Both are Gemma4-related, but one is a multi-GPU CUDA OOM and the other is a device-map mismatch in forward; different failure modes and code paths." + }, + { + "left": "issue:43994", + "right": "issue:44493", + "accept": false, + "reason": "SigLIP2 incorrect outputs vs a position-id key warning across many models; too different to be the same bug." + }, + { + "left": "issue:45588", + "right": "issue:46050", + "accept": false, + "reason": "Flash-attention sinkless-model crash vs sinq save/load corruption; unrelated subsystems and symptoms." + }, + { + "left": "issue:42490", + "right": "issue:43696", + "accept": false, + "reason": "Local-path import warning and GPT-oss OOM are unrelated." + }, + { + "left": "issue:29159", + "right": "issue:32090", + "accept": false, + "reason": "Tokenizer slow/fast inconsistency is not the same issue as a Trainer GPU broadcast NoneType error." + }, + { + "left": "issue:44977", + "right": "issue:46701", + "accept": false, + "reason": "Qwen3.5 flash-attention generation failure vs a DiffusionGemma pipeline output issue; different models and execution paths." + }, + { + "left": "issue:45084", + "right": "issue:45520", + "accept": false, + "reason": "Non-template-node compile error and flash_attn import KeyError are separate import/compilation problems." + }, + { + "left": "issue:28218", + "right": "issue:34567", + "accept": false, + "reason": "Added-token spacing behavior and TrainerState token-count tracking are unrelated." + }, + { + "left": "issue:43065", + "right": "issue:46369", + "accept": false, + "reason": "A dummy Conv2d implementation and a memory overflow report on sam3-litetext are not the same defect." + }, + { + "left": "issue:44554", + "right": "issue:45446", + "accept": false, + "reason": "MPS attention correctness with mismatched head dims is unrelated to a PyTorch version check bug." + }, + { + "left": "issue:44423", + "right": "issue:44734", + "accept": false, + "reason": "Both are serve crashes, but they hit different continuations and different tensor/object errors; not the same underlying bug." + }, + { + "left": "issue:15354", + "right": "issue:38175", + "accept": false, + "reason": "TorchScript GeneratorExp export failure and SigLIP2 zero probabilities are unrelated." + }, + { + "left": "issue:43493", + "right": "issue:43701", + "accept": false, + "reason": "SigLIP2 HF-vs-JAX discrepancy and resume_from_checkpoint key mismatch do not share a concrete code path." + }, + { + "left": "issue:41720", + "right": "issue:46710", + "accept": false, + "reason": "A800 cudaErrorAssert under auto device mapping is different from an accuracy regression after upgrading Transformers." + }, + { + "left": "issue:45406", + "right": "issue:46121", + "accept": false, + "reason": "Gemma4Processor _tokenizer crash in serve and a rope-parameter JSON list type error are unrelated." + }, + { + "left": "issue:44977", + "right": "issue:46710", + "accept": false, + "reason": "Flash-attention generation breakage in Qwen3.5 is not the same as a model accuracy regression after version upgrade." + }, + { + "left": "issue:45072", + "right": "issue:45198", + "accept": false, + "reason": "BF16 dtype mismatches in SwitchTransformers/TimmWrapperModel and Wav2Vec2 save/tokenization failures are unrelated." + }, + { + "left": "issue:43295", + "right": "issue:43450", + "accept": false, + "reason": "Processor.tokenizer regression with image inputs and batched video shape errors are different multimodal bugs." + }, + { + "left": "issue:37428", + "right": "issue:43957", + "accept": false, + "reason": "Missing flash-support import and meta-device model loading failures are not the same issue." + }, + { + "left": "issue:45588", + "right": "issue:45759", + "accept": false, + "reason": "Flash-attention sinkless-model crash and from_config text_config unwrapping are separate processor/config problems." + }, + { + "left": "issue:43540", + "right": "issue:45200", + "accept": false, + "reason": "Video-input ValueError in Qwen3OmniMoe and missing mm_token_type_ids in Gemma4 are different multimodal preprocessing bugs." + }, + { + "left": "issue:43295", + "right": "issue:43873", + "accept": false, + "reason": "Processor/tokenizer regression and quantization offloading behavior are far apart in both symptom and code path." + }, + { + "left": "issue:44186", + "right": "issue:46192", + "accept": false, + "reason": "LayoutLMv2Tokenizer NER/padding crash and return_tensors being ignored are distinct tokenizer API issues." + }, + { + "left": "issue:43572", + "right": "issue:46821", + "accept": false, + "reason": "StableLmConfig pad_token_idx regression and TimesFM window_size AttributeError are unrelated config bugs." + }, + { + "left": "issue:43295", + "right": "issue:44610", + "accept": false, + "reason": "Tokenizer/processor regression and OmDet-Turbo processor size mismatch are different model-specific preprocessing problems." + }, + { + "left": "issue:45823", + "right": "issue:46192", + "accept": false, + "reason": "Gemma4 device-mapping mismatch and return_tensors being ignored are unrelated." + }, + { + "left": "issue:45200", + "right": "issue:45823", + "accept": false, + "reason": "One is a missing default mm_token_type_ids, the other is a device placement mismatch in forward; not the same bug." + }, + { + "left": "issue:29870", + "right": "issue:29942", + "accept": false, + "reason": "Bart generate decoder_attention_mask bug and Flash Attention 2 test failures 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": 29, + "review_unit_count": 60, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-500", + "nodes": [ + "issue:28218", + "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:44792", + "issue:44849", + "issue:44987", + "issue:45127", + "issue:45198", + "issue:45200", + "issue:45290", + "issue:45356", + "issue:45381", + "issue:45520", + "issue:45563", + "issue:46178", + "issue:46192", + "issue:46209", + "issue:46285", + "issue:46512", + "issue:46894", + "issue:46899" + ], + "soft_pairs": [ + "issue:43550|issue:43844", + "issue:45520|issue:46178", + "issue:43450|issue:43575", + "issue:43572|issue:44987", + "issue:44554|issue:45356", + "issue:42490|issue:43493", + "issue:37428|issue:43824", + "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:44737|issue:46894", + "issue:28218|issue:35141", + "issue:43278|issue:43856", + "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", + "issue:45200|issue:46899", + "issue:44792|issue:45563" + ], + "prepared_review_unit_hash": "12676e1e6536853b46f180aea05bc526a8306f7591ff034d825b4ec2982b20db", + "component_budget": { + "node_count": 500, + "item_count": 500, + "soft_pair_count": 1839, + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 26, + "serialized_chars": 35887, + "estimated_input_tokens": 8972, + "estimated_eval_tokens": 18200 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is a loose semantic mix of unrelated Transformers regressions and docs/test issues; none of the proposed soft pairs look like true duplicates, so I would not merge any of them. I\u2019d treat issue 44521 as the weakest umbrella representative only because it has the most engagement, but the group is too heterogeneous for a strong canonical duplicate target.", + "confidence": 0.9, + "canonical_issue_reason": "Issue 44521 has the strongest engagement signals (discussion and inbound references) and sits on a high-traffic API path, but the cluster is too heterogeneous to support a truly canonical duplicate issue.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 44521 is the best single representative by impact/engagement, though only weakly so because the cluster spans many unrelated bugs.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43550", + "right": "issue:43844", + "accept": false, + "reason": "Different failures: Bamba torch.compile/SDPA correctness vs DeepSpeed ZeRO-3 gradient growth." + }, + { + "left": "issue:45520", + "right": "issue:46178", + "accept": false, + "reason": "Unrelated import/backend problems: flash_attn KeyError vs tensorflow_text missing from BACKENDS_MAPPING." + }, + { + "left": "issue:43450", + "right": "issue:43575", + "accept": false, + "reason": "Video processor batched-shape bug vs Qwen2 TP loading OOM; different code paths and symptoms." + }, + { + "left": "issue:43572", + "right": "issue:44987", + "accept": false, + "reason": "Both are loading/config regressions, but one is StableLm pad_token_idx and the other is a package/version load failure for a specific model." + }, + { + "left": "issue:44554", + "right": "issue:45356", + "accept": false, + "reason": "MPS attention correctness issue vs tokenizer codec/regression in Kimi-K2.5; not the same bug." + }, + { + "left": "issue:42490", + "right": "issue:43493", + "accept": false, + "reason": "Trailing-slash local path warning is unrelated to the SigLIP2 HF-vs-JAX implementation discrepancy." + }, + { + "left": "issue:37428", + "right": "issue:43824", + "accept": false, + "reason": "Both are import errors, but they involve different missing symbols and different modules/code paths." + }, + { + "left": "issue:43232", + "right": "issue:44849", + "accept": false, + "reason": "Generation kwargs update after sync_gpus is unrelated to the Qwen3.5 output_hidden_states bug." + }, + { + "left": "issue:43295", + "right": "issue:43493", + "accept": false, + "reason": "Processor.tokenizer/image passthrough regression is a different bug from the SigLIP2 implementation mismatch." + }, + { + "left": "issue:44466", + "right": "issue:45127", + "accept": false, + "reason": "Serialization inconsistency for lm_head.weight is not the same underlying bug as LoRA merge collapse after vocab extension." + }, + { + "left": "issue:39692", + "right": "issue:46192", + "accept": false, + "reason": "Docs/example mistakes and silently ignored return_tensors are different issues." + }, + { + "left": "issue:44521", + "right": "issue:45290", + "accept": false, + "reason": "Both touch apply_chat_template, but one is zeroed assistant masks and the other is a crash on tool-call messages; not the same failure." + }, + { + "left": "issue:43653", + "right": "issue:44112", + "accept": false, + "reason": "BigBirdTokenizer special-token registration bug vs GraniteSpeech CI device-override test flake; unrelated." + }, + { + "left": "issue:45381", + "right": "issue:46209", + "accept": false, + "reason": "Qwen2.5-VL video position ids bug is unrelated to FP8 inference failure on B200." + }, + { + "left": "issue:44737", + "right": "issue:46894", + "accept": false, + "reason": "XLNet CPU tensor allocation bug vs CohereASR double-shift training-loss bug; different subsystems." + }, + { + "left": "issue:28218", + "right": "issue:35141", + "accept": false, + "reason": "Tokenizer adds extra space after added token, while the other is output embedding reinit after resizing; distinct problems." + }, + { + "left": "issue:43278", + "right": "issue:43856", + "accept": false, + "reason": "Embedding dtype drift in evaluate is not the same as Qwen3 MoE memory inefficiency." + }, + { + "left": "issue:42915", + "right": "issue:43824", + "accept": false, + "reason": "FineGrainedFP8Config failure for Qwen3Moe and Qwen2_5_VL import error are unrelated." + }, + { + "left": "issue:43950", + "right": "issue:46285", + "accept": false, + "reason": "Non-persistent buffer corruption in from_pretrained is a different regression from the CLIP text-model structure change." + }, + { + "left": "issue:43824", + "right": "issue:43975", + "accept": false, + "reason": "ImportError for Qwen2_5_VL is unrelated to DeepSeek detokenization regression." + }, + { + "left": "issue:43452", + "right": "issue:46512", + "accept": false, + "reason": "gguf_file loading breakage and stale config.dtype after from_config are different lifecycle bugs." + }, + { + "left": "issue:43334", + "right": "issue:43525", + "accept": false, + "reason": "Same symptom class (missing pad_token_id), but different models/configs and no evidence of one shared underlying fix." + }, + { + "left": "issue:43262", + "right": "issue:43450", + "accept": false, + "reason": "Audio apply_chat_template sampling-rate default bug is unrelated to batched video output shape." + }, + { + "left": "issue:44442", + "right": "issue:45198", + "accept": false, + "reason": "AutoTokenizer failing FastSpeech2ConformerTokenizer is a different issue than Wav2Vec2 save/tokenization failure." + }, + { + "left": "issue:45200", + "right": "issue:46899", + "accept": false, + "reason": "Missing mm_token_type_ids default and unsafe dtype casting in multimodal embedders are distinct multimodal bugs." + }, + { + "left": "issue:44792", + "right": "issue:45563", + "accept": false, + "reason": "Janus image-generation test failure and stale paged-generate warning are not the same code-path problem." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 30, + "review_unit_count": 60, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-500", + "nodes": [ + "issue:28282", + "issue:29942", + "issue:32090", + "issue:33357", + "issue:37428", + "issue:38175", + "issue:41720", + "issue:43278", + "issue:43335", + "issue:43388", + "issue:43425", + "issue:43504", + "issue:43526", + "issue:43550", + "issue:43606", + "issue:43611", + "issue:43653", + "issue:43819", + "issue:43825", + "issue:43827", + "issue:43901", + "issue:43931", + "issue:44112", + "issue:44206", + "issue:44448", + "issue:44561", + "issue:44610", + "issue:44743", + "issue:44855", + "issue:44857", + "issue:44898", + "issue:45083", + "issue:45200", + "issue:45325", + "issue:45412", + "issue:45520", + "issue:45584", + "issue:45593", + "issue:45759", + "issue:46170", + "issue:46192", + "issue:46209", + "issue:46234", + "issue:46285", + "issue:46424", + "issue:46512", + "issue:46821", + "issue:46858" + ], + "soft_pairs": [ + "issue:43819|issue:44112", + "issue:44898|issue:45412", + "issue:43526|issue:46192", + "issue:45200|issue:45325", + "issue:43827|issue:43901", + "issue:43335|issue:43550", + "issue:41720|issue:46209", + "issue:44561|issue:46234", + "issue:43388|issue:43526", + "issue:43819|issue:45083", + "issue:44857|issue:45593", + "issue:44610|issue:44857", + "issue:46424|issue:46858", + "issue:45759|issue:46285", + "issue:45520|issue:46170", + "issue:46512|issue:46821", + "issue:43606|issue:45083", + "issue:38175|issue:45412", + "issue:44206|issue:45584", + "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", + "issue:43335|issue:43504", + "issue:44855|issue:45083", + "issue:43278|issue:43825" + ], + "prepared_review_unit_hash": "9e864900eb2e405c07ec049e758e6b4646bf2ed5b0d2a0ac236515f274ffbbef", + "component_budget": { + "node_count": 500, + "item_count": 500, + "soft_pair_count": 1839, + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 29, + "serialized_chars": 36749, + "estimated_input_tokens": 9188, + "estimated_eval_tokens": 18632 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is heterogeneous: most pairs share only a broad subsystem or theme, not the same concrete bug/change, so all soft edges should be rejected. I\u2019d use issue 43611 as the best representative regression in the set.", + "confidence": 0.62, + "canonical_issue_reason": "issue:43611 is the most broadly impactful concrete regression here: a v5.0 loading break tied to `base_model_prefix`, with clear user-facing impact across multiple model classes.", + "canonical_pr_reason": null, + "best_issue_reason": "Same reason: it is a well-scoped, high-impact runtime regression rather than a narrow model-specific edge case or a docs-only issue.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43819", + "right": "issue:44112", + "accept": false, + "reason": "Different bugs: DAC latent reconstruction mismatch vs a GraniteSpeech CI device-override test failure." + }, + { + "left": "issue:44898", + "right": "issue:45412", + "accept": false, + "reason": "Perceiver interpolation failure and RT-DETR memory leak are unrelated code paths and failure modes." + }, + { + "left": "issue:43526", + "right": "issue:46192", + "accept": false, + "reason": "Both touch processors, but one is label reduction and the other is `text_kwargs`/`return_tensors` handling." + }, + { + "left": "issue:45200", + "right": "issue:45325", + "accept": false, + "reason": "Different multimodal bugs: missing default `mm_token_type_ids` vs RoPE position scaling for still images." + }, + { + "left": "issue:43827", + "right": "issue:43901", + "accept": false, + "reason": "Both are docs issues, but they concern different APIs and different deprecations." + }, + { + "left": "issue:43335", + "right": "issue:43550", + "accept": false, + "reason": "SwitchTransformers config bug and Bamba torch.compile/SDPA failure are distinct problems." + }, + { + "left": "issue:41720", + "right": "issue:46209", + "accept": false, + "reason": "Different GPU/hardware failures with no shared concrete code path." + }, + { + "left": "issue:44561", + "right": "issue:46234", + "accept": false, + "reason": "Both involve missing utils in v5, but the missing symbols and breakage mechanisms differ." + }, + { + "left": "issue:43388", + "right": "issue:43526", + "accept": false, + "reason": "Metric gathering drops tuple labels; Beit image processor returns only one label. Different layers and symptoms." + }, + { + "left": "issue:43819", + "right": "issue:45083", + "accept": false, + "reason": "Different helper/model logic: latent STE mismatch vs feature-length helper behavior." + }, + { + "left": "issue:44857", + "right": "issue:45593", + "accept": false, + "reason": "AMP crash in LwDetrImageLoss is unrelated to D-FINE's auxiliary-loss omission." + }, + { + "left": "issue:44610", + "right": "issue:44857", + "accept": false, + "reason": "Processor image-size mismatch and AMP/CUDA loss crash are separate bugs." + }, + { + "left": "issue:46424", + "right": "issue:46858", + "accept": false, + "reason": "Cache-length config handling and static-cache slice arithmetic are different defects." + }, + { + "left": "issue:45759", + "right": "issue:46285", + "accept": false, + "reason": "Composite Qwen config unwrapping is unrelated to CLIP structure inconsistency." + }, + { + "left": "issue:45520", + "right": "issue:46170", + "accept": false, + "reason": "Both are import/platform issues, but they stem from different dependencies and environments." + }, + { + "left": "issue:46512", + "right": "issue:46821", + "accept": false, + "reason": "Stale dtype config state and TimesFM `window_size` AttributeError are unrelated." + }, + { + "left": "issue:43606", + "right": "issue:45083", + "accept": false, + "reason": "CPU offload device mismatch is unrelated to qwen3_omni_moe feature-length logic." + }, + { + "left": "issue:38175", + "right": "issue:45412", + "accept": false, + "reason": "SigLIP2 zero probabilities and RT-DETR memory leakage are unrelated." + }, + { + "left": "issue:44206", + "right": "issue:45584", + "accept": false, + "reason": "Unsupported `center` arg crash and Whisper empty-transcription failure are different audio bugs." + }, + { + "left": "issue:43335", + "right": "issue:43653", + "accept": false, + "reason": "Sparse-layer creation and BigBird special-token registration are separate issues." + }, + { + "left": "issue:37428", + "right": "issue:43425", + "accept": false, + "reason": "Flash-attention import-symbol failure and generic Torch 2.10 incompatibility are not the same bug." + }, + { + "left": "issue:43550", + "right": "issue:44112", + "accept": false, + "reason": "Bamba compile/SDPA runtime failure is unrelated to a stale GraniteSpeech CI test." + }, + { + "left": "issue:29942", + "right": "issue:32090", + "accept": false, + "reason": "Flash Attention test failures and Trainer `_gpu_broadcast_one` TypeError are unrelated." + }, + { + "left": "issue:28282", + "right": "issue:33357", + "accept": false, + "reason": "PyTorch-missing import error and MacOS bus error in CLIP are different failure modes." + }, + { + "left": "issue:43611", + "right": "issue:43931", + "accept": false, + "reason": "Broad v5 loading regression vs a model-specific Qwen3-VL shape mismatch." + }, + { + "left": "issue:44448", + "right": "issue:44743", + "accept": false, + "reason": "Pegasus v4/v5 output drift and Qwen recurrent-state reset are unrelated." + }, + { + "left": "issue:43335", + "right": "issue:43504", + "accept": false, + "reason": "SwitchTransformers sparse-layer config bug vs Beit pretrained-load legacy-field bug." + }, + { + "left": "issue:44855", + "right": "issue:45083", + "accept": false, + "reason": "Python 3.13 DebertaV2 import parsing failure is unrelated to qwen3_omni_moe helper behavior." + }, + { + "left": "issue:43278", + "right": "issue:43825", + "accept": false, + "reason": "Embedding dtype drift in train/eval is unrelated to the v5 pipeline error-message regression." + } + ] + }, + "evaluator_result": 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": 60, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-500", + "nodes": [ + "issue:26413", + "issue:28218", + "issue:29127", + "issue:31501", + "issue:32090", + "issue:33357", + "issue:34689", + "issue:35141", + "issue:37428", + "issue:39692", + "issue:42490", + "issue:43335", + "issue:43526", + "issue:43575", + "issue:43606", + "issue:43701", + "issue:43720", + "issue:43819", + "issue:43854", + "issue:44112", + "issue:44206", + "issue:44351", + "issue:44360", + "issue:44448", + "issue:44493", + "issue:44514", + "issue:44521", + "issue:44561", + "issue:44617", + "issue:44779", + "issue:44857", + "issue:45072", + "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:29127|issue:32090", + "issue:43335|issue:45083", + "issue:43720|issue:44112", + "issue:43335|issue:45593", + "issue:43335|issue:44112", + "issue:37428|issue:44351", + "issue:44514|issue:44521", + "issue:45290|issue:46326", + "issue:45127|issue:45245", + "issue:45588|issue:45859", + "issue:34689|issue:35141", + "issue:42490|issue:43854", + "issue:46431|issue:46566", + "issue:45310|issue:45478", + "issue:44448|issue:44779", + "issue:43819|issue:44493", + "issue:39692|issue:43526", + "issue:26413|issue:31501", + "issue:43606|issue:44857", + "issue:28218|issue:32090", + "issue:45459|issue:46234", + "issue:43701|issue:45083", + "issue:39692|issue:44493", + "issue:44617|issue:46697", + "issue:44206|issue:45823", + "issue:32090|issue:33357", + "issue:42490|issue:43575", + "issue:44360|issue:44561", + "issue:44857|issue:45072" + ], + "prepared_review_unit_hash": "a543cfbcad867e46d9104f1c9badd65c4c065d01f18d0430a1ab32e0e8d2b8b1", + "component_budget": { + "node_count": 500, + "item_count": 500, + "soft_pair_count": 1839, + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 29, + "serialized_chars": 36466, + "estimated_input_tokens": 9117, + "estimated_eval_tokens": 18490 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a set of unrelated bug reports. The only clear duplicate pair is the repeated Qwen3.5 Moe `from_pretrained` regression; the rest are similar only at a broad subsystem level and should stay separate.", + "confidence": 0.98, + "canonical_issue_reason": "issue:45310 is the earlier and more detailed report, and issue:45478 repeats the same Qwen3.5 Moe `from_pretrained` error verbatim.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:45310 is the strongest canonical issue candidate because it predates the duplicate and has the fuller original discussion context.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:29127", + "right": "issue:32090", + "accept": false, + "reason": "Different failures in different paths: LayoutLMv3 error-message clarity vs Trainer `_gpu_broadcast_one` rejecting `NoneType`." + }, + { + "left": "issue:43335", + "right": "issue:45083", + "accept": false, + "reason": "Different models and bugs: SwitchTransformers sparse-layer creation vs qwen3_omni_moe feature-length helper behavior." + }, + { + "left": "issue:43720", + "right": "issue:44112", + "accept": false, + "reason": "Both are CI/device-related, but one is BitNet weight unpacking during accelerate loading and the other is a stale GraniteSpeech device override test." + }, + { + "left": "issue:43335", + "right": "issue:45593", + "accept": false, + "reason": "SwitchTransformers config bug and D-FINE auxiliary-loss behavior are unrelated." + }, + { + "left": "issue:43335", + "right": "issue:44112", + "accept": false, + "reason": "Sparse-layer config bug vs GraniteSpeech stale-device CI failure; not the same code path." + }, + { + "left": "issue:37428", + "right": "issue:44351", + "accept": false, + "reason": "Both are import-related, but they concern different missing symbols and different modules." + }, + { + "left": "issue:44514", + "right": "issue:44521", + "accept": false, + "reason": "Both involve `apply_chat_template`, but one is batching/padding crash while the other is incorrect assistant masks for multimodal inputs." + }, + { + "left": "issue:45290", + "right": "issue:46326", + "accept": false, + "reason": "Different chat-template bugs: tool-call assistant messages with no content vs `continue_final_message` on `mistral_common` tokenizers." + }, + { + "left": "issue:45127", + "right": "issue:45245", + "accept": false, + "reason": "Extended-vocabulary LoRA merge collapse is unrelated to the 2^24 category limit runtime error." + }, + { + "left": "issue:45588", + "right": "issue:45859", + "accept": false, + "reason": "Different failure modes in different components: flash-attention `s_aux=None` crash vs CPU-offload incompatibility in Qwen3_5MoeTextRotaryEmbedding." + }, + { + "left": "issue:34689", + "right": "issue:35141", + "accept": false, + "reason": "Model-loading regression for Llama 3.2 Vision is not the same as output embedding reinitialization after resizing embeddings." + }, + { + "left": "issue:42490", + "right": "issue:43854", + "accept": false, + "reason": "Local-path import warning and GLM-4.7-Flash loading failure are distinct issues." + }, + { + "left": "issue:46431", + "right": "issue:46566", + "accept": false, + "reason": "Processor missing a chat template is unrelated to DiffusionGemma weights staying on `meta` with `device_map='auto'`." + }, + { + "left": "issue:45310", + "right": "issue:45478", + "accept": true, + "reason": "Same exact Qwen3.5 Moe `from_pretrained` error; `45478` is a duplicate of `45310`." + }, + { + "left": "issue:44448", + "right": "issue:44779", + "accept": false, + "reason": "Separate version-regression reports for different models and tokenization/generation paths." + }, + { + "left": "issue:43819", + "right": "issue:44493", + "accept": false, + "reason": "DAC latent/forward mismatch and unexpected position-id keys are different bugs." + }, + { + "left": "issue:39692", + "right": "issue:43526", + "accept": false, + "reason": "SigLIP2 documentation example errors are unrelated to BeitImageProcessorFast `reduce_labels` behavior." + }, + { + "left": "issue:26413", + "right": "issue:31501", + "accept": false, + "reason": "Checkpoint-resume epoch-iterator failure is not the same as FineWeb SLM training not starting." + }, + { + "left": "issue:43606", + "right": "issue:44857", + "accept": false, + "reason": "CPU-offload device mismatch in Bark is unrelated to LwDetrImageLoss AMP CUDA crash." + }, + { + "left": "issue:28218", + "right": "issue:32090", + "accept": false, + "reason": "Tokenizer extra-space behavior is a different bug from Trainer broadcast type handling." + }, + { + "left": "issue:45459", + "right": "issue:46234", + "accept": false, + "reason": "Tokenizer protobuf error masking and missing `is_falcon_mamba_ssm_available` are distinct utility-layer issues." + }, + { + "left": "issue:43701", + "right": "issue:45083", + "accept": false, + "reason": "Checkpoint key mismatch and qwen3_omni_moe helper behavior do not describe the same bug." + }, + { + "left": "issue:39692", + "right": "issue:44493", + "accept": false, + "reason": "Documentation example issues are unrelated to the unexpected position-id key regression." + }, + { + "left": "issue:44617", + "right": "issue:46697", + "accept": false, + "reason": "CUDA OOM in Sam3Video and Swin2SR over-padding are unrelated." + }, + { + "left": "issue:44206", + "right": "issue:45823", + "accept": false, + "reason": "Unsupported `center` arg crash and Gemma4 PLE device mismatch are different code paths." + }, + { + "left": "issue:32090", + "right": "issue:33357", + "accept": false, + "reason": "Trainer `_gpu_broadcast_one` type error is unrelated to the MacOS CLIP bus error." + }, + { + "left": "issue:42490", + "right": "issue:43575", + "accept": false, + "reason": "Import warning for local paths is unrelated to tensor-parallel OOM on Qwen2-57B-A14B-Instruct." + }, + { + "left": "issue:44360", + "right": "issue:44561", + "accept": false, + "reason": "DSA indexer discussion and removal of `is_torch_fx_available` breaking trust-remote-code models are unrelated." + }, + { + "left": "issue:44857", + "right": "issue:45072", + "accept": false, + "reason": "Different dtype/precision failures in different models; not the same underlying bug." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet, and the only accepted pair (issue:45310/issue:45478) is a clear duplicate based on identical titles and highly similar content. The remaining soft-edge verdicts stay conservative by treating broad subsystem similarity as insufficient for duplication." + }, + "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": 60, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-500", + "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:44448", + "issue:44466", + "issue:44493", + "issue:44561", + "issue:44568", + "issue:44779", + "issue:44855", + "issue:44857", + "issue:44871", + "issue:44977", + "issue:45083", + "issue:45229", + "issue:45290", + "issue:45381", + "issue:45440", + "issue:45491", + "issue:45588", + "issue:45698", + "issue:45920", + "issue:46192", + "issue:46399", + "issue:46421", + "issue:46655", + "issue:46858" + ], + "soft_pairs": [ + "issue:43550|issue:45083", + "issue:27957|issue:34689", + "issue:29942|issue:35141", + "issue:38175|issue:44493", + "issue:29870|issue:32090", + "issue:43278|issue:43329", + "issue:44448|issue:44977", + "issue:45920|issue:46192", + "issue:45290|issue:45381", + "issue:10000|issue:15354", + "issue:44977|issue:45229", + "issue:43278|issue:43381", + "issue:29870|issue:33290", + "issue:43927|issue:45698", + "issue:44855|issue:46399", + "issue:45290|issue:46655", + "issue:40444|issue:43526", + "issue:44857|issue:45588", + "issue:37428|issue:42175", + "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", + "issue:44568|issue:45491", + "issue:43278|issue:43575", + "issue:44871|issue:44977" + ], + "prepared_review_unit_hash": "5a2f54f25b95b5bdd80872bcb8bf6211efca454a03d380bbd7e7d1519a965201", + "component_budget": { + "node_count": 500, + "item_count": 500, + "soft_pair_count": 1839, + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 30, + "serialized_chars": 37072, + "estimated_input_tokens": 9268, + "estimated_eval_tokens": 18792 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "These reports are largely unrelated, model-specific regressions spanning tokenizer behavior, flash-attention, cache handling, serialization, and packaging. The similarity is mostly thematic, not the same underlying bug/change, so none of the soft pairs should be merged.", + "confidence": 0.95, + "canonical_issue_reason": "issue:45920 is the broadest, most representative standalone bug here: a multi-model v5 tokenizer regression with inbound references and strong discussion activity.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:45920 has the best combination of breadth, cross-model impact, and external references, making it the strongest issue to anchor this cluster.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43550", + "right": "issue:45083", + "accept": false, + "reason": "Different failures: Bamba torch.compile/SDPA vs qwen3_omni_moe feature-length helper bug." + }, + { + "left": "issue:27957", + "right": "issue:34689", + "accept": false, + "reason": "Flash-Attn2 support request vs Llama 3.2 Vision model-loading regression." + }, + { + "left": "issue:29942", + "right": "issue:35141", + "accept": false, + "reason": "Failing Flash-Attn2 tests are unrelated to embedding reinitialization after resize." + }, + { + "left": "issue:38175", + "right": "issue:44493", + "accept": false, + "reason": "Zero probabilities in siglip2 is not the same as unexpected position-id keys across many models." + }, + { + "left": "issue:29870", + "right": "issue:32090", + "accept": false, + "reason": "Generate-time decoder mask bug vs Trainer/NoneType broadcast error." + }, + { + "left": "issue:43278", + "right": "issue:43329", + "accept": false, + "reason": "Embedding dtype drift on eval is unrelated to undefined vars in multimodal token counting." + }, + { + "left": "issue:44448", + "right": "issue:44977", + "accept": false, + "reason": "Different output regression for Pegasus vs Qwen3.5 flash-attention generation crash." + }, + { + "left": "issue:45920", + "right": "issue:46192", + "accept": false, + "reason": "Tokenizer ID regression vs return_tensors being ignored in text_kwargs." + }, + { + "left": "issue:45290", + "right": "issue:45381", + "accept": false, + "reason": "Chat-template tool-call crash and Qwen2.5-VL video position-id bug affect different paths." + }, + { + "left": "issue:10000", + "right": "issue:15354", + "accept": false, + "reason": "German DistilBertModel issue is unrelated to torch.jit.script GeneratorExp export failure." + }, + { + "left": "issue:44977", + "right": "issue:45229", + "accept": false, + "reason": "Flash-attention generation problem is unrelated to Gemma4 multi-GPU OOM." + }, + { + "left": "issue:43278", + "right": "issue:43381", + "accept": false, + "reason": "Eval-mode gradient checkpointing is a different bug from BF16-to-FP32 dtype changes." + }, + { + "left": "issue:29870", + "right": "issue:33290", + "accept": false, + "reason": "Decoder attention-mask assignment bug is unrelated to deepspeed adafactor OOM." + }, + { + "left": "issue:43927", + "right": "issue:45698", + "accept": false, + "reason": "Custom token IDs lost on save/load is different from loading the wrong custom module after save_pretrained." + }, + { + "left": "issue:44855", + "right": "issue:46399", + "accept": false, + "reason": "Python 3.13 import parsing error is unrelated to Qwen3.5 kernel registration/decorator failure." + }, + { + "left": "issue:45290", + "right": "issue:46655", + "accept": false, + "reason": "Tool-call chat-template crash is not the same as SmolVLM video padding calculation bug." + }, + { + "left": "issue:40444", + "right": "issue:43526", + "accept": false, + "reason": "IterableDataset multi-image finetuning failure is unrelated to BeitImageProcessorFast reduce_labels." + }, + { + "left": "issue:44857", + "right": "issue:45588", + "accept": false, + "reason": "AMP/CUDA loss crash and flash_attention s_aux=None attribute error are different code paths." + }, + { + "left": "issue:37428", + "right": "issue:42175", + "accept": false, + "reason": "Missing flash-attention helper import is unrelated to torch-only pip backend packaging." + }, + { + "left": "issue:32090", + "right": "issue:33290", + "accept": false, + "reason": "Trainer broadcast NoneType error and quantized deepspeed OOM are different symptoms and subsystems." + }, + { + "left": "issue:46421", + "right": "issue:46858", + "accept": false, + "reason": "Chunked-prefill recompilation with lazy cache init is distinct from static-cache seq_length tensor arithmetic." + }, + { + "left": "issue:44561", + "right": "issue:44779", + "accept": false, + "reason": "Removed is_torch_fx_available helper breakage is unrelated to DeepSeek tokenizer regression." + }, + { + "left": "issue:44448", + "right": "issue:44466", + "accept": false, + "reason": "Pegasus output differences do not match the lm_head.weight serialization/tied-weights issue." + }, + { + "left": "issue:43873", + "right": "issue:45229", + "accept": false, + "reason": "Quantization/offloading behavior is unrelated to Gemma4 multi-GPU CUDA OOM." + }, + { + "left": "issue:25251", + "right": "issue:32090", + "accept": false, + "reason": "Pipeline top_k nesting bug is unrelated to the Trainer broadcast NoneType error." + }, + { + "left": "issue:43976", + "right": "issue:45440", + "accept": false, + "reason": "Python-version compatibility bug is unrelated to DeepSeekV3MoE divergence from remote implementation." + }, + { + "left": "issue:42490", + "right": "issue:43873", + "accept": false, + "reason": "Trailing-slash import warning is not the same as offloading failing with quantization." + }, + { + "left": "issue:44568", + "right": "issue:45491", + "accept": false, + "reason": "Tokenizer add_special_tokens regression is unrelated to Gemma3 sliding-window NaN embeddings." + }, + { + "left": "issue:43278", + "right": "issue:43575", + "accept": false, + "reason": "Eval embedding dtype drift is unrelated to tensor-parallel OOM on Qwen2-57B." + }, + { + "left": "issue:44871", + "right": "issue:44977", + "accept": false, + "reason": "Gemma3 eos_token_id inconsistency is a different bug from Qwen3.5 flash-attention generation failure." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 33, + "review_unit_count": 60, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-500", + "nodes": [ + "issue:29127", + "issue:29870", + "issue:29942", + "issue:30064", + "issue:33357", + "issue:34567", + "issue:34689", + "issue:39692", + "issue:41720", + "issue:42490", + "issue:42915", + "issue:42994", + "issue:43065", + "issue:43066", + "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:44462", + "issue:44479", + "issue:44743", + "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:45593", + "issue:45712", + "issue:45803", + "issue:45823", + "issue:46121", + "issue:46693" + ], + "soft_pairs": [ + "issue:43381|issue:43701", + "issue:45588|issue:45823", + "issue:33357|issue:39692", + "issue:29127|issue:34567", + "issue:41720|issue:45229", + "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", + "issue:43066|issue:44462", + "issue:45440|issue:45593", + "issue:45372|issue:45459", + "issue:42490|issue:43381" + ], + "prepared_review_unit_hash": "71cd6015c21faf7a4f6f7e40763cf034a244b7872ef2c0f33546c3eb90daa161", + "component_budget": { + "node_count": 500, + "item_count": 500, + "soft_pair_count": 1839, + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 32, + "serialized_chars": 37287, + "estimated_input_tokens": 9322, + "estimated_eval_tokens": 18900 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a grab-bag of mostly unrelated Transformers issues. The soft links are generally based on broad lexical similarity (same model family, same subsystem, or same error class) rather than the same underlying bug, so none should be merged as duplicates.", + "confidence": 0.95, + "canonical_issue_reason": "No clear canonical issue: the items span unrelated model-specific regressions, tokenizer/load bugs, training-state issues, quantization problems, and documentation/examples.", + "canonical_pr_reason": null, + "best_issue_reason": "No issue is representative enough to act as a reliable duplicate target for the whole cluster.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43381", + "right": "issue:43701", + "accept": false, + "reason": "Both involve training utilities, but one is eval-mode gradient checkpointing and the other is resume-from-checkpoint key mismatch; different code paths." + }, + { + "left": "issue:45588", + "right": "issue:45823", + "accept": false, + "reason": "Different failures in Gemma4/flash-attention vs device-map forwarding; same broad area, not the same bug." + }, + { + "left": "issue:33357", + "right": "issue:39692", + "accept": false, + "reason": "Both mention vision models, but one is a MacOS CLIP bus error and the other is a SigLIP2 documentation/example issue." + }, + { + "left": "issue:29127", + "right": "issue:34567", + "accept": false, + "reason": "Unrelated problems: LayoutLMv3 error messaging vs TrainerState token-count tracking." + }, + { + "left": "issue:41720", + "right": "issue:45229", + "accept": false, + "reason": "Both are inference/OOM-like reports, but they affect different models and different failure modes (cuda assert vs CUDA OOM)." + }, + { + "left": "issue:44977", + "right": "issue:45563", + "accept": false, + "reason": "Qwen3.5 flash-attention generation bug vs a stale warning in paged generate(); not the same defect." + }, + { + "left": "issue:45198", + "right": "issue:45803", + "accept": false, + "reason": "Different tokenizer/preprocessing paths and different bugs: save/tokenization failure vs broad exception masking in Fuyu conversion." + }, + { + "left": "issue:44361", + "right": "issue:45198", + "accept": false, + "reason": "Different tokenizer families and unrelated crashes." + }, + { + "left": "issue:44186", + "right": "issue:45198", + "accept": false, + "reason": "LayoutLMv2 tokenizer padding/NER crash is unrelated to Wav2Vec2 save/tokenization failure." + }, + { + "left": "issue:43653", + "right": "issue:44743", + "accept": false, + "reason": "BigBird tokenizer special-token registration vs Qwen3.5 recurrent-state reset; no shared underlying bug." + }, + { + "left": "issue:43611", + "right": "issue:46121", + "accept": false, + "reason": "Base-model-prefix loading regression is unrelated to rope-parameter validation and JSON list handling." + }, + { + "left": "issue:29870", + "right": "issue:34689", + "accept": false, + "reason": "BART generation kwargs bug vs Llama 3.2 Vision model loading regression; different subsystems and fixes." + }, + { + "left": "issue:45372", + "right": "issue:45588", + "accept": false, + "reason": "Gemma4 processor import/dependency issue vs flash-attention AttributeError; same model family, different concrete failure." + }, + { + "left": "issue:42915", + "right": "issue:42994", + "accept": false, + "reason": "Both are quantization-related, but one is FineGrainedFP8 loading and the other is quantized-model saving." + }, + { + "left": "issue:43756", + "right": "issue:45440", + "accept": false, + "reason": "Different model implementations and different symptoms (RoPE-layer mismatch vs DeepSeekV3MoE divergence)." + }, + { + "left": "issue:43653", + "right": "issue:44442", + "accept": false, + "reason": "BigBird tokenizer special-token bug is unrelated to AutoTokenizer loading FastSpeech2ConformerTokenizer." + }, + { + "left": "issue:29942", + "right": "issue:30064", + "accept": false, + "reason": "Flash Attention 2 test failures and void-segmentation-map processing are unrelated." + }, + { + "left": "issue:44206", + "right": "issue:44479", + "accept": false, + "reason": "Unsupported center argument in image feature extraction is unrelated to a video-input regression in Qwen VL models." + }, + { + "left": "issue:45372", + "right": "issue:45406", + "accept": false, + "reason": "Related Gemma4 area, but one is a mistral_common import failure and the other is a missing _tokenizer attribute in serve; not the same code-path bug." + }, + { + "left": "issue:43412", + "right": "issue:45588", + "accept": false, + "reason": "Executorch export guard/support issues are unrelated to flash-attention sink-less model crashes." + }, + { + "left": "issue:45459", + "right": "issue:45712", + "accept": false, + "reason": "Tokenizer protobuf error handling vs leftover dummy classes; no shared bug." + }, + { + "left": "issue:45491", + "right": "issue:46693", + "accept": false, + "reason": "Both mention FA2/long context, but one is NaN embeddings and the other is a performance regression." + }, + { + "left": "issue:44265", + "right": "issue:45198", + "accept": false, + "reason": "torch.export/torch_compilable_check failure is unrelated to Wav2Vec2 save/tokenization failure." + }, + { + "left": "issue:29127", + "right": "issue:33357", + "accept": false, + "reason": "LayoutLMv3 error-message clarity and CLIP MacOS bus error are unrelated." + }, + { + "left": "issue:43065", + "right": "issue:43994", + "accept": false, + "reason": "Dummy Conv2d placeholder in SAM3 pixel decoder is unrelated to SigLIP2 nonsensical AutoModel/pipeline results." + }, + { + "left": "issue:43065", + "right": "issue:45412", + "accept": false, + "reason": "Dummy SAM3 layer cleanup is unrelated to RT-DETR memory not being released." + }, + { + "left": "issue:45245", + "right": "issue:45563", + "accept": false, + "reason": "Category-count limit error and stale generate warning are unrelated." + }, + { + "left": "issue:43575", + "right": "issue:43701", + "accept": false, + "reason": "Tensor-parallel OOM for a large Qwen model is unrelated to checkpoint key mismatch." + }, + { + "left": "issue:43066", + "right": "issue:44462", + "accept": false, + "reason": "Tokenizer decoder-type migration bug vs AutoTokenizer ignoring tokenizer.json; different loading paths and fixes." + }, + { + "left": "issue:45440", + "right": "issue:45593", + "accept": false, + "reason": "DeepSeekV3MoE divergence is unrelated to D-FINE auxiliary-loss logic." + }, + { + "left": "issue:45372", + "right": "issue:45459", + "accept": false, + "reason": "Gemma4 processor import regression is unrelated to protobuf decode-error handling in tokenizers." + }, + { + "left": "issue:42490", + "right": "issue:43381", + "accept": false, + "reason": "Local-path slash import warning is unrelated to eval-mode gradient checkpointing 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": 34, + "review_unit_count": 60, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-500", + "nodes": [ + "issue:27957", + "issue:29127", + "issue:33290", + "issue:34689", + "issue:35141", + "issue:42490", + "issue:42915", + "issue:43262", + "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:44206", + "issue:44279", + "issue:44387", + "issue:44442", + "issue:44464", + "issue:44514", + "issue:44743", + "issue:44811", + "issue:44898", + "issue:44977", + "issue:45072", + "issue:45245", + "issue:45250", + "issue:45381", + "issue:45412", + "issue:45491", + "issue:45584", + "issue:45588", + "issue:45600", + "issue:45834", + "issue:45859", + "issue:46121", + "issue:46369", + "issue:46399", + "issue:46640", + "issue:46693" + ], + "soft_pairs": [ + "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:43526|issue:46693", + "issue:43824|issue:44279", + "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:43334|issue:43531", + "issue:46121|issue:46399", + "issue:45588|issue:45600", + "issue:43381|issue:43873", + "issue:44514|issue:45381", + "issue:43761|issue:45859", + "issue:43646|issue:44077", + "issue:44977|issue:45245", + "issue:44743|issue:44811", + "issue:44442|issue:44898", + "issue:43262|issue:44206", + "issue:42490|issue:44743" + ], + "prepared_review_unit_hash": "8d1828c9158d36f878f86610b93a8cc64216a40d85b77f22cacc11c4556c931b", + "component_budget": { + "node_count": 500, + "item_count": 500, + "soft_pair_count": 1839, + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 30, + "serialized_chars": 36563, + "estimated_input_tokens": 9141, + "estimated_eval_tokens": 18538 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "All proposed pairs look like topical false positives rather than true duplicates. They share broad keywords (e.g. Flash Attention, OOM, Qwen, audio) but not the same concrete code path or fix.", + "confidence": 0.96, + "canonical_issue_reason": "No clear canonical issue emerges: the items are semantically mixed, and none of the soft pairs describe the same underlying bug/change strongly enough to anchor a duplicate cluster.", + "canonical_pr_reason": null, + "best_issue_reason": "No single issue is a good global representative because the set spans unrelated regressions and feature requests rather than one coherent problem area.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:29127", + "right": "issue:35141", + "accept": false, + "reason": "Different bugs: LayoutLMv3 error-message clarity vs token embedding reinitialization in post_init." + }, + { + "left": "issue:43421", + "right": "issue:45491", + "accept": false, + "reason": "One is about runtime special-token/post-processor updates; the other is a Gemma3 NaN embedding issue." + }, + { + "left": "issue:43994", + "right": "issue:45412", + "accept": false, + "reason": "Different subsystems and symptoms: Siglip2 bad outputs vs RT-DETR memory not being released." + }, + { + "left": "issue:45412", + "right": "issue:46369", + "accept": false, + "reason": "Both are memory-related, but one is a GC/memory leak report and the other is model call memory overflow; not the same bug." + }, + { + "left": "issue:43575", + "right": "issue:44387", + "accept": false, + "reason": "Both mention OOM, but the causes differ: tensor-parallel loading vs int4 quantization reserved-memory regression." + }, + { + "left": "issue:44442", + "right": "issue:45072", + "accept": false, + "reason": "Tokenizer loading failure vs dtype mismatch in bfloat16 inference; unrelated failures." + }, + { + "left": "issue:44464", + "right": "issue:45584", + "accept": false, + "reason": "Chunked generation with compiled forward is unrelated to Whisper empty-transcription failure after align_special_tokens." + }, + { + "left": "issue:42915", + "right": "issue:43931", + "accept": false, + "reason": "Qwen3Moe FP8 config failure and Qwen3-VL shape-mismatch loading error are different bugs." + }, + { + "left": "issue:43526", + "right": "issue:45834", + "accept": false, + "reason": "BeitImageProcessorFast label reduction bug is unrelated to Kosmos2.5 long-OCR index errors." + }, + { + "left": "issue:43526", + "right": "issue:46693", + "accept": false, + "reason": "Label reduction bug vs FA2 long-context performance regression; no shared code path." + }, + { + "left": "issue:43824", + "right": "issue:44279", + "accept": false, + "reason": "Specific missing import error vs a vague dependency issue; not the same failure." + }, + { + "left": "issue:27957", + "right": "issue:29127", + "accept": false, + "reason": "Flash Attention 2 support for XLM-Roberta is unrelated to LayoutLMv3 box-validation messaging." + }, + { + "left": "issue:27957", + "right": "issue:33290", + "accept": false, + "reason": "Flash Attention 2 support/problem vs deepspeed Adafactor OOM; different areas and fixes." + }, + { + "left": "issue:45250", + "right": "issue:46640", + "accept": false, + "reason": "Too generic on both sides; they only share the Flash Attention 2 topic, not a concrete shared bug/change." + }, + { + "left": "issue:43994", + "right": "issue:46369", + "accept": false, + "reason": "Siglip2 incorrect results and SAM3 LiteText memory overflow are unrelated model-specific problems." + }, + { + "left": "issue:33290", + "right": "issue:34689", + "accept": false, + "reason": "OOM in deepspeed Adafactor is not the same as a model-loading regression for Llama 3.2 Vision." + }, + { + "left": "issue:43526", + "right": "issue:44186", + "accept": false, + "reason": "Different image/tokenization bugs: label reduction in BeitImageProcessorFast vs LayoutLMv2Tokenizer crashes." + }, + { + "left": "issue:33290", + "right": "issue:35141", + "accept": false, + "reason": "Optimizer OOM and embedding reinitialization are unrelated issues." + }, + { + "left": "issue:43334", + "right": "issue:43531", + "accept": false, + "reason": "Missing pad_token_id in Qwen3-VL text config is unrelated to a sliding_window bug in Qwen3-MoE." + }, + { + "left": "issue:46121", + "right": "issue:46399", + "accept": false, + "reason": "RoPE validation TypeError and rotary-pos-emb registration failure are distinct code paths." + }, + { + "left": "issue:45588", + "right": "issue:45600", + "accept": false, + "reason": "Flash-attention sink-less AttributeError is unrelated to stale auto_mappings config references." + }, + { + "left": "issue:43381", + "right": "issue:43873", + "accept": false, + "reason": "Gradient checkpointing in eval mode and quantization offloading are different concerns." + }, + { + "left": "issue:44514", + "right": "issue:45381", + "accept": false, + "reason": "Batched apply_chat_template padding crash is not the same as wrong vision_position_ids in Qwen2.5-VL video input." + }, + { + "left": "issue:43761", + "right": "issue:45859", + "accept": false, + "reason": "CLIP hidden_states regression and Qwen3_5Moe CPU-offload incompatibility are unrelated." + }, + { + "left": "issue:43646", + "right": "issue:44077", + "accept": false, + "reason": "Generic custom-model initialization break is broader than the specific patchtsmixer post_init policy issue." + }, + { + "left": "issue:44977", + "right": "issue:45245", + "accept": false, + "reason": "Qwen3.5 flash-attention generation and a categories-limit runtime error are unrelated." + }, + { + "left": "issue:44743", + "right": "issue:44811", + "accept": false, + "reason": "Recurrent-state reset with cache is unrelated to Whisper batch_decode skip_special_tokens handling." + }, + { + "left": "issue:44442", + "right": "issue:44898", + "accept": false, + "reason": "AutoTokenizer loading failure and Perceiver non-default-resolution failure are different bugs." + }, + { + "left": "issue:43262", + "right": "issue:44206", + "accept": false, + "reason": "Audio chat-template sampling-rate defaulting and an unsupported center arg in LasrFeatureExtractor are separate issues." + }, + { + "left": "issue:42490", + "right": "issue:44743", + "accept": false, + "reason": "Trailing-slash local-model import warning has nothing to do with recurrent state reset in Qwen3." + } + ] + }, + "evaluator_result": 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": 60, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-500", + "nodes": [ + "issue:27957", + "issue:29870", + "issue:32090", + "issue:42175", + "issue:42617", + "issue:42907", + "issue:43232", + "issue:43316", + "issue:43388", + "issue:43493", + "issue:43525", + "issue:43550", + "issue:43575", + "issue:43606", + "issue:43643", + "issue:43701", + "issue:43720", + "issue:43742", + "issue:43749", + "issue:43761", + "issue:43827", + "issue:43873", + "issue:43931", + "issue:44062", + "issue:44077", + "issue:44079", + "issue:44188", + "issue:44190", + "issue:44442", + "issue:44479", + "issue:44496", + "issue:44855", + "issue:44871", + "issue:44913", + "issue:44987", + "issue:45003", + "issue:45137", + "issue:45216", + "issue:45325", + "issue:45440", + "issue:45698", + "issue:45759", + "issue:45803", + "issue:45907", + "issue:46178", + "issue:46234", + "issue:46506", + "issue:46697" + ], + "soft_pairs": [ + "issue:45003|issue:45440", + "issue:43720|issue:44442", + "issue:42175|issue:46234", + "issue:27957|issue:29870", + "issue:43493|issue:43827", + "issue:43316|issue:43525", + "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:43761|issue:44079", + "issue:44190|issue:44496", + "issue:43761|issue:44188", + "issue:44987|issue:46178", + "issue:43931|issue:45216", + "issue:44442|issue:45803", + "issue:43232|issue:43388", + "issue:42907|issue:45698", + "issue:43701|issue:46506", + "issue:42617|issue:43575", + "issue:27957|issue:32090", + "issue:43550|issue:43701", + "issue:43575|issue:43742", + "issue:43606|issue:43701" + ], + "prepared_review_unit_hash": "1a00ed5b77468b8943edc35a61c3a957c8ea10ec0b0a291dfa3de6abccb653a9", + "component_budget": { + "node_count": 500, + "item_count": 500, + "soft_pair_count": 1839, + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 28, + "serialized_chars": 36217, + "estimated_input_tokens": 9055, + "estimated_eval_tokens": 18366 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is highly heterogeneous: it mixes unrelated bugs in generation, loading, configs, quantization/offloading, tokenizer behavior, docs, and distributed training. None of the soft pairs appears to be the same underlying issue, so I rejected them all.", + "confidence": 0.95, + "canonical_issue_reason": "If a single representative is needed, issue 43701 is the clearest self-contained bug report in the set, but the cluster is too diverse to justify a true canonical duplicate.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 43701 is the most broadly readable and concrete report, but only as a weak representative; it is not meaningfully connected to the rest of the cluster.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:45003", + "right": "issue:45440", + "accept": false, + "reason": "Different bug classes and code paths: unsafe sys.modules access vs divergence in DeepseekV3MoE implementation." + }, + { + "left": "issue:43720", + "right": "issue:44442", + "accept": false, + "reason": "BitNet accelerate-loading failure and tokenizer-loading failure are unrelated." + }, + { + "left": "issue:42175", + "right": "issue:46234", + "accept": false, + "reason": "Package extra dependency coverage vs missing import utility symbol; not the same bug." + }, + { + "left": "issue:27957", + "right": "issue:29870", + "accept": false, + "reason": "FlashAttention support for XLMRoberta is unrelated to BART generate() decoder_attention_mask initialization." + }, + { + "left": "issue:43493", + "right": "issue:43827", + "accept": false, + "reason": "Model implementation discrepancy vs stale docs referencing pipeline(); different issues." + }, + { + "left": "issue:43316", + "right": "issue:43525", + "accept": false, + "reason": "Both are config-related, but they concern different models and different missing/changed fields." + }, + { + "left": "issue:44062", + "right": "issue:45907", + "accept": false, + "reason": "Tokenizer AddedToken argument duplication and list_repo_files revision forwarding are unrelated." + }, + { + "left": "issue:44077", + "right": "issue:44479", + "accept": false, + "reason": "Optional post_init validation for patchtsmixer is not the same as the Qwen video-input regression." + }, + { + "left": "issue:45325", + "right": "issue:46697", + "accept": false, + "reason": "Qwen2.5-VL RoPE scaling bug and Swin2SR padding bug affect different preprocessing logic." + }, + { + "left": "issue:43873", + "right": "issue:44871", + "accept": false, + "reason": "Quantization/offloading behavior and Gemma eos_token_id mismatch are not the same underlying defect." + }, + { + "left": "issue:44855", + "right": "issue:45137", + "accept": false, + "reason": "Python 3.13 torch.jit parsing/indentation issue is unrelated to DeepSpeed ZeRO3 deque underflow." + }, + { + "left": "issue:43643", + "right": "issue:44913", + "accept": false, + "reason": "Remote-code config field loss vs GPTNeoX rotary_pct persistence are different config-loading problems." + }, + { + "left": "issue:43749", + "right": "issue:43931", + "accept": false, + "reason": "FSDP CPU RAM-efficient loading breakage and Qwen3-VL shape mismatch are distinct loading failures." + }, + { + "left": "issue:43316", + "right": "issue:45759", + "accept": false, + "reason": "Gemma3TextConfig API discrepancy and composite Qwen multimodal config unwrapping are different config/model-selection bugs." + }, + { + "left": "issue:43761", + "right": "issue:44079", + "accept": false, + "reason": "CLIPVisionModel hidden_states regression and ModelOutput key assignment when None are separate runtime behaviors." + }, + { + "left": "issue:44190", + "right": "issue:44496", + "accept": false, + "reason": "Local dataset loading in a training script is unrelated to AutoConfig failing on an unrecognized model." + }, + { + "left": "issue:44442", + "right": "issue:45803", + "accept": false, + "reason": "Tokenizer loading failure for FastSpeech2Conformer and bare except swallowing interrupts/errors in FuyuBatchFeature are unrelated." + }, + { + "left": "issue:43232", + "right": "issue:43388", + "accept": false, + "reason": "Generation kwargs update after sync_gpus and gather_for_metrics tuple-label dropping affect different subsystems." + }, + { + "left": "issue:42907", + "right": "issue:45698", + "accept": false, + "reason": "Saving dequantized Ministrals/Devstrals and loading the wrong custom module after save_pretrained are distinct save/load issues." + }, + { + "left": "issue:43701", + "right": "issue:46506", + "accept": false, + "reason": "Checkpoint-key mismatch and CPU/CUDA output divergence are not the same bug." + }, + { + "left": "issue:42617", + "right": "issue:43575", + "accept": false, + "reason": "A script launch problem and a tensor-parallel OOM on a specific model are different failures." + }, + { + "left": "issue:27957", + "right": "issue:32090", + "accept": false, + "reason": "FlashAttention2 support for XLMRoberta is unrelated to Trainer GPU broadcast rejecting None." + }, + { + "left": "issue:43550", + "right": "issue:43701", + "accept": false, + "reason": "torch.compile SDPA failure in Bamba-9B-v2 is unrelated to resume_from_checkpoint key mismatch." + }, + { + "left": "issue:43575", + "right": "issue:43742", + "accept": false, + "reason": "Qwen2-57B tensor-parallel OOM and MobileLLM key error do not share a concrete code-path." + }, + { + "left": "issue:43606", + "right": "issue:43701", + "accept": false, + "reason": "CPU offload device mismatch and checkpoint key mismatch are different bugs in different flows." + }, + { + "left": "issue:43761", + "right": "issue:44188", + "accept": false, + "reason": "Hidden_states regression in CLIPVisionModel and attention-kernel divergence under allow_is_bidirectional_skip are unrelated." + }, + { + "left": "issue:44987", + "right": "issue:46178", + "accept": false, + "reason": "Version-specific model-loading failure and missing BACKENDS_MAPPING entry for tensorflow_text are not the same issue." + }, + { + "left": "issue:43931", + "right": "issue:45216", + "accept": false, + "reason": "Shape mismatch when loading Qwen3-VL and incorrect save_pretrained output for Qwen3.5 are separate load/save 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": 36, + "review_unit_count": 60, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-500", + "nodes": [ + "issue:16998", + "issue:22355", + "issue:25251", + "issue:28282", + "issue:42175", + "issue:43295", + "issue:43316", + "issue:43352", + "issue:43404", + "issue:43441", + "issue:43452", + "issue:43504", + "issue:43540", + "issue:43572", + "issue:43646", + "issue:43650", + "issue:43716", + "issue:43761", + "issue:43825", + "issue:44077", + "issue:44162", + "issue:44242", + "issue:44464", + "issue:44479", + "issue:44496", + "issue:44514", + "issue:44534", + "issue:44610", + "issue:44792", + "issue:44871", + "issue:44913", + "issue:45542", + "issue:45584", + "issue:45632", + "issue:45698", + "issue:45803", + "issue:46050", + "issue:46170", + "issue:46178", + "issue:46202", + "issue:46234", + "issue:46291", + "issue:46365", + "issue:46369", + "issue:46421", + "issue:46424", + "issue:46640", + "issue:46894" + ], + "soft_pairs": [ + "issue:25251|issue:28282", + "issue:46234|issue:46291", + "issue:43295|issue:43825", + "issue:16998|issue:22355", + "issue:43295|issue:44871", + "issue:44871|issue:45584", + "issue:44464|issue:46421", + "issue:43316|issue:43716", + "issue:44077|issue:44534", + "issue:42175|issue:43316", + "issue:44871|issue:46050", + "issue:43646|issue:44913", + "issue:45803|issue:46365", + "issue:44162|issue:44792", + "issue:43452|issue:44496", + "issue:43404|issue:43441", + "issue:44242|issue:46894", + "issue:45632|issue:46424", + "issue:45803|issue:46170", + "issue:44610|issue:46369", + "issue:43650|issue:46202", + "issue:43761|issue:44479", + "issue:43504|issue:45698", + "issue:45542|issue:46234", + "issue:43352|issue:46640", + "issue:43540|issue:44514", + "issue:43572|issue:46178" + ], + "prepared_review_unit_hash": "d088350cedc2175181464014623910b4e31cfc927d685804f0da835cd54e8bcc", + "component_budget": { + "node_count": 500, + "item_count": 500, + "soft_pair_count": 1839, + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 27, + "serialized_chars": 35662, + "estimated_input_tokens": 8916, + "estimated_eval_tokens": 18088 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is a loose grab-bag of unrelated Transformers issues. None of the soft pairs look like the same concrete bug or change, so all are rejected.", + "confidence": 0.87, + "canonical_issue_reason": "issue:45698 is the strongest standalone bug report here: it is open, has the most discussion, and describes a concrete reproducible regression in from_pretrained/save_pretrained behavior.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:45698 is the best overall issue candidate because it is specific, user-facing, and already has active investigation; the other issues are narrower or clearly unrelated to one another.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:25251", + "right": "issue:28282", + "accept": false, + "reason": "Different bugs: pipeline output shape regression vs missing PyTorch import for AutoModel." + }, + { + "left": "issue:46234", + "right": "issue:46291", + "accept": false, + "reason": "One is a missing import_utils symbol; the other is a kernels dependency/version check on importing pipelines." + }, + { + "left": "issue:43295", + "right": "issue:43825", + "accept": false, + "reason": "Both are v5 regressions, but one is about processor/tokenizer access for custom models and the other is an incorrect pipeline task message." + }, + { + "left": "issue:16998", + "right": "issue:22355", + "accept": false, + "reason": "Unrelated: model_max_length handling for DeBERTa-V3 vs missing transformers.onnx module." + }, + { + "left": "issue:43295", + "right": "issue:44871", + "accept": false, + "reason": "Different model families and failures: processor/tokenizer regression vs Gemma eos_token_id mismatch." + }, + { + "left": "issue:44871", + "right": "issue:45584", + "accept": false, + "reason": "Gemma eos token configuration bug vs Whisper empty-transcription failure after token alignment." + }, + { + "left": "issue:44464", + "right": "issue:46421", + "accept": false, + "reason": "Chunked generation inconsistency with compiled forward is not the same as spurious recompilation from cache lazy init." + }, + { + "left": "issue:43316", + "right": "issue:43716", + "accept": false, + "reason": "API discrepancy in Gemma3TextConfig is unrelated to Mistral-3 image preprocessor/model dtype mismatch." + }, + { + "left": "issue:44077", + "right": "issue:44534", + "accept": false, + "reason": "PatchTSMixer post_init restriction is a model-class API issue, not the non-persistent buffer junk-fill regression." + }, + { + "left": "issue:42175", + "right": "issue:43316", + "accept": false, + "reason": "Backend packaging/install issue vs config API discrepancy; no shared underlying defect." + }, + { + "left": "issue:44871", + "right": "issue:46050", + "accept": false, + "reason": "Gemma eos_token_id inconsistency is distinct from quantized model save/load breakage." + }, + { + "left": "issue:43646", + "right": "issue:44913", + "accept": false, + "reason": "General custom model initialization breakage is not the same as GPTNeoX rotary_pct not persisting on reload." + }, + { + "left": "issue:45803", + "right": "issue:46365", + "accept": false, + "reason": "Bare except swallowing errors in Fuyu conversion is unrelated to Trainer import being blocked by datasets." + }, + { + "left": "issue:44162", + "right": "issue:44792", + "accept": false, + "reason": "ESM2 breakage and a janus image-generation test failure are different models and failure modes." + }, + { + "left": "issue:43452", + "right": "issue:44496", + "accept": false, + "reason": "Both concern model loading, but one is gguf_file handling for from_pretrained and the other is missing model_type in config.json." + }, + { + "left": "issue:43404", + "right": "issue:43441", + "accept": false, + "reason": "lm_head tying in Mistral3ForConditionalGeneration is a different bug from a FlashAttention support failure in Ministral-3." + }, + { + "left": "issue:44242", + "right": "issue:46894", + "accept": false, + "reason": "Both are loss-related, but they affect different models and mechanisms; the training-loss double-shift bug is not the same as missing load-balancing loss." + }, + { + "left": "issue:45632", + "right": "issue:46424", + "accept": false, + "reason": "Cache path collision for local trust_remote_code models is unrelated to max_cache_length only applying to quantized cache." + }, + { + "left": "issue:45803", + "right": "issue:46170", + "accept": false, + "reason": "Different environment failures: Fuyu exception handling vs Windows/ROCm torch.distributed dependency." + }, + { + "left": "issue:44610", + "right": "issue:46369", + "accept": false, + "reason": "Processor input-size mismatch for OmDet-Turbo is unrelated to sam3-litetext memory overflow." + }, + { + "left": "issue:43650", + "right": "issue:46202", + "accept": false, + "reason": "A malformed issue title about data is unrelated to Gemma pipeline roundtrip failures." + }, + { + "left": "issue:43761", + "right": "issue:44479", + "accept": false, + "reason": "CLIPVisionModel hidden_states regression is not the same as the Qwen video-input regression." + }, + { + "left": "issue:43504", + "right": "issue:45698", + "accept": false, + "reason": "Legacy field handling in Beit loading is a different concrete bug from wrong custom module loading after save_pretrained." + }, + { + "left": "issue:45542", + "right": "issue:46234", + "accept": false, + "reason": "TensorBoard/TensorFlow backend packaging issue is unrelated to a missing import_utils helper." + }, + { + "left": "issue:43352", + "right": "issue:46640", + "accept": false, + "reason": "The FlashAttention-2 support complaint is too vague to equate with the NemotronHForCausalLM support error." + }, + { + "left": "issue:43540", + "right": "issue:44514", + "accept": false, + "reason": "Qwen3OmniMoe video-processing ValueError and Qwen2_5_VLProcessor batched apply_chat_template crash are different paths." + }, + { + "left": "issue:43572", + "right": "issue:46178", + "accept": false, + "reason": "StableLmConfig pad_token_idx regression is unrelated to the tensorflow_text backend mapping 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": 37, + "review_unit_count": 60, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-500", + "nodes": [ + "issue:25251", + "issue:27957", + "issue:29870", + "issue:35141", + "issue:41950", + "issue:42490", + "issue:43065", + "issue:43388", + "issue:43475", + "issue:43525", + "issue:43531", + "issue:43644", + "issue:43673", + "issue:43688", + "issue:43827", + "issue:43931", + "issue:43976", + "issue:43994", + "issue:44060", + "issue:44190", + "issue:44242", + "issue:44534", + "issue:44617", + "issue:44734", + "issue:44743", + "issue:44792", + "issue:44855", + "issue:44869", + "issue:44913", + "issue:44933", + "issue:44964", + "issue:44977", + "issue:45127", + "issue:45200", + "issue:45245", + "issue:45362", + "issue:45406", + "issue:45584", + "issue:45588", + "issue:45600", + "issue:45632", + "issue:46082", + "issue:46421", + "issue:46617", + "issue:46697", + "issue:46832", + "issue:46858", + "issue:46897" + ], + "soft_pairs": [ + "issue:44792|issue:45600", + "issue:43065|issue:44617", + "issue:43388|issue:43688", + "issue:44792|issue:44977", + "issue:45127|issue:45200", + "issue:43673|issue:44743", + "issue:41950|issue:43475", + "issue:44190|issue:44964", + "issue:43525|issue:43531", + "issue:44534|issue:44913", + "issue:35141|issue:42490", + "issue:44734|issue:45406", + "issue:46082|issue:46697", + "issue:45632|issue:46617", + "issue:44869|issue:46832", + "issue:44855|issue:45588", + "issue:45245|issue:46858", + "issue:43827|issue:43994", + "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" + ], + "prepared_review_unit_hash": "aed5e4635fb2ee774cfb77df6f6442aab4951b77dd104f895847af99376ab2c9", + "component_budget": { + "node_count": 500, + "item_count": 500, + "soft_pair_count": 1839, + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 27, + "serialized_chars": 35949, + "estimated_input_tokens": 8988, + "estimated_eval_tokens": 18232 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is mostly heterogeneous; several items are only loosely related by subsystem, but none of the soft-edge pairs look like the same concrete bug or change. The best canonical issue among the set is the non-persistent-buffer regression report.", + "confidence": 0.45, + "canonical_issue_reason": "Issue 43644 is the clearest and earliest report of a concrete regression ('non-persistent buffers filled with junk'), and it is the most canonical-looking candidate in the set.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 43644 is the most specific, actionable, and representative report among the issues here; it describes a clear runtime regression rather than a broad symptom or a secondary follow-up.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44792", + "right": "issue:45600", + "accept": false, + "reason": "Different bug surfaces: a janus test failure versus a removed Sam3LiteText config reference." + }, + { + "left": "issue:43065", + "right": "issue:44617", + "accept": false, + "reason": "Unrelated symptoms: a dummy Conv2d implementation detail versus a CUDA OOM report." + }, + { + "left": "issue:43388", + "right": "issue:43688", + "accept": false, + "reason": "Different training/eval bugs: metrics label truncation versus auxiliary loss normalization." + }, + { + "left": "issue:44792", + "right": "issue:44977", + "accept": false, + "reason": "Same general area of generation, but different models and code paths; not the same bug." + }, + { + "left": "issue:45127", + "right": "issue:45200", + "accept": false, + "reason": "One is about LoRA merge/tied embeddings, the other about mm_token_type_ids defaults; separate issues." + }, + { + "left": "issue:43673", + "right": "issue:44743", + "accept": false, + "reason": "Both involve cache-like behavior, but one is missing cache in chunked prefill and the other resets recurrent states." + }, + { + "left": "issue:41950", + "right": "issue:43475", + "accept": false, + "reason": "Different SAM3 failures: pipeline processor lookup versus a missing vision encoder attribute." + }, + { + "left": "issue:44190", + "right": "issue:44964", + "accept": false, + "reason": "Different loading paths: local dataset script failure versus a specific multimodal model load failure." + }, + { + "left": "issue:43525", + "right": "issue:43531", + "accept": false, + "reason": "Unrelated model config/attention bugs: missing pad_token_id versus sliding_window behavior." + }, + { + "left": "issue:44534", + "right": "issue:44913", + "accept": false, + "reason": "Both are configuration/state issues, but non-persistent buffers corruption is not the same as rotary_pct reload behavior." + }, + { + "left": "issue:35141", + "right": "issue:42490", + "accept": false, + "reason": "Different problems: embedding reinitialization versus a path-with-slash import warning." + }, + { + "left": "issue:44734", + "right": "issue:45406", + "accept": false, + "reason": "Separate serve/model failures: KV-cache continuation indexing versus a missing processor attribute." + }, + { + "left": "issue:46082", + "right": "issue:46697", + "accept": false, + "reason": "Completely different domains: head_dim validation versus image padding behavior." + }, + { + "left": "issue:45632", + "right": "issue:46617", + "accept": false, + "reason": "Both touch trust_remote_code caching, but one is a leaf-directory collision and the other is a FileNotFoundError; not the same concrete bug." + }, + { + "left": "issue:44869", + "right": "issue:46832", + "accept": false, + "reason": "Unrelated model paths: Whisper timestamp decoding versus RT-DETR feature-level mismatch." + }, + { + "left": "issue:44855", + "right": "issue:45588", + "accept": false, + "reason": "Different failure modes and code paths: Python 3.13 JIT parsing versus flash_attention handling of s_aux=None." + }, + { + "left": "issue:45245", + "right": "issue:46858", + "accept": false, + "reason": "A category-count limit error is unrelated to a device-tensor slice arithmetic bug in cache prefill." + }, + { + "left": "issue:43827", + "right": "issue:43994", + "accept": false, + "reason": "One is a docs-removal mismatch, the other is a model/pipeline inference bug; not the same issue." + }, + { + "left": "issue:43673", + "right": "issue:46421", + "accept": false, + "reason": "Related chunked-prefill/cache area, but missing cache and lazy-init recompilation are distinct concrete regressions." + }, + { + "left": "issue:43931", + "right": "issue:44060", + "accept": false, + "reason": "Different Qwen model issues: weight-shape mismatch versus tied-weights warning contamination." + }, + { + "left": "issue:44242", + "right": "issue:46897", + "accept": false, + "reason": "Both concern loss computation, but one is missing load-balancing loss and the other is a double-shift training bug." + }, + { + "left": "issue:29870", + "right": "issue:35141", + "accept": false, + "reason": "Unrelated bugs: BART generate-time decoder mask initialization versus embedding reinitialization after resize." + }, + { + "left": "issue:44617", + "right": "issue:45584", + "accept": false, + "reason": "Different models and failures: SAM3 video OOM versus Whisper empty-transcription decoding." + }, + { + "left": "issue:43644", + "right": "issue:44913", + "accept": false, + "reason": "Non-persistent buffer corruption and rotary_pct reload are separate persistence/config bugs." + }, + { + "left": "issue:44933", + "right": "issue:45362", + "accept": false, + "reason": "Different domains entirely: image_utils import cleanup versus a Qwen3.5 chat crash." + }, + { + "left": "issue:43976", + "right": "issue:44855", + "accept": false, + "reason": "Separate version-compatibility issues: Python 3.9/3.10 support versus a Python 3.13 IndentationError." + } + ] + }, + "evaluator_result": 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": 60, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-500", + "nodes": [ + "issue:25251", + "issue:29870", + "issue:39692", + "issue:41720", + "issue:43065", + "issue:43295", + "issue:43335", + "issue:43526", + "issue:43540", + "issue:43550", + "issue:43673", + "issue:43854", + "issue:44077", + "issue:44112", + "issue:44190", + "issue:44462", + "issue:44464", + "issue:44479", + "issue:44521", + "issue:44610", + "issue:44617", + "issue:44734", + "issue:44737", + "issue:44779", + "issue:44877", + "issue:44938", + "issue:44964", + "issue:45127", + "issue:45245", + "issue:45276", + "issue:45290", + "issue:45357", + "issue:45381", + "issue:45542", + "issue:45563", + "issue:45584", + "issue:45600", + "issue:45632", + "issue:45803", + "issue:45907", + "issue:46234", + "issue:46369", + "issue:46489", + "issue:46655", + "issue:46697", + "issue:46752", + "issue:46858" + ], + "soft_pairs": [ + "issue:45563|issue:45584", + "issue:43550|issue:44190", + "issue:45381|issue:46369", + "issue:44077|issue:44877", + "issue:44479|issue:44734", + "issue:41720|issue:45245", + "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:43295|issue:45276", + "issue:44617|issue:46369", + "issue:43526|issue:46697", + "issue:45290|issue:46752", + "issue:44462|issue:45907", + "issue:45276|issue:45584", + "issue:44737|issue:45803", + "issue:43540|issue:44734", + "issue:44938|issue:45542", + "issue:25251|issue:29870", + "issue:44521|issue:45381", + "issue:43673|issue:46858", + "issue:43854|issue:44964", + "issue:45632|issue:46234" + ], + "prepared_review_unit_hash": "929d7d544870e43fcf896e34039c982d97152a521ef5a36aa80d6c8763764f66", + "component_budget": { + "node_count": 500, + "item_count": 500, + "soft_pair_count": 1839, + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 28, + "serialized_chars": 35695, + "estimated_input_tokens": 8924, + "estimated_eval_tokens": 18104 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a grab bag of mostly unrelated single-issue bugs across generation, tokenization, multimodal processing, configs, and CI. None of the soft-edge pairs look like true duplicates.", + "confidence": 0.94, + "canonical_issue_reason": "No clear duplicate family emerges from the cluster, so there is no strong canonical issue. If a representative must be picked, issue 45584 is the cleanest self-contained bug report with a specific failure mode and reproducible symptoms.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 45584 is the best representative issue in this set: it has a focused title, a concrete user-facing crash, and enough detail to stand in as a canonical example of the cluster\u2019s overall bug style.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:45563", + "right": "issue:45584", + "accept": false, + "reason": "Different generation bugs: a stale warning for num_return_sequences vs an empty-transcription crash after align_special_tokens." + }, + { + "left": "issue:43550", + "right": "issue:44190", + "accept": false, + "reason": "Unrelated failures: Bamba torch.compile/SDPA assert vs a dataset-loading script error." + }, + { + "left": "issue:45381", + "right": "issue:46369", + "accept": false, + "reason": "Both touch multimodal models, but one is qwen video vision_position_ids and the other is Sam3 LiteText memory overflow." + }, + { + "left": "issue:44077", + "right": "issue:44877", + "accept": false, + "reason": "Different config problems: optional post_init handling vs strict config loading for granite_speech." + }, + { + "left": "issue:44479", + "right": "issue:44734", + "accept": false, + "reason": "Different code paths: video-input regression in Qwen VL variants vs serve KV-cache continuation indexing." + }, + { + "left": "issue:41720", + "right": "issue:45245", + "accept": false, + "reason": "CUDA auto device-mapping assert and category-count overflow are unrelated bugs." + }, + { + "left": "issue:43335", + "right": "issue:44190", + "accept": false, + "reason": "SwitchTransformers sparse-layer construction is unrelated to the local dataset-loading script issue." + }, + { + "left": "issue:43065", + "right": "issue:44610", + "accept": false, + "reason": "Sam3PixelDecoder dummy Conv2d and OmDet-Turbo input-size mismatch are different model/processor bugs." + }, + { + "left": "issue:44464", + "right": "issue:45563", + "accept": false, + "reason": "Different generation issues: inconsistent compiled chunked generation vs a stale warning for num_return_sequences." + }, + { + "left": "issue:39692", + "right": "issue:44190", + "accept": false, + "reason": "Documentation example errors are unrelated to the image-classification script failing on local datasets." + }, + { + "left": "issue:45127", + "right": "issue:45357", + "accept": false, + "reason": "LoRA/extended-vocab collapse and incorrect visual encoder keys on save_pretrained are separate save/load problems." + }, + { + "left": "issue:45381", + "right": "issue:46655", + "accept": false, + "reason": "Both are video-related, but they affect different models and different failure modes." + }, + { + "left": "issue:44112", + "right": "issue:45600", + "accept": false, + "reason": "Both mention CI, but one is a stale device override test and the other is a broken auto_mappings reference." + }, + { + "left": "issue:44779", + "right": "issue:46489", + "accept": false, + "reason": "Related DeepSeek family, but not the same bug: generic tokenizer regression vs a DeepSeek-Coder v1 gap in a prior fix." + }, + { + "left": "issue:43295", + "right": "issue:45276", + "accept": false, + "reason": "Processor.tokenizer regression and resize_token_embeddings propagation are different mechanisms." + }, + { + "left": "issue:44617", + "right": "issue:46369", + "accept": false, + "reason": "Sam3Video OOM and Sam3-LiteText memory overflow are different resource issues in different variants." + }, + { + "left": "issue:43526", + "right": "issue:46697", + "accept": false, + "reason": "ImageProcessorFast label reduction and Swin2SR over-padding are unrelated preprocessing bugs." + }, + { + "left": "issue:45290", + "right": "issue:46752", + "accept": false, + "reason": "apply_chat_template crashes in different scenarios: tool-call assistant messages vs an empty list." + }, + { + "left": "issue:44462", + "right": "issue:45907", + "accept": false, + "reason": "AutoTokenizer ignoring tokenizer.json is unrelated to list_repo_files missing the revision argument." + }, + { + "left": "issue:45276", + "right": "issue:45584", + "accept": false, + "reason": "Embedding resize propagation and Whisper empty transcription are unrelated bugs." + }, + { + "left": "issue:44737", + "right": "issue:45803", + "accept": false, + "reason": "XLNet CPU tensor allocation and Fuyu bare except swallowing errors are different issues." + }, + { + "left": "issue:43540", + "right": "issue:44734", + "accept": false, + "reason": "Qwen3OmniMoe video-input ValueError and serve KV-cache continuation indexing are unrelated." + }, + { + "left": "issue:44938", + "right": "issue:45542", + "accept": false, + "reason": "Python 3.14 import failure and the tensorboard/TensorFlow backend issue do not share a code path." + }, + { + "left": "issue:25251", + "right": "issue:29870", + "accept": false, + "reason": "Pipeline top_k nesting and Bart decoder_attention_mask initialization are unrelated bugs." + }, + { + "left": "issue:44521", + "right": "issue:45381", + "accept": false, + "reason": "Both are multimodal issues, but assistant_masks being all zeros is not the same as incorrect vision_position_ids." + }, + { + "left": "issue:43673", + "right": "issue:46858", + "accept": false, + "reason": "Chunked_prefill cache missing and static-cache seq_length tensor arithmetic are different cache bugs." + }, + { + "left": "issue:43854", + "right": "issue:44964", + "accept": false, + "reason": "Two separate model-loading failures for different models and different underlying causes." + }, + { + "left": "issue:45632", + "right": "issue:46234", + "accept": false, + "reason": "Cache-path collision for local models is unrelated to a missing import_utils helper symbol." + } + ] + }, + "evaluator_result": 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": 60, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-500", + "nodes": [ + "issue:27957", + "issue:32090", + "issue:35141", + "issue:36246", + "issue:38617", + "issue:41628", + "issue:41950", + "issue:42175", + "issue:42796", + "issue:42907", + "issue:43295", + "issue:43381", + "issue:43404", + "issue:43441", + "issue:43479", + "issue:43526", + "issue:43531", + "issue:43643", + "issue:43650", + "issue:43673", + "issue:43901", + "issue:43976", + "issue:44016", + "issue:44112", + "issue:44220", + "issue:44246", + "issue:44279", + "issue:44295", + "issue:44336", + "issue:44496", + "issue:44521", + "issue:44683", + "issue:44792", + "issue:44964", + "issue:44987", + "issue:45003", + "issue:45042", + "issue:45083", + "issue:45216", + "issue:45335", + "issue:45375", + "issue:45446", + "issue:45593", + "issue:45600", + "issue:46082", + "issue:46121", + "issue:46424", + "issue:46506" + ], + "soft_pairs": [ + "issue:44246|issue:45003", + "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", + "issue:43976|issue:44987", + "issue:43295|issue:45335", + "issue:44016|issue:44220", + "issue:45083|issue:46506", + "issue:42907|issue:43441", + "issue:44683|issue:45446", + "issue:43526|issue:43901", + "issue:43976|issue:44279", + "issue:43381|issue:45593", + "issue:43479|issue:45375", + "issue:44496|issue:45042", + "issue:43901|issue:44496", + "issue:43673|issue:44336", + "issue:44792|issue:46424" + ], + "prepared_review_unit_hash": "f693f288b874fc3bd32465a300d4f84d656ee87b4d206492211ca77b29fb743d", + "component_budget": { + "node_count": 500, + "item_count": 500, + "soft_pair_count": 1839, + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 31, + "serialized_chars": 36931, + "estimated_input_tokens": 9233, + "estimated_eval_tokens": 18722 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a loose grab bag of Transformers issues with only superficial topical overlap (imports, configs, multimodal models, caches, etc.). None of the proposed soft pairs look like the same underlying bug or change, so all soft edges should be rejected.", + "confidence": 0.84, + "canonical_issue_reason": "issue:44964 is the most concrete, reproducible, and triage-friendly report in the set: it names a specific model, a specific regression point, and a clear load failure mode.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:44964 is the best single issue anchor because it is narrowly scoped and actionable, unlike the broader packaging/import/doc or vague regression reports elsewhere in the cluster.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44246", + "right": "issue:45003", + "accept": false, + "reason": "Import-performance complaint vs unsafe sys.modules access: related area, but different bugs and different fixes." + }, + { + "left": "issue:44964", + "right": "issue:45600", + "accept": false, + "reason": "Different model families and failure modes: Phi-4 multimodal load regression vs removed Sam3LiteText config refs breaking CI." + }, + { + "left": "issue:43643", + "right": "issue:44496", + "accept": false, + "reason": "Both are model/config loading errors, but one is missing fields under trust_remote_code and the other is missing model_type for a specific repo." + }, + { + "left": "issue:44112", + "right": "issue:44964", + "accept": false, + "reason": "CI test flake in GraniteSpeech is unrelated to the Phi-4 multimodal loading regression." + }, + { + "left": "issue:27957", + "right": "issue:35141", + "accept": false, + "reason": "FlashAttention support for XLM-R is unrelated to embedding reinitialization when resizing token embeddings." + }, + { + "left": "issue:42175", + "right": "issue:44279", + "accept": false, + "reason": "Both mention dependencies, but one is a pip extras/backend packaging complaint and the other is a separate general dependency issue." + }, + { + "left": "issue:46082", + "right": "issue:46121", + "accept": false, + "reason": "Both are config-related, but they hit different validation paths and different inputs." + }, + { + "left": "issue:36246", + "right": "issue:41628", + "accept": false, + "reason": "Different missing symbols: Qwen2_5_VLImageProcessor vs AutoImageProcessor." + }, + { + "left": "issue:42796", + "right": "issue:44521", + "accept": false, + "reason": "Tokenizer BPE-marker handling and multimodal assistant_mask generation are different code paths and bugs." + }, + { + "left": "issue:36246", + "right": "issue:38617", + "accept": false, + "reason": "Both are import errors, but for different removed/missing names with no shared concrete fix." + }, + { + "left": "issue:43404", + "right": "issue:45216", + "accept": false, + "reason": "Weight-tying bug in a model class is unrelated to a save_pretrained checkpoint regression." + }, + { + "left": "issue:32090", + "right": "issue:35141", + "accept": false, + "reason": "Trainer broadcast NoneType error is unrelated to post_init reinitializing output embeddings." + }, + { + "left": "issue:44295", + "right": "issue:46506", + "accept": false, + "reason": "Buffer access error and CPU/CUDA output divergence are different problems with no clear shared root cause." + }, + { + "left": "issue:41950", + "right": "issue:44016", + "accept": false, + "reason": "Video-classification pipeline image-processor lookup is unrelated to a notebook syntax error." + }, + { + "left": "issue:43531", + "right": "issue:45600", + "accept": false, + "reason": "Qwen3-MoE sliding-window behavior and auto_mappings CI breakage are unrelated." + }, + { + "left": "issue:43650", + "right": "issue:44220", + "accept": false, + "reason": "The titles do not describe the same bug; one is too vague to tie to the audio feature extraction issue." + }, + { + "left": "issue:43673", + "right": "issue:46424", + "accept": false, + "reason": "Chunked-prefill cache regression and max_cache_length being honored only for Quantized Cache are separate cache behaviors." + }, + { + "left": "issue:43976", + "right": "issue:44987", + "accept": false, + "reason": "Python-version compatibility failure is unrelated to loading a specific model repo." + }, + { + "left": "issue:43295", + "right": "issue:45335", + "accept": false, + "reason": "Processor/tokenizer regression and t5gemma embedding resize bug are different model pipeline issues." + }, + { + "left": "issue:44016", + "right": "issue:44220", + "accept": false, + "reason": "Notebook syntax error is unrelated to the fbank feature extraction bug." + }, + { + "left": "issue:45083", + "right": "issue:46506", + "accept": false, + "reason": "Helper output-length logic in qwen3_omni_moe is not the same as a CPU/CUDA parity discrepancy." + }, + { + "left": "issue:42907", + "right": "issue:43441", + "accept": false, + "reason": "Dequantized checkpoint saving and FlashAttention runtime failures in the Ministral family are distinct bugs." + }, + { + "left": "issue:44683", + "right": "issue:45446", + "accept": false, + "reason": "Compiled flex_attention failure on torch>=2.9 and an incorrect version check for AuxRequest import are different flex_attention issues." + }, + { + "left": "issue:43526", + "right": "issue:43901", + "accept": false, + "reason": "Image-processor label reduction bug and pipeline docs mismatch are unrelated." + }, + { + "left": "issue:43381", + "right": "issue:45593", + "accept": false, + "reason": "Gradient checkpointing eval-mode restriction has nothing to do with D-FINE auxiliary losses." + }, + { + "left": "issue:43479", + "right": "issue:45375", + "accept": false, + "reason": "Both are multimodal config issues, but they concern different models and different missing/default fields." + }, + { + "left": "issue:44496", + "right": "issue:45042", + "accept": false, + "reason": "Unrecognized model/config loading and PIL backend torchvision requirements are different bugs." + }, + { + "left": "issue:43901", + "right": "issue:44496", + "accept": false, + "reason": "Docs behavior mismatch is unrelated to model-type/config import failure." + }, + { + "left": "issue:43673", + "right": "issue:44336", + "accept": false, + "reason": "Cache handling during generation and ANSI codes in loading_report are unrelated." + }, + { + "left": "issue:44792", + "right": "issue:46424", + "accept": false, + "reason": "Janus image-generation test failure is unrelated to cache length configuration semantics." + }, + { + "left": "issue:44964", + "right": "issue:45003", + "accept": false, + "reason": "Phi-4 model loading regression and unsafe sys.modules access are different code-path problems with different fixes." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 40, + "review_unit_count": 60, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-500", + "nodes": [ + "issue:39692", + "issue:41950", + "issue:42796", + "issue:43295", + "issue:43299", + "issue:43441", + "issue:43577", + "issue:43650", + "issue:43653", + "issue:43824", + "issue:43854", + "issue:43901", + "issue:43976", + "issue:44016", + "issue:44062", + "issue:44075", + "issue:44242", + "issue:44279", + "issue:44327", + "issue:44351", + "issue:44368", + "issue:44466", + "issue:44496", + "issue:44521", + "issue:44779", + "issue:44933", + "issue:45020", + "issue:45200", + "issue:45216", + "issue:45276", + "issue:45278", + "issue:45335", + "issue:45357", + "issue:45405", + "issue:45584", + "issue:45588", + "issue:45593", + "issue:45759", + "issue:45812", + "issue:45823", + "issue:46082", + "issue:46123", + "issue:46369", + "issue:46399", + "issue:46512", + "issue:46515", + "issue:46693", + "issue:46752" + ], + "soft_pairs": [ + "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:39692|issue:43901", + "issue:43976|issue:45405", + "issue:43577|issue:45216", + "issue:44351|issue:45278", + "issue:46082|issue:46512", + "issue:42796|issue:43650", + "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" + ], + "prepared_review_unit_hash": "e85eb4cab8d6019c04994f248c245c0bac98b304af495ac512d62f1251643973", + "component_budget": { + "node_count": 500, + "item_count": 500, + "soft_pair_count": 1839, + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 28, + "serialized_chars": 36213, + "estimated_input_tokens": 9054, + "estimated_eval_tokens": 18364 + }, + "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 similar by topic (tokenizers, multimodal processors, save/load regressions, pipelines, imports, etc.) but describe different bugs and code paths. No soft edge looks like a true duplicate merge candidate.", + "confidence": 0.33, + "canonical_issue_reason": "If one issue must anchor the cluster, issue 44521 is the strongest candidate: it is a concrete, reproducible multimodal API regression, has the most discussion/inbound references, and is broader than the other items. That said, the set does not form a real duplicate cluster.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 44521 is the best standalone issue in the set because it is active, well-scoped, and has clear downstream impact on multimodal chat-template behavior. It is still not a duplicate of the others.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44016", + "right": "issue:44062", + "accept": false, + "reason": "Both involve notebook/tokenizer problems, but one is a notebook syntax error and the other is an AddedToken keyword conflict; different failure modes and fix paths." + }, + { + "left": "issue:44496", + "right": "issue:45759", + "accept": false, + "reason": "Both are model-loading/config issues, but one is an unrecognized model/config problem and the other is about unwraping text_config in AutoModelForCausalLM.from_config; not the same bug." + }, + { + "left": "issue:43824", + "right": "issue:44933", + "accept": false, + "reason": "Both are import errors, but they concern different symbols/modules entirely (Qwen2_5_VLForConditionalGeneration vs image_utils)." + }, + { + "left": "issue:43441", + "right": "issue:45588", + "accept": false, + "reason": "Both touch flash attention, but they hit different code paths: Ministral FlashAttention failure vs flash_attention.py crashing on s_aux=None for sink-less models." + }, + { + "left": "issue:46123", + "right": "issue:46369", + "accept": false, + "reason": "One is a pipeline batching bug that drops the last partial batch; the other is a model memory overflow. Different layers of the stack." + }, + { + "left": "issue:44327", + "right": "issue:46752", + "accept": false, + "reason": "Both are runtime errors in pipeline utilities, but one is top_k bounds handling in QA span decoding and the other is empty-list handling in apply_chat_template." + }, + { + "left": "issue:44521", + "right": "issue:46693", + "accept": false, + "reason": "Multimodal assistant mask generation and FA2 long-context performance regression are unrelated bugs, even if both affect generation workflows." + }, + { + "left": "issue:44279", + "right": "issue:46515", + "accept": false, + "reason": "These are both dependency/environment complaints, but one is a vague transformer dependency issue and the other is a specific torch 2.4.1 incompatibility." + }, + { + "left": "issue:45335", + "right": "issue:45584", + "accept": false, + "reason": "Both involve model/token embedding behavior, but one is resize_token_embeddings not propagating in t5gemma while the other is Whisper generation failing after align_special_tokens." + }, + { + "left": "issue:43299", + "right": "issue:45216", + "accept": false, + "reason": "Both are Qwen3.5 regressions, but one is a loading failure for MoE models and the other is incorrect save_pretrained checkpoint contents." + }, + { + "left": "issue:46082", + "right": "issue:46399", + "accept": false, + "reason": "A config validation issue for Llama head_dim is unrelated to Qwen3.5 kernel registration failures in use_kernels=True." + }, + { + "left": "issue:43295", + "right": "issue:44779", + "accept": false, + "reason": "Both are tokenizer/processor regressions, but one is about processor.tokenizer/images handling and the other is Deepseek tokenizer output correctness." + }, + { + "left": "issue:44075", + "right": "issue:44368", + "accept": false, + "reason": "SGD args being ignored and a tie_word_embeddings warning are unrelated training/config issues." + }, + { + "left": "issue:39692", + "right": "issue:43901", + "accept": false, + "reason": "Both mention documentation, but they cover different pipelines and different problems: SigLIP2 example mismatches vs TextClassificationPipeline return_all_scores docs drift." + }, + { + "left": "issue:43976", + "right": "issue:45405", + "accept": false, + "reason": "Python version support failure and a PEFT minimum-version bump are different dependency-management issues." + }, + { + "left": "issue:43577", + "right": "issue:45216", + "accept": false, + "reason": "Blip2 dtype propagation and Qwen3.5 save_pretrained serialization are different regressions with different affected code paths." + }, + { + "left": "issue:44351", + "right": "issue:45278", + "accept": false, + "reason": "A missing HybridCache import and a broad set of import errors after upgrade are related only at a very high level; not the same concrete issue." + }, + { + "left": "issue:46082", + "right": "issue:46512", + "accept": false, + "reason": "Both involve config attributes, but one rejects explicit head_dim while the other leaves model.config.dtype stale after from_config(dtype=...)." + }, + { + "left": "issue:42796", + "right": "issue:43650", + "accept": false, + "reason": "Tokenizer BPE marker handling and a nonspecific 'ADD THE DATA' report are not the same bug, and the latter is too vague to match." + }, + { + "left": "issue:44779", + "right": "issue:45812", + "accept": false, + "reason": "Both are tokenizer regressions, but they concern different model families and different symptoms (Deepseek incorrect results vs Granite wrong token IDs)." + }, + { + "left": "issue:41950", + "right": "issue:45020", + "accept": false, + "reason": "A video-classification processor lookup bug and broad remote_code breakage are different compatibility issues." + }, + { + "left": "issue:43299", + "right": "issue:45357", + "accept": false, + "reason": "Both concern Qwen3.5 save/load behavior, but one is a loading breakage and the other is incorrect visual encoder keys on save_pretrained." + }, + { + "left": "issue:44466", + "right": "issue:45276", + "accept": false, + "reason": "Both involve weight/embedding serialization, but one is device-dependent lm_head.weight serialization while the other is resize_token_embeddings not updating gemma4 submodules." + }, + { + "left": "issue:43299", + "right": "issue:43854", + "accept": false, + "reason": "Different multimodal model families and different failures: Qwen3VL MoE loading vs GLM-4.7-Flash unit-test loading." + }, + { + "left": "issue:43653", + "right": "issue:45593", + "accept": false, + "reason": "BigBirdTokenizer special-token registration and D-FINE auxiliary-loss gating are completely unrelated bugs." + }, + { + "left": "issue:41950", + "right": "issue:45823", + "accept": false, + "reason": "The first is a pipeline processor-selection bug; the second is a Gemma4 device placement mismatch during forward. Different subsystems." + }, + { + "left": "issue:44242", + "right": "issue:45200", + "accept": false, + "reason": "Load-balancing loss gating and Gemma4 mm_token_type_ids defaults are different multimodal training issues with 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": 41, + "review_unit_count": 60, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-500", + "nodes": [ + "issue:33290", + "issue:42175", + "issue:42796", + "issue:42994", + "issue:43317", + "issue:43441", + "issue:43479", + "issue:43525", + "issue:43650", + "issue:43673", + "issue:43723", + "issue:43761", + "issue:43824", + "issue:43881", + "issue:43901", + "issue:43976", + "issue:44016", + "issue:44162", + "issue:44246", + "issue:44261", + "issue:44263", + "issue:44291", + "issue:44297", + "issue:44336", + "issue:44361", + "issue:44462", + "issue:44623", + "issue:44671", + "issue:44683", + "issue:44749", + "issue:44779", + "issue:45081", + "issue:45103", + "issue:45372", + "issue:45397", + "issue:45542", + "issue:45593", + "issue:45600", + "issue:45656", + "issue:45712", + "issue:45800", + "issue:45834", + "issue:45859", + "issue:45907", + "issue:46209", + "issue:46291", + "issue:46399", + "issue:46851" + ], + "soft_pairs": [ + "issue:43650|issue:46209", + "issue:44297|issue:44462", + "issue:43479|issue:43525", + "issue:44683|issue:45542", + "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:44336|issue:46851", + "issue:44261|issue:44263", + "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" + ], + "prepared_review_unit_hash": "d068d17fb9dd3a5ff0b4f6231d8c000d3bc575c31981429d01e47d250462d336", + "component_budget": { + "node_count": 500, + "item_count": 500, + "soft_pair_count": 1839, + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 27, + "serialized_chars": 35701, + "estimated_input_tokens": 8926, + "estimated_eval_tokens": 18108 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is heterogeneous: the soft pairs mostly share only superficial subsystem overlap (tokenizers, loading, generation, multimodal, Deepspeed, etc.) but do not describe the same concrete bug. I would keep a broad, well-discussed issue like import/performance regressions as the best representative issue, but there is no PR in the set.", + "confidence": 0.87, + "canonical_issue_reason": "issue:44246 is the best representative issue here: it has broad user impact, clear reproduction around import-time performance, and higher discussion/inbound activity than most other candidates.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:44246 is the strongest general-purpose issue candidate for triage because it is clearly described, has broader scope than model-specific failures, and appears more central than the narrower one-off regressions.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43650", + "right": "issue:46209", + "accept": false, + "reason": "Unrelated: a vague \"ADD THE DATA\" issue is not the same as an FP8 inference failure on B200." + }, + { + "left": "issue:44297", + "right": "issue:44462", + "accept": false, + "reason": "Different tokenizer bugs: save_pretrained metadata mismatch vs AutoTokenizer ignoring repository tokenizer.json." + }, + { + "left": "issue:43479", + "right": "issue:43525", + "accept": false, + "reason": "Config initialization defaults for multimodal components is unrelated to Llama4 missing pad_token_id." + }, + { + "left": "issue:44683", + "right": "issue:45542", + "accept": false, + "reason": "Different areas and failure modes: torch 2.9 flex_attention vs missing TensorFlow backend installation." + }, + { + "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." + }, + { + "left": "issue:44361", + "right": "issue:45593", + "accept": false, + "reason": "Tokenizer AttributeError on tasks is not the same as D-FINE auxiliary-loss behavior." + }, + { + "left": "issue:45081", + "right": "issue:45907", + "accept": false, + "reason": "Mistral regex patch crash and list_repo_files revision forwarding are different code paths and bugs." + }, + { + "left": "issue:33290", + "right": "issue:45656", + "accept": false, + "reason": "Both involve Deepspeed/optimizer, but one is OOM with Adafactor and the other is double optimizer stepping." + }, + { + "left": "issue:43441", + "right": "issue:45834", + "accept": false, + "reason": "Different models and issues: FlashAttention failure for Ministral-3 vs OCR index error in Kosmos2.5." + }, + { + "left": "issue:42994", + "right": "issue:43317", + "accept": false, + "reason": "Quantized model saving failure is distinct from loading a dequantized model with gpu+cpu offload." + }, + { + "left": "issue:44623", + "right": "issue:44749", + "accept": false, + "reason": "Processor save_pretrained missing files is unrelated to a Chinese report about inference/data filtering slowdown." + }, + { + "left": "issue:44336", + "right": "issue:46851", + "accept": false, + "reason": "ANSI-code output in loading_report has nothing to do with Qwen3.5 variable-length position_ids." + }, + { + "left": "issue:44261", + "right": "issue:44263", + "accept": false, + "reason": "Numerical precision issue in MLA q_a_layernorm is unrelated to torch.split return values in GlmMoeDsaIndexer." + }, + { + "left": "issue:45103", + "right": "issue:45372", + "accept": false, + "reason": "Auto-docstring future-annotations crash and ReasoningEffort import failure are different bugs in different modules." + }, + { + "left": "issue:45600", + "right": "issue:45712", + "accept": false, + "reason": "Broken auto_mappings references and leftover dummy_pt_objects cleanup are separate maintenance issues." + }, + { + "left": "issue:45859", + "right": "issue:46851", + "accept": false, + "reason": "CPU offload incompatibility in rotary embedding is not the same as variable-length input handling." + }, + { + "left": "issue:42175", + "right": "issue:43976", + "accept": false, + "reason": "Packaging TensorFlow omission is unrelated to a Python 3.9+/3.10+ compatibility break." + }, + { + "left": "issue:44623", + "right": "issue:45397", + "accept": false, + "reason": "Missing files on processor save_pretrained is not the same as gemma-4 zero3 from_pretrained loading." + }, + { + "left": "issue:44671", + "right": "issue:44779", + "accept": false, + "reason": "Incorrect masked LM predictions in CamemBERT and incorrect Deepseek tokenizer outputs are separate model/tokenizer regressions." + }, + { + "left": "issue:43723", + "right": "issue:44162", + "accept": false, + "reason": "AutoTokenizer loading failure and ESM2 breakage are different model-loading problems." + }, + { + "left": "issue:43901", + "right": "issue:44246", + "accept": false, + "reason": "Docs mismatch for return_all_scores is unrelated to intermittent slow imports." + }, + { + "left": "issue:43673", + "right": "issue:43761", + "accept": false, + "reason": "GenerationMixin cache missing during chunked_prefill and CLIPVision hidden_states regression are different components and symptoms." + }, + { + "left": "issue:43824", + "right": "issue:43976", + "accept": false, + "reason": "Missing Qwen2_5_VL import and Python version support issues are not the same bug." + }, + { + "left": "issue:45542", + "right": "issue:45800", + "accept": false, + "reason": "Missing TensorFlow backend from a torch install is distinct from a torch/transformers version incompatibility." + }, + { + "left": "issue:42796", + "right": "issue:43441", + "accept": false, + "reason": "Tokenizer BPE marker handling and FlashAttention failure are different classes of bugs." + }, + { + "left": "issue:43525", + "right": "issue:43881", + "accept": false, + "reason": "Missing pad_token_id on Llama4Config is unrelated to glm-4v-9b loading failure." + }, + { + "left": "issue:46291", + "right": "issue:46399", + "accept": false, + "reason": "kernels v0.15 version/revision requirement and Qwen3.5 rotary embedding registration crash are different integration failures." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 42, + "review_unit_count": 60, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-500", + "nodes": [ + "issue:41628", + "issue:42491", + "issue:42886", + "issue:43299", + "issue:43525", + "issue:43526", + "issue:43531", + "issue:43540", + "issue:43643", + "issue:43784", + "issue:43976", + "issue:44016", + "issue:44188", + "issue:44242", + "issue:44246", + "issue:44295", + "issue:44297", + "issue:44336", + "issue:44373", + "issue:44451", + "issue:44485", + "issue:44496", + "issue:44589", + "issue:44871", + "issue:45003", + "issue:45072", + "issue:45083", + "issue:45103", + "issue:45325", + "issue:45362", + "issue:45397", + "issue:45542", + "issue:45561", + "issue:45563", + "issue:45588", + "issue:45593", + "issue:45823", + "issue:46132", + "issue:46399", + "issue:46614", + "issue:46655", + "issue:46665", + "issue:46726", + "issue:46752", + "issue:46821", + "issue:46829", + "issue:46851" + ], + "soft_pairs": [ + "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", + "issue:46614|issue:46665", + "issue:43540|issue:46752", + "issue:45072|issue:46829", + "issue:44295|issue:45397", + "issue:43531|issue:45362", + "issue:42491|issue:43299", + "issue:44246|issue:44871", + "issue:44336|issue:44485", + "issue:44242|issue:45593", + "issue:45588|issue:46851", + "issue:45003|issue:46821", + "issue:42886|issue:46132", + "issue:44451|issue:46752" + ], + "prepared_review_unit_hash": "a9476120dcbdbde30727f0907ed815a5295a9a36b9a7af73fed4b5397b3ca748", + "component_budget": { + "node_count": 500, + "item_count": 500, + "soft_pair_count": 1839, + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 29, + "serialized_chars": 35841, + "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 a loose collection of unrelated Transformers issues that only share superficial keywords (models, config, loading, position_ids, etc.). I do not see any true duplicate pairs among the soft edges.", + "confidence": 0.95, + "canonical_issue_reason": "Issue 46851 is the best representative of the cluster only in the sense that it is a concrete, reproducible runtime regression with a clear code path. But the cluster is not a duplicate set, so no single issue is a true canonical for the whole group.", + "canonical_pr_reason": null, + "best_issue_reason": "46851 has the clearest, most actionable bug report and a specific model-family regression; it is the strongest stand-alone issue in this mixed cluster.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44188", + "right": "issue:44336", + "accept": false, + "reason": "Completely different subsystems: attention kernel behavior under torch.compile vs terminal ANSI handling in loading_report." + }, + { + "left": "issue:44016", + "right": "issue:45823", + "accept": false, + "reason": "Notebook syntax error and Gemma4 device_map forward mismatch are unrelated bug classes." + }, + { + "left": "issue:44336", + "right": "issue:44373", + "accept": false, + "reason": "ANSI escaping in loading_report and a wrong docstring for position_ids are not the same underlying issue." + }, + { + "left": "issue:44589", + "right": "issue:45362", + "accept": false, + "reason": "Float8 storage lookup failure is unrelated to a Qwen3.5 chat crash." + }, + { + "left": "issue:43525", + "right": "issue:43643", + "accept": false, + "reason": "Both involve config fields, but one is a missing pad_token_id attribute and the other is trust_remote_code returning incomplete AutoConfig fields; different failures and fixes." + }, + { + "left": "issue:46655", + "right": "issue:46726", + "accept": false, + "reason": "Video padding calculation bug vs missing vision backend reporting; same area, but distinct code paths and symptoms." + }, + { + "left": "issue:45325", + "right": "issue:46851", + "accept": false, + "reason": "Both mention position_ids, but they affect different Qwen model families and different logic (RoPE scaling vs variable-length input handling)." + }, + { + "left": "issue:41628", + "right": "issue:43976", + "accept": false, + "reason": "Importing AutoImageProcessor and Python version compatibility are unrelated." + }, + { + "left": "issue:44297", + "right": "issue:45397", + "accept": false, + "reason": "Tokenizer save_pretrained metadata mismatch is unrelated to Gemma4 zero3 from_pretrained behavior." + }, + { + "left": "issue:43784", + "right": "issue:45542", + "accept": false, + "reason": "Both are import-time backend issues, but one is missing nn for sentence-transformers and the other is undefined tf backend when only tensorboard is installed; not the same bug." + }, + { + "left": "issue:45083", + "right": "issue:46851", + "accept": false, + "reason": "Different Qwen3 submodels and different helper paths; not the same position_ids bug." + }, + { + "left": "issue:44246", + "right": "issue:44496", + "accept": false, + "reason": "Moe load-balancing loss gating and unrecognized model_type are unrelated." + }, + { + "left": "issue:46399", + "right": "issue:46851", + "accept": false, + "reason": "Kernel registration decorator crash is a separate issue from variable-length position_ids handling." + }, + { + "left": "issue:45561", + "right": "issue:45563", + "accept": false, + "reason": "A pytest-xdist file race and a stale generation warning are unrelated." + }, + { + "left": "issue:45103", + "right": "issue:46399", + "accept": false, + "reason": "Auto-docstring annotation parsing crash and Qwen kernel registration are unrelated tooling/runtime bugs." + }, + { + "left": "issue:43526", + "right": "issue:45003", + "accept": false, + "reason": "BEiT reduce_labels behavior and unsafe sys.modules access are unrelated." + }, + { + "left": "issue:46614", + "right": "issue:46665", + "accept": false, + "reason": "Stale phonemizer language state and inconsistent from_pretrained behavior are different bugs." + }, + { + "left": "issue:43540", + "right": "issue:46752", + "accept": false, + "reason": "Qwen3OmniMoe video input validation and empty-list chat template handling are unrelated." + }, + { + "left": "issue:45072", + "right": "issue:46829", + "accept": false, + "reason": "bfloat16 dtype mismatch issues are unrelated to VideoPrism hidden_states/output packaging." + }, + { + "left": "issue:44295", + "right": "issue:45397", + "accept": false, + "reason": "Reading a bufferized position_ids tensor and Gemma4 zero3 loading are different failure modes." + }, + { + "left": "issue:43531", + "right": "issue:45362", + "accept": false, + "reason": "Both are Qwen-family regressions, but one is sliding_window handling in Qwen3-MoE and the other is chat behavior in Qwen3.5; not the same concrete bug." + }, + { + "left": "issue:42491", + "right": "issue:43299", + "accept": false, + "reason": "Both involve Qwen3 MoE compatibility, but one is LoRA reuse across major versions and the other is model loading breakage in dev0; different root causes." + }, + { + "left": "issue:44246", + "right": "issue:44871", + "accept": false, + "reason": "Load-balancing loss gating and Gemma-3 eos_token_id inconsistency are unrelated." + }, + { + "left": "issue:44336", + "right": "issue:44485", + "accept": false, + "reason": "Terminal ANSI output bug and GLM-5 RoPE implementation discussion are unrelated." + }, + { + "left": "issue:44242", + "right": "issue:45593", + "accept": false, + "reason": "Both mention auxiliary losses, but they are different models and different gating conditions; too broad to merge." + }, + { + "left": "issue:45588", + "right": "issue:46851", + "accept": false, + "reason": "flash_attention sink-less model crash is unrelated to Qwen3.5 variable-length position_ids." + }, + { + "left": "issue:45003", + "right": "issue:46821", + "accept": false, + "reason": "Unsafe sys.modules access and TimesFM window_size AttributeError are unrelated." + }, + { + "left": "issue:42886", + "right": "issue:46132", + "accept": false, + "reason": "Offline tokenizer cache loading and AttentionInterface.register behavior are unrelated." + }, + { + "left": "issue:44451", + "right": "issue:46752", + "accept": false, + "reason": "ScandiBERT loading failure and empty-list chat template crash 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": 43, + "review_unit_count": 60, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-500", + "nodes": [ + "issue:29870", + "issue:30333", + "issue:36246", + "issue:41950", + "issue:42757", + "issue:43257", + "issue:43262", + "issue:43295", + "issue:43475", + "issue:43526", + "issue:43540", + "issue:43550", + "issue:43596", + "issue:43901", + "issue:44038", + "issue:44162", + "issue:44220", + "issue:44242", + "issue:44246", + "issue:44295", + "issue:44351", + "issue:44466", + "issue:44479", + "issue:44496", + "issue:44623", + "issue:44749", + "issue:44779", + "issue:44857", + "issue:44877", + "issue:44987", + "issue:45003", + "issue:45137", + "issue:45200", + "issue:45292", + "issue:45335", + "issue:45375", + "issue:45459", + "issue:45561", + "issue:45584", + "issue:45600", + "issue:45759", + "issue:45859", + "issue:46369", + "issue:46399", + "issue:46506", + "issue:46655", + "issue:46726", + "issue:46897" + ], + "soft_pairs": [ + "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:44479|issue:46655", + "issue:44295|issue:44749", + "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" + ], + "prepared_review_unit_hash": "878fac6693f8e40a18b5696fb830cb80d80b6407edfc4b04ade152691b94bacb", + "component_budget": { + "node_count": 500, + "item_count": 500, + "soft_pair_count": 1839, + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 29, + "serialized_chars": 36559, + "estimated_input_tokens": 9140, + "estimated_eval_tokens": 18536 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is highly heterogeneous: most items are distinct bug reports across unrelated subsystems, and none of the soft pairs are strong enough to treat as duplicates. The closest themes are video-processing regressions and strict/config handling, but they still point to different root causes and code paths.", + "confidence": 0.22, + "canonical_issue_reason": "issue:44479 is the broadest and most representative issue here: it describes a concrete regression affecting multiple Qwen video models in v5.3.0, which is the closest thing to a shared bug theme in the set.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:44479 is the best issue candidate because it covers a widespread, user-visible regression with multiple affected models and a clear version-specific breakage; the others are narrower or unrelated.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43550", + "right": "issue:46506", + "accept": false, + "reason": "Different models, different failure modes, and different subsystems; both mention runtime issues but not the same concrete bug." + }, + { + "left": "issue:44779", + "right": "issue:45292", + "accept": false, + "reason": "Tokenizer correctness regression and resize_token_embeddings behavior are unrelated code paths." + }, + { + "left": "issue:36246", + "right": "issue:42757", + "accept": false, + "reason": "Import errors from different symbols/modules with no shared underlying defect." + }, + { + "left": "issue:36246", + "right": "issue:44351", + "accept": false, + "reason": "Both are import errors, but for different APIs and causes; not the same bug." + }, + { + "left": "issue:41950", + "right": "issue:44220", + "accept": false, + "reason": "Video-classification processor lookup and fbank feature extraction are separate processing bugs." + }, + { + "left": "issue:43901", + "right": "issue:45003", + "accept": false, + "reason": "Documentation mismatch for a pipeline setting is unrelated to unsafe sys.modules access." + }, + { + "left": "issue:43257", + "right": "issue:45137", + "accept": false, + "reason": "Both involve DeepSpeed, but one is MoE weight conversion and the other is a ZeRO3 deque/indexing crash." + }, + { + "left": "issue:45003", + "right": "issue:45561", + "accept": false, + "reason": "Unsafe module lookup and pytest-xdist file race are unrelated." + }, + { + "left": "issue:44496", + "right": "issue:45003", + "accept": false, + "reason": "Model config loading failure and sys.modules access are not the same issue." + }, + { + "left": "issue:43540", + "right": "issue:46655", + "accept": false, + "reason": "Both are video-input bugs, but in different models with different failure mechanisms." + }, + { + "left": "issue:44162", + "right": "issue:45200", + "accept": false, + "reason": "ESM2 breakage and Gemma 4 token-type defaults are unrelated." + }, + { + "left": "issue:45375", + "right": "issue:45759", + "accept": false, + "reason": "Both are Qwen multimodal config issues, but one is strict-field dropping and the other is AutoModelForCausalLM text_config unwrapping." + }, + { + "left": "issue:43262", + "right": "issue:44246", + "accept": false, + "reason": "Audio chat-template sampling-rate defaulting and import latency are unrelated." + }, + { + "left": "issue:44479", + "right": "issue:46655", + "accept": false, + "reason": "Both are video regressions, but they affect different model families and appear to stem from different processor bugs." + }, + { + "left": "issue:44295", + "right": "issue:44749", + "accept": false, + "reason": "A runtime position_ids error and a performance regression during filtering are not the same bug." + }, + { + "left": "issue:43475", + "right": "issue:46369", + "accept": false, + "reason": "SAM 3 video encoder attribute error and SAM 3 LiteText memory overflow are distinct failures." + }, + { + "left": "issue:45859", + "right": "issue:46399", + "accept": false, + "reason": "Both mention Qwen3.5 rotary/embed-related code, but one is CPU offload incompatibility and the other is a kernel registration decorator crash." + }, + { + "left": "issue:45137", + "right": "issue:45600", + "accept": false, + "reason": "DeepSpeed ZeRO3 deque error and auto_mappings CI breakage are unrelated." + }, + { + "left": "issue:36246", + "right": "issue:43596", + "accept": false, + "reason": "Different import/runtime errors with no shared root cause." + }, + { + "left": "issue:44877", + "right": "issue:45375", + "accept": false, + "reason": "Both involve strict config handling, but they concern different models and different outcomes." + }, + { + "left": "issue:43295", + "right": "issue:45292", + "accept": false, + "reason": "Processor API regression and token-embedding resize behavior are unrelated." + }, + { + "left": "issue:45292", + "right": "issue:45584", + "accept": false, + "reason": "Token embedding resizing and Whisper empty-transcription generation are different problems." + }, + { + "left": "issue:44466", + "right": "issue:45335", + "accept": false, + "reason": "Serialization inconsistency for lm_head.weight and decoder.embed_tokens resize behavior are not the same underlying bug." + }, + { + "left": "issue:44623", + "right": "issue:46726", + "accept": false, + "reason": "Missing files on save_pretrained and unrecognized image processor on missing backend are different processor issues." + }, + { + "left": "issue:44242", + "right": "issue:44857", + "accept": false, + "reason": "MoE load-balancing loss omission and AMP/CUDA crash in image loss are unrelated." + }, + { + "left": "issue:43526", + "right": "issue:46897", + "accept": false, + "reason": "Label reduction bug in BeitImageProcessorFast and Florence2 training-loss shift issue are not the same change." + }, + { + "left": "issue:44038", + "right": "issue:46506", + "accept": false, + "reason": "Qwen3-VL-Moe bug and PPDocLayoutV3 CPU/CUDA discrepancy are unrelated." + }, + { + "left": "issue:44987", + "right": "issue:45459", + "accept": false, + "reason": "Model loading failure for a specific repo and tokenizer protobuf error masking are different bugs." + }, + { + "left": "issue:29870", + "right": "issue:30333", + "accept": false, + "reason": "Bart generate() UnboundLocalError and MLflow job-status reporting 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": 44, + "review_unit_count": 60, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-500", + "nodes": [ + "issue:42199", + "issue:42886", + "issue:43065", + "issue:43352", + "issue:43408", + "issue:43475", + "issue:43526", + "issue:43540", + "issue:43653", + "issue:43723", + "issue:43906", + "issue:43931", + "issue:44060", + "issue:44077", + "issue:44162", + "issue:44188", + "issue:44297", + "issue:44368", + "issue:44479", + "issue:44483", + "issue:44623", + "issue:44877", + "issue:44977", + "issue:44995", + "issue:45103", + "issue:45200", + "issue:45229", + "issue:45250", + "issue:45397", + "issue:45412", + "issue:45563", + "issue:45588", + "issue:45656", + "issue:45685", + "issue:45987", + "issue:46082", + "issue:46123", + "issue:46170", + "issue:46202", + "issue:46421", + "issue:46431", + "issue:46710", + "issue:46829", + "issue:46832", + "issue:46851", + "issue:46858", + "issue:46894" + ], + "soft_pairs": [ + "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", + "issue:43475|issue:43540", + "issue:44877|issue:46170", + "issue:46170|issue:46832", + "issue:44368|issue:45563", + "issue:43723|issue:46710", + "issue:44995|issue:45200", + "issue:43931|issue:44368", + "issue:44188|issue:46202", + "issue:44483|issue:46202", + "issue:46829|issue:46858", + "issue:45987|issue:46082", + "issue:43906|issue:46202", + "issue:44995|issue:45588", + "issue:44188|issue:46421", + "issue:45685|issue:46431", + "issue:45103|issue:46858", + "issue:43065|issue:46123", + "issue:44297|issue:44623" + ], + "prepared_review_unit_hash": "7f378e615ded3dc5f9655bdb81d40ef945a63c31f18af692ec24323c99fd2721", + "component_budget": { + "node_count": 500, + "item_count": 500, + "soft_pair_count": 1839, + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 29, + "serialized_chars": 35960, + "estimated_input_tokens": 8990, + "estimated_eval_tokens": 18236 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is highly heterogeneous; the soft pairs are mostly superficially similar but not the same underlying bug or change. No PRs are present, so the best anchor is an issue, not a PR.", + "confidence": 0.41, + "canonical_issue_reason": "issue:46858 is the best single anchor: it is a concrete, code-path-specific bug with a clear failure mode and is the closest thing here to a reusable bug family (cache length / static cache behavior). The rest of the cluster is largely unrelated.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:46858 is the strongest issue candidate because it is explicit, actionable, and narrowly scoped, whereas most other items are broader feature regressions, warnings, or unrelated model-specific failures.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43352", + "right": "issue:45250", + "accept": false, + "reason": "Both mention Flash Attention 2.0, but one is a model-specific unsupported-path error and the other is a generic topic issue; not the same bug." + }, + { + "left": "issue:42199", + "right": "issue:46851", + "accept": false, + "reason": "DET R cardinality/background-class handling and Qwen3.5 variable-length position_ids are unrelated code paths and symptoms." + }, + { + "left": "issue:43526", + "right": "issue:46894", + "accept": false, + "reason": "BEiT label reduction and CohereASR loss shifting are different model families and different failure modes." + }, + { + "left": "issue:44060", + "right": "issue:44077", + "accept": false, + "reason": "One is a tied-weights warning in Qwen3-Next; the other is a patchtsmixer post_init policy issue. Same broad theme only at a high level." + }, + { + "left": "issue:44368", + "right": "issue:44977", + "accept": false, + "reason": "A tied-embeddings warning is not the same as a flash-attention generation regression in Qwen3.5." + }, + { + "left": "issue:43653", + "right": "issue:44162", + "accept": false, + "reason": "Tokenizer special-token registration and ESM2 breakage are different subsystems and different root causes." + }, + { + "left": "issue:45412", + "right": "issue:46123", + "accept": false, + "reason": "RT-DETR memory cleanup and MaskGenerationPipeline batching logic are unrelated bugs." + }, + { + "left": "issue:44479", + "right": "issue:46170", + "accept": false, + "reason": "Video-input regression in Qwen models and Windows/ROCm import dependency failures are unrelated." + }, + { + "left": "issue:42886", + "right": "issue:45229", + "accept": false, + "reason": "Offline tokenizer cache loading and Gemma4 multi-GPU CUDA OOM are different classes of failure." + }, + { + "left": "issue:45397", + "right": "issue:45656", + "accept": false, + "reason": "Gemma-4 ZeRO3 loading and double optimizer stepping under deepspeed are not the same concrete bug." + }, + { + "left": "issue:43065", + "right": "issue:43408", + "accept": false, + "reason": "SAM3 pixel-decoder implementation detail and sam3_video vs sam3_tracker model-type warning are distinct issues." + }, + { + "left": "issue:43475", + "right": "issue:43540", + "accept": false, + "reason": "SAM3 vision encoder output attribute error and Qwen3OmniMoe video-input ValueError are unrelated." + }, + { + "left": "issue:44877", + "right": "issue:46170", + "accept": false, + "reason": "Strict config loading for granite_speech and ROCm/Windows distributed import errors do not share the same bug." + }, + { + "left": "issue:46170", + "right": "issue:46832", + "accept": false, + "reason": "A platform/dependency import issue is not the same as an RT-DETR feature-level mismatch in forward." + }, + { + "left": "issue:44368", + "right": "issue:45563", + "accept": false, + "reason": "These are both warnings, but they concern different generation settings and different code paths." + }, + { + "left": "issue:43723", + "right": "issue:46710", + "accept": false, + "reason": "Tokenizer loading in v5 and an accuracy regression after upgrade are too different to be duplicates." + }, + { + "left": "issue:44995", + "right": "issue:45200", + "accept": false, + "reason": "GlmMoeDsa stale cache crash and Gemma 4 mm_token_type_ids defaults are different model behaviors." + }, + { + "left": "issue:43931", + "right": "issue:44368", + "accept": false, + "reason": "Qwen3-VL weight-shape mismatch and tied-embeddings warnings are unrelated." + }, + { + "left": "issue:44188", + "right": "issue:46202", + "accept": false, + "reason": "torch.compile attention-kernel divergence and Gemma token roundtrip failure are not the same issue." + }, + { + "left": "issue:44483", + "right": "issue:46202", + "accept": false, + "reason": "Chat-completions API rejection and Gemma pipeline roundtrip failure are different regressions." + }, + { + "left": "issue:46829", + "right": "issue:46858", + "accept": false, + "reason": "VideoPrism output field mix-up and cache.get_seq_length returning a device tensor are different bugs." + }, + { + "left": "issue:45987", + "right": "issue:46082", + "accept": false, + "reason": "A cache return-type contract bug and a LlamaConfig head_dim validation issue are unrelated." + }, + { + "left": "issue:43906", + "right": "issue:46202", + "accept": false, + "reason": "An isolated reproduction of another issue is not enough to treat it as the same bug as the Gemma roundtrip failure." + }, + { + "left": "issue:44995", + "right": "issue:45588", + "accept": false, + "reason": "A stale cache crash in GlmMoeDsa and a flash_attention.py None-handling crash are separate code paths." + }, + { + "left": "issue:44188", + "right": "issue:46421", + "accept": false, + "reason": "Both mention compile/cache behavior, but one is attention-kernel divergence and the other is lazy cache init causing recompiles; not the same fix." + }, + { + "left": "issue:45685", + "right": "issue:46431", + "accept": false, + "reason": "MPS histogram integer support and missing chat template validation are unrelated." + }, + { + "left": "issue:45103", + "right": "issue:46858", + "accept": false, + "reason": "auto_docstring annotation handling and cache length arithmetic are completely different concerns." + }, + { + "left": "issue:43065", + "right": "issue:46123", + "accept": false, + "reason": "SAM3 pixel-decoder internals and partial-batch masking logic do not describe the same defect." + }, + { + "left": "issue:44297", + "right": "issue:44623", + "accept": false, + "reason": "Tokenizer save_pretrained metadata mismatch and processor save_pretrained missing files are both persistence-related, but they are different failure modes and artifacts." + } + ] + }, + "evaluator_result": 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": 60, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-500", + "nodes": [ + "issue:28282", + "issue:30990", + "issue:39692", + "issue:41628", + "issue:42371", + "issue:42994", + "issue:43232", + "issue:43441", + "issue:43475", + "issue:43482", + "issue:43525", + "issue:43643", + "issue:43646", + "issue:43673", + "issue:43701", + "issue:43723", + "issue:43986", + "issue:43994", + "issue:44038", + "issue:44162", + "issue:44188", + "issue:44361", + "issue:44373", + "issue:44623", + "issue:44749", + "issue:44869", + "issue:44871", + "issue:44933", + "issue:44964", + "issue:45092", + "issue:45137", + "issue:45230", + "issue:45310", + "issue:45356", + "issue:45375", + "issue:45397", + "issue:45406", + "issue:45478", + "issue:45656", + "issue:46132", + "issue:46202", + "issue:46326", + "issue:46463", + "issue:46726", + "issue:46752", + "issue:46821", + "issue:46851", + "issue:46877" + ], + "soft_pairs": [ + "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:43475|issue:43643", + "issue:43232|issue:43673", + "issue:44188|issue:46851", + "issue:28282|issue:30990", + "issue:41628|issue:44933", + "issue:43475|issue:46821", + "issue:45230|issue:45397", + "issue:44964|issue:45137", + "issue:45310|issue:45375", + "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" + ], + "prepared_review_unit_hash": "f8827e362c7661cdd0dcc697950e8ba2fdb86000c6f81eeef3ed42af8b8a0730", + "component_budget": { + "node_count": 500, + "item_count": 500, + "soft_pair_count": 1839, + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 29, + "serialized_chars": 36055, + "estimated_input_tokens": 9014, + "estimated_eval_tokens": 18284 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a loose grab-bag of Transformers v5 regressions and model-specific loading/behavior issues. The soft links are generally only broad topical similarity, not the same underlying bug or change, so they should stay separate.", + "confidence": 0.38, + "canonical_issue_reason": "Issue 43646 is the broadest, most representative regression report in the set: it frames a general Transformers 5.0 compatibility break around custom model initialization, which matches the cluster\u2019s overall theme better than the narrower model-specific reports.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 43646 is the best single issue representative because it is general, clearly regression-oriented, and spans the same v5 compatibility class as many of the other reports.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44361", + "right": "issue:44749", + "accept": false, + "reason": "Different failures: MLukeTokenizer task-time AttributeError vs a performance regression after upgrading Transformers. Same library, but not the same bug path." + }, + { + "left": "issue:43994", + "right": "issue:46726", + "accept": false, + "reason": "Both involve vision/image tooling, but one reports bad outputs from AutoModel/pipeline and the other is a missing-backend error in AutoProcessor. Different problems and fixes." + }, + { + "left": "issue:44749", + "right": "issue:45656", + "accept": false, + "reason": "A data-processing slowdown after upgrade is unrelated to DeepSpeed causing optimizer.step to run twice." + }, + { + "left": "issue:43701", + "right": "issue:44749", + "accept": false, + "reason": "Checkpoint key mismatch and slowdown after upgrade are unrelated symptoms with different code paths." + }, + { + "left": "issue:44871", + "right": "issue:46132", + "accept": false, + "reason": "Gemma eos_token_id inconsistency is unrelated to AttentionInterface.register changing callable behavior." + }, + { + "left": "issue:44623", + "right": "issue:45656", + "accept": false, + "reason": "Missing files from processor.save_pretrained is a serialization bug, not a DeepSpeed optimizer-step issue." + }, + { + "left": "issue:43646", + "right": "issue:44162", + "accept": false, + "reason": "Custom model initialization breakage and ESM2 being broken are both broad regressions, but they are distinct model-specific failures." + }, + { + "left": "issue:43441", + "right": "issue:46752", + "accept": false, + "reason": "FlashAttention failure for Ministral-3 is unrelated to apply_chat_template crashing on an empty list." + }, + { + "left": "issue:45375", + "right": "issue:45478", + "accept": false, + "reason": "Both are Qwen3.5/MoE related, but one is a missing config field under @strict while the other is a generic from_pretrained regression; not clearly the same concrete bug." + }, + { + "left": "issue:42994", + "right": "issue:43482", + "accept": false, + "reason": "Quantized model saving failure and GGUF loading failure are different quantization pathways with different failure modes." + }, + { + "left": "issue:43475", + "right": "issue:43643", + "accept": false, + "reason": "SAM 3 Video missing attribute and AutoConfig trust_remote_code field loss are unrelated issues." + }, + { + "left": "issue:43232", + "right": "issue:43673", + "accept": false, + "reason": "Both are generation-related, but sync_gpus kwarg handling and chunked_prefill cache availability are different bugs." + }, + { + "left": "issue:44188", + "right": "issue:46851", + "accept": false, + "reason": "Attention kernel divergence under torch.compile is unrelated to Qwen3.5 variable-length position_ids handling." + }, + { + "left": "issue:28282", + "right": "issue:30990", + "accept": false, + "reason": "PyTorch-not-found import error and Sentence Transformers hanging on load are different issues." + }, + { + "left": "issue:41628", + "right": "issue:44933", + "accept": false, + "reason": "Both are import/export problems, but they concern different missing symbols and different modules; too coarse to merge." + }, + { + "left": "issue:43475", + "right": "issue:46821", + "accept": false, + "reason": "SAM 3 Video attribute error and TimesFM window_size AttributeError are different model-specific failures." + }, + { + "left": "issue:45230", + "right": "issue:45397", + "accept": false, + "reason": "A generic bug report and Gemma-4 zero3 from_pretrained failure are unrelated." + }, + { + "left": "issue:44964", + "right": "issue:45137", + "accept": false, + "reason": "Phi-4 multimodal loading failure and DeepSpeed ZeRO3 deque underflow are unrelated subsystems." + }, + { + "left": "issue:45310", + "right": "issue:45375", + "accept": false, + "reason": "Both touch Qwen3.5/MoE, but one is a generic from_pretrained error and the other is a specific missing deepstack_visual_indexes field; not enough evidence they are the same underlying defect." + }, + { + "left": "issue:43986", + "right": "issue:46726", + "accept": false, + "reason": "AutoProcessor crash without torchvision and AutoProcessor reporting an unrecognized image processor are different missing-dependency / error-message behaviors." + }, + { + "left": "issue:44869", + "right": "issue:45356", + "accept": false, + "reason": "Whisper timestamp decode crash and Kimi-K2.5 tokenizer codec regression are unrelated tokenizer issues." + }, + { + "left": "issue:44038", + "right": "issue:44623", + "accept": false, + "reason": "Qwen3-VL-Moe from_pretrained failures and processor.save_pretrained missing files are different operations and failure modes." + }, + { + "left": "issue:39692", + "right": "issue:46726", + "accept": false, + "reason": "Documentation example errors for SigLIP2 and a runtime missing-vision-backend crash are not the same bug." + }, + { + "left": "issue:46463", + "right": "issue:46877", + "accept": false, + "reason": "Incorrect T5 decoder-input validation messaging and auto_docstring emitting diagnostics at import time are unrelated docstring/lint behaviors." + }, + { + "left": "issue:44373", + "right": "issue:46851", + "accept": false, + "reason": "A wrong docstring for position_ids is unrelated to a runtime bug in variable-length inputs using position_ids." + }, + { + "left": "issue:42994", + "right": "issue:43525", + "accept": false, + "reason": "Quantized model saving failure and Llama4Config missing pad_token_id are different issues." + }, + { + "left": "issue:42371", + "right": "issue:45092", + "accept": false, + "reason": "TF32 settings guidance and InternVL2 meta-initialization incompatibility are unrelated." + }, + { + "left": "issue:43723", + "right": "issue:46326", + "accept": false, + "reason": "Tokenizer loading in v5 and continue_final_message behavior for mistral_common are different tokenizer bugs." + }, + { + "left": "issue:45406", + "right": "issue:46202", + "accept": false, + "reason": "Gemma4Processor missing _tokenizer and Gemma pipeline roundtrip failure on \u2581 are different bugs in different layers." + } + ] + }, + "evaluator_result": 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": 46, + "review_unit_count": 60, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-500", + "nodes": [ + "issue:30333", + "issue:30990", + "issue:33290", + "issue:35141", + "issue:36010", + "issue:41628", + "issue:42175", + "issue:42673", + "issue:42831", + "issue:42898", + "issue:42994", + "issue:43232", + "issue:43278", + "issue:43479", + "issue:43504", + "issue:43704", + "issue:43784", + "issue:43824", + "issue:43873", + "issue:43976", + "issue:44038", + "issue:44222", + "issue:44351", + "issue:44496", + "issue:44568", + "issue:44623", + "issue:44792", + "issue:44857", + "issue:44933", + "issue:44945", + "issue:45081", + "issue:45103", + "issue:45137", + "issue:45310", + "issue:45405", + "issue:45561", + "issue:45593", + "issue:45600", + "issue:45712", + "issue:45753", + "issue:46077", + "issue:46123", + "issue:46202", + "issue:46399", + "issue:46693", + "issue:46701", + "issue:46829", + "issue:46877" + ], + "soft_pairs": [ + "issue:43873|issue:44945", + "issue:44945|issue:45753", + "issue:43784|issue:46399", + "issue:41628|issue:44351", + "issue:44857|issue:46877", + "issue:45593|issue:46877", + "issue:42673|issue:43704", + "issue:44038|issue:45310", + "issue:36010|issue:41628", + "issue:44222|issue:45310", + "issue:43278|issue:45137", + "issue:46202|issue:46701", + "issue:42994|issue:43824", + "issue:45081|issue:45103", + "issue:42831|issue:44222", + "issue:43232|issue:44792", + "issue:43504|issue:45561", + "issue:42175|issue:44933", + "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" + ], + "prepared_review_unit_hash": "950da8a2926f00589bfc9e633d9d5b7ebb65c3add4a1fe373138a7466d403e88", + "component_budget": { + "node_count": 500, + "item_count": 500, + "soft_pair_count": 1839, + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 30, + "serialized_chars": 36539, + "estimated_input_tokens": 9135, + "estimated_eval_tokens": 18526 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a grab bag of unrelated issue reports. The clearest duplicate pair is the repeated Qwen3ForCausalLM VRAM leak report (#42673/#43704); the rest do not look like the same underlying bug.", + "confidence": 0.95, + "canonical_issue_reason": "#42673 is the best canonical issue because it is the older report in the clearest duplicate pair with #43704, and it has more discussion activity plus the same title/symptom.", + "canonical_pr_reason": null, + "best_issue_reason": "#42673 is the strongest representative issue in this cluster: it matches #43704 on the Qwen3ForCausalLM VRAM leak symptom and has the richer discussion context.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43873", + "right": "issue:44945", + "accept": false, + "reason": "Quantization/offloading failure vs pipeline-parallel output correctness; different code paths and symptoms." + }, + { + "left": "issue:44945", + "right": "issue:45753", + "accept": false, + "reason": "Pipeline-parallel output issue vs image-specific infinite loop; not the same bug." + }, + { + "left": "issue:43784", + "right": "issue:46399", + "accept": false, + "reason": "Import-time NameError in sentence-transformers integration vs Qwen3.5 kernel registration crash; unrelated failures." + }, + { + "left": "issue:41628", + "right": "issue:44351", + "accept": false, + "reason": "Different missing symbols (`AutoImageProcessor` vs `HybridCache`) and likely separate API breakages." + }, + { + "left": "issue:44857", + "right": "issue:46877", + "accept": false, + "reason": "CUDA/AMP loss crash vs docstring lint logging at import time; no shared bug." + }, + { + "left": "issue:45593", + "right": "issue:46877", + "accept": false, + "reason": "Training loss/auxiliary-loss behavior vs auto_docstring stdout noise; unrelated." + }, + { + "left": "issue:42673", + "right": "issue:43704", + "accept": true, + "reason": "Same title and same symptom: Qwen3ForCausalLM VRAM leak when used in multiple dataloader threads; clearly the same underlying bug." + }, + { + "left": "issue:44038", + "right": "issue:45310", + "accept": false, + "reason": "Different Qwen MoE models, different version constraints, and different failure modes." + }, + { + "left": "issue:36010", + "right": "issue:41628", + "accept": false, + "reason": "Separate import errors for different exported symbols; not one bug." + }, + { + "left": "issue:44222", + "right": "issue:45310", + "accept": false, + "reason": "FP8 save_pretrained failure vs MoE from_pretrained error; different concrete code paths." + }, + { + "left": "issue:43278", + "right": "issue:45137", + "accept": false, + "reason": "Embedding dtype drift during eval vs DeepSpeed ZeRO3 deque IndexError; unrelated." + }, + { + "left": "issue:46202", + "right": "issue:46701", + "accept": false, + "reason": "Gemma token roundtrip issue vs DiffusionGemma pipeline failure; too different to merge." + }, + { + "left": "issue:42994", + "right": "issue:43824", + "accept": false, + "reason": "Quantized model saving failure vs missing Qwen2_5_VL class import; unrelated." + }, + { + "left": "issue:45081", + "right": "issue:45103", + "accept": false, + "reason": "Tokenizer regex patch crash vs auto_docstring annotation handling crash; different subsystems." + }, + { + "left": "issue:42831", + "right": "issue:44222", + "accept": false, + "reason": "FP8 accuracy regression vs FP8 model saving bug; same broad area but not the same bug." + }, + { + "left": "issue:43232", + "right": "issue:44792", + "accept": false, + "reason": "Generation kwargs/sync_gpus bug vs Janus image-generation test failure; not the same failure." + }, + { + "left": "issue:43504", + "right": "issue:45561", + "accept": false, + "reason": "Legacy-field model loading issue vs pytest-xdist race on a temp file; unrelated." + }, + { + "left": "issue:42175", + "right": "issue:44933", + "accept": false, + "reason": "Backend extra packaging problem vs missing image_utils import; different packaging/API issues." + }, + { + "left": "issue:45561", + "right": "issue:45600", + "accept": false, + "reason": "File-race in patched testing utils vs stale auto_mappings references; separate CI/test problems." + }, + { + "left": "issue:43976", + "right": "issue:44933", + "accept": false, + "reason": "Python version compatibility issue vs missing import from image_utils; unrelated." + }, + { + "left": "issue:44496", + "right": "issue:45561", + "accept": false, + "reason": "Unrecognized model/config loading failure vs xdist file race; no shared code path." + }, + { + "left": "issue:30333", + "right": "issue:33290", + "accept": false, + "reason": "MLflow reporting failure vs DeepSpeed OOM; different runtime issues." + }, + { + "left": "issue:45405", + "right": "issue:46877", + "accept": false, + "reason": "Dependency version bump complaint vs docstring import-time output; unrelated." + }, + { + "left": "issue:30990", + "right": "issue:35141", + "accept": false, + "reason": "Loading hang in Sentence Transformers vs embedding reinitialization after resize; different bugs." + }, + { + "left": "issue:42898", + "right": "issue:44568", + "accept": false, + "reason": "Tokenization cleanup behavior change vs add_special_tokens/BOS-EOS regression; related domain but different behavior." + }, + { + "left": "issue:42175", + "right": "issue:45712", + "accept": false, + "reason": "Torch extra packaging issue vs leftover dummy classes leaking into dir(); different maintenance problems." + }, + { + "left": "issue:43479", + "right": "issue:46829", + "accept": false, + "reason": "Multimodal config default-init bug vs hidden_states field mismatch in video classification output; unrelated." + }, + { + "left": "issue:46077", + "right": "issue:46123", + "accept": false, + "reason": "Bidirectional attention mask bug vs final partial batch dropping results; different pipelines." + }, + { + "left": "issue:44623", + "right": "issue:45310", + "accept": false, + "reason": "processor.save_pretrained missing files vs Qwen3.5 MoE from_pretrained error; not the same bug." + }, + { + "left": "issue:42673", + "right": "issue:46693", + "accept": false, + "reason": "VRAM leak in multi-threaded Qwen3ForCausalLM loading vs FA2 long-context performance regression; unrelated performance problems." + } + ] + }, + "evaluator_result": { + "accept": false, + "feedback": "The duplicate-pair judgment is mostly conservative, but the canonical issue reason overstates the packet in two places: it claims this is the \"only true duplicate pair\" and says #42673 has \"more references,\" neither of which is directly supported by the packet (both issues show 0 inbound references). Please soften those claims to what is actually evidenced, e.g. older issue with more discussion activity and matching title/symptom." + }, + "evaluator_used": true, + "retried": true, + "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": 60, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-500", + "nodes": [ + "issue:22355", + "issue:27957", + "issue:28282", + "issue:30333", + "issue:34689", + "issue:35141", + "issue:39692", + "issue:42222", + "issue:42796", + "issue:42831", + "issue:42898", + "issue:43116", + "issue:43295", + "issue:43479", + "issue:43723", + "issue:43784", + "issue:43824", + "issue:43874", + "issue:44016", + "issue:44060", + "issue:44075", + "issue:44079", + "issue:44190", + "issue:44222", + "issue:44295", + "issue:44297", + "issue:44360", + "issue:44568", + "issue:44623", + "issue:44625", + "issue:44734", + "issue:44779", + "issue:44869", + "issue:44933", + "issue:45092", + "issue:45103", + "issue:45125", + "issue:45200", + "issue:45356", + "issue:45405", + "issue:45656", + "issue:45712", + "issue:45812", + "issue:46231", + "issue:46762", + "issue:46877", + "issue:46909" + ], + "soft_pairs": [ + "issue:43479|issue:44869", + "issue:43723|issue:44568", + "issue:44016|issue:46231", + "issue:42898|issue:45812", + "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", + "issue:42898|issue:43295", + "issue:42898|issue:45356", + "issue:44295|issue:45656", + "issue:43116|issue:44190", + "issue:30333|issue:35141", + "issue:45405|issue:45712", + "issue:44222|issue:44297", + "issue:22355|issue:28282", + "issue:44060|issue:45125", + "issue:44075|issue:46231", + "issue:44360|issue:46762", + "issue:44933|issue:45712", + "issue:42831|issue:44623", + "issue:44297|issue:45656", + "issue:42796|issue:43723", + "issue:44222|issue:44623" + ], + "prepared_review_unit_hash": "fae5594fb13d5113b23540c680f434d8fbb0e67d0c6a75ff44b6b66cdff312fd", + "component_budget": { + "node_count": 500, + "item_count": 500, + "soft_pair_count": 1839, + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 30, + "serialized_chars": 36263, + "estimated_input_tokens": 9066, + "estimated_eval_tokens": 18388 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "No strong duplicate group emerges here: the items are mostly unrelated Transformers issues spanning tokenizers, multimodal configs, docs, model loading, and training utilities. The closest theme is general v5/tokenizer regression, but the soft pairs are still too different to merge as duplicates.", + "confidence": 0.86, + "canonical_issue_reason": "issue:42898 is the most umbrella-like representative of the recurring tokenizer/v5 regression theme, with a broad user-facing behavior change rather than a narrow model-specific symptom.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:42898 is the best single issue to anchor this noisy cluster because it is broad, clearly stated, and closest to the shared regression theme across several nearby candidates.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43479", + "right": "issue:44869", + "accept": false, + "reason": "Different bugs: default multimodal config initialization vs a Whisper word-timestamp decode crash." + }, + { + "left": "issue:43723", + "right": "issue:44568", + "accept": false, + "reason": "Both are v5/tokenizer-related, but one is an AutoTokenizer loading failure and the other is missing BOS/EOS insertion for a specific tokenizer." + }, + { + "left": "issue:44016", + "right": "issue:46231", + "accept": false, + "reason": "A notebook syntax error and a docs parameter-name mismatch are unrelated documentation issues." + }, + { + "left": "issue:42898", + "right": "issue:45812", + "accept": false, + "reason": "Both involve tokenizers, but one is a general clean_up_tokenization_spaces v5 behavior change and the other is Granite token IDs being wrong." + }, + { + "left": "issue:44625", + "right": "issue:46877", + "accept": false, + "reason": "Qwen3.5 label propagation and auto_docstring stdout logging are unrelated subsystems." + }, + { + "left": "issue:42898", + "right": "issue:44779", + "accept": false, + "reason": "Both are tokenizer regressions, but they affect different models and different failure modes." + }, + { + "left": "issue:39692", + "right": "issue:46909", + "accept": false, + "reason": "Different broken documentation examples for different models; not the same underlying change." + }, + { + "left": "issue:43479", + "right": "issue:44079", + "accept": false, + "reason": "Multimodal config defaults and ModelOutput key assignment are unrelated code paths." + }, + { + "left": "issue:43874", + "right": "issue:45092", + "accept": false, + "reason": "A missing image-processor method and remote-code/meta-init incompatibility are distinct multimodal loading issues." + }, + { + "left": "issue:43784", + "right": "issue:43824", + "accept": false, + "reason": "Different import errors from different causes: missing nn import vs missing Qwen2.5-VL export." + }, + { + "left": "issue:27957", + "right": "issue:30333", + "accept": false, + "reason": "Flash Attention model support and MLflow reporting on failed jobs are unrelated." + }, + { + "left": "issue:44734", + "right": "issue:45103", + "accept": false, + "reason": "Serve/KV-cache indexing crash and auto_docstring annotation handling are unrelated." + }, + { + "left": "issue:30333", + "right": "issue:34689", + "accept": false, + "reason": "MLflow failure reporting and Llama model loading regression are different problems." + }, + { + "left": "issue:42222", + "right": "issue:45200", + "accept": false, + "reason": "VitPose model breakage and Gemma 4 mm_token_type_ids defaults are different model-specific bugs." + }, + { + "left": "issue:42898", + "right": "issue:43295", + "accept": false, + "reason": "Both are v5 regressions involving tokenizer/processor behavior, but the concrete failures are different." + }, + { + "left": "issue:42898", + "right": "issue:45356", + "accept": false, + "reason": "Tokenizer regressions in v5, but one is clean_up_tokenization_spaces behavior and the other is Kimi codec/regex handling." + }, + { + "left": "issue:44295", + "right": "issue:45656", + "accept": false, + "reason": "Reading a registered buffer and deepspeed double optimizer steps are unrelated." + }, + { + "left": "issue:43116", + "right": "issue:44190", + "accept": false, + "reason": "Two different example scripts with different failures; not the same bug." + }, + { + "left": "issue:30333", + "right": "issue:35141", + "accept": false, + "reason": "Failed MLflow reporting and embedding reinitialization after resize are unrelated." + }, + { + "left": "issue:45405", + "right": "issue:45712", + "accept": false, + "reason": "A dependency version bump problem and leftover dummy class cleanup are not the same issue." + }, + { + "left": "issue:44222", + "right": "issue:44297", + "accept": false, + "reason": "Both mention save_pretrained, but one is FP8 MoE model saving and the other is tokenizer config metadata." + }, + { + "left": "issue:22355", + "right": "issue:28282", + "accept": false, + "reason": "Import errors, but one is missing transformers.onnx and the other is missing PyTorch entirely." + }, + { + "left": "issue:44060", + "right": "issue:45125", + "accept": false, + "reason": "Different Qwen model bugs: a wrong tied-weights warning versus missing tensor-parallel plan metadata." + }, + { + "left": "issue:44075", + "right": "issue:46231", + "accept": false, + "reason": "Optimizer behavior and documentation wording mismatch are unrelated." + }, + { + "left": "issue:44360", + "right": "issue:46762", + "accept": false, + "reason": "A DSA indexer implementation concern and Minimax attention numerical differences are different algorithmic issues." + }, + { + "left": "issue:44933", + "right": "issue:45712", + "accept": false, + "reason": "Both touch exports/symbol visibility, but one is a nonexistent import and the other is leftover dummy classes in a different module." + }, + { + "left": "issue:42831", + "right": "issue:44623", + "accept": false, + "reason": "Fine-grained FP8 accuracy and processor.save_pretrained file omission are unrelated." + }, + { + "left": "issue:44297", + "right": "issue:45656", + "accept": false, + "reason": "Tokenizer save metadata and deepspeed optimizer stepping are unrelated code paths." + }, + { + "left": "issue:42796", + "right": "issue:43723", + "accept": false, + "reason": "Both are tokenizer-related, but one is BPE marker handling for a specific model and the other is a generic AutoTokenizer loading issue." + }, + { + "left": "issue:44222", + "right": "issue:44623", + "accept": false, + "reason": "Both involve persistence APIs, but one is FP8 MoE model saving and the other is processor file output; 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": 48, + "review_unit_count": 60, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-500", + "nodes": [ + "issue:30990", + "issue:38175", + "issue:41628", + "issue:42757", + "issue:42831", + "issue:42907", + "issue:43122", + "issue:43232", + "issue:43475", + "issue:43479", + "issue:43644", + "issue:43723", + "issue:44190", + "issue:44222", + "issue:44261", + "issue:44263", + "issue:44279", + "issue:44295", + "issue:44297", + "issue:44360", + "issue:44373", + "issue:44448", + "issue:44483", + "issue:44484", + "issue:44485", + "issue:44623", + "issue:44977", + "issue:45071", + "issue:45092", + "issue:45245", + "issue:45310", + "issue:45357", + "issue:45381", + "issue:45406", + "issue:45834", + "issue:46050", + "issue:46082", + "issue:46132", + "issue:46209", + "issue:46305", + "issue:46328", + "issue:46369", + "issue:46424", + "issue:46431", + "issue:46655", + "issue:46710", + "issue:46909" + ], + "soft_pairs": [ + "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:43122|issue:46209", + "issue:44263|issue:45092", + "issue:44190|issue:46082", + "issue:43232|issue:46424", + "issue:43232|issue:45834", + "issue:44295|issue:44623", + "issue:43475|issue:46305", + "issue:44263|issue:44360", + "issue:46209|issue:46431", + "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:44483|issue:45406" + ], + "prepared_review_unit_hash": "fcb8986b7140d2eae39939c643fc793a6d2518b1657b67aaeeb3924c1c3f4555", + "component_budget": { + "node_count": 500, + "item_count": 500, + "soft_pair_count": 1839, + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 30, + "serialized_chars": 35671, + "estimated_input_tokens": 8918, + "estimated_eval_tokens": 18092 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This set is a loose collection of mostly unrelated Transformers issues. The soft edges are driven by surface-level similarity around v5 regressions, config/save/load problems, and model-specific failures, but the underlying bugs are distinct in almost every pair.", + "confidence": 0.92, + "canonical_issue_reason": "issue:45071 is the best cluster anchor because it is the broadest library-level regression here (core PretrainedConfig/type-checking behavior in v5), whereas most other issues are narrower model- or feature-specific failures.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:45071 is the strongest representative issue for the cluster\u2019s apparent theme of v5 regression breakage: it is user-facing, cross-cutting, and more general than the model-specific or one-off API errors in the rest of the set.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44448", + "right": "issue:46132", + "accept": false, + "reason": "Both mention v4/v5 behavior, but one is Pegasus output drift and the other is AttentionInterface.register semantics; different code paths and fixes." + }, + { + "left": "issue:45245", + "right": "issue:46132", + "accept": false, + "reason": "Category-count overflow is a data/encoding limit issue, not an attention-interface behavior change." + }, + { + "left": "issue:43479", + "right": "issue:43644", + "accept": false, + "reason": "Both involve config/init behavior, but one is multimodal config defaults and the other is non-persistent buffer initialization; not the same bug." + }, + { + "left": "issue:41628", + "right": "issue:42757", + "accept": false, + "reason": "Two separate import errors from different packages/symbols; only superficially similar as dependency breakage." + }, + { + "left": "issue:44373", + "right": "issue:44485", + "accept": false, + "reason": "Wrong docstring for position_ids is documentation-only; GLM-5 RoPE is a real implementation issue." + }, + { + "left": "issue:44297", + "right": "issue:45310", + "accept": false, + "reason": "One is tokenizer save_pretrained metadata mismatch, the other is a from_pretrained loading error; related area but different failures." + }, + { + "left": "issue:46132", + "right": "issue:46710", + "accept": false, + "reason": "Both are regressions, but one is an attention registration behavior change and the other is an accuracy regression in a specific model after upgrade." + }, + { + "left": "issue:45071", + "right": "issue:46082", + "accept": false, + "reason": "Both touch config validation, but one is type-checking behavior and the other is explicit head_dim acceptance rules; not the same defect." + }, + { + "left": "issue:44222", + "right": "issue:44484", + "accept": false, + "reason": "FP8 moe save_pretrained corruption and max_shard_size defaults are unrelated save_pretrained concerns." + }, + { + "left": "issue:43723", + "right": "issue:44977", + "accept": false, + "reason": "Tokenizer loading failure and flash-attention generation failure affect different subsystems and different models." + }, + { + "left": "issue:43122", + "right": "issue:46209", + "accept": false, + "reason": "Tokenization drift across versions and FP8 inference failure on B200 are unrelated issues." + }, + { + "left": "issue:44263", + "right": "issue:45092", + "accept": false, + "reason": "torch.split return values in a DSA indexer are unrelated to remote-code checkpoint incompatibility with meta initialization." + }, + { + "left": "issue:44190", + "right": "issue:46082", + "accept": false, + "reason": "A local dataset loading script failure is not the same bug as LlamaConfig head_dim validation." + }, + { + "left": "issue:43232", + "right": "issue:46424", + "accept": false, + "reason": "Generation kwargs update after sync_gpus and cache max length handling are different generation/cache behaviors." + }, + { + "left": "issue:43232", + "right": "issue:45834", + "accept": false, + "reason": "The generation bug and the Kosmos2.5 OCR index error are unrelated." + }, + { + "left": "issue:44295", + "right": "issue:44623", + "accept": false, + "reason": "A position_ids buffer access error is not the same as processor.save_pretrained missing files." + }, + { + "left": "issue:43475", + "right": "issue:46305", + "accept": false, + "reason": "SAM3 missing an attribute and MoonshineStreaming off-by-one masking are different model-specific bugs." + }, + { + "left": "issue:44263", + "right": "issue:44360", + "accept": false, + "reason": "Both mention the DSA indexer, but one is about split return values and the other about a missing ReLU; different fixes." + }, + { + "left": "issue:46209", + "right": "issue:46431", + "accept": false, + "reason": "FP8 inference on B200 and missing chat template are unrelated issues." + }, + { + "left": "issue:46305", + "right": "issue:46655", + "accept": false, + "reason": "Both are video-processing bugs, but one is a sliding-window off-by-one and the other is a padding calculation error; not the same defect." + }, + { + "left": "issue:44263", + "right": "issue:44373", + "accept": false, + "reason": "Indexer split behavior and a wrong docstring for position_ids are unrelated." + }, + { + "left": "issue:30990", + "right": "issue:38175", + "accept": false, + "reason": "Sentence Transformers loading hangs and SigLIP2 zero probabilities are different model-level problems." + }, + { + "left": "issue:42831", + "right": "issue:45310", + "accept": false, + "reason": "FP8 accuracy regression and Qwen3.5 from_pretrained failure are distinct issues." + }, + { + "left": "issue:45092", + "right": "issue:45357", + "accept": false, + "reason": "InternVL2 meta-init incompatibility and Qwen3.5 save_pretrained key mismatch involve different models and different failure modes." + }, + { + "left": "issue:46369", + "right": "issue:46909", + "accept": false, + "reason": "SAM3 memory overflow and a broken DIA documentation example are unrelated." + }, + { + "left": "issue:42907", + "right": "issue:46050", + "accept": false, + "reason": "Both involve save/load of quantized models, but one is dequantized Ministrals/Devstrals saving and the other is sinq round-trip breakage; different quantization paths." + }, + { + "left": "issue:44279", + "right": "issue:46328", + "accept": false, + "reason": "A dependency problem and a NumPy-to-list performance issue are unrelated." + }, + { + "left": "issue:45381", + "right": "issue:46305", + "accept": false, + "reason": "Both concern video position/masking, but they affect different models and code paths; not the same underlying bug." + }, + { + "left": "issue:44261", + "right": "issue:45092", + "accept": false, + "reason": "MLA q_a_layernorm epsilon/config precision issue is unrelated to InternVL2 meta-initialization incompatibility." + }, + { + "left": "issue:44483", + "right": "issue:45406", + "accept": false, + "reason": "A chat/completions request rejection and a Gemma4Processor missing _tokenizer attribute are different serving failures." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 49, + "review_unit_count": 60, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-500", + "nodes": [ + "issue:22355", + "issue:27957", + "issue:28218", + "issue:30990", + "issue:33357", + "issue:39401", + "issue:41628", + "issue:42371", + "issue:42757", + "issue:42796", + "issue:42831", + "issue:42886", + "issue:42907", + "issue:42994", + "issue:43066", + "issue:43122", + "issue:43232", + "issue:43408", + "issue:43540", + "issue:43576", + "issue:43644", + "issue:43646", + "issue:43784", + "issue:43824", + "issue:43937", + "issue:43976", + "issue:44038", + "issue:44188", + "issue:44263", + "issue:44295", + "issue:44297", + "issue:44373", + "issue:44488", + "issue:44568", + "issue:44829", + "issue:44945", + "issue:44964", + "issue:45278", + "issue:45381", + "issue:45593", + "issue:45753", + "issue:46002", + "issue:46172", + "issue:46305", + "issue:46369", + "issue:46402", + "issue:46762", + "issue:46909" + ], + "soft_pairs": [ + "issue:22355|issue:27957", + "issue:44829|issue:46002", + "issue:43644|issue:46305", + "issue:42831|issue:44038", + "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", + "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:43576|issue:43784", + "issue:44295|issue:44297", + "issue:42371|issue:44263", + "issue:39401|issue:44488", + "issue:42907|issue:44964", + "issue:46172|issue:46762", + "issue:43976|issue:44945", + "issue:42907|issue:43937" + ], + "prepared_review_unit_hash": "64bc82c6a86de145aab6f5b4c345da8a7d8f30c26aeec00acdc0db8852a7f9ed", + "component_budget": { + "node_count": 500, + "item_count": 500, + "soft_pair_count": 1839, + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 30, + "serialized_chars": 36050, + "estimated_input_tokens": 9013, + "estimated_eval_tokens": 18282 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is heterogeneous: most items are distinct regressions or feature-specific bugs, with a few superficially similar import/tokenizer issues but no clear duplicate pair. The best umbrella issue is the broad version-upgrade import breakage report.", + "confidence": 0.82, + "canonical_issue_reason": "issue:45278 is the broadest and most representative issue in the set, covering multiple import failures after upgrading transformers, unlike the more narrow or unrelated reports.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:45278 is the most useful global issue candidate because it describes a broader, user-visible regression pattern that can subsume several import-related symptoms.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44829", + "right": "issue:46002", + "accept": false, + "reason": "Different problems: flash_attention_3 training collapse vs NaNs after checkout/init." + }, + { + "left": "issue:43644", + "right": "issue:46305", + "accept": false, + "reason": "Unrelated bugs: non-persistent buffer junk values vs a sliding-window off-by-one in audio masking." + }, + { + "left": "issue:42831", + "right": "issue:44038", + "accept": false, + "reason": "Different symptoms and code paths: FineGrainedFP8 accuracy issue vs Qwen3-VL-Moe bug." + }, + { + "left": "issue:41628", + "right": "issue:43824", + "accept": false, + "reason": "Both are import errors, but for different missing symbols and likely different export/regression causes." + }, + { + "left": "issue:43408", + "right": "issue:46402", + "accept": false, + "reason": "Different SAM3-related issues: model-type warning vs missing text encoder weights." + }, + { + "left": "issue:42831", + "right": "issue:45753", + "accept": false, + "reason": "Accuracy regression in FP8 is not the same as an infinite-loop bug on a specific image." + }, + { + "left": "issue:22355", + "right": "issue:28218", + "accept": false, + "reason": "Module import failure vs tokenizer spacing behavior; no shared bug." + }, + { + "left": "issue:39401", + "right": "issue:44568", + "accept": false, + "reason": "Both tokenizer bugs, but one is offset_mapping for Qwen3 and the other is BOS/EOS insertion for mdeberta-v3-base." + }, + { + "left": "issue:42886", + "right": "issue:43066", + "accept": false, + "reason": "Offline cache loading failure is unrelated to a wrong tokenizer decoder type." + }, + { + "left": "issue:42886", + "right": "issue:42994", + "accept": false, + "reason": "Tokenizer cache/offline loading and quantized model saving are different failure modes." + }, + { + "left": "issue:42757", + "right": "issue:45278", + "accept": false, + "reason": "44257-like single missing import is not the same concrete bug as the broader multi-import regression; too broad to merge as duplicates." + }, + { + "left": "issue:27957", + "right": "issue:30990", + "accept": false, + "reason": "Flash Attention 2 on XLMRoberta and Sentence Transformers loading hang are unrelated." + }, + { + "left": "issue:44188", + "right": "issue:44373", + "accept": false, + "reason": "Attention-kernel divergence and a wrong docstring are different kinds of issues." + }, + { + "left": "issue:43408", + "right": "issue:46369", + "accept": false, + "reason": "SAM3 model-type warning is unrelated to memory overflow in sam3-litetext." + }, + { + "left": "issue:43232", + "right": "issue:45593", + "accept": false, + "reason": "Generation kwargs handling after sync_gpus is unrelated to D-FINE auxiliary-loss behavior." + }, + { + "left": "issue:42796", + "right": "issue:43122", + "accept": false, + "reason": "Ministral BPE marker handling differs from a version-to-version tokenization drift report." + }, + { + "left": "issue:43646", + "right": "issue:46002", + "accept": false, + "reason": "Custom model initialization breakage is not the same as NaNs in classification heads." + }, + { + "left": "issue:45381", + "right": "issue:46909", + "accept": false, + "reason": "Qwen2.5-VL video position IDs and a broken documentation example for DIA are unrelated." + }, + { + "left": "issue:43540", + "right": "issue:46909", + "accept": false, + "reason": "Video-input ValueError in Qwen3OmniMoe is unrelated to a documentation example bug." + }, + { + "left": "issue:30990", + "right": "issue:33357", + "accept": false, + "reason": "Loading hang for Sentence Transformers and a MacOS bus error in CLIP are different failures." + }, + { + "left": "issue:42907", + "right": "issue:44945", + "accept": false, + "reason": "Dequantized Ministral save failure is unrelated to incorrect outputs under pipeline parallelism." + }, + { + "left": "issue:43576", + "right": "issue:43784", + "accept": false, + "reason": "Broken env command and a NameError importing sentence-transformers are different issues." + }, + { + "left": "issue:44295", + "right": "issue:44297", + "accept": false, + "reason": "Position_ids buffer read error is unrelated to tokenizer_class mismatch on save_pretrained." + }, + { + "left": "issue:42371", + "right": "issue:44263", + "accept": false, + "reason": "TF32 settings warning and torch.split return values in GlmMoeDsaIndexer do not share a bug." + }, + { + "left": "issue:39401", + "right": "issue:44488", + "accept": false, + "reason": "Qwen3 offset_mapping bug and failure to load cjvt/sleng-bert are different tokenizer/model issues." + }, + { + "left": "issue:42907", + "right": "issue:44964", + "accept": false, + "reason": "Quantized model saving failure is unrelated to a multimodal model load failure." + }, + { + "left": "issue:46172", + "right": "issue:46762", + "accept": false, + "reason": "DeepSeek-V4 comb transpose bug and Minimax attention numerical drift are not the same code-path problem." + }, + { + "left": "issue:43976", + "right": "issue:44945", + "accept": false, + "reason": "Python version compatibility issue is unrelated to pipeline-parallelism output corruption." + }, + { + "left": "issue:42907", + "right": "issue:43937", + "accept": false, + "reason": "Quantized save failure is unrelated to invalid GenerationConfig handling for GLM-5." + } + ] + }, + "evaluator_result": 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": 60, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-500", + "nodes": [ + "issue:22355", + "issue:29127", + "issue:29159", + "issue:36296", + "issue:39401", + "issue:41628", + "issue:42175", + "issue:42757", + "issue:42907", + "issue:43116", + "issue:43122", + "issue:43278", + "issue:43408", + "issue:43452", + "issue:43475", + "issue:43493", + "issue:43756", + "issue:43784", + "issue:43824", + "issue:44038", + "issue:44060", + "issue:44117", + "issue:44188", + "issue:44222", + "issue:44295", + "issue:44297", + "issue:44351", + "issue:44360", + "issue:44485", + "issue:44521", + "issue:44625", + "issue:44704", + "issue:44779", + "issue:44829", + "issue:44938", + "issue:44945", + "issue:45200", + "issue:45216", + "issue:45278", + "issue:45290", + "issue:45478", + "issue:45600", + "issue:45753", + "issue:46231", + "issue:46326", + "issue:46328", + "issue:46614", + "issue:46877" + ], + "soft_pairs": [ + "issue:43756|issue:44060", + "issue:41628|issue:45278", + "issue:44117|issue:44704", + "issue:43784|issue:45600", + "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", + "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:22355|issue:29127", + "issue:42175|issue:44938", + "issue:43493|issue:45753", + "issue:46231|issue:46877", + "issue:44188|issue:46328" + ], + "prepared_review_unit_hash": "d8638ad84b6d7b2e8de3643f450eb3f02de8483ca5b73400b546873670c8db49", + "component_budget": { + "node_count": 500, + "item_count": 500, + "soft_pair_count": 1839, + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 29, + "serialized_chars": 36237, + "estimated_input_tokens": 9060, + "estimated_eval_tokens": 18376 + }, + "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 and import/behavior issues. None of the soft pairs look like the same underlying bug or change, so all are rejected. The broadest representative issue is the general import-regression report.", + "confidence": 0.82, + "canonical_issue_reason": "issue:45278 is the best umbrella-style representative because it describes a broad upgrade-related import regression, which is the closest thing to a cluster-wide theme here.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:45278 is the most general and high-signal issue in the set, but it still only covers one narrow family of failures rather than the whole cluster.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43756", + "right": "issue:44060", + "accept": false, + "reason": "Different models and failure modes: RoPE layer counting vs a tied-weights warning in Qwen3-Next." + }, + { + "left": "issue:41628", + "right": "issue:45278", + "accept": false, + "reason": "Both are import problems, but they are different symbols and likely different causes; too broad to merge." + }, + { + "left": "issue:44117", + "right": "issue:44704", + "accept": false, + "reason": "One is TOKENIZER_MAPPING_NAMES returning None; the other is AutoProcessor not forwarding kwargs. Different code paths." + }, + { + "left": "issue:43784", + "right": "issue:45600", + "accept": false, + "reason": "Sentence-transformers import NameError is unrelated to stale auto_mappings entries for Sam3LiteText." + }, + { + "left": "issue:39401", + "right": "issue:44521", + "accept": false, + "reason": "Tokenizer offset_mapping bug vs multimodal assistant_mask generation bug; not the same underlying defect." + }, + { + "left": "issue:44704", + "right": "issue:45290", + "accept": false, + "reason": "Cached-file kwargs forwarding and chat-template tool-call crashing are separate API paths." + }, + { + "left": "issue:44625", + "right": "issue:44829", + "accept": false, + "reason": "Config propagation for num_labels is unrelated to a flash_attention_3 training degeneration bug." + }, + { + "left": "issue:44360", + "right": "issue:44485", + "accept": false, + "reason": "DSA indexer/ReLU discussion and GLM-5 RoPE implementation are different model/kernel issues." + }, + { + "left": "issue:43408", + "right": "issue:43475", + "accept": false, + "reason": "Both mention SAM3, but one is a model-type warning and the other is a missing output attribute." + }, + { + "left": "issue:43122", + "right": "issue:44521", + "accept": false, + "reason": "Tokenizer version drift is not the same as assistant_masks becoming all zero for multimodal inputs." + }, + { + "left": "issue:44038", + "right": "issue:44297", + "accept": false, + "reason": "Qwen3-VL-Moe import/compatibility issue vs tokenizer save_pretrained metadata mismatch; different bugs." + }, + { + "left": "issue:43122", + "right": "issue:46326", + "accept": false, + "reason": "Different tokenizer regressions affecting different behaviors and APIs." + }, + { + "left": "issue:42907", + "right": "issue:43452", + "accept": false, + "reason": "Saving dequantized model weights and gguf_file loading breakage are distinct workflows." + }, + { + "left": "issue:44038", + "right": "issue:44295", + "accept": false, + "reason": "Qwen3-VL-Moe regression and position_ids buffer access error do not describe the same bug." + }, + { + "left": "issue:45216", + "right": "issue:45478", + "accept": false, + "reason": "Related family and model name, but save_pretrained regression and from_pretrained error are different concrete failures." + }, + { + "left": "issue:22355", + "right": "issue:29159", + "accept": false, + "reason": "Missing transformers.onnx import is unrelated to slow/fast tokenizer inconsistency." + }, + { + "left": "issue:39401", + "right": "issue:46614", + "accept": false, + "reason": "Qwen3 offset_mapping and stale phonemizer language state are unrelated tokenization features." + }, + { + "left": "issue:44038", + "right": "issue:44222", + "accept": false, + "reason": "Both involve Qwen3-VL-Moe, but one is import/model wiring and the other is FP8 save_pretrained behavior." + }, + { + "left": "issue:43824", + "right": "issue:44351", + "accept": false, + "reason": "Different missing imports from different modules and symbols." + }, + { + "left": "issue:43116", + "right": "issue:45200", + "accept": false, + "reason": "Example-script multi-label output bug and Gemma 4 token-type defaults are separate issues." + }, + { + "left": "issue:43122", + "right": "issue:44779", + "accept": false, + "reason": "Both are tokenizer regressions, but for different models and symptoms; not mergeable as one fix." + }, + { + "left": "issue:42757", + "right": "issue:44351", + "accept": false, + "reason": "Both are import errors, but one is huggingface_hub and the other is transformers; different broken exports." + }, + { + "left": "issue:36296", + "right": "issue:43116", + "accept": false, + "reason": "Tensor parallel training and multi-label classification output are unrelated." + }, + { + "left": "issue:43278", + "right": "issue:44945", + "accept": false, + "reason": "Dtype mismatch during evaluate is unrelated to incorrect pipeline-parallel outputs." + }, + { + "left": "issue:22355", + "right": "issue:29127", + "accept": false, + "reason": "Missing module import and LayoutLMv3 error-message clarity are unrelated." + }, + { + "left": "issue:42175", + "right": "issue:44938", + "accept": false, + "reason": "Backend packaging omission and Python 3.14 load failure are different installation/runtime problems." + }, + { + "left": "issue:43493", + "right": "issue:45753", + "accept": false, + "reason": "SigLIP2 implementation discrepancy and Qwen3_5 image loop are unrelated model-specific bugs." + }, + { + "left": "issue:46231", + "right": "issue:46877", + "accept": false, + "reason": "Documentation parameter-name mismatch and import-time doc-lint stdout output are separate docs tooling issues." + }, + { + "left": "issue:44188", + "right": "issue:46328", + "accept": false, + "reason": "torch.compile attention-kernel branching and a tokenizer phonemizer padding issue 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": 51, + "review_unit_count": 60, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-500", + "nodes": [ + "issue:28348", + "issue:33453", + "issue:38617", + "issue:39401", + "issue:42222", + "issue:42371", + "issue:42673", + "issue:42898", + "issue:43066", + "issue:43116", + "issue:43441", + "issue:43493", + "issue:43611", + "issue:43618", + "issue:43825", + "issue:43827", + "issue:43874", + "issue:43937", + "issue:43976", + "issue:43992", + "issue:44038", + "issue:44075", + "issue:44263", + "issue:44355", + "issue:44485", + "issue:44610", + "issue:44625", + "issue:44704", + "issue:44779", + "issue:44936", + "issue:44995", + "issue:45072", + "issue:45083", + "issue:45092", + "issue:45278", + "issue:45357", + "issue:45446", + "issue:45478", + "issue:46002", + "issue:46231", + "issue:46614", + "issue:46640", + "issue:46697", + "issue:46710", + "issue:46752", + "issue:46762", + "issue:46829", + "issue:46909" + ], + "soft_pairs": [ + "issue:43937|issue:44485", + "issue:38617|issue:45278", + "issue:42898|issue:46614", + "issue:43066|issue:44779", + "issue:43611|issue:46002", + "issue:44038|issue:45478", + "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", + "issue:43618|issue:46829", + "issue:43992|issue:44704", + "issue:44075|issue:44936", + "issue:43116|issue:43825", + "issue:28348|issue:46640", + "issue:43066|issue:46710", + "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" + ], + "prepared_review_unit_hash": "fac701cfef7f999166e3036fc9d2fe69d6e223945ca70344007483add8b44e17", + "component_budget": { + "node_count": 500, + "item_count": 500, + "soft_pair_count": 1839, + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 28, + "serialized_chars": 35637, + "estimated_input_tokens": 8910, + "estimated_eval_tokens": 18076 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "All soft-edge candidates look like false positives: they share broad themes (tokenizers, Qwen/GLM, FlashAttention, upgrade regressions) but point to different concrete bugs, models, or code paths. No pull requests are present.", + "confidence": 0.88, + "canonical_issue_reason": "issue:45278 is the best umbrella-style anchor: it describes a broad, reproducible import regression after upgrade with multiple failing imports, making it the most representative issue in this set.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:45278 is the strongest global issue candidate because it is a clear, high-signal regression report with broad impact and an obvious upgrade trigger, so it is the most useful triage anchor.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43937", + "right": "issue:44485", + "accept": false, + "reason": "Same GLM-5 family, but one is GenerationConfig validation and the other is RoPE implementation; different failure points and not the same bug." + }, + { + "left": "issue:38617", + "right": "issue:45278", + "accept": false, + "reason": "Both are import-related, but 38617 is a specific missing symbol while 45278 is a broad upgrade regression with many import errors; not the same underlying issue." + }, + { + "left": "issue:42898", + "right": "issue:46614", + "accept": false, + "reason": "Tokenization-space behavior change and stale phonemizer backend state are unrelated code paths." + }, + { + "left": "issue:43066", + "right": "issue:44779", + "accept": false, + "reason": "Both involve tokenizer regressions, but one is decoder-type selection and the other is DeepSeek output correctness; different concrete bugs." + }, + { + "left": "issue:43611", + "right": "issue:46002", + "accept": false, + "reason": "Model loading/base_model_prefix breakage and NaNs in classification heads are distinct problems." + }, + { + "left": "issue:44038", + "right": "issue:45478", + "accept": false, + "reason": "They involve different Qwen3 variants and different failures: VL-Moe bug vs from_pretrained error for Qwen3.5 MoE." + }, + { + "left": "issue:43976", + "right": "issue:45446", + "accept": false, + "reason": "Python version compatibility regression and a wrong PyTorch version gate for AuxRequest import are not the same issue." + }, + { + "left": "issue:44355", + "right": "issue:46231", + "accept": false, + "reason": "Compiled Python file errors and a docs parameter-name mismatch are unrelated." + }, + { + "left": "issue:43116", + "right": "issue:43493", + "accept": false, + "reason": "Example script multi-label output bug and SigLIP2 implementation discrepancy are different components and symptoms." + }, + { + "left": "issue:42673", + "right": "issue:45072", + "accept": false, + "reason": "VRAM leak across dataloader threads and bfloat16 dtype mismatches are separate runtime failures." + }, + { + "left": "issue:45357", + "right": "issue:45478", + "accept": false, + "reason": "Both are Qwen3.5 MoE-related, but one is save_pretrained key serialization and the other is from_pretrained loading; not the same concrete code-path problem." + }, + { + "left": "issue:44625", + "right": "issue:46002", + "accept": false, + "reason": "Propagating num_labels into text config and NaNs in classification heads are different defects." + }, + { + "left": "issue:42371", + "right": "issue:43874", + "accept": false, + "reason": "TF32 API guidance/regression and a missing image-processor method are unrelated." + }, + { + "left": "issue:45083", + "right": "issue:45446", + "accept": false, + "reason": "Helper-length computation behavior and a PyTorch version check bug are not the same underlying change." + }, + { + "left": "issue:44610", + "right": "issue:46909", + "accept": false, + "reason": "Processor over-padding and a broken documentation example are different kinds of issues." + }, + { + "left": "issue:43618", + "right": "issue:46829", + "accept": false, + "reason": "Both are output-field problems, but they affect different model outputs and different fields; not mergeable as one bug." + }, + { + "left": "issue:43992", + "right": "issue:44704", + "accept": false, + "reason": "Missing embed_tokens.weight and incomplete kwargs forwarding to cached_file are separate loader issues." + }, + { + "left": "issue:44075", + "right": "issue:44936", + "accept": false, + "reason": "SGD argument handling and trainer.evaluate failing after train are unrelated." + }, + { + "left": "issue:43116", + "right": "issue:43825", + "accept": false, + "reason": "Classification example behavior and pipeline translation error messaging are different subsystems." + }, + { + "left": "issue:28348", + "right": "issue:46640", + "accept": false, + "reason": "Both mention FlashAttention 2, but one is GPT2LMHeadModel support and the other is a generic flash-attention issue; different model/code paths." + }, + { + "left": "issue:43066", + "right": "issue:46710", + "accept": false, + "reason": "Tokenizer decoder-type regression and DeepSeek accuracy regression after upgrade are not the same bug." + }, + { + "left": "issue:44263", + "right": "issue:46762", + "accept": false, + "reason": "A torch.split return-value bug in GlmMoeDsaIndexer is different from Minimax MSA attention numerical differences." + }, + { + "left": "issue:46614", + "right": "issue:46752", + "accept": false, + "reason": "Stale phonemizer backend state and an empty-list crash in apply_chat_template are unrelated." + }, + { + "left": "issue:44995", + "right": "issue:46697", + "accept": false, + "reason": "Second-forward stale cache crash and image over-padding are different code paths." + }, + { + "left": "issue:45092", + "right": "issue:46762", + "accept": false, + "reason": "Old InternVL2 meta-init incompatibility and Minimax attention numerical discrepancies are unrelated." + }, + { + "left": "issue:42222", + "right": "issue:43441", + "accept": false, + "reason": "VitPose model breakage and Ministral-3 FlashAttention failure are different model-specific regressions." + }, + { + "left": "issue:43827", + "right": "issue:45446", + "accept": false, + "reason": "Documentation still referencing pipeline() and a PyTorch version check for flex_attention are not the same issue." + }, + { + "left": "issue:33453", + "right": "issue:39401", + "accept": false, + "reason": "Generic tokenizer loading regression and Qwen3 offset_mapping are both tokenizer-related but describe different failures." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 52, + "review_unit_count": 60, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-500", + "nodes": [ + "issue:38617", + "issue:39401", + "issue:41628", + "issue:42222", + "issue:42617", + "issue:42673", + "issue:42796", + "issue:43064", + "issue:43122", + "issue:43408", + "issue:43450", + "issue:43673", + "issue:43742", + "issue:43937", + "issue:43986", + "issue:44117", + "issue:44291", + "issue:44380", + "issue:44410", + "issue:44466", + "issue:44484", + "issue:44779", + "issue:44829", + "issue:44877", + "issue:44977", + "issue:45071", + "issue:45081", + "issue:45216", + "issue:45310", + "issue:45341", + "issue:45356", + "issue:45362", + "issue:45405", + "issue:45446", + "issue:45561", + "issue:45563", + "issue:45588", + "issue:46002", + "issue:46077", + "issue:46178", + "issue:46231", + "issue:46298", + "issue:46328", + "issue:46512", + "issue:46614", + "issue:46655", + "issue:46752" + ], + "soft_pairs": [ + "issue:43064|issue:43673", + "issue:42673|issue:42796", + "issue:43937|issue:45563", + "issue:44380|issue:46077", + "issue:43673|issue:46328", + "issue:45356|issue:46614", + "issue:42673|issue:46655", + "issue:42222|issue:46752", + "issue:44291|issue:45071", + "issue:45405|issue:45446", + "issue:44117|issue:45588", + "issue:44466|issue:46002", + "issue:44410|issue:45310", + "issue:39401|issue:44779", + "issue:43937|issue:44977", + "issue:45341|issue:46328", + "issue:43450|issue:43986", + "issue:45341|issue:45561", + "issue:44829|issue:45216", + "issue:45071|issue:46512", + "issue:44117|issue:45081", + "issue:43408|issue:43742", + "issue:38617|issue:41628", + "issue:43122|issue:46298", + "issue:42617|issue:45341", + "issue:45362|issue:46328", + "issue:42222|issue:46178", + "issue:44484|issue:45310", + "issue:44877|issue:46231" + ], + "prepared_review_unit_hash": "50670bdb5f784d2654cf074465b15fd3ab392faeb023185430ab437c1a922e2f", + "component_budget": { + "node_count": 500, + "item_count": 500, + "soft_pair_count": 1839, + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 29, + "serialized_chars": 35864, + "estimated_input_tokens": 8966, + "estimated_eval_tokens": 18188 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Loose cluster of mostly unrelated transformers issues that only share broad theme similarity (v5 regressions, tokenization, generation, model-loading, and training). None of the soft pairs look like the same underlying bug/change, so they should stay separate.", + "confidence": 0.82, + "canonical_issue_reason": "If forced to pick one issue as the umbrella representative, issue #43122 is the broadest and most central: it describes a cross-version tokenizer regression with a clear repro, and it is closer to the other tokenizer-focused reports than the model-specific or utility-specific bugs. That said, this is still a noisy mixed cluster, not a true duplicate set.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue #43122 is the best single issue to retain because it is the most general high-signal regression in the set and the most plausible umbrella for nearby tokenizer complaints; the rest are narrower, unrelated, or touch different code paths.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43064", + "right": "issue:43673", + "accept": false, + "reason": "Different failures: optimizer-state mismatch under FSDP2+PEFT+cpu_ram_efficient_loading vs missing GenerationMixin cache during chunked prefill." + }, + { + "left": "issue:42673", + "right": "issue:42796", + "accept": false, + "reason": "One is a VRAM leak in multi-threaded dataloader use; the other is tokenizer BPE-marker handling. Same broad area, different bug." + }, + { + "left": "issue:43937", + "right": "issue:45563", + "accept": false, + "reason": "GenerationConfig validation error vs stale warning text for num_return_sequences in paged generate; not the same defect." + }, + { + "left": "issue:44380", + "right": "issue:46077", + "accept": false, + "reason": "Different attention bugs: GPT2 attention scaling ignored under SDPA/FlashAttention vs Gemma4 building causal masks when bidirectional attention is requested." + }, + { + "left": "issue:43673", + "right": "issue:46328", + "accept": false, + "reason": "Generation cache missing in chunked prefill vs SequenceFeatureExtractor.pad inefficiency; unrelated code paths." + }, + { + "left": "issue:45356", + "right": "issue:46614", + "accept": false, + "reason": "Tokenizer codec/BPE regression vs stale phonemizer backend after language switch; different tokenizer subsystems and symptoms." + }, + { + "left": "issue:42673", + "right": "issue:46655", + "accept": false, + "reason": "VRAM leak in Qwen3ForCausalLM threads vs video processor padding calculation error; not the same bug." + }, + { + "left": "issue:42222", + "right": "issue:46752", + "accept": false, + "reason": "VitPose model breakage vs apply_chat_template crashing on empty list; unrelated failures." + }, + { + "left": "issue:44291", + "right": "issue:45071", + "accept": false, + "reason": "Both are v5 regressions, but one is an unexpected _is_hf_initialized argument during init_empty_weights and the other is PretrainedConfig type checking. Different issues." + }, + { + "left": "issue:45405", + "right": "issue:45446", + "accept": false, + "reason": "PEFT minimum-version bump metadata problem vs incorrect torch version guard for AuxRequest import; unrelated." + }, + { + "left": "issue:44117", + "right": "issue:45588", + "accept": false, + "reason": "TOKENIZER_MAPPING_NAMES None handling vs flash_attention crash on s_aux=None; different layers of the stack." + }, + { + "left": "issue:44466", + "right": "issue:46002", + "accept": false, + "reason": "Serialization of lm_head.weight depending on device vs NaNs in classification heads after checkout/init; different failure modes." + }, + { + "left": "issue:44410", + "right": "issue:45310", + "accept": false, + "reason": "Missing qwen3next projections and Qwen3.5 from_pretrained errors are both model-family issues, but not the same concrete code-path bug." + }, + { + "left": "issue:39401", + "right": "issue:44779", + "accept": false, + "reason": "Both concern tokenizer regressions, but one is wrong offset_mapping and the other is incorrect DeepSeek tokenization results. Similar theme, not the same bug." + }, + { + "left": "issue:43937", + "right": "issue:44977", + "accept": false, + "reason": "GLM-5 invalid GenerationConfig and Qwen3.5 flash-attention generation problems are distinct model-specific issues." + }, + { + "left": "issue:45341", + "right": "issue:46328", + "accept": false, + "reason": "A tiny testing_utils bug vs a padding performance issue; no shared underlying defect." + }, + { + "left": "issue:43450", + "right": "issue:43986", + "accept": false, + "reason": "Video processor shape bug vs confusing crash loading a video model without torchvision; different root causes." + }, + { + "left": "issue:45341", + "right": "issue:45561", + "accept": false, + "reason": "Testing_utils bug vs pytest-xdist race on captured_info.txt; different concurrency and file-handling problems." + }, + { + "left": "issue:44829", + "right": "issue:45216", + "accept": false, + "reason": "Degenerate training with flash_attention_3 vs Qwen3.5 save_pretrained checkpoint regression; different code paths." + }, + { + "left": "issue:45071", + "right": "issue:46512", + "accept": false, + "reason": "PreTrainedConfig type-checking breakage vs stale model.config.dtype after from_config(dtype=...); related only by version timing." + }, + { + "left": "issue:44117", + "right": "issue:45081", + "accept": false, + "reason": "Tokenizer mapping fallback bug vs _patch_mistral_regex backend_tokenizer AttributeError; different tokenizer plumbing bugs." + }, + { + "left": "issue:43408", + "right": "issue:43742", + "accept": false, + "reason": "Model-type warning mismatch vs key error loading facebook/MobileLLM-125M; not the same defect." + }, + { + "left": "issue:38617", + "right": "issue:41628", + "accept": false, + "reason": "Both are import errors, but for different missing exports and different call sites; not duplicates." + }, + { + "left": "issue:43122", + "right": "issue:46298", + "accept": false, + "reason": "Tokenization regression vs noisy logging message; no shared underlying bug." + }, + { + "left": "issue:42617", + "right": "issue:45341", + "accept": false, + "reason": "Script execution failure vs a small testing_utils bug; unrelated." + }, + { + "left": "issue:45362", + "right": "issue:46328", + "accept": false, + "reason": "Qwen3.5 chat crash vs SequenceFeatureExtractor.pad inefficiency; different components." + }, + { + "left": "issue:42222", + "right": "issue:46178", + "accept": false, + "reason": "VitPose model breakage vs backend mapping rejecting tensorflow_text; unrelated." + }, + { + "left": "issue:44484", + "right": "issue:45310", + "accept": false, + "reason": "Question about shard size default vs Qwen3.5 MoE from_pretrained error; not the same bug." + }, + { + "left": "issue:44877", + "right": "issue:46231", + "accept": false, + "reason": "Strict config loading failure vs a documentation parameter-name typo; not duplicates." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 53, + "review_unit_count": 60, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-500", + "nodes": [ + "issue:27957", + "issue:38617", + "issue:39401", + "issue:41628", + "issue:42222", + "issue:42371", + "issue:43012", + "issue:43066", + "issue:43097", + "issue:43122", + "issue:43208", + "issue:43352", + "issue:43441", + "issue:43452", + "issue:43577", + "issue:43673", + "issue:43874", + "issue:43931", + "issue:43937", + "issue:43986", + "issue:44038", + "issue:44075", + "issue:44220", + "issue:44261", + "issue:44263", + "issue:44360", + "issue:44410", + "issue:44483", + "issue:44484", + "issue:44485", + "issue:44493", + "issue:44496", + "issue:44829", + "issue:44933", + "issue:45003", + "issue:45071", + "issue:45278", + "issue:45341", + "issue:45464", + "issue:45478", + "issue:45542", + "issue:45656", + "issue:46172", + "issue:46209", + "issue:46617", + "issue:46762", + "issue:46894" + ], + "soft_pairs": [ + "issue:42371|issue:44261", + "issue:43937|issue:44496", + "issue:39401|issue:46209", + "issue:44410|issue:45478", + "issue:42371|issue:46617", + "issue:44261|issue:46762", + "issue:43208|issue:46894", + "issue:44933|issue:45278", + "issue:38617|issue:44220", + "issue:44075|issue:45656", + "issue:43066|issue:43122", + "issue:44263|issue:44485", + "issue:43097|issue:44493", + "issue:43673|issue:45341", + "issue:43931|issue:44410", + "issue:41628|issue:45071", + "issue:42222|issue:45542", + "issue:45071|issue:45542", + "issue:44483|issue:45464", + "issue:44484|issue:45478", + "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" + ], + "prepared_review_unit_hash": "c81712ba752605d86db3ab09da09478366df739be8919be1f3d5e016bf2128d4", + "component_budget": { + "node_count": 500, + "item_count": 500, + "soft_pair_count": 1839, + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 31, + "serialized_chars": 35867, + "estimated_input_tokens": 8967, + "estimated_eval_tokens": 18190 + }, + "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 regression reports, with a few loose v5/tokenizer/import similarities. The broadest issue is the generic upgrade/import-regression report; only the tokenizer-v5 pair looks like a plausible duplicate.", + "confidence": 0.42, + "canonical_issue_reason": "issue:45278 is the broadest, most umbrella-style report in the set: it describes widespread import failures after upgrading to v5.5.0, which best matches the cluster\u2019s overall regression theme better than the many model-specific bugs.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:45278 is the best single representative issue because it is the most general and most likely to subsume other compatibility/import regressions in this cluster.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:42371", + "right": "issue:44261", + "accept": false, + "reason": "Different problems: TF32 configuration behavior vs an MLA layernorm epsilon precision bug." + }, + { + "left": "issue:43937", + "right": "issue:44496", + "accept": false, + "reason": "One is a GenerationConfig validation error; the other is an unrecognized model/config loading failure." + }, + { + "left": "issue:39401", + "right": "issue:46209", + "accept": false, + "reason": "Tokenizer offset-mapping bug and FP8 inference failure are unrelated symptoms and code paths." + }, + { + "left": "issue:44410", + "right": "issue:45478", + "accept": false, + "reason": "Both mention Qwen3-family models, but one is missing projections in a layer while the other is a from_pretrained loading error." + }, + { + "left": "issue:42371", + "right": "issue:46617", + "accept": false, + "reason": "TF32 settings are unrelated to cache-loading FileNotFoundError behavior." + }, + { + "left": "issue:44261", + "right": "issue:46762", + "accept": false, + "reason": "Layernorm epsilon precision mismatch is not the same underlying issue as Minimax attention numerical differences." + }, + { + "left": "issue:43208", + "right": "issue:46894", + "accept": false, + "reason": "Both are training bugs, but they involve different models and different failure mechanisms." + }, + { + "left": "issue:44933", + "right": "issue:45278", + "accept": false, + "reason": "Both are import-related, but one is a specific missing export and the other is a broad upgrade regression report; not the same bug." + }, + { + "left": "issue:38617", + "right": "issue:44220", + "accept": false, + "reason": "An ImportError for a missing symbol is not the same as the fbank feature extraction issue." + }, + { + "left": "issue:44075", + "right": "issue:45656", + "accept": false, + "reason": "SGD argument handling and double optimizer stepping under deepspeed are different optimizer bugs." + }, + { + "left": "issue:43066", + "right": "issue:43122", + "accept": true, + "reason": "Both point to the same v5 tokenizer regression: changed decoder/tokenization behavior after the 4.57.x -> 5.0 transition." + }, + { + "left": "issue:44263", + "right": "issue:44485", + "accept": false, + "reason": "One is a torch.split/indexing bug in a MoE indexer; the other is a RoPE implementation issue." + }, + { + "left": "issue:43097", + "right": "issue:44493", + "accept": false, + "reason": "Removing tie_embeddings_and_encoder_decoder and unexpected position-id keys are different configuration/model-loading regressions." + }, + { + "left": "issue:43673", + "right": "issue:45341", + "accept": false, + "reason": "Chunked-prefill GenerationMixin cache behavior and a testing_utils bug do not share the same failure mode." + }, + { + "left": "issue:43931", + "right": "issue:44410", + "accept": false, + "reason": "Different Qwen3-related model-loading failures: weight-shape mismatch vs missing attention projections." + }, + { + "left": "issue:41628", + "right": "issue:45071", + "accept": false, + "reason": "Missing AutoImageProcessor import and PreTrainedConfig type-checking breakage are unrelated API regressions." + }, + { + "left": "issue:42222", + "right": "issue:45542", + "accept": false, + "reason": "Vitpose model breakage and tensorboard/TensorFlow backend detection are unrelated." + }, + { + "left": "issue:45071", + "right": "issue:45542", + "accept": false, + "reason": "These are different subsystems: config type checking versus backend initialization." + }, + { + "left": "issue:44483", + "right": "issue:45464", + "accept": false, + "reason": "Both involve chat/completions, but one is request acceptance in v5.3 and the other is streaming inference on a specific model." + }, + { + "left": "issue:44484", + "right": "issue:45478", + "accept": false, + "reason": "Saving shard size defaults and Qwen3.5 from_pretrained loading errors are unrelated." + }, + { + "left": "issue:43874", + "right": "issue:44261", + "accept": false, + "reason": "Missing image-processor method and MLA layernorm epsilon precision are different bugs." + }, + { + "left": "issue:44263", + "right": "issue:46172", + "accept": false, + "reason": "Torch split return handling in GlmMoeDsaIndexer is unrelated to deepseek-v4 transpose behavior." + }, + { + "left": "issue:44360", + "right": "issue:46172", + "accept": false, + "reason": "A missing ReLU in the DSA indexer and a dtype/transpose issue are not the same bug." + }, + { + "left": "issue:27957", + "right": "issue:43352", + "accept": false, + "reason": "Both mention Flash Attention 2, but one is a model support gap and the other is a model-specific incompatibility." + }, + { + "left": "issue:43931", + "right": "issue:44038", + "accept": false, + "reason": "Both are Qwen3-VL-family reports, but the concrete failures differ: shape mismatch versus a separate model bug." + }, + { + "left": "issue:43012", + "right": "issue:43577", + "accept": false, + "reason": "A compile-time precision warning and a load-time dtype mismatch are related only loosely, not the same code-path bug." + }, + { + "left": "issue:43441", + "right": "issue:44829", + "accept": false, + "reason": "FlashAttention 2/3 issues in different models and different symptoms are not the same fix." + }, + { + "left": "issue:43097", + "right": "issue:43452", + "accept": false, + "reason": "A removed config option and a gguf_file from_pretrained regression are separate issues." + }, + { + "left": "issue:45003", + "right": "issue:45341", + "accept": false, + "reason": "Unsafe sys.modules access and a testing_utils bug are unrelated." + }, + { + "left": "issue:43986", + "right": "issue:45341", + "accept": false, + "reason": "Video AutoProcessor crashing without torchvision and a testing_utils bug do not share the same underlying cause." + }, + { + "left": "issue:44261", + "right": "issue:44485", + "accept": false, + "reason": "Layernorm epsilon precision and GLM-5 RoPE implementation are different numerical issues, not one bug." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The evaluation is grounded in the packet and stays conservative. Using issue:45278 as the umbrella representative for the broad import-regression cluster is reasonable, and the only accepted soft pair (43066/43122) is well supported by the v5 tokenizer regression theme. The rejected pairs are mostly defensible as distinct bugs or at least not clearly mergeable into a single fix." + }, + "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": 54, + "review_unit_count": 60, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-500", + "nodes": [ + "issue:38617", + "issue:42757", + "issue:42890", + "issue:43012", + "issue:43064", + "issue:43097", + "issue:43208", + "issue:43502", + "issue:43531", + "issue:43673", + "issue:43873", + "issue:43874", + "issue:43986", + "issue:43992", + "issue:44038", + "issue:44062", + "issue:44117", + "issue:44263", + "issue:44315", + "issue:44351", + "issue:44380", + "issue:44410", + "issue:44484", + "issue:44704", + "issue:44936", + "issue:44987", + "issue:45103", + "issue:45341", + "issue:45372", + "issue:45375", + "issue:45440", + "issue:45588", + "issue:45753", + "issue:46050", + "issue:46123", + "issue:46172", + "issue:46298", + "issue:46365", + "issue:46369", + "issue:46421", + "issue:46655", + "issue:46664", + "issue:46701", + "issue:46710", + "issue:46752", + "issue:46821", + "issue:46897" + ], + "soft_pairs": [ + "issue:45753|issue:46369", + "issue:44117|issue:44987", + "issue:43874|issue:44263", + "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:42890|issue:46298", + "issue:46655|issue:46664", + "issue:46172|issue:46710", + "issue:38617|issue:44351", + "issue:46298|issue:46752", + "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" + ], + "prepared_review_unit_hash": "50e9ffb6950379da37eaf38b4682bcc445ee31f3fe550c56264a748e5ca1982e", + "component_budget": { + "node_count": 500, + "item_count": 500, + "soft_pair_count": 1839, + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 28, + "serialized_chars": 35338, + "estimated_input_tokens": 8835, + "estimated_eval_tokens": 17926 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "All soft pairs look like false positives: they share broad subsystems or similar error wording, but not the same concrete bug or change. No pull requests are present in the set.", + "confidence": 0.81, + "canonical_issue_reason": "Issue 43873 is the strongest anchor: it is still open, has the most discussion/inbound references, and describes a substantive user-impacting bug with a clear reproduction surface.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 43873 is the best overall issue candidate because it is detailed, high-signal, and central enough to attract follow-up reports.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:45753", + "right": "issue:46369", + "accept": false, + "reason": "Different failures and models: infinite looping on a specific image vs memory overflow in sam3-litetext." + }, + { + "left": "issue:44117", + "right": "issue:44987", + "accept": false, + "reason": "One is a generic tokenizer mapping assumption bug; the other is a specific model-loading failure. Similar loading context, but not the same defect." + }, + { + "left": "issue:43874", + "right": "issue:44263", + "accept": false, + "reason": "Multimodal token counting AttributeError and a torch.split return-value issue are separate code paths." + }, + { + "left": "issue:45440", + "right": "issue:46172", + "accept": false, + "reason": "Both are DeepSeek-related, but one is native-vs-remote divergence and the other is a specific dtype/transpose bug." + }, + { + "left": "issue:43531", + "right": "issue:45341", + "accept": false, + "reason": "Qwen3-MoE sliding-window behavior and a testing utility bug are unrelated." + }, + { + "left": "issue:43673", + "right": "issue:46123", + "accept": false, + "reason": "Cache initialization during chunked prefill and a final-batch pipeline flag bug are different behaviors." + }, + { + "left": "issue:43097", + "right": "issue:45588", + "accept": false, + "reason": "A removed config option and a flash-attention crash on sink-less models are distinct issues." + }, + { + "left": "issue:38617", + "right": "issue:42757", + "accept": false, + "reason": "Both are import errors, but they involve different missing symbols from different modules and likely different dependency breaks." + }, + { + "left": "issue:44038", + "right": "issue:44410", + "accept": false, + "reason": "Same broad model family, but one is a load-time bug and the other is missing projections in layer 0." + }, + { + "left": "issue:44410", + "right": "issue:45375", + "accept": false, + "reason": "Missing attention projections and a missing config field are separate defects in different parts of the model stack." + }, + { + "left": "issue:43208", + "right": "issue:46897", + "accept": false, + "reason": "xLSTM training blockers and a Florence2 loss-shift bug do not describe the same underlying problem." + }, + { + "left": "issue:43064", + "right": "issue:44315", + "accept": false, + "reason": "Optimizer state corruption under FSDP2/PEFT is unrelated to Liger Kernel not being applied with model_init." + }, + { + "left": "issue:43986", + "right": "issue:45372", + "accept": false, + "reason": "Different dependency failures: missing torchvision vs missing ReasoningEffort from mistral_common." + }, + { + "left": "issue:43097", + "right": "issue:46050", + "accept": false, + "reason": "One concerns a removed embedding-tie option; the other is save/load corruption for quantized models." + }, + { + "left": "issue:43986", + "right": "issue:44484", + "accept": false, + "reason": "Processor import crash and max_shard_size semantics are unrelated." + }, + { + "left": "issue:42890", + "right": "issue:46298", + "accept": false, + "reason": "A flaky test due to missing set_seed and a noisy logging message are not the same bug." + }, + { + "left": "issue:46655", + "right": "issue:46664", + "accept": false, + "reason": "Both hit SmolVLM video processing, but one is padding calculation and the other is num_frames semantics; related area, different bug." + }, + { + "left": "issue:46172", + "right": "issue:46710", + "accept": false, + "reason": "A specific tensor transformation bug in DeepSeek-v4 is not the same as a broader accuracy regression after upgrade." + }, + { + "left": "issue:38617", + "right": "issue:44351", + "accept": false, + "reason": "Both are import-name failures, but they refer to different symbols and modules, so they are separate breakages." + }, + { + "left": "issue:46298", + "right": "issue:46752", + "accept": false, + "reason": "Logging noise and an empty-list IndexError in apply_chat_template are unrelated." + }, + { + "left": "issue:44062", + "right": "issue:44117", + "accept": false, + "reason": "AddedToken keyword duplication and TOKENIZER_MAPPING_NAMES returning None are different tokenizer issues." + }, + { + "left": "issue:43097", + "right": "issue:43873", + "accept": false, + "reason": "A removed config field and a quantization/offloading malfunction do not share the same code-path problem." + }, + { + "left": "issue:43012", + "right": "issue:46821", + "accept": false, + "reason": "A bfloat16 warning during compilation and a window_size AttributeError are unrelated." + }, + { + "left": "issue:44117", + "right": "issue:46701", + "accept": false, + "reason": "Tokenizer-mapping assumptions and a DiffusionGemma pipeline failure are different bugs." + }, + { + "left": "issue:43502", + "right": "issue:43992", + "accept": false, + "reason": "local_files_only network leakage and a missing embed_tokens weight are separate from-pretrained problems." + }, + { + "left": "issue:44936", + "right": "issue:46365", + "accept": false, + "reason": "trainer.evaluate() after train() and importing Trainer conflicting with datasets are unrelated." + }, + { + "left": "issue:44704", + "right": "issue:45103", + "accept": false, + "reason": "cached_file kwarg forwarding and auto_docstring crashing under future annotations are different subsystems." + }, + { + "left": "issue:44380", + "right": "issue:46421", + "accept": false, + "reason": "Attention scaling being ignored and lazy cache initialization causing recompilation are distinct issues." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 55, + "review_unit_count": 60, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-500", + "nodes": [ + "issue:36683", + "issue:38617", + "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:43618", + "issue:43756", + "issue:43824", + "issue:43825", + "issue:43906", + "issue:43976", + "issue:43986", + "issue:43992", + "issue:44038", + "issue:44077", + "issue:44117", + "issue:44162", + "issue:44222", + "issue:44230", + "issue:44368", + "issue:44380", + "issue:44483", + "issue:44484", + "issue:44485", + "issue:44514", + "issue:44655", + "issue:44938", + "issue:45478", + "issue:45479", + "issue:45584", + "issue:45712", + "issue:45753", + "issue:46132", + "issue:46172", + "issue:46178", + "issue:46298", + "issue:46365" + ], + "soft_pairs": [ + "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", + "issue:44483|issue:44514", + "issue:43618|issue:44380", + "issue:38617|issue:43502", + "issue:44038|issue:44230", + "issue:43756|issue:44485" + ], + "prepared_review_unit_hash": "2d423037086e011d074020447f48145678ad2da55ddae6092b940622724de9f0", + "component_budget": { + "node_count": 500, + "item_count": 500, + "soft_pair_count": 1839, + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 31, + "serialized_chars": 36744, + "estimated_input_tokens": 9186, + "estimated_eval_tokens": 18628 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is heterogeneous: the soft pairs mostly share only broad Transformers surface areas, not the same concrete bug. I rejected every soft edge and picked a single nominal issue representative from the most actionable load-path regression.", + "confidence": 0.42, + "canonical_issue_reason": "Issue 43502 is the most concrete, user-facing regression in the set, with a narrow loading/offline code path and a clear reproduction. The cluster itself is not a true duplicate group, so this is only a nominal representative.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 43502 is the strongest standalone report here: clear symptom, specific API path, and a plausible localized fix surface. It is the best fit if one issue must represent the set.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44938", + "right": "issue:46178", + "accept": false, + "reason": "Both are import/loading-related, but one is a Python 3.14 load failure and the other is an invalid backend mapping error for tensorflow_text; different bugs." + }, + { + "left": "issue:43986", + "right": "issue:44117", + "accept": false, + "reason": "Different failure modes: missing torchvision handling in AutoProcessor versus TOKENIZER_MAPPING_NAMES returning None in from_pretrained." + }, + { + "left": "issue:42907", + "right": "issue:44655", + "accept": false, + "reason": "One is about failing to save dequantized Ministral/Devstral models; the other is pipeline object save_pretrained behavior. Not the same bug." + }, + { + "left": "issue:42907", + "right": "issue:45753", + "accept": false, + "reason": "Save/dequantization failure versus an infinite loop on a specific image in Qwen3_5; unrelated symptoms and code paths." + }, + { + "left": "issue:43825", + "right": "issue:44655", + "accept": false, + "reason": "Pipeline translation-task warning and pipeline save_pretrained are separate issues; same subsystem only at a very broad level." + }, + { + "left": "issue:43012", + "right": "issue:43550", + "accept": false, + "reason": "One is a compile-time precision warning for bfloat16 models; the other is a Bamba SDPA torch.compile failure. Different concrete bugs." + }, + { + "left": "issue:45712", + "right": "issue:46298", + "accept": false, + "reason": "Repo hygiene around dummy classes leaking without torch is unrelated to a noisy logging message." + }, + { + "left": "issue:43208", + "right": "issue:43550", + "accept": false, + "reason": "xLSTM training bugs and Bamba torch.compile/SDPA failure are distinct model-specific problems." + }, + { + "left": "issue:43012", + "right": "issue:44368", + "accept": false, + "reason": "A PyTorch precision warning on compile is not the same as a tie_word_embeddings warning during Qwen3.5 LoRA fine-tuning." + }, + { + "left": "issue:43992", + "right": "issue:44484", + "accept": false, + "reason": "Missing embed_tokens.weight in UMT5Encoder.from_pretrained is unrelated to the default max_shard_size question." + }, + { + "left": "issue:43352", + "right": "issue:43976", + "accept": false, + "reason": "Flash Attention support in NemotronH and Python-version compatibility for Transformers 5.1.0 are unrelated." + }, + { + "left": "issue:43352", + "right": "issue:44380", + "accept": false, + "reason": "Nemotron Flash Attention support and GPT2 attention scaling being ignored are different model/backend bugs." + }, + { + "left": "issue:43066", + "right": "issue:43906", + "accept": false, + "reason": "Wrong tokenizer decoder type in v5 is not the same as an isolated reproduction of another issue." + }, + { + "left": "issue:44162", + "right": "issue:46172", + "accept": false, + "reason": "ESM2 breakage and a deepseek-v4 dtype/transpose bug are different model-specific failures." + }, + { + "left": "issue:43097", + "right": "issue:46298", + "accept": false, + "reason": "Removal of tie_embeddings_and_encoder_decoder and a noisy logging message are unrelated." + }, + { + "left": "issue:44380", + "right": "issue:46132", + "accept": false, + "reason": "GPT2 SDPA/FlashAttention scaling is unrelated to AttentionInterface.register changing the registered function behavior." + }, + { + "left": "issue:45584", + "right": "issue:46365", + "accept": false, + "reason": "Whisper empty-transcription generation failure and an import conflict with datasets are unrelated." + }, + { + "left": "issue:44077", + "right": "issue:46365", + "accept": false, + "reason": "patchtsmixer post_init validation and Trainer import issues caused by datasets are not the same bug." + }, + { + "left": "issue:44222", + "right": "issue:45478", + "accept": false, + "reason": "FP8 save_pretrained problems for moe models are distinct from a Qwen3.5 MoE from_pretrained error." + }, + { + "left": "issue:36683", + "right": "issue:43582", + "accept": false, + "reason": "Gemma3Config missing vocab_size and an Apple Silicon caching_allocator_warmup TypeError are unrelated." + }, + { + "left": "issue:44380", + "right": "issue:45479", + "accept": false, + "reason": "GPT2 attention scaling under SDPA/FlashAttention is a different bug from zero-loss classification with num_labels=1." + }, + { + "left": "issue:42175", + "right": "issue:46365", + "accept": false, + "reason": "TensorFlow not being included in the torch extra and Trainer import failing with datasets are distinct dependency issues." + }, + { + "left": "issue:42757", + "right": "issue:43824", + "accept": false, + "reason": "Missing is_offline_mode import and missing Qwen2_5_VLForConditionalGeneration export are different import regressions." + }, + { + "left": "issue:43064", + "right": "issue:44380", + "accept": false, + "reason": "FSDP2/PEFT optimizer state corruption and GPT2 attention scaling are unrelated training/backend issues." + }, + { + "left": "issue:42490", + "right": "issue:43012", + "accept": false, + "reason": "Trailing-slash local model loading warning is unrelated to the bfloat16 compile precision warning." + }, + { + "left": "issue:43208", + "right": "issue:43526", + "accept": false, + "reason": "xLSTM training bugs and BeitImageProcessorFast reduce_labels returning one label are different code paths." + }, + { + "left": "issue:44483", + "right": "issue:44514", + "accept": false, + "reason": "A v5 chat/completions request acceptance problem is not the same as a batched apply_chat_template crash." + }, + { + "left": "issue:43618", + "right": "issue:44380", + "accept": false, + "reason": "CLIPOutput attentions no longer assigned and GPT2 attention scaling ignored are separate model behavior bugs." + }, + { + "left": "issue:38617", + "right": "issue:43502", + "accept": false, + "reason": "Importing layer_type_validation and making API requests despite local_files_only=True are both load-path symptoms, but they are clearly different defects." + }, + { + "left": "issue:44038", + "right": "issue:44230", + "accept": false, + "reason": "Qwen3-VL-Moe breakage and fp8 support for Qwen3-VL/Qwen3.5 MoE inference are different feature/compatibility issues." + }, + { + "left": "issue:43756", + "right": "issue:44485", + "accept": false, + "reason": "Smollm3 RoPE layer mismatch and GLM-5 RoPE implementation discussion are not the same concrete bug." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 56, + "review_unit_count": 60, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-500", + "nodes": [ + "issue:29870", + "issue:33453", + "issue:36010", + "issue:36296", + "issue:36683", + "issue:42757", + "issue:42890", + "issue:42994", + "issue:43010", + "issue:43064", + "issue:43208", + "issue:43299", + "issue:43352", + "issue:43475", + "issue:43525", + "issue:43611", + "issue:43618", + "issue:43761", + "issue:43867", + "issue:43901", + "issue:43937", + "issue:43976", + "issue:43992", + "issue:44075", + "issue:44077", + "issue:44355", + "issue:44483", + "issue:44492", + "issue:44625", + "issue:44655", + "issue:44855", + "issue:44908", + "issue:44938", + "issue:45071", + "issue:45216", + "issue:45276", + "issue:45278", + "issue:45310", + "issue:45375", + "issue:45561", + "issue:46077", + "issue:46082", + "issue:46202", + "issue:46365", + "issue:46369", + "issue:46829", + "issue:46851", + "issue:46877" + ], + "soft_pairs": [ + "issue:42994|issue:44655", + "issue:43618|issue:43761", + "issue:43299|issue:43867", + "issue:36010|issue:42757", + "issue:44625|issue:46365", + "issue:45276|issue:46077", + "issue:44355|issue:46877", + "issue:43352|issue:43937", + "issue:44483|issue:46369", + "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", + "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" + ], + "prepared_review_unit_hash": "3f6f3809204bc985243ec8b22433072cc9bcf6afed97c77f091751b0ee3b7e5e", + "component_budget": { + "node_count": 500, + "item_count": 500, + "soft_pair_count": 1839, + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 34, + "serialized_chars": 37584, + "estimated_input_tokens": 9396, + "estimated_eval_tokens": 19048 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is mostly a set of unrelated regression/compatibility issues across different models, loaders, and utilities. None of the soft-edge pairs look like true duplicates under a strict same-bug standard, though a few are superficially adjacent (especially CLIP and Gemma/Qwen model-family regressions).", + "confidence": 0.83, + "canonical_issue_reason": "issue:43618 is the clearest, most concrete regression report in the set: it names a specific model/output field and points to a focused code path, making it a strong canonical issue anchor.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:43618 is the best standalone issue candidate because it is narrowly scoped, actionable, and has a clear symptom-to-code-path mapping compared with the broader or more ambiguous reports.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:42994", + "right": "issue:44655", + "accept": false, + "reason": "Both mention saving, but one is quantized model saving and the other is pipeline save_pretrained; different objects and likely different fixes." + }, + { + "left": "issue:43618", + "right": "issue:43761", + "accept": false, + "reason": "Same CLIP family, but one is missing attentions and the other is missing hidden_states; related regression theme, not clearly the same bug." + }, + { + "left": "issue:43299", + "right": "issue:43867", + "accept": false, + "reason": "Both concern model loading regressions, but they affect different Qwen/serialization scenarios and do not point to the same code path." + }, + { + "left": "issue:36010", + "right": "issue:42757", + "accept": false, + "reason": "Different import failures from different dependencies; no shared underlying defect is evident." + }, + { + "left": "issue:44625", + "right": "issue:46365", + "accept": false, + "reason": "A Qwen config propagation bug versus a Trainer import problem; unrelated subsystems and failure modes." + }, + { + "left": "issue:45276", + "right": "issue:46077", + "accept": false, + "reason": "Both are Gemma4 issues, but resizing embeddings and bidirectional attention masks are separate behaviors and code paths." + }, + { + "left": "issue:44355", + "right": "issue:46877", + "accept": false, + "reason": "Compiled-file import errors versus import-time docstring lint output; unrelated." + }, + { + "left": "issue:43352", + "right": "issue:43937", + "accept": false, + "reason": "Different model-specific generation/attention failures; not the same concrete bug." + }, + { + "left": "issue:44483", + "right": "issue:46369", + "accept": false, + "reason": "API request validation versus memory overflow in a model forward path; clearly different problems." + }, + { + "left": "issue:36683", + "right": "issue:43475", + "accept": false, + "reason": "Different missing-attribute errors on different model families (Gemma3 vs SAM 3 Video)." + }, + { + "left": "issue:43611", + "right": "issue:44655", + "accept": false, + "reason": "Base_model_prefix loading regression and pipeline save_pretrained failure are separate serialization issues." + }, + { + "left": "issue:43901", + "right": "issue:44655", + "accept": false, + "reason": "Docs mismatch for TextClassificationPipeline versus pipeline persistence; not the same bug." + }, + { + "left": "issue:43867", + "right": "issue:45216", + "accept": false, + "reason": "One is loading failure with sorted state_dict, the other is incorrect checkpoint saving; related area but different directions and likely different fixes." + }, + { + "left": "issue:43976", + "right": "issue:44938", + "accept": false, + "reason": "Both are Python-version compatibility issues, but for different interpreter versions and different failure modes." + }, + { + "left": "issue:36683", + "right": "issue:44077", + "accept": false, + "reason": "Different model/config attribute bug versus optimizer argument handling; unrelated." + }, + { + "left": "issue:44625", + "right": "issue:46077", + "accept": false, + "reason": "Qwen config propagation and Gemma4 attention-mask construction are unrelated." + }, + { + "left": "issue:46077", + "right": "issue:46082", + "accept": false, + "reason": "Gemma4 attention masking versus LlamaConfig head_dim validation; different code paths." + }, + { + "left": "issue:43618", + "right": "issue:45216", + "accept": false, + "reason": "CLIP output regression versus Qwen3.5 save_pretrained regression; different models and bug classes." + }, + { + "left": "issue:43208", + "right": "issue:46829", + "accept": false, + "reason": "xLSTM training blockers versus VideoPrism output formatting; no shared bug." + }, + { + "left": "issue:36683", + "right": "issue:43525", + "accept": false, + "reason": "Missing config fields in different model families; not the same defect." + }, + { + "left": "issue:36683", + "right": "issue:44625", + "accept": false, + "reason": "Both are config-related, but they affect different models and different missing fields." + }, + { + "left": "issue:29870", + "right": "issue:36296", + "accept": false, + "reason": "Bart generate-time UnboundLocalError versus tensor-parallel training bug; unrelated." + }, + { + "left": "issue:43992", + "right": "issue:45310", + "accept": false, + "reason": "Both involve Qwen3.5 from_pretrained, but one is a missing encoder weight and the other is a broader model loading error; not clearly the same issue." + }, + { + "left": "issue:43064", + "right": "issue:44492", + "accept": false, + "reason": "Optimizer-state corruption in distributed training versus a typo in cache strategies; unrelated." + }, + { + "left": "issue:43010", + "right": "issue:45278", + "accept": false, + "reason": "no_grad decoration for cache updates versus broad import errors after upgrade; different fixes and code paths." + }, + { + "left": "issue:44075", + "right": "issue:46365", + "accept": false, + "reason": "Optimizer argument handling versus Trainer import failure; unrelated." + }, + { + "left": "issue:44908", + "right": "issue:46851", + "accept": false, + "reason": "Scheduler kwargs bug versus variable-length position_ids handling; different components." + }, + { + "left": "issue:29870", + "right": "issue:33453", + "accept": false, + "reason": "Bart generation mask bug versus tokenizer loading regression; different subsystems." + }, + { + "left": "issue:44938", + "right": "issue:45071", + "accept": false, + "reason": "Python 3.14 load failure versus PretrainedConfig type-checking regression; different issues." + }, + { + "left": "issue:44355", + "right": "issue:44855", + "accept": false, + "reason": "Compiled Python import errors versus a Python 3.13 @torch.jit.script parsing problem; unrelated." + }, + { + "left": "issue:43867", + "right": "issue:46202", + "accept": false, + "reason": "State_dict loading failure versus Gemma pipeline roundtrip on a specific token; different failure modes." + }, + { + "left": "issue:36683", + "right": "issue:43064", + "accept": false, + "reason": "A config attribute bug versus distributed optimizer state corruption; unrelated." + }, + { + "left": "issue:43867", + "right": "issue:45375", + "accept": false, + "reason": "Different Qwen/Gemma serialization issues with different missing fields and paths." + }, + { + "left": "issue:42890", + "right": "issue:45561", + "accept": false, + "reason": "Missing set_seed in tests versus xdist race on a captured info file; test stability issues, but not the same bug." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 57, + "review_unit_count": 60, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-500", + "nodes": [ + "issue:28282", + "issue:30333", + "issue:33453", + "issue:36296", + "issue:41669", + "issue:42548", + "issue:42890", + "issue:42898", + "issue:43010", + "issue:43065", + "issue:43232", + "issue:43502", + "issue:43575", + "issue:43576", + "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:44351", + "issue:44355", + "issue:44360", + "issue:44387", + "issue:44485", + "issue:44556", + "issue:44734", + "issue:44749", + "issue:44961", + "issue:44964", + "issue:45125", + "issue:45357", + "issue:45375", + "issue:45440", + "issue:45478", + "issue:45600", + "issue:45850", + "issue:46132", + "issue:46285", + "issue:46617", + "issue:46664", + "issue:46726" + ], + "soft_pairs": [ + "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", + "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:43010|issue:43232", + "issue:43502|issue:44961", + "issue:43502|issue:44222" + ], + "prepared_review_unit_hash": "c4565b6fd063ed56f5fbfd45c72718c85d2e246c36114b58ac4eb80e260733ef", + "component_budget": { + "node_count": 500, + "item_count": 500, + "soft_pair_count": 1839, + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 33, + "serialized_chars": 37181, + "estimated_input_tokens": 9296, + "estimated_eval_tokens": 18848 + }, + "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: import errors, tokenizer/model loading regressions, tensor-parallel problems, and model-specific bugs. The only clear soft duplicate is the import-time slowdown pair around `import transformers` / `import *` in models.", + "confidence": 0.83, + "canonical_issue_reason": "Issue 41669 is the clearest, most actionable root-cause report: it identifies a specific import-time regression from `import *` in models and matches the only near-duplicate slow-import report.", + "canonical_pr_reason": null, + "best_issue_reason": "Best overall representative because it is concrete, scoped to one code path, and describes a fixable performance regression rather than a vague symptom or unrelated model-specific bug.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43010", + "right": "issue:45850", + "accept": false, + "reason": "Different problems: `@torch.no_grad` on cache/layer updates vs a broad complaint that v5 breaks many models." + }, + { + "left": "issue:41669", + "right": "issue:44246", + "accept": true, + "reason": "Both describe the same underlying issue: slow `transformers` import. 41669 gives the specific root cause (`import *` in models), while 44246 is the generic symptom report." + }, + { + "left": "issue:36296", + "right": "issue:43575", + "accept": false, + "reason": "Both mention tensor parallelism, but one is a training bug and the other is an OOM while loading a specific model; too different to treat as the same bug." + }, + { + "left": "issue:43010", + "right": "issue:43976", + "accept": false, + "reason": "Unrelated: generation/cache mutation behavior vs Python version compatibility." + }, + { + "left": "issue:43618", + "right": "issue:44077", + "accept": false, + "reason": "Different model issues with different symptoms and code paths." + }, + { + "left": "issue:43992", + "right": "issue:45478", + "accept": false, + "reason": "Both are `from_pretrained` failures, but they involve different models and different failure modes." + }, + { + "left": "issue:43867", + "right": "issue:45357", + "accept": false, + "reason": "Different regression classes: state-dict loading order issue vs incorrect visual encoder keys during save." + }, + { + "left": "issue:28282", + "right": "issue:33453", + "accept": false, + "reason": "Completely different issues: missing PyTorch dependency vs tokenizer loading regression." + }, + { + "left": "issue:43618", + "right": "issue:46285", + "accept": false, + "reason": "`CLIPOutput.attentions` assignment bug is unrelated to the CLIP text model structural inconsistency." + }, + { + "left": "issue:44360", + "right": "issue:45440", + "accept": false, + "reason": "Different architectures and failures: DSA indexer activation issue vs DeepSeekV3 implementation divergence." + }, + { + "left": "issue:42890", + "right": "issue:44964", + "accept": false, + "reason": "Test flakiness due to missing `set_seed()` is not the same as a model load failure for Phi-4 multimodal." + }, + { + "left": "issue:42890", + "right": "issue:44112", + "accept": false, + "reason": "Both involve tests/CI, but the failures are different and not the same underlying bug." + }, + { + "left": "issue:42890", + "right": "issue:45600", + "accept": false, + "reason": "Test seed instability and broken auto-mappings reference are unrelated." + }, + { + "left": "issue:42898", + "right": "issue:46664", + "accept": false, + "reason": "Tokenizer cleanup behavior change and video processor `num_frames` semantics are unrelated." + }, + { + "left": "issue:43502", + "right": "issue:44062", + "accept": false, + "reason": "`local_files_only=True` still making requests is unrelated to `AddedToken(... special=...)` argument duplication." + }, + { + "left": "issue:44734", + "right": "issue:46617", + "accept": false, + "reason": "KV-cache continuation crash in serve is unrelated to cache-loading `FileNotFoundError` for trust-remote-code models." + }, + { + "left": "issue:43723", + "right": "issue:44556", + "accept": false, + "reason": "Related only at a very high level; one is tokenizer loading and the other is checkpoint reload after version upgrade, not clearly the same bug." + }, + { + "left": "issue:44222", + "right": "issue:45125", + "accept": false, + "reason": "FP8 save-pretrained behavior and missing `_tp_plan` for tensor parallelism are different issues." + }, + { + "left": "issue:44387", + "right": "issue:44556", + "accept": false, + "reason": "Int4 memory regression/OOM and checkpoint reload compatibility are different failure modes." + }, + { + "left": "issue:43824", + "right": "issue:46617", + "accept": false, + "reason": "Missing exported model class vs cache `FileNotFoundError` are unrelated." + }, + { + "left": "issue:43502", + "right": "issue:44485", + "accept": false, + "reason": "`local_files_only` network calls and GLM-5 RoPE implementation are unrelated." + }, + { + "left": "issue:43874", + "right": "issue:46617", + "accept": false, + "reason": "Missing image-processor method causing multimodal token counting failure is unrelated to cache file lookup failure." + }, + { + "left": "issue:43576", + "right": "issue:46132", + "accept": false, + "reason": "Broken CLI `transformers env` and `AttentionInterface.register` behavior change are different issues." + }, + { + "left": "issue:45125", + "right": "issue:45478", + "accept": false, + "reason": "Both concern Qwen3.5 MoE, but one is tensor-parallel metadata and the other is a broader `from_pretrained` error; not enough to merge as one bug." + }, + { + "left": "issue:30333", + "right": "issue:33453", + "accept": false, + "reason": "MLflow job-reporting failure and tokenizer loading regression are unrelated." + }, + { + "left": "issue:44355", + "right": "issue:44749", + "accept": false, + "reason": "Compiled Python-file errors and a performance regression after version upgrade are unrelated." + }, + { + "left": "issue:42548", + "right": "issue:44351", + "accept": false, + "reason": "Different import errors for different symbols; same package, but not the same underlying bug." + }, + { + "left": "issue:45125", + "right": "issue:45375", + "accept": false, + "reason": "Missing tensor-parallel plan and missing vision config field are different Qwen3.5-MoE issues." + }, + { + "left": "issue:46664", + "right": "issue:46726", + "accept": false, + "reason": "Video processor `num_frames` semantics and AutoProcessor backend-missing error are unrelated." + }, + { + "left": "issue:43010", + "right": "issue:43232", + "accept": false, + "reason": "Cache/layer `no_grad` concern is unrelated to generation kwargs update after `sync_gpus`." + }, + { + "left": "issue:43502", + "right": "issue:44961", + "accept": false, + "reason": "`local_files_only` network requests and the unrelated `racoon` issue are not duplicates." + }, + { + "left": "issue:43502", + "right": "issue:44222", + "accept": false, + "reason": "Network requests despite `local_files_only` is unrelated to FP8 MoE save/pretrained behavior." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet, and the one accepted soft duplicate (issue:41669 vs issue:44246) is conservative and well supported by the titles. The remaining soft-edge rejections are also appropriately cautious; no overstatement or mergeability issue is apparent." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 58, + "review_unit_count": 60, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-500", + "nodes": [ + "issue:29870", + "issue:30333", + "issue:30990", + "issue:31795", + "issue:33453", + "issue:34689", + "issue:36296", + "issue:38617", + "issue:41628", + "issue:42548", + "issue:42757", + "issue:43299", + "issue:43475", + "issue:43576", + "issue:43673", + "issue:43824", + "issue:43827", + "issue:43901", + "issue:44038", + "issue:44222", + "issue:44230", + "issue:44273", + "issue:44315", + "issue:44351", + "issue:44355", + "issue:44360", + "issue:44393", + "issue:44485", + "issue:44492", + "issue:44556", + "issue:44623", + "issue:44908", + "issue:44938", + "issue:44961", + "issue:45003", + "issue:45071", + "issue:45125", + "issue:45230", + "issue:45362", + "issue:45468", + "issue:45542", + "issue:45753", + "issue:45850", + "issue:46143", + "issue:46144", + "issue:46209", + "issue:46617", + "issue:46664" + ], + "soft_pairs": [ + "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:43576|issue:43824", + "issue:44961|issue:45468", + "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", + "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" + ], + "prepared_review_unit_hash": "61fdb544773a2c6576db7c378a82481f9c0003e5c4bb77143bc1c4dd4ac1c301", + "component_budget": { + "node_count": 500, + "item_count": 500, + "soft_pair_count": 1839, + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 34, + "serialized_chars": 36924, + "estimated_input_tokens": 9231, + "estimated_eval_tokens": 18718 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The items are mostly a heterogeneous mix of unrelated regressions, import errors, docs issues, and model-specific bugs. None of the soft-similarity pairs look like the same underlying defect/change, so all candidate duplicate merges should be rejected.", + "confidence": 0.94, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": "issue:43576 is the clearest concrete regression report in the set and the most plausible representative issue, but the cluster is too heterogeneous to treat it as a true canonical duplicate target.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:30990", + "right": "issue:31795", + "accept": false, + "reason": "Different problems: a loading hang vs documentation confusion in `model.forward`." + }, + { + "left": "issue:43827", + "right": "issue:44961", + "accept": false, + "reason": "One is a docs cleanup after pipeline removals; the other is an unrelated issue with no matching bug signal." + }, + { + "left": "issue:30333", + "right": "issue:36296", + "accept": false, + "reason": "MLFlow reporting on training failure is unrelated to tensor-parallel training behavior." + }, + { + "left": "issue:43299", + "right": "issue:44556", + "accept": false, + "reason": "Both involve v5 regressions, but one is Qwen3VL MoE loading and the other is checkpoint reload compatibility; not the same bug." + }, + { + "left": "issue:29870", + "right": "issue:31795", + "accept": false, + "reason": "A BART generation `UnboundLocalError` is not the same as a documentation inconsistency." + }, + { + "left": "issue:44355", + "right": "issue:45071", + "accept": false, + "reason": "Compiled-file execution errors are different from `PretrainedConfig` type-checking breakage." + }, + { + "left": "issue:33453", + "right": "issue:34689", + "accept": false, + "reason": "Both are loading regressions, but they affect different model-loading paths and likely different causes." + }, + { + "left": "issue:42757", + "right": "issue:43576", + "accept": false, + "reason": "Importing `is_offline_mode` from `huggingface_hub` is a separate import regression from the broken `transformers env` command." + }, + { + "left": "issue:43576", + "right": "issue:45071", + "accept": false, + "reason": "A CLI command regression and a config type-checking regression are different failure modes." + }, + { + "left": "issue:44315", + "right": "issue:44360", + "accept": false, + "reason": "`model_init` losing Liger Kernel application and a DSA/ReLU discussion are unrelated." + }, + { + "left": "issue:31795", + "right": "issue:38617", + "accept": false, + "reason": "Docs about `input_ids`/`past_key_values` do not match an `ImportError` for `layer_type_validation`." + }, + { + "left": "issue:43576", + "right": "issue:43824", + "accept": false, + "reason": "Broken environment CLI and missing `Qwen2_5_VLForConditionalGeneration` export are distinct issues." + }, + { + "left": "issue:44961", + "right": "issue:45468", + "accept": false, + "reason": "The placeholder 'racoon' issue is not the same as the Gemma-4 audio positional encoding bug." + }, + { + "left": "issue:44485", + "right": "issue:44961", + "accept": false, + "reason": "GLM-5 RoPE implementation discussion is unrelated to the placeholder issue." + }, + { + "left": "issue:44351", + "right": "issue:46617", + "accept": false, + "reason": "A missing `HybridCache` import is different from a cache-file `FileNotFoundError` when loading from cache." + }, + { + "left": "issue:44908", + "right": "issue:44961", + "accept": false, + "reason": "Scheduler kwargs being ignored has nothing to do with the placeholder issue." + }, + { + "left": "issue:43576", + "right": "issue:44938", + "accept": false, + "reason": "Both mention v5 breakage, but one is the `transformers env` command and the other is Python 3.14 load failure." + }, + { + "left": "issue:36296", + "right": "issue:43901", + "accept": false, + "reason": "Tensor-parallel training bugs are unrelated to `return_all_scores` documentation mismatch." + }, + { + "left": "issue:46143", + "right": "issue:46144", + "accept": false, + "reason": "Missing `**kwargs` forwarding and RoFormer attention-interface usage are related broadly, but not the same concrete bug." + }, + { + "left": "issue:43475", + "right": "issue:46617", + "accept": false, + "reason": "A missing SAM3 vision output attribute is unrelated to trust-remote-code cache loading failures." + }, + { + "left": "issue:44222", + "right": "issue:44230", + "accept": false, + "reason": "One is about FP8 `save_pretrained`, the other about FP8 inference support; different code paths." + }, + { + "left": "issue:41628", + "right": "issue:42548", + "accept": false, + "reason": "`AutoImageProcessor` and `PreTrainedModel` import errors are separate symbol/export problems." + }, + { + "left": "issue:44315", + "right": "issue:46209", + "accept": false, + "reason": "Liger Kernel application with `model_init` is unrelated to FP8 inference on B200." + }, + { + "left": "issue:44485", + "right": "issue:46664", + "accept": false, + "reason": "GLM-5 RoPE implementation and SmolVLM `num_frames` semantics are different model-specific issues." + }, + { + "left": "issue:43673", + "right": "issue:44492", + "accept": false, + "reason": "A functional cache-missing bug is not the same as a typo in cache strategy text." + }, + { + "left": "issue:45362", + "right": "issue:45850", + "accept": false, + "reason": "A Qwen3.5-35B chat crash is much more specific than a broad complaint that v5 breaks many models." + }, + { + "left": "issue:44273", + "right": "issue:45542", + "accept": false, + "reason": "Lazy loading problems and TensorBoard/TensorFlow backend detection are different defects." + }, + { + "left": "issue:44623", + "right": "issue:45230", + "accept": false, + "reason": "Missing files from `processor.save_pretrained()` is not the same as a generic bug report." + }, + { + "left": "issue:44360", + "right": "issue:46664", + "accept": false, + "reason": "A DSA/ReLU discussion is unrelated to SmolVLM video-frame semantics." + }, + { + "left": "issue:44038", + "right": "issue:45125", + "accept": false, + "reason": "A Qwen3-VL loading bug and missing tensor-parallel plan for Qwen3_5Moe are separate model-family issues." + }, + { + "left": "issue:44273", + "right": "issue:45003", + "accept": false, + "reason": "Lazy loading not working and unsafe `sys.modules` access are different loading-path bugs." + }, + { + "left": "issue:44393", + "right": "issue:45071", + "accept": false, + "reason": "Qwen3-VL 2D bbox output errors are unrelated to `PretrainedConfig` type-checking failures." + }, + { + "left": "issue:44393", + "right": "issue:45753", + "accept": false, + "reason": "Both touch Qwen3-VL/Qwen3.5, but the symptoms differ: bbox output errors vs an infinite loop on a specific image." + }, + { + "left": "issue:43576", + "right": "issue:44556", + "accept": false, + "reason": "The v5 `transformers env` CLI break and checkpoint reload incompatibility 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": 59, + "review_unit_count": 60, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-500", + "nodes": [ + "issue:31795", + "issue:33453", + "issue:34689", + "issue:41669", + "issue:42371", + "issue:42890", + "issue:43352", + "issue:43408", + "issue:43519", + "issue:43525", + "issue:43572", + "issue:43576", + "issue:43673", + "issue:43704", + "issue:43976", + "issue:44077", + "issue:44230", + "issue:44246", + "issue:44273", + "issue:44295", + "issue:44297", + "issue:44315", + "issue:44351", + "issue:44360", + "issue:44393", + "issue:44485", + "issue:44492", + "issue:44556", + "issue:44655", + "issue:44704", + "issue:44829", + "issue:44861", + "issue:44908", + "issue:45071", + "issue:45092", + "issue:45230", + "issue:45278", + "issue:45310", + "issue:45341", + "issue:45468", + "issue:45478", + "issue:45850", + "issue:46095", + "issue:46218", + "issue:46328", + "issue:46655", + "issue:46894" + ], + "soft_pairs": [ + "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", + "issue:43519|issue:46655", + "issue:44273|issue:44908", + "issue:44655|issue:44861", + "issue:45230|issue:45310", + "issue:43576|issue:46218", + "issue:43519|issue:44485", + "issue:45468|issue:46894", + "issue:43519|issue:44908", + "issue:41669|issue:43704", + "issue:31795|issue:34689" + ], + "prepared_review_unit_hash": "b256177e7a3e7fb4bafee74583a503cc8fb6994ec6c762550612f040eeb298a5", + "component_budget": { + "node_count": 500, + "item_count": 500, + "soft_pair_count": 1839, + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 40, + "serialized_chars": 38535, + "estimated_input_tokens": 9634, + "estimated_eval_tokens": 19524 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a loose collection of Transformers v5 regression reports around model loading, import behavior, and config/API compatibility. The soft pairs are semantically related in places, but none are strong enough to treat as the same underlying bug/change.", + "confidence": 0.78, + "canonical_issue_reason": "45850 is the broadest umbrella in the set: it summarizes the general v5 breakage theme affecting many models and code paths.", + "canonical_pr_reason": null, + "best_issue_reason": "45278 is the clearest representative of the recurring concrete pattern here: upgrade-related import/loading regressions across multiple models.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:41669", + "right": "issue:43673", + "accept": false, + "reason": "Both relate to import/loading behavior, but one is about `import *` slowing imports and the other is a missing cache object during chunked prefill; different bugs." + }, + { + "left": "issue:43525", + "right": "issue:45850", + "accept": false, + "reason": "`pad_token_id` missing in one config class is a specific attribute regression; the other is a broad complaint about many models breaking after v5." + }, + { + "left": "issue:44556", + "right": "issue:45092", + "accept": false, + "reason": "Both are v5 loading incompatibilities, but one is checkpoint reload after upgrade and the other is remote-code/meta-init incompatibility for a specific model family." + }, + { + "left": "issue:43572", + "right": "issue:45850", + "accept": false, + "reason": "A missing `pad_token_idx` in one config is not the same underlying issue as a general 'many models break' regression report." + }, + { + "left": "issue:44492", + "right": "issue:45850", + "accept": false, + "reason": "A typo in cache-strategy docs/labels is unrelated to the broad v5 model-breakage report." + }, + { + "left": "issue:45278", + "right": "issue:45850", + "accept": false, + "reason": "Related theme, but one is a concrete import-error regression after upgrade while the other is a generic umbrella complaint; not the same bug." + }, + { + "left": "issue:44230", + "right": "issue:46095", + "accept": false, + "reason": "Different requests entirely: FP8 inference support for Qwen models versus a DeepSeekV4 weight-conversion question." + }, + { + "left": "issue:42371", + "right": "issue:44492", + "accept": false, + "reason": "TF32 API guidance and a cache-strategy typo are unrelated issues." + }, + { + "left": "issue:41669", + "right": "issue:46218", + "accept": false, + "reason": "`import *` performance complaints are not the same as a missing `Qwen3_5ForCausalLM` import/export error." + }, + { + "left": "issue:43408", + "right": "issue:44315", + "accept": false, + "reason": "Different model/config problems: a wrong model-type warning versus Liger Kernel not being applied with `model_init`." + }, + { + "left": "issue:42890", + "right": "issue:44908", + "accept": false, + "reason": "One is missing `set_seed()` in tests; the other is a scheduler ignoring kwargs. No shared underlying bug." + }, + { + "left": "issue:44246", + "right": "issue:44273", + "accept": false, + "reason": "Both mention loading/lazy-loading, but one is intermittent import slowness and the other is a general lazy-loading failure; too broad to merge." + }, + { + "left": "issue:43576", + "right": "issue:44908", + "accept": false, + "reason": "`transformers env` is broken in v5, while the other is a scheduler keyword-argument bug; different code paths." + }, + { + "left": "issue:43352", + "right": "issue:44393", + "accept": false, + "reason": "Flash Attention 2 support gating for Nemotron is unrelated to a Qwen3-VL bounding-box output bug." + }, + { + "left": "issue:43976", + "right": "issue:44393", + "accept": false, + "reason": "A Python version compatibility issue is not the same as a Qwen3-VL hallucination/output problem." + }, + { + "left": "issue:44908", + "right": "issue:45071", + "accept": false, + "reason": "A scheduler kwargs bug and a `PretrainedConfig` type-checking regression are separate problems." + }, + { + "left": "issue:44295", + "right": "issue:45230", + "accept": false, + "reason": "Reading `position_ids` after buffer registration is a different failure mode than an unspecified bug report." + }, + { + "left": "issue:44297", + "right": "issue:45230", + "accept": false, + "reason": "Tokenizer save metadata mismatch is a specific serialization bug, not the same as the generic bug report." + }, + { + "left": "issue:44704", + "right": "issue:44908", + "accept": false, + "reason": "`AutoProcessor.from_pretrained` not forwarding kwargs to `cached_file` is unrelated to an inverse-sqrt scheduler parameter bug." + }, + { + "left": "issue:44492", + "right": "issue:46328", + "accept": false, + "reason": "A typo in cache strategies is not the same as `SequenceFeatureExtractor.pad` converting arrays inefficiently." + }, + { + "left": "issue:41669", + "right": "issue:44351", + "accept": false, + "reason": "Import-star cleanup and a `HybridCache` import failure are distinct import-related problems." + }, + { + "left": "issue:41669", + "right": "issue:45341", + "accept": false, + "reason": "A broad import-performance issue and a small testing-utils bug are not the same underlying change." + }, + { + "left": "issue:43976", + "right": "issue:45850", + "accept": false, + "reason": "Python-version incompatibility is more specific than the broad 'many models break' v5 complaint, and they do not point to the same defect." + }, + { + "left": "issue:44393", + "right": "issue:45478", + "accept": false, + "reason": "Both are Qwen3.5/Qwen3-VL-related, but the symptoms are different model errors with no evidence of a shared root cause." + }, + { + "left": "issue:43976", + "right": "issue:46218", + "accept": false, + "reason": "A Python support regression and a missing module import are unrelated." + }, + { + "left": "issue:44077", + "right": "issue:44861", + "accept": false, + "reason": "One is an API/validation rule change for `post_init`; the other is an AttributeError in tied-weight key handling." + }, + { + "left": "issue:31795", + "right": "issue:33453", + "accept": false, + "reason": "Documentation confusion about forward args is unrelated to a tokenizer-loading regression." + }, + { + "left": "issue:43576", + "right": "issue:44273", + "accept": false, + "reason": "A broken CLI environment command and a lazy-loading failure are too different to be duplicates." + }, + { + "left": "issue:44315", + "right": "issue:44829", + "accept": false, + "reason": "Liger Kernel integration not applied with `model_init` is a separate issue from degenerate training under flash_attention_3." + }, + { + "left": "issue:44360", + "right": "issue:45468", + "accept": false, + "reason": "A missing ReLU in the DSA indexer and Gemma-4 positional encoding are unrelated model-implementation issues." + }, + { + "left": "issue:43519", + "right": "issue:46655", + "accept": false, + "reason": "Both are processor math bugs, but one is timestamp calculation and the other is video padding calculation; different code paths." + }, + { + "left": "issue:44273", + "right": "issue:44908", + "accept": false, + "reason": "Lazy loading failures and scheduler kwargs handling are unrelated." + }, + { + "left": "issue:44655", + "right": "issue:44861", + "accept": false, + "reason": "Saving pipeline objects and tied-weight key crashes are separate serialization/internal-state bugs." + }, + { + "left": "issue:45230", + "right": "issue:45310", + "accept": false, + "reason": "An empty bug report does not identify the same issue as the Qwen3.5 MoE from_pretrained error." + }, + { + "left": "issue:43576", + "right": "issue:46218", + "accept": false, + "reason": "A broken environment CLI command and a missing Qwen3_5 model import are distinct failures." + }, + { + "left": "issue:43519", + "right": "issue:44485", + "accept": false, + "reason": "Qwen3-VL timestamp math and GLM-5 RoPE implementation are both model-specific, but not the same bug." + }, + { + "left": "issue:45468", + "right": "issue:46894", + "accept": false, + "reason": "Gemma-4 audio positional encoding and CohereASR training-loss shifting are different model/training bugs." + }, + { + "left": "issue:41669", + "right": "issue:43704", + "accept": false, + "reason": "Import-star cleanup and a VRAM leak in multi-dataloader-thread use are unrelated." + }, + { + "left": "issue:31795", + "right": "issue:34689", + "accept": false, + "reason": "Forward-API documentation confusion is not the same as a model-loading failure for Llama 3.2 Vision." + } + ] + }, + "evaluator_result": 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": 60, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-500", + "nodes": [ + "issue:28348", + "issue:29870", + "issue:30333", + "issue:31515", + "issue:31795", + "issue:36296", + "issue:37934", + "issue:38617", + "issue:42371", + "issue:42913", + "issue:42936", + "issue:43066", + "issue:43240", + "issue:43298", + "issue:43352", + "issue:43502", + "issue:43519", + "issue:43525", + "issue:43572", + "issue:43618", + "issue:43668", + "issue:43704", + "issue:43717", + "issue:43813", + "issue:44031", + "issue:44230", + "issue:44261", + "issue:44263", + "issue:44351", + "issue:44485", + "issue:44492", + "issue:44623", + "issue:44861", + "issue:44908", + "issue:44938", + "issue:44998", + "issue:45250", + "issue:45290", + "issue:45310", + "issue:45468", + "issue:45850", + "issue:46018", + "issue:46095", + "issue:46144", + "issue:46172", + "issue:46218" + ], + "soft_pairs": [ + "issue:46018|issue:46172", + "issue:43704|issue:44485", + "issue:44485|issue:45468", + "issue:42371|issue:43704", + "issue:43502|issue:43519", + "issue:45850|issue:46218", + "issue:43519|issue:44623", + "issue:44263|issue:45468", + "issue:44261|issue:45468", + "issue:43298|issue:46095", + "issue:43519|issue:43704", + "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:28348|issue:43352", + "issue:44230|issue:44623", + "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", + "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": "d608f88c678707decacc54b86967c736f1ba99820bec13aacec1a26fcd2f60ee", + "component_budget": { + "node_count": 500, + "item_count": 500, + "soft_pair_count": 1839, + "serialized_chars": 953882, + "estimated_input_tokens": 238471, + "estimated_eval_tokens": 477198 + }, + "budget": { + "node_count": 46, + "item_count": 46, + "soft_pair_count": 44, + "serialized_chars": 38877, + "estimated_input_tokens": 9720, + "estimated_eval_tokens": 19696 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is noisy and does not contain any clear duplicate issue/PR groupings. The soft pairs mostly share broad vocabulary (v5, tokenizer, attention, import, loading) but describe different models, subsystems, and failure modes. I rejected all soft edges and picked the most representative concrete issue as a weak anchor.", + "confidence": 0.58, + "canonical_issue_reason": "issue:31515 is the most concrete and broadly framed user-facing regression in the set, and it has the strongest soft connectivity, but there is still no true duplicate cluster here.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:31515 is the best single issue anchor because it is specific, actionable, and closest to a general underlying runtime/loading problem, even though it does not truly duplicate the others.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:46018", + "right": "issue:46172", + "accept": false, + "reason": "Different models and bugs: DeepSeek-V4 gating vs RoFormer attention implementation." + }, + { + "left": "issue:43704", + "right": "issue:44485", + "accept": false, + "reason": "VRAM leak in dataloader threads is unrelated to GLM-5 RoPE math." + }, + { + "left": "issue:44485", + "right": "issue:45468", + "accept": false, + "reason": "Both are model-specific, but the bugs are different: RoPE implementation vs audio positional encoding." + }, + { + "left": "issue:42371", + "right": "issue:43704", + "accept": false, + "reason": "TF32 configuration API behavior is unrelated to a multi-thread VRAM leak." + }, + { + "left": "issue:43502", + "right": "issue:43519", + "accept": false, + "reason": "local_files_only network requests and Qwen3VL timestamp calculation are different code paths." + }, + { + "left": "issue:45850", + "right": "issue:46218", + "accept": false, + "reason": "A vague version-wide breakage report is not the same bug as a specific Qwen3_5 import failure." + }, + { + "left": "issue:43519", + "right": "issue:44623", + "accept": false, + "reason": "Processor timestamp math and save_pretrained missing files are unrelated." + }, + { + "left": "issue:44263", + "right": "issue:45468", + "accept": false, + "reason": "torch.split return handling is unrelated to Gemma-4 audio positional encoding." + }, + { + "left": "issue:44261", + "right": "issue:45468", + "accept": false, + "reason": "Missing rms_norm_eps in MLA q_a_layernorm does not match Gemma-4 audio positional encoding." + }, + { + "left": "issue:43298", + "right": "issue:46095", + "accept": false, + "reason": "Continuous batching for audio models and DeepSeek-V4 weight conversion are unrelated." + }, + { + "left": "issue:43519", + "right": "issue:43704", + "accept": false, + "reason": "Timestamp calculation and VRAM leak in dataloader threads are different issues." + }, + { + "left": "issue:44351", + "right": "issue:46218", + "accept": false, + "reason": "Both are import errors, but for different missing symbols/modules and different underlying causes." + }, + { + "left": "issue:43352", + "right": "issue:46218", + "accept": false, + "reason": "Flash Attention support for NemotronH is unrelated to importing Qwen3_5ForCausalLM." + }, + { + "left": "issue:42913", + "right": "issue:43066", + "accept": false, + "reason": "Both mention tokenizer v5, but one is a broad behavior regression and the other is a specific decoder-type bug." + }, + { + "left": "issue:28348", + "right": "issue:45250", + "accept": false, + "reason": "Both mention Flash Attention 2.0, but they target different models and different requests." + }, + { + "left": "issue:44908", + "right": "issue:44998", + "accept": false, + "reason": "Scheduler kwargs handling is unrelated to the nonsensical 'Unemployment' issue." + }, + { + "left": "issue:44492", + "right": "issue:44998", + "accept": false, + "reason": "A cache-strategy typo is not related to the unrelated 'Unemployment' issue." + }, + { + "left": "issue:43618", + "right": "issue:44998", + "accept": false, + "reason": "CLIPOutput attentions regression is unrelated to the unrelated 'Unemployment' issue." + }, + { + "left": "issue:44998", + "right": "issue:45850", + "accept": false, + "reason": "No relation between the unrelated 'Unemployment' issue and a broad v5 breakage report." + }, + { + "left": "issue:44938", + "right": "issue:44998", + "accept": false, + "reason": "Python 3.14 loading failure is unrelated to the unrelated 'Unemployment' issue." + }, + { + "left": "issue:44861", + "right": "issue:45290", + "accept": false, + "reason": "Tied-weight key crash and chat template tool-call crash are different failures." + }, + { + "left": "issue:31515", + "right": "issue:38617", + "accept": false, + "reason": "Slow from_pretrained loading and missing layer_type_validation import are unrelated." + }, + { + "left": "issue:43572", + "right": "issue:44861", + "accept": false, + "reason": "Missing pad_token_idx in config is unrelated to tied-weight key handling." + }, + { + "left": "issue:43525", + "right": "issue:44861", + "accept": false, + "reason": "Missing pad_token_id in Llama4Config is unrelated to tied-weight key handling." + }, + { + "left": "issue:28348", + "right": "issue:43352", + "accept": false, + "reason": "Both concern Flash Attention 2.0, but they are different models and different kinds of support requests." + }, + { + "left": "issue:44230", + "right": "issue:44623", + "accept": false, + "reason": "FP8 support for Qwen3-VL/Qwen3.5 MoE is unrelated to processor.save_pretrained missing files." + }, + { + "left": "issue:29870", + "right": "issue:31515", + "accept": false, + "reason": "Bart generate decoder mask crash and from_pretrained slowdown are unrelated." + }, + { + "left": "issue:30333", + "right": "issue:31515", + "accept": false, + "reason": "MLflow reporting on failed jobs is unrelated to checkpoint loading slowness." + }, + { + "left": "issue:31515", + "right": "issue:36296", + "accept": false, + "reason": "from_pretrained performance and tensor-parallel training bug are different problems." + }, + { + "left": "issue:31515", + "right": "issue:31795", + "accept": false, + "reason": "Loading performance and documentation confusion are not the same issue." + }, + { + "left": "issue:44230", + "right": "issue:45310", + "accept": false, + "reason": "FP8 support and a from_pretrained error for Qwen3.5 MoE are not the same concrete bug." + }, + { + "left": "issue:43298", + "right": "issue:44230", + "accept": false, + "reason": "Continuous batching for audio models is unrelated to FP8 support for Qwen3-VL/Qwen3.5 MoE." + }, + { + "left": "issue:43717", + "right": "issue:46144", + "accept": false, + "reason": "Mamba init_weights usage is unrelated to RoFormer not using the attention interface." + }, + { + "left": "issue:42936", + "right": "issue:46018", + "accept": false, + "reason": "Mask2former ignore_value handling is unrelated to DeepSeek-V4 gating." + }, + { + "left": "issue:43668", + "right": "issue:44031", + "accept": false, + "reason": "A config type hint bug is unrelated to tokenizer regex warning spam." + }, + { + "left": "issue:42936", + "right": "issue:43240", + "accept": false, + "reason": "Ignored mask value handling and kwargs propagation to loss are different bugs." + }, + { + "left": "issue:43240", + "right": "issue:46144", + "accept": false, + "reason": "Loss kwargs propagation is unrelated to RoFormer attention interface use." + }, + { + "left": "issue:42936", + "right": "issue:44031", + "accept": false, + "reason": "Mask2former ignore_value and tokenizer regex warnings are unrelated." + }, + { + "left": "issue:42936", + "right": "issue:46144", + "accept": false, + "reason": "Mask2former ignore_value handling is unrelated to RoFormer attention interface usage." + }, + { + "left": "issue:42936", + "right": "issue:43813", + "accept": false, + "reason": "A missing ignore_value use site is unrelated to a typo in quantization_operation." + }, + { + "left": "issue:43240", + "right": "issue:43668", + "accept": false, + "reason": "Loss kwargs propagation and a config type-hint bug are unrelated." + }, + { + "left": "issue:43668", + "right": "issue:43813", + "accept": false, + "reason": "Incorrect type hint and typo in quantization_operation are unrelated." + }, + { + "left": "issue:37934", + "right": "issue:43668", + "accept": false, + "reason": "Llama4TextL2Norm/RMS norm question is unrelated to a ModernBERT type-hint issue." + }, + { + "left": "issue:42936", + "right": "issue:43668", + "accept": false, + "reason": "Mask2former ignore_value and ModernBERT norm_eps typing are unrelated." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 2, + "component_count": 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": "1c745a834526bfbbd579ab48506844f5b1d94c4370237f758623c5e5f215598c", + "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 asks for a truncation-from-beginning option, one asks for ReduceLROnPlateau-like behavior, and one asks for Wav2Vec2 word boundary timestamps. They target different features and code paths.", + "confidence": 0.96, + "canonical_issue_reason": "No canonical issue: the items are about unrelated requests, so there is no single underlying bug or feature to consolidate.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue #10519 is the most concrete and actionable of the set, with a clear feature request and specific truncation behavior, even though it is not duplicate of the others.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:10519", + "right": "issue:11307", + "accept": false, + "reason": "Different domains: token truncation behavior vs Wav2Vec2 word offset extraction. No shared underlying bug or change." + }, + { + "left": "issue:10519", + "right": "issue:11005", + "accept": false, + "reason": "One is about tokenizer truncation direction; the other is about learning-rate scheduling functionality. These are unrelated feature requests." + }, + { + "left": "issue:11005", + "right": "issue:11307", + "accept": false, + "reason": "ReduceLROnPlateau-like training control and Wav2Vec2 timing offsets are unrelated and would not be merged as one issue." + } + ] + }, + "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": "1c7c6c2b77cdc567735227dcb3a7a69e721f7082446520adda4dfb574dc6744f", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 2477, + "estimated_input_tokens": 620, + "estimated_eval_tokens": 1496 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 2477, + "estimated_input_tokens": 620, + "estimated_eval_tokens": 1496 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "This cluster is mostly a loose semantic group around model/tokenization topics, but the Tapas tokenization issue is unrelated to the model-loading reports. The two model-loading issues are only superficially similar and do not look like the same underlying bug.", + "confidence": 0.72, + "canonical_issue_reason": "Issue #16618 is the broadest and most representative of the only plausible overlap in the cluster, but the duplication evidence is weak.", + "canonical_pr_reason": null, + "best_issue_reason": "#16618 is the best issue candidate only because it is the closest to the cluster's shared model-loading theme; #15980 is a narrower auth/error-message report and #13244 is unrelated.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:13244", + "right": "issue:16618", + "accept": false, + "reason": "Tapas tokenization vs. loading bert-base-uncased are different code paths and different user problems; this is not the same bug." + }, + { + "left": "issue:15980", + "right": "issue:16618", + "accept": false, + "reason": "One is specifically about the error message when downloading a private model while logged out; the other is a generic inability to load bert-base-uncased. Same general area, but not the same underlying issue." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 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": "75c2f88b8a6c6012032f577cfdb1814882bcf520a49bef3c26f30f166547807f", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1589, + "estimated_input_tokens": 398, + "estimated_eval_tokens": 1052 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1589, + "estimated_input_tokens": 398, + "estimated_eval_tokens": 1052 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are not duplicates: one requests Flash Attention 2 support for more architectures, the other requests GGUF support for more architectures. They target different features and code paths.", + "confidence": 0.97, + "canonical_issue_reason": "No single canonical issue fits here; the two issues are distinct feature requests with different underlying implementation work.", + "canonical_pr_reason": null, + "best_issue_reason": "Neither issue is a better canonical for the other, since they do not describe the same bug or change.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:26350", + "right": "issue:33260", + "accept": false, + "reason": "Different enhancement requests (Flash Attention 2 vs GGUF support); not the same underlying change or bug." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "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-27088-8", + "nodes": [ + "issue:27088", + "issue:28936", + "issue:38435", + "issue:43834", + "issue:44365", + "issue:44370", + "issue:46222", + "issue:46226" + ], + "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:44365|issue:44370", + "issue:44365|issue:46222", + "issue:44365|issue:46226", + "issue:44370|issue:46222", + "issue:44370|issue:46226", + "issue:46222|issue:46226" + ], + "prepared_review_unit_hash": "1a401dd5788e7fb5821128907bd5c740e2fa01cf59b774f1a476c09d6211d8c3", + "component_budget": { + "node_count": 8, + "item_count": 8, + "soft_pair_count": 24, + "serialized_chars": 13529, + "estimated_input_tokens": 3383, + "estimated_eval_tokens": 7022 + }, + "budget": { + "node_count": 8, + "item_count": 8, + "soft_pair_count": 24, + "serialized_chars": 13529, + "estimated_input_tokens": 3383, + "estimated_eval_tokens": 7022 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "This cluster is not one true duplicate set: most issues are separate language-specific documentation translation requests, while only the repeated placeholder-template issues and the identical Thai pair look like real duplicates.", + "confidence": 0.77, + "canonical_issue_reason": "issue:43834 is the best representative of the duplicate-like subgroup because it is the earliest of the near-identical placeholder-title issues and sits in the densest duplicate neighborhood.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:43834 is the strongest global representative for the cluster\u2019s duplicate core, but only for the placeholder-template subgroup, not for the distinct language-specific requests.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:27088", + "right": "issue:44370", + "accept": false, + "reason": "Different language request (Turkish vs generic placeholder template); similar body structure is not enough to treat as the same underlying issue." + }, + { + "left": "issue:27088", + "right": "issue:44365", + "accept": false, + "reason": "Turkish translation request is distinct from the generic placeholder-template issue." + }, + { + "left": "issue:27088", + "right": "issue:43834", + "accept": false, + "reason": "Different concrete translation request; same broad docs/i18n theme only." + }, + { + "left": "issue:27088", + "right": "issue:38435", + "accept": false, + "reason": "Turkish and Romanian translation requests are separate language tasks, not the same underlying issue." + }, + { + "left": "issue:27088", + "right": "issue:28936", + "accept": false, + "reason": "Turkish and Spanish translation requests are distinct by target language." + }, + { + "left": "issue:28936", + "right": "issue:43834", + "accept": false, + "reason": "Spanish translation request is not the same issue as the generic placeholder-template ticket." + }, + { + "left": "issue:28936", + "right": "issue:44365", + "accept": false, + "reason": "Different target language; not a duplicate underlying change/request." + }, + { + "left": "issue:28936", + "right": "issue:38435", + "accept": false, + "reason": "Spanish and Romanian translation requests are separate items." + }, + { + "left": "issue:28936", + "right": "issue:44370", + "accept": false, + "reason": "Spanish translation request does not match the generic placeholder-template issue." + }, + { + "left": "issue:38435", + "right": "issue:44370", + "accept": false, + "reason": "Romanian translation request is separate from the placeholder-template ticket." + }, + { + "left": "issue:38435", + "right": "issue:44365", + "accept": false, + "reason": "Different language-specific request; same subsystem is too broad to merge." + }, + { + "left": "issue:38435", + "right": "issue:43834", + "accept": false, + "reason": "Romanian request and generic placeholder-template issue are not the same underlying issue." + }, + { + "left": "issue:38435", + "right": "issue:46222", + "accept": false, + "reason": "Romanian translation request is unrelated to the Thai-titled issue." + }, + { + "left": "issue:38435", + "right": "issue:46226", + "accept": false, + "reason": "Romanian translation request is unrelated to the Thai-titled issue." + }, + { + "left": "issue:43834", + "right": "issue:44370", + "accept": true, + "reason": "Same generic placeholder title and near-identical issue shape; these look like duplicated copies of the same template-based issue." + }, + { + "left": "issue:43834", + "right": "issue:44365", + "accept": true, + "reason": "Same placeholder-template issue with no meaningful distinguishing information; likely duplicate." + }, + { + "left": "issue:43834", + "right": "issue:46222", + "accept": false, + "reason": "Placeholder-template issue and Thai-text issue do not look like the same underlying request." + }, + { + "left": "issue:43834", + "right": "issue:46226", + "accept": false, + "reason": "Placeholder-template issue and Thai-text issue are not the same concrete issue." + }, + { + "left": "issue:44365", + "right": "issue:44370", + "accept": true, + "reason": "Same placeholder title and extremely similar body profile; plausible duplicate of one template issue." + }, + { + "left": "issue:44365", + "right": "issue:46222", + "accept": false, + "reason": "Generic placeholder-template issue vs Thai-text issue; not the same request." + }, + { + "left": "issue:44365", + "right": "issue:46226", + "accept": false, + "reason": "Generic placeholder-template issue vs Thai-text issue; not a duplicate." + }, + { + "left": "issue:44370", + "right": "issue:46222", + "accept": false, + "reason": "Generic placeholder-template issue and Thai-text issue appear unrelated." + }, + { + "left": "issue:44370", + "right": "issue:46226", + "accept": false, + "reason": "Generic placeholder-template issue and Thai-text issue appear unrelated." + }, + { + "left": "issue:46222", + "right": "issue:46226", + "accept": true, + "reason": "These have the same title and appear to be exact copies of the same issue." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary stays grounded in the packet: it distinguishes the clearly different language-specific translation requests from the near-identical placeholder-template issues and the identical Thai pair. The soft-edge verdicts are conservative and do not overclaim duplicates beyond what the titles/structure support." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "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": "0927151adf127f16702dd2d443533d3dc2fd11e8886287482f7d722bff4b1a7e", + "component_budget": { + "node_count": 19, + "item_count": 19, + "soft_pair_count": 40, + "serialized_chars": 24676, + "estimated_input_tokens": 6169, + "estimated_eval_tokens": 12594 + }, + "budget": { + "node_count": 19, + "item_count": 19, + "soft_pair_count": 40, + "serialized_chars": 24676, + "estimated_input_tokens": 6169, + "estimated_eval_tokens": 12594 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "This cluster is mostly a pile of unrelated model-addition requests. The similarity signal is driven by generic wording like \"Add/Support model\" rather than a shared underlying bug or code change, so none of the soft edges look mergeable as duplicates.", + "confidence": 0.97, + "canonical_issue_reason": "Issue 43671 is the best representative anchor here because it is a clear, active model-support request and sits near the densest soft-similarity area, but the overall set is too heterogeneous to justify true duplicate canonization.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 43671 is the most useful global representative: it is concrete, active, and thematically closest to the only obvious mini-cluster (Qwen3 voice support). Even so, it should not be treated as a duplicate of the other requests.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:36718", + "right": "issue:41115", + "accept": false, + "reason": "Different requested artifacts: a RoMa keypoint matcher versus a MiniCPM3 model architecture. Shared \"add model\" phrasing only." + }, + { + "left": "issue:36817", + "right": "issue:41211", + "accept": false, + "reason": "Both are model/config requests, but EuroBert config and DEIMv2 are unrelated model additions with no shared code-path change." + }, + { + "left": "issue:36817", + "right": "issue:43671", + "accept": false, + "reason": "EuroBert config work is unrelated to Qwen3-TTS support; same broad request pattern, different model and modality." + }, + { + "left": "issue:36817", + "right": "issue:36879", + "accept": false, + "reason": "EuroBert config and RF-DETR are distinct model-support requests with different targets and implementation work." + }, + { + "left": "issue:36879", + "right": "issue:41211", + "accept": false, + "reason": "RF-DETR and DEIMv2 are separate detection-model additions, not the same underlying bug or change." + }, + { + "left": "issue:36879", + "right": "issue:43909", + "accept": false, + "reason": "RF-DETR is a detection model; LFM2.5 Audio 1.5B is an audio model. Different feature requests." + }, + { + "left": "issue:36879", + "right": "issue:44121", + "accept": false, + "reason": "RF-DETR and a weight-sparse transformer request are different model families and different implementation work." + }, + { + "left": "issue:36879", + "right": "issue:43837", + "accept": false, + "reason": "RF-DETR versus Qwen3-ASR are unrelated model-support requests across different domains." + }, + { + "left": "issue:36879", + "right": "issue:43671", + "accept": false, + "reason": "RF-DETR is a vision detector request, not the same change as Qwen3-TTS support." + }, + { + "left": "issue:37015", + "right": "issue:43837", + "accept": false, + "reason": "NeoBERT and Qwen3-ASR are unrelated models; no shared code-path problem." + }, + { + "left": "issue:37015", + "right": "issue:44121", + "accept": false, + "reason": "NeoBERT support and a weight-sparse transformer are different model additions, not duplicates." + }, + { + "left": "issue:37015", + "right": "issue:43909", + "accept": false, + "reason": "NeoBERT and LFM2.5 Audio 1.5B are unrelated requests." + }, + { + "left": "issue:36879", + "right": "issue:37015", + "accept": false, + "reason": "RF-DETR and NeoBERT are different model-support issues in different domains." + }, + { + "left": "issue:37015", + "right": "issue:42503", + "accept": false, + "reason": "NeoBERT and ModernVBERT models are separate model-addition requests, not one concrete change." + }, + { + "left": "issue:41115", + "right": "issue:43837", + "accept": false, + "reason": "MiniCPM3 architecture support and Qwen3-ASR support are different model requests." + }, + { + "left": "issue:41115", + "right": "issue:44121", + "accept": false, + "reason": "MiniCPM3 architecture work is not the same as adding a weight-sparse transformer model." + }, + { + "left": "issue:41115", + "right": "issue:42503", + "accept": false, + "reason": "MiniCPM3 and ModernVBERT are distinct models with different integration work." + }, + { + "left": "issue:41115", + "right": "issue:45027", + "accept": false, + "reason": "MiniCPM3 versus Voxtral-4B-TTS support are unrelated model requests." + }, + { + "left": "issue:41211", + "right": "issue:43671", + "accept": false, + "reason": "DEIMv2 and Qwen3-TTS support are different model additions, not the same change." + }, + { + "left": "issue:42503", + "right": "issue:43837", + "accept": false, + "reason": "ModernVBERT and Qwen3-ASR are separate model-support tasks." + }, + { + "left": "issue:42503", + "right": "issue:43671", + "accept": false, + "reason": "ModernVBERT models are unrelated to Qwen3-TTS support." + }, + { + "left": "issue:42503", + "right": "issue:42738", + "accept": false, + "reason": "ModernVBERT and BERT-like models with RoPE are broadly related in wording only; they are not the same concrete request." + }, + { + "left": "issue:42738", + "right": "issue:45027", + "accept": false, + "reason": "RoPE-related BERT-like models and Voxtral TTS support are unrelated." + }, + { + "left": "issue:42738", + "right": "issue:46653", + "accept": false, + "reason": "The second item is not even a model request, so these cannot be duplicates." + }, + { + "left": "issue:43671", + "right": "issue:43837", + "accept": false, + "reason": "Both are Qwen3 support requests, but TTS and ASR are different modalities and would not plausibly merge into one PR." + }, + { + "left": "issue:43671", + "right": "issue:45027", + "accept": false, + "reason": "Qwen3-TTS and Voxtral-4B-TTS are different model integrations, not the same underlying change." + }, + { + "left": "issue:43671", + "right": "issue:45177", + "accept": false, + "reason": "Qwen3-TTS support and DeepSeek-OCR2 are unrelated model requests." + }, + { + "left": "issue:43671", + "right": "issue:44121", + "accept": false, + "reason": "Qwen3-TTS support is not the same change as adding a weight-sparse transformer model." + }, + { + "left": "issue:43837", + "right": "issue:43909", + "accept": false, + "reason": "Qwen3-ASR and LFM2.5 Audio 1.5B are both audio-related, but they are different model support requests." + }, + { + "left": "issue:43837", + "right": "issue:45322", + "accept": false, + "reason": "Qwen3-ASR support and EUPE are unrelated model integrations." + }, + { + "left": "issue:43837", + "right": "issue:44121", + "accept": false, + "reason": "Qwen3-ASR support is unrelated to the weight-sparse transformer request." + }, + { + "left": "issue:43909", + "right": "issue:44205", + "accept": false, + "reason": "LFM2.5 Audio 1.5B and SAM3-LiteText are different model families and domains." + }, + { + "left": "issue:43909", + "right": "issue:46645", + "accept": false, + "reason": "Audio model support and a Granite SWA+sink option are not the same code-path change." + }, + { + "left": "issue:44121", + "right": "issue:45322", + "accept": false, + "reason": "Weight-sparse transformer support and EUPE are separate model requests with no concrete overlap." + }, + { + "left": "issue:43837", + "right": "issue:45027", + "accept": false, + "reason": "Qwen3-ASR and Voxtral-4B-TTS are different speech-model support requests." + }, + { + "left": "issue:37015", + "right": "issue:45027", + "accept": false, + "reason": "NeoBERT and Voxtral TTS support are unrelated requests." + }, + { + "left": "issue:43837", + "right": "issue:45177", + "accept": false, + "reason": "Qwen3-ASR and DeepSeek-OCR2 are different model integrations." + }, + { + "left": "issue:45322", + "right": "issue:46133", + "accept": false, + "reason": "EUPE and TIPSv2 are distinct vision-language model requests; same broad area, not the same change." + }, + { + "left": "issue:42503", + "right": "issue:46653", + "accept": false, + "reason": "ModernVBERT models and \"Mirko Privitera\" are not even comparable as the same artifact." + }, + { + "left": "issue:45027", + "right": "issue:46653", + "accept": false, + "reason": "Voxtral-4B-TTS support and \"Mirko Privitera\" are unrelated; no duplicate relationship." + } + ] + }, + "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": "3cdb9c76e66faae3acf6dd34157f4a7f78e469e1a970923dd8ab3257dd93b582", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1506, + "estimated_input_tokens": 377, + "estimated_eval_tokens": 1010 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1506, + "estimated_input_tokens": 377, + "estimated_eval_tokens": 1010 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These two issues are unrelated: one is a docs request to add Pruna as an optimization framework, the other is a Vision RoPE refactor. There is no duplicate relationship or shared code path.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": "No canonical issue is appropriate here; the issues describe different topics and different parts of the project.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:38740", + "right": "issue:46443", + "accept": false, + "reason": "Not the same underlying bug or change: one is a documentation request about Pruna, the other is a Vision RoPE refactor in model code." + } + ] + }, + "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": "31b3d78ee4bc5e7a83396d91ede3d3fff2e984a35754c6a68f9d5705070ae2af", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1480, + "estimated_input_tokens": 370, + "estimated_eval_tokens": 996 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1480, + "estimated_input_tokens": 370, + "estimated_eval_tokens": 996 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These two issues are not duplicates: one is about block decoding behavior, the other is about DeepSeek v3.2 model support.", + "confidence": 0.98, + "canonical_issue_reason": "No canonical issue: the topics are unrelated and do not describe the same underlying bug or feature request.", + "canonical_pr_reason": null, + "best_issue_reason": "No single best issue for this cluster, because the pair is heterogeneous rather than a duplicate set.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:41084", + "right": "issue:43037", + "accept": false, + "reason": "Different requests: block decoding vs. DeepSeek v3.2 support. They do not point to the same code-path 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": 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": "3ba5a29fad91e31c7cd5dd19c6889b497cdce05790702486a446d0f61a583285", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1584, + "estimated_input_tokens": 396, + "estimated_eval_tokens": 1048 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1584, + "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 not duplicates: one is a specific API/flag behavior request for `from_pretrained()` with TiledFusedLogitsLoss/FLCE, while the other is a separate feature request to integrate TiledMLP for lower memory usage.", + "confidence": 0.95, + "canonical_issue_reason": "Issue 41306 is the more concrete, narrowly scoped report and better represents the cluster than the separate TiledMLP feature request.", + "canonical_pr_reason": null, + "best_issue_reason": "41306 is the best issue to surface because it describes a specific, actionable behavior change with a clear target API.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:41306", + "right": "issue:41826", + "accept": false, + "reason": "Different underlying requests: one concerns a FLCE flag in `from_pretrained()`, the other concerns integrating TiledMLP for memory savings. They do not appear to be 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": 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": "0549078a92ee910dc9ef85b71802bbf8f4046f0836e07e1641b8402ef0c7cb36", + "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 two issues are thematically related around custom trainers and Trainer behavior, but they describe different underlying problems: one is about `remove_unused_column` and customizability, the other is about FLOPS estimation for custom trainers. I would not cluster them as duplicates.", + "confidence": 0.86, + "canonical_issue_reason": "Issue 43598 is the better representative because it states a concrete Trainer behavior change (`remove_unused_column`) and a broader customizability concern, while the other issue is a separate FLOPS-estimation edge case.", + "canonical_pr_reason": null, + "best_issue_reason": "43598 is the more central and actionable issue in this pair; 43600 is narrower and concerns a different code path.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43598", + "right": "issue:43600", + "accept": false, + "reason": "Related theme, but not the same bug/change: one is about column removal/customization in Trainer, the other about FLOPS accounting for custom trainers. They could be fixed separately and would not naturally merge into one issue." + } + ] + }, + "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": "ccc0f405441b4d4f9fe12e5dabfafe05510b4637cda319e1d7c6b6a834be9540", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1611, + "estimated_input_tokens": 403, + "estimated_eval_tokens": 1062 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1611, + "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 Qwen model families and different AutoModel entry points, so they do not look like duplicates.", + "confidence": 0.96, + "canonical_issue_reason": "Issue #43684 is the broader registration request for Qwen3-Omni across AutoModel and AutoModelForConditionalGeneration, so it is the more central issue in this pair.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue #43684 is the better representative because it targets a core model-registration gap rather than a single task-specific head.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43684", + "right": "issue:44405", + "accept": false, + "reason": "Different underlying requests: Qwen3-Omni model registration for AutoModel/AutoModelForConditionalGeneration vs. AutoModelForSequenceClassification support for Qwen3.5. They concern different configs and different model heads, so they should not be merged as the same bug." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 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": "a2a96ad7f6b8a935fe151f73e45175f342d4388cf7fa7d412839d3455ba059f8", + "component_budget": { + "node_count": 4, + "item_count": 4, + "soft_pair_count": 6, + "serialized_chars": 4266, + "estimated_input_tokens": 1067, + "estimated_eval_tokens": 2390 + }, + "budget": { + "node_count": 4, + "item_count": 4, + "soft_pair_count": 6, + "serialized_chars": 4266, + "estimated_input_tokens": 1067, + "estimated_eval_tokens": 2390 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Three issues share the same Hugging Face title and look like duplicate placeholder reports; the \"Fett\" issue appears unrelated.", + "confidence": 0.66, + "canonical_issue_reason": "issue #43845 is the earliest of the near-identical Hugging Face issues and is the best stand-in for the duplicate set.", + "canonical_pr_reason": null, + "best_issue_reason": "issue #43845 is the cleanest representative of the repeated Hugging Face report cluster; the other matching issues can fold into it.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43835", + "right": "issue:43845", + "accept": false, + "reason": "\"Fett\" does not match the Hugging Face report and there is no evidence they describe the same issue." + }, + { + "left": "issue:43835", + "right": "issue:43846", + "accept": false, + "reason": "\"Fett\" looks unrelated to the Hugging Face issue cluster." + }, + { + "left": "issue:43835", + "right": "issue:43859", + "accept": false, + "reason": "No shared topic beyond generic similarity; this does not look like the same underlying issue." + }, + { + "left": "issue:43845", + "right": "issue:43846", + "accept": true, + "reason": "Same title and very similar metadata/body size; these look like duplicate reports of the same Hugging Face issue." + }, + { + "left": "issue:43845", + "right": "issue:43859", + "accept": true, + "reason": "All are the same Hugging Face report topic, and this pair appears to be duplicate issue filing." + }, + { + "left": "issue:43846", + "right": "issue:43859", + "accept": true, + "reason": "Same repeated Hugging Face issue title and closely related report content; likely duplicates." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is supported by the packet: 43845/43846/43859 share the same title and have relatively high soft-similarity, while 43835 is notably less similar and looks separate. The canonical-choice rationale is also grounded in creation order. Soft-edge verdicts are conservative enough overall." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "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": "c08bac667a3ecb1bbee2ab87c35add7f46ee6f6c5b4ce93827498346d48e5c92", + "component_budget": { + "node_count": 5, + "item_count": 5, + "soft_pair_count": 9, + "serialized_chars": 5794, + "estimated_input_tokens": 1449, + "estimated_eval_tokens": 3154 + }, + "budget": { + "node_count": 5, + "item_count": 5, + "soft_pair_count": 9, + "serialized_chars": 5794, + "estimated_input_tokens": 1449, + "estimated_eval_tokens": 3154 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "This cluster looks like low-signal spam / unrelated placeholder issues, with no evidence of a shared underlying bug or change. The similar titles are not enough to treat them as duplicates.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": "Issue 44418 is the least obviously spammy and the only one that reads like a real user-facing report, but there is still no evidence it is a duplicate of the others.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44371", + "right": "issue:44572", + "accept": false, + "reason": "Both are just '' placeholders with no body or targets; similarity is not enough to infer the same bug." + }, + { + "left": "issue:44371", + "right": "issue:44573", + "accept": false, + "reason": "Both are generic spam placeholders, but there is no concrete issue content tying them to the same underlying problem." + }, + { + "left": "issue:44371", + "right": "issue:44418", + "accept": false, + "reason": "One is spam and the other is a documentation suggestion; the titles are not evidence of the same issue." + }, + { + "left": "issue:44371", + "right": "issue:44937", + "accept": false, + "reason": "A spam placeholder and a titled issue about 'Google LLC' are not the same underlying bug or change." + }, + { + "left": "issue:44418", + "right": "issue:44572", + "accept": false, + "reason": "Documentation enhancement and spam placeholder are not a plausible duplicate pair." + }, + { + "left": "issue:44418", + "right": "issue:44573", + "accept": false, + "reason": "No concrete overlap beyond weak text similarity; these do not look like the same issue." + }, + { + "left": "issue:44572", + "right": "issue:44573", + "accept": false, + "reason": "Both are spam placeholders, but there is no substantive evidence they report the same problem." + }, + { + "left": "issue:44572", + "right": "issue:44937", + "accept": false, + "reason": "A spam placeholder and a titled issue about 'Google LLC' are unrelated." + }, + { + "left": "issue:44573", + "right": "issue:44937", + "accept": false, + "reason": "The titles and lack of body details do not support treating these as duplicates." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 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": "41cc2a898d4f3def52174adb836e789492e43e0df040d18800ee3d85791aad0e", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1645, + "estimated_input_tokens": 412, + "estimated_eval_tokens": 1080 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1645, + "estimated_input_tokens": 412, + "estimated_eval_tokens": 1080 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "The two closed issues appear to be duplicate reports of the same energy-reduction proposal, with matching titles and very similar content.", + "confidence": 0.98, + "canonical_issue_reason": "Issue #45145 is the better canonical issue because it was created first and already has an inbound reference, so it is slightly more established as the primary thread.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue #45145 is the best issue representative for this cluster for the same reasons: earliest creation and a reference signal, while #45151 appears to be a near-identical duplicate.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:45145", + "right": "issue:45151", + "accept": true, + "reason": "Same title, same apparent topic, and highly similar body size/content indicate the same underlying issue rather than merely related reports." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The analyst output is grounded in the packet: the two issues share the exact title, have very similar body lengths, and the packet provides soft_similarity evidence. The canonical/best-issue reasoning is conservative and supported by the earlier creation time plus the inbound reference on #45145. No unsupported mergeability claims are made." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "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": "d4155d6c0f7c2028b26ea423ee47ae1700d44f1845642cd94b2bc8c4ac76ee83", + "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": "The two closed issues are not duplicates; they are unrelated, title-only posts that appear to be personal names/dates rather than the same bug or request.", + "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": "No evidence of a shared bug, request, or code path; the titles are different names with different dates and there is no body or linkage tying them together." + } + ] + }, + "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": "c81581221d08c355e208dbce7ab99bdfb342c0a088cbdcdd915ab2972c665334", + "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": "The two closed issues appear to be unrelated, bare-name entries with different people and dates; there is no indication of a shared bug, feature, or code change.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": "Neither issue is a good canonical candidate because both are similarly content-free and do not describe an actionable product problem.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:45764", + "right": "issue:45765", + "accept": false, + "reason": "These are just two different personal-name/date entries with no shared technical subject, so they do not look like duplicate reports of 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": "4baa76b9aed350d90ff64f26f96ba08b8bfcb20cbb8cd3f42b194070f0d0c3d3", + "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": "The two issues are unrelated feature requests: one asks for Qwen3_5 token classification support, the other asks for FP8 training support. They should not be clustered as duplicates.", + "confidence": 0.98, + "canonical_issue_reason": "Issue 45810 is the more concrete and actionable request, with a specific model class named in the title.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 45810 is the best representative because it is narrowly scoped and clearly describes the missing functionality.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:45810", + "right": "issue:46295", + "accept": false, + "reason": "Different feature areas and code paths: model-class support versus FP8 training support. They do not describe the same underlying bug or change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 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": "ffcdecb03bf3e0c570233bb2220bbc9a49b883508a690d480a621d79264b82ac", + "component_budget": { + "node_count": 5, + "item_count": 5, + "soft_pair_count": 8, + "serialized_chars": 5726, + "estimated_input_tokens": 1432, + "estimated_eval_tokens": 3120 + }, + "budget": { + "node_count": 5, + "item_count": 5, + "soft_pair_count": 8, + "serialized_chars": 5726, + "estimated_input_tokens": 1432, + "estimated_eval_tokens": 3120 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "This is not one duplicate cluster but two separate exact-duplicate pairs: 45870/45871 and 45872/45873. Issue 45869 is a related dataset marketing/request item but not the same underlying ask as the others.", + "confidence": 0.97, + "canonical_issue_reason": "45872 is the best single representative if one must be chosen: it has an exact duplicate (45873) and is the most concrete, actionable DataForge request. The other exact-dupe pair (45870/45871) is a separate issue, and 45869 is a different dataset-bundle request.", + "canonical_pr_reason": null, + "best_issue_reason": "45872 is the strongest anchor for this set because it is an exact duplicate pair with a more specific feature-oriented title ('financial time series + scraping service').", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:45869", + "right": "issue:45870", + "accept": false, + "reason": "Both mention financial data/time-series themes, but one is a dataset bundle offer and the other is a DataForge example request; not the same underlying issue." + }, + { + "left": "issue:45869", + "right": "issue:45871", + "accept": false, + "reason": "Same as 45869/45870: related topic, but different artifact type and request scope." + }, + { + "left": "issue:45870", + "right": "issue:45871", + "accept": true, + "reason": "Exact title match and same apparent example request; these look like duplicate issues." + }, + { + "left": "issue:45870", + "right": "issue:45872", + "accept": false, + "reason": "Both are DataForge/financial-time-series themed, but 45870 is an example while 45872 asks for a scraping service; not the same concrete change." + }, + { + "left": "issue:45870", + "right": "issue:45873", + "accept": false, + "reason": "Same reasoning as 45870/45872; related subject, but different underlying request." + }, + { + "left": "issue:45871", + "right": "issue:45872", + "accept": false, + "reason": "Exact duplicate of 45870, but still a different request from the scraping-service issue." + }, + { + "left": "issue:45871", + "right": "issue:45873", + "accept": false, + "reason": "Different underlying request despite overlapping domain terminology." + }, + { + "left": "issue:45872", + "right": "issue:45873", + "accept": true, + "reason": "Exact title match and same feature request; these are clear duplicates." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet: 45870/45871 and 45872/45873 are exact title matches, while 45869 is clearly a different dataset-bundle request. The non-duplicate verdicts are appropriately conservative given only soft similarity 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": "e0011f7dd5e2206b7cb3b2e4207ae2ccadff8fbb842738e7fe711dfdcde82118", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1636, + "estimated_input_tokens": 409, + "estimated_eval_tokens": 1074 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1636, + "estimated_input_tokens": 409, + "estimated_eval_tokens": 1074 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These two issues are related to static-cache handling for linear attention, but the reported failures differ enough that they should not be merged as duplicates without more evidence.", + "confidence": 0.66, + "canonical_issue_reason": "Issue 46439 is the more general/root-cause-looking report: it describes static-cache early initialization corrupting linear-attention layers and producing shape errors, which could encompass narrower crash variants.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 46439 is the best representative because it names the suspected mechanism and the broader failure mode, making it more useful as the primary tracking issue.", + "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 reports shape corruption during early initialization while the other reports a Qwen3.5 KeyError for 'linear_attention'; they may be related, but they do not clearly describe the same concrete bug." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 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": "ae20f63216fbbe7cb3a90b3cb740641f9051c6eec93da73cd39884d1aa704961", + "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 describe the same Wav2Vec2CTCTokenizer v5 regression verbatim. #46552 is the better canonical tracker because it is open and already has inbound references; #46551 appears to be an earlier overlapping report, but the packet does not explicitly confirm the closure reason.", + "confidence": 0.96, + "canonical_issue_reason": "#46552 is the better canonical issue because it is still open and has inbound references, while #46551 is the same bug report verbatim and appears to be the older overlapping record.", + "canonical_pr_reason": null, + "best_issue_reason": "#46552 is the best issue to keep as the primary tracker: it is open, already referenced, and matches #46551\u2019s title/body exactly.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:46551", + "right": "issue:46552", + "accept": true, + "reason": "Same exact title and same underlying tokenizer bug description; they are effectively the same issue, with no contrary evidence in the packet." + } + ] + }, + "evaluator_result": { + "accept": false, + "feedback": "The pairing is well grounded as a strong duplicate-like match, but the writeup overstates the evidence by calling #46551 a \"closed duplicate.\" The packet only shows that #46551 is closed and #46552 is open with inbound references; it does not explicitly prove the closure reason. Keep the canonical recommendation, but phrase the duplicate/closure status more conservatively." + }, + "evaluator_used": true, + "retried": true, + "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": "a5ac32533ba3db2c6aac740cfed3a70322dabd37c21ebf3e2df84cac217200a6", + "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 are two separate training issues in the same general area, but they describe different failure modes: one is a tensor shape mismatch during object detection training, the other is an out-of-memory problem after several epochs with RT-DETR v2. I would not merge them.", + "confidence": 0.94, + "canonical_issue_reason": "Issue #46657 is the more concrete bug report: it names a specific failure mode (tensor size mismatch) and a broader object-detection training context, making it a better representative of the cluster if forced to choose one issue.", + "canonical_pr_reason": null, + "best_issue_reason": "#46657 is the best issue candidate because it describes an actionable code-path problem rather than a resource-pressure symptom, and its title is more likely to map to a single underlying defect.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:46657", + "right": "issue:46692", + "accept": false, + "reason": "They share a training-related theme, but the underlying problems differ: tensor dimension mismatch versus OOM during later epochs. These are not the same bug and would not plausibly be fixed by one change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 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": "6d78b03b27273ebd16d91ce5482824b1cdebe6c83a90e4df6536f20bbda4753b", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1671, + "estimated_input_tokens": 418, + "estimated_eval_tokens": 1092 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1671, + "estimated_input_tokens": 418, + "estimated_eval_tokens": 1092 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "The two issues are both about model loading/device placement, but they describe different failure modes in different code paths, so they should not be clustered as duplicates.", + "confidence": 0.95, + "canonical_issue_reason": "Issue 46822 is the stronger anchor: it has more discussion and inbound references, and it describes a concrete ZeRO-3 zero.Init multi-GPU OOM scenario with a specific composite model submodule.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 46822 is the best representative issue for the cluster because it is the more developed report and the symptom/cause are more specific and actionable.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:46822", + "right": "issue:46823", + "accept": false, + "reason": "Both involve loading models onto devices, but 46822 is a ZeRO-3 partitioning/OOM bug during zero.Init on a composite submodule, while 46823 is a 4-bit + device_map/max_memory routing bug that incorrectly dispatches to CPU and raises ValueError. Different mechanisms, symptoms, and likely fixes." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 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": "cdd9e8ed7426e434ef423c6bb012b3959945c1c4e4d6dafb66c294041f567d27", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 7223, + "estimated_input_tokens": 1806, + "estimated_eval_tokens": 3868 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 7223, + "estimated_input_tokens": 1806, + "estimated_eval_tokens": 3868 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Two open PRs target the VibeVoice family, but they represent different layers of work: the first is the initial VibeVoice integration, while the second adds VibeVoice Realtime and related model/processor changes.", + "confidence": 0.91, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #40546 is the better canonical PR because it is the broader, foundational VibeVoice implementation and has substantially more review activity, making it the primary representative of this feature set.", + "best_issue_reason": null, + "best_pr_reason": "PR #40546 is the best overall representative for the cluster since it introduces the main VibeVoice support; #42781 is a later, more specialized follow-on for realtime functionality rather than the same mergeable change.", + "soft_edge_verdicts": [ + { + "left": "pull_request:40546", + "right": "pull_request:42781", + "accept": false, + "reason": "Reject: despite overlapping VibeVoice files, #40546 is the base VibeVoice implementation and #42781 is a separate VibeVoice Realtime extension with additional feature-extractor/processing and generation changes. They are related, but not the same concrete code-path fix/change and would not plausibly be merged as one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 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": "c9d06f029a67398b6900ae3f9ca79a82b46b9389ade3a6fd44591ac7d500d1ef", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4676, + "estimated_input_tokens": 1169, + "estimated_eval_tokens": 2594 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4676, + "estimated_input_tokens": 1169, + "estimated_eval_tokens": 2594 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "The cluster contains two unrelated pull requests: one adds TimesFM covariate forecasting, the other adds DINOv3 image-classification support. They do not appear to fix the same underlying bug or implement the same change.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "pull_request:40755", + "right": "pull_request:41224", + "accept": false, + "reason": "Different model families and different feature areas; TimesFM covariate forecasting is unrelated to DINOv3 image-classification support." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 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": "474094e2db7289b57fa7fce08a3971ef3abdddc72fe56555d6e0ce9b5f84477b", + "component_budget": { + "node_count": 15, + "item_count": 15, + "soft_pair_count": 26, + "serialized_chars": 42243, + "estimated_input_tokens": 10561, + "estimated_eval_tokens": 21378 + }, + "budget": { + "node_count": 15, + "item_count": 15, + "soft_pair_count": 26, + "serialized_chars": 42243, + "estimated_input_tokens": 10561, + "estimated_eval_tokens": 21378 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "This cluster is mostly a grab bag of unrelated PRs. The only true duplicate-like pair is the two `fixed_cross_entropy` kwargs fixes; everything else targets different models, utilities, or tests and should stay separate.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 43251 is the best canonical representative for the duplicate pair because it directly references the tracked bug (43240) and cleanly implements the `cross_entropy` kwargs passthrough in `fixed_cross_entropy`.", + "best_issue_reason": null, + "best_pr_reason": "PR 43251 is the strongest global representative of the cluster\u2019s only real duplicate candidate; it is narrowly scoped, bug-driven, and the closest match to the underlying fix.", + "soft_edge_verdicts": [ + { + "left": "pull_request:41855", + "right": "pull_request:43823", + "accept": false, + "reason": "Tokenizer missing-methods work for Mistral is unrelated to adding the MobileLLM model." + }, + { + "left": "pull_request:41855", + "right": "pull_request:43424", + "accept": false, + "reason": "Mistral tokenizer API additions and an executorch export test address different code paths and different symptoms." + }, + { + "left": "pull_request:42668", + "right": "pull_request:43823", + "accept": false, + "reason": "Processor auto-loading robustness changes are unrelated to introducing MobileLLM." + }, + { + "left": "pull_request:43251", + "right": "pull_request:43823", + "accept": false, + "reason": "A loss-helper kwargs fix is unrelated to MobileLLM model addition." + }, + { + "left": "pull_request:43251", + "right": "pull_request:44827", + "accept": false, + "reason": "`fixed_cross_entropy` kwargs support and Mistral4 test fixes are not the same underlying bug." + }, + { + "left": "pull_request:43424", + "right": "pull_request:46911", + "accept": false, + "reason": "Executorch dynamic-shape export testing and Olmo3 RoPE handling are different features in different subsystems." + }, + { + "left": "pull_request:43424", + "right": "pull_request:46707", + "accept": false, + "reason": "A dynamic-shape export test is unrelated to adding native FSDP2 support." + }, + { + "left": "pull_request:43424", + "right": "pull_request:44827", + "accept": false, + "reason": "Executorch export testing and Mistral4 test/model fixes are separate concerns." + }, + { + "left": "pull_request:43424", + "right": "pull_request:43823", + "accept": false, + "reason": "Executorch export tests and MobileLLM model scaffolding are unrelated." + }, + { + "left": "pull_request:43823", + "right": "pull_request:45655", + "accept": false, + "reason": "MobileLLM model creation is unrelated to resolving `cls.config` annotation/attribute order." + }, + { + "left": "pull_request:43823", + "right": "pull_request:43995", + "accept": false, + "reason": "MobileLLM addition and Falcon output-collection refactoring do not fix the same code path." + }, + { + "left": "pull_request:43823", + "right": "pull_request:45213", + "accept": false, + "reason": "A model implementation PR is unrelated to the new-model skill/doc/tooling PR." + }, + { + "left": "pull_request:43823", + "right": "pull_request:46707", + "accept": false, + "reason": "MobileLLM model code and FSDP2 migration are unrelated changes." + }, + { + "left": "pull_request:43823", + "right": "pull_request:46911", + "accept": false, + "reason": "MobileLLM addition and Olmo3 RoPE-per-layer support are different model implementations." + }, + { + "left": "pull_request:43995", + "right": "pull_request:45213", + "accept": false, + "reason": "Falcon output-capture refactoring and the model-creation skill are not the same bug or change." + }, + { + "left": "pull_request:43995", + "right": "pull_request:44827", + "accept": false, + "reason": "Falcon standardized output collection and Mistral4 test fixes are unrelated." + }, + { + "left": "pull_request:44827", + "right": "pull_request:45170", + "accept": false, + "reason": "Mistral4 test fixes and the `layrnorm` typo renaming in CLIP-like models are different changes." + }, + { + "left": "pull_request:44827", + "right": "pull_request:45213", + "accept": false, + "reason": "A model test-fix PR is unrelated to the model-creation skill repository changes." + }, + { + "left": "pull_request:44827", + "right": "pull_request:45655", + "accept": false, + "reason": "Mistral4 test updates and config-resolution order fixes are unrelated." + }, + { + "left": "pull_request:44827", + "right": "pull_request:46048", + "accept": false, + "reason": "Mistral4 test fixes and TDT loss-kernel integration are separate code paths." + }, + { + "left": "pull_request:44827", + "right": "pull_request:46267", + "accept": false, + "reason": "Mistral4 test fixes and chat-response parsing updates are unrelated." + }, + { + "left": "pull_request:44827", + "right": "pull_request:46707", + "accept": false, + "reason": "Mistral4 test/model changes and native FSDP2 migration are not the same underlying issue." + }, + { + "left": "pull_request:44827", + "right": "pull_request:46911", + "accept": false, + "reason": "Mistral4 test fixes and Olmo3 RoPE changes are unrelated model-specific work." + }, + { + "left": "pull_request:45170", + "right": "pull_request:46707", + "accept": false, + "reason": "A LayerNorm naming/weight-renaming fix in CLIP-like models is unrelated to FSDP2 migration." + }, + { + "left": "pull_request:46048", + "right": "pull_request:46707", + "accept": false, + "reason": "TDT loss-kernel integration and FSDP2 migration do not address the same bug or implementation change." + }, + { + "left": "pull_request:44827", + "right": "pull_request:45213", + "accept": false, + "reason": "Mistral4 test/model work is unrelated to the model-creation skill PR." + }, + { + "left": "pull_request:43823", + "right": "pull_request:45213", + "accept": false, + "reason": "MobileLLM model addition is unrelated to the model-creation skill and scaffolding changes." + }, + { + "left": "pull_request:43251", + "right": "pull_request:43254", + "accept": true, + "reason": "Both PRs change `fixed_cross_entropy` in the same file to pass supported kwargs through to `nn.functional.cross_entropy`; they fix the same concrete loss-helper bug and could be merged into one PR." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The assessment is well grounded in the packet. The only accepted pair, 43251/43254, is a conservative match: both touch `fixed_cross_entropy` in the same file, point to the same issue target (43240), and are plausibly mergeable into one bugfix PR. The rest of the pairwise rejections are appropriately cautious and do not overstate similarity." + }, + "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": "26826399be55932318e38a3b55498429ddf195d36278f6a692bde16f602ca9b5", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5935, + "estimated_input_tokens": 1484, + "estimated_eval_tokens": 3224 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5935, + "estimated_input_tokens": 1484, + "estimated_eval_tokens": 3224 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Two open PRs both upgrade GitHub Actions, but they target different action families and change sets: one is a broad Node 24 compatibility pass, the other is a Docker build action version bump. They are related but not duplicates.", + "confidence": 0.92, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #42975 is the better canonical representative because it has the clearest, broadest scope: upgrading multiple workflow actions for Node 24 compatibility across many CI workflows.", + "best_issue_reason": null, + "best_pr_reason": "PR #42975 is the strongest representative of the cluster because its title and file coverage summarize the overarching workflow-action upgrade theme more completely than the narrower Docker-action bump in #42976.", + "soft_edge_verdicts": [ + { + "left": "pull_request:42975", + "right": "pull_request:42976", + "accept": false, + "reason": "Both are workflow dependency upgrades, but they affect different underlying actions and concrete code paths (checkout/cache/upload-artifact vs docker/build-push-action). They could be part of a larger maintenance sweep, but not the same fix or mergeable as one duplicate PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 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": "d41c076ac823ac20841c19dac202ffd2266d4266abbeaf52bd1a3d72d4f228f2", + "component_budget": { + "node_count": 6, + "item_count": 6, + "soft_pair_count": 8, + "serialized_chars": 21198, + "estimated_input_tokens": 5300, + "estimated_eval_tokens": 10856 + }, + "budget": { + "node_count": 6, + "item_count": 6, + "soft_pair_count": 8, + "serialized_chars": 21198, + "estimated_input_tokens": 5300, + "estimated_eval_tokens": 10856 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Mostly a bundle of unrelated model-addition PRs. The only close-looking pair is Step3p7-related, but they still differ in scope: one integrates the model into Transformers, the other is a WIP scripts/mini-model conversion repo. No true duplicate pair stands out.", + "confidence": 0.9, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 46562 is the strongest representative of the only near-duplicate area: a full Step 3.7 Flash Transformers integration with docs, auto-mappings, and tests. The other Step3p7 PR is a separate WIP script/conversion effort, not the same change.", + "best_issue_reason": null, + "best_pr_reason": "PR 46562 is the best overall PR to anchor this cluster because it is the most complete, self-contained model integration and the only one that plausibly overlaps with another item in the set. The rest are different model additions.", + "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 just boilerplate for model additions, not the same underlying change." + }, + { + "left": "pull_request:42978", + "right": "pull_request:43451", + "accept": false, + "reason": "Different model families (ViT NEPA vs Molmo2). Similar scaffolding changes do not make this the same bug or change." + }, + { + "left": "pull_request:42978", + "right": "pull_request:45490", + "accept": false, + "reason": "Different model families (ViT NEPA vs CTSM). They touch shared auto/docs plumbing but implement unrelated additions." + }, + { + "left": "pull_request:43448", + "right": "pull_request:43451", + "accept": false, + "reason": "Molmo and Molmo2 are related by naming but are distinct model integrations with different files and code paths." + }, + { + "left": "pull_request:43448", + "right": "pull_request:45490", + "accept": false, + "reason": "Molmo vs CTSM are unrelated model additions; overlapping framework files are not enough to treat them as duplicates." + }, + { + "left": "pull_request:43451", + "right": "pull_request:45490", + "accept": false, + "reason": "Molmo2 vs CTSM are separate model integrations with different implementations and test coverage." + }, + { + "left": "pull_request:45490", + "right": "pull_request:46562", + "accept": false, + "reason": "CTSM and Step 3.7 Flash are unrelated model additions; the shared auto/doc changes are generic plumbing, not one underlying fix." + }, + { + "left": "pull_request:46562", + "right": "pull_request:46658", + "accept": false, + "reason": "Both mention Step3p7, but 46562 adds the Transformers model while 46658 is a WIP mini-model/script conversion repo. They are different code paths and not plausibly 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-43291-8", + "nodes": [ + "pull_request:43291", + "pull_request:43665", + "pull_request:44815", + "pull_request:46380", + "pull_request:46539", + "pull_request:46870", + "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:46870", + "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": "137c06b664034453c1b095ccb0d3aacf26c0c77f44f9c3c8d6b56f618668591d", + "component_budget": { + "node_count": 8, + "item_count": 8, + "soft_pair_count": 16, + "serialized_chars": 21660, + "estimated_input_tokens": 5415, + "estimated_eval_tokens": 11086 + }, + "budget": { + "node_count": 8, + "item_count": 8, + "soft_pair_count": 16, + "serialized_chars": 21660, + "estimated_input_tokens": 5415, + "estimated_eval_tokens": 11086 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Only one soft-edge pair looks duplicate-like: the two YOLOS device-mismatch fixes, where 46922 is a fuller version of 46886. The rest are different models, workflows, or unrelated code paths and should remain separate.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 46922 is the best canonical PR because it contains the more complete YOLOS device-placement fix, including the mid-position embedding device move and the no_split_modules adjustment, whereas 46886 is the narrower earlier variant.", + "best_issue_reason": null, + "best_pr_reason": "PR 46922 is the strongest representative of the only true duplicate-like pair in the cluster; it addresses the same YOLOS auto-device-map bug as 46886 and includes the broader fix surface.", + "soft_edge_verdicts": [ + { + "left": "pull_request:43291", + "right": "pull_request:43665", + "accept": false, + "reason": "Both are test-oriented, but they target different model suites and different failure modes (Whisper vs CLIP/ViT)." + }, + { + "left": "pull_request:43291", + "right": "pull_request:44815", + "accept": false, + "reason": "Whisper tokenizer/test changes are unrelated to FP8 dequantization and model-loading logic." + }, + { + "left": "pull_request:43291", + "right": "pull_request:46380", + "accept": false, + "reason": "Whisper decoding/tests and generation-config validation are different subsystems with different bugs." + }, + { + "left": "pull_request:43291", + "right": "pull_request:46922", + "accept": false, + "reason": "Whisper test/tokenizer work is unrelated to the YOLOS device-placement bug." + }, + { + "left": "pull_request:43665", + "right": "pull_request:46539", + "accept": false, + "reason": "Generic CLIP/ViT test edits do not match the Phimoe CI memory-sharing fix." + }, + { + "left": "pull_request:43665", + "right": "pull_request:46870", + "accept": false, + "reason": "Different areas: model test edits versus ASR pipeline classification changes." + }, + { + "left": "pull_request:43665", + "right": "pull_request:44815", + "accept": false, + "reason": "These are not the same change or bug; one is model tests, the other is FP8/model-loading work." + }, + { + "left": "pull_request:43665", + "right": "pull_request:46380", + "accept": false, + "reason": "Both touch broad model code/testing, but they address distinct issues and are not mergeable as one fix." + }, + { + "left": "pull_request:43665", + "right": "pull_request:46922", + "accept": false, + "reason": "CLIP/ViT test edits are unrelated to YOLOS device mismatch." + }, + { + "left": "pull_request:44815", + "right": "pull_request:46539", + "accept": false, + "reason": "FP8 dequantization/model-loading changes are unrelated to CI runner memory limits and integration-test setup." + }, + { + "left": "pull_request:44815", + "right": "pull_request:46380", + "accept": false, + "reason": "FP8 kernel/loading fixes and generation watermarking validation are different code paths and bugs." + }, + { + "left": "pull_request:44815", + "right": "pull_request:46922", + "accept": false, + "reason": "FP8 dequantization work is unrelated to YOLOS device placement." + }, + { + "left": "pull_request:46380", + "right": "pull_request:46539", + "accept": false, + "reason": "Generation-config validation and Phimoe CI environment fixes are separate problems." + }, + { + "left": "pull_request:46380", + "right": "pull_request:46922", + "accept": false, + "reason": "Watermarking/config validation changes are unrelated to YOLOS device mismatch." + }, + { + "left": "pull_request:46539", + "right": "pull_request:46922", + "accept": false, + "reason": "CI instance-sharing fixes and YOLOS model device-placement fixes are not the same underlying issue." + }, + { + "left": "pull_request:46886", + "right": "pull_request:46922", + "accept": true, + "reason": "Both target the same YOLOS device-mismatch bug with device_map=\"auto\"; 46922 is the broader version that also fixes the forward-path embedding placement." + } + ] + }, + "evaluator_result": { + "accept": false, + "feedback": "The summary is mostly grounded, but several soft-edge verdicts are not conservative and are internally inconsistent: e.g. `43665` vs `44815`, `46380`, and `46922` are marked `accept: true` even though the reasons explicitly say they are different subsystems / not duplicates. That overstates evidence. The only well-supported duplicate-like pair in the packet is `46886`/`46922`, which is plausibly mergeable as the same YOLOS bugfix." + }, + "evaluator_used": true, + "retried": true, + "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": "efa2f9436b91922ea3bbfae1e580a08be1193e208ff953122ddefdc78a09ae8f", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 6348, + "estimated_input_tokens": 1587, + "estimated_eval_tokens": 3430 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 6348, + "estimated_input_tokens": 1587, + "estimated_eval_tokens": 3430 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are three unrelated '[don't merge]' repository-bot / workflow test PRs with only superficial similarity in title and purpose. They touch different code paths: packaging/setup, comment/workflow automation, and security-gate workflow behavior. No duplicate pair is a strong same-change match.", + "confidence": 0.93, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": "PR 43532 is the broadest representative of the cluster if one must be picked: it modifies multiple workflow files plus testing utilities and test code, whereas the others are narrower and target different areas.", + "soft_edge_verdicts": [ + { + "left": "pull_request:43488", + "right": "pull_request:46747", + "accept": false, + "reason": "Different changes: 43488 is a setup/version and dependency tweak plus a test change, while 46747 only switches the PR CI security gate workflow trigger from pull_request_target to pull_request. Similar '[don't merge]' framing is not enough to make them the same underlying fix." + }, + { + "left": "pull_request:43488", + "right": "pull_request:43532", + "accept": false, + "reason": "43488 is about packaging/versioning and adding pytest-random-order, whereas 43532 rewrites comment-driven CI/workflow automation and testing utilities. They do not fix the same concrete bug and would not plausibly merge into one PR." + }, + { + "left": "pull_request:43532", + "right": "pull_request:46747", + "accept": false, + "reason": "Both touch workflows, but the actual code paths differ: 43532 updates self-comment CI, scheduled/slack workflows, testing helpers, and a vit test; 46747 only changes the security gate workflow trigger. This is too broad to treat as one duplicate change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 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": "0c04026f2ad3c05ed44a2cf158e0232a2a95daead3a20cf6ce7d2f9a2628f927", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 6083, + "estimated_input_tokens": 1521, + "estimated_eval_tokens": 3298 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 6083, + "estimated_input_tokens": 1521, + "estimated_eval_tokens": 3298 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Two separate feature PRs: one adds a Promptable Concept Segmentation pipeline, the other adds a Promptable Visual Segmentation pipeline. They share documentation/registry plumbing, but they implement different pipelines and model integrations, so they are not duplicates.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single canonical PR: these are distinct feature additions with different pipeline names, model hooks, and tests, despite overlapping docs/metadata updates.", + "best_issue_reason": null, + "best_pr_reason": "Neither PR is a global duplicate canonical; if a representative must be chosen, PR #43613 is slightly broader in model coverage, but it is still a separate feature from #43612.", + "soft_edge_verdicts": [ + { + "left": "pull_request:43612", + "right": "pull_request:43613", + "accept": false, + "reason": "Shared docs and registration files are incidental. One adds Promptable Concept Segmentation, the other adds Promptable Visual Segmentation; they target different tasks and code paths, so they should not be merged 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": "39f1a9a98253c83e682041f8254d7b47def04c8f42bb72b29eabdf731be9b5ec", + "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 open PRs both touch `trainer.py`, but they implement different trainer extension hooks: one changes gradient-accumulation loss scaling, the other refactors signature-column filtering. They are not the same underlying change and should not be merged as duplicates.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": "No canonical PR: the patches address different code paths and different custom-trainer extensibility points.", + "best_issue_reason": null, + "best_pr_reason": "PR #43651 is the clearer self-contained behavior change, but it is still unrelated to PR #43663; it is not a duplicate candidate for the cluster.", + "soft_edge_verdicts": [ + { + "left": "pull_request:43651", + "right": "pull_request:43663", + "accept": false, + "reason": "Different functionality in the same file: #43651 changes gradient accumulation loss scaling, while #43663 adds a helper for model signature column filtering. They do not fix the same bug or implement the same change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 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": "217d61badd7b399b69bbedfc98d80aaf3b96473363c652bf420d4e64509118ab", + "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": "These two open PRs touch the same file and both mention TypeAdapter-related NameErrors, but they address different root causes: one is a broader annotation/runtime typing cleanup in the CLI, the other is a narrow fix for optional pydantic imports. They are related, but not duplicates.", + "confidence": 0.91, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 43656 is the broader CLI typing fix and likely the better canonical representative for this cluster; it addresses runtime annotation evaluation in serve.py rather than only the pydantic-import edge case.", + "best_issue_reason": null, + "best_pr_reason": "PR 43656 is the strongest single representative because it appears to fix the underlying annotation NameError pattern more generally in the CLI code path, while PR 43842 is a narrower follow-up about TypeAdapter being unavailable when pydantic is missing.", + "soft_edge_verdicts": [ + { + "left": "pull_request:43656", + "right": "pull_request:43842", + "accept": false, + "reason": "Reject: both are in serve.py and both mention NameError, but the diffs point to different bugs\u201443656 changes annotation evaluation with __future__ annotations and typing cleanup, while 43842 guards a pydantic TypeAdapter import behind TYPE_CHECKING. They are related but not the same underlying fix." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 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:44059|pull_request:44722", + "pull_request:44018|pull_request:44071", + "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": "c687705031019804ae71c272450db66ee25afa8a0a1a6838cf2eeb99a4e8f792", + "component_budget": { + "node_count": 28, + "item_count": 28, + "soft_pair_count": 77, + "serialized_chars": 84505, + "estimated_input_tokens": 21127, + "estimated_eval_tokens": 42510 + }, + "budget": { + "node_count": 28, + "item_count": 28, + "soft_pair_count": 72, + "serialized_chars": 76936, + "estimated_input_tokens": 19234, + "estimated_eval_tokens": 38724 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a batch of model-specific refactors for standardized output tracing/capture_outputs. The only true duplicate-like overlaps are within the same concrete model file or a broader PR that subsumes the same file-level change; cross-model \u201csame issue target\u201d pairs should stay separate.", + "confidence": 0.92, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 44722 is the best canonical PR for the GPT-J/CodeGen duplicate subset: it is the latest and most complete variant, covers the shared GPT-J tracing refactor plus the CodeGen companion change, and aligns directly with the tracking issue.", + "best_issue_reason": null, + "best_pr_reason": "PR 44722 is the strongest representative of the duplicate subset because it consolidates the shared GPT-J/CodeGen output-tracing change rather than a narrower or earlier variant.", + "soft_edge_verdicts": [ + { + "left": "pull_request:43996", + "right": "pull_request:44085", + "accept": false, + "reason": "Different model families (CVT/FNet vs GPT-J); same broad output-tracing theme is not enough." + }, + { + "left": "pull_request:43996", + "right": "pull_request:44044", + "accept": false, + "reason": "Different models and code paths (CVT/FNet vs DeBERTa v2)." + }, + { + "left": "pull_request:44066", + "right": "pull_request:44085", + "accept": true, + "reason": "Both target the same GPT-J output-tracing refactor in the same file; 44085 looks like a narrower variant of the same change." + }, + { + "left": "pull_request:44007", + "right": "pull_request:44072", + "accept": false, + "reason": "Both are output-tracing refactors, but they are different model implementations (ResNet/RegNet/RT-DETR vs EfficientNet)." + }, + { + "left": "pull_request:44072", + "right": "pull_request:44722", + "accept": false, + "reason": "EfficientNet and GPT-J/CodeGen are unrelated code paths despite similar refactor style." + }, + { + "left": "pull_request:44066", + "right": "pull_request:44072", + "accept": false, + "reason": "GPT-J/CodeGen vs EfficientNet; shared issue tracking does not make them duplicates." + }, + { + "left": "pull_request:44013", + "right": "pull_request:44044", + "accept": false, + "reason": "MobileNetV2 and DeBERTa v2 are unrelated implementations." + }, + { + "left": "pull_request:44066", + "right": "pull_request:44086", + "accept": false, + "reason": "GPT-J/CodeGen versus MGP-STR; different model internals and fixes." + }, + { + "left": "pull_request:44018", + "right": "pull_request:44068", + "accept": true, + "reason": "Same model/file family (GPT-Neo) and the diffs are the same output-tracing refactor shape." + }, + { + "left": "pull_request:44066", + "right": "pull_request:44071", + "accept": false, + "reason": "GPT-J/CodeGen vs MPT are different model code paths." + }, + { + "left": "pull_request:44066", + "right": "pull_request:44068", + "accept": false, + "reason": "GPT-J/CodeGen versus GPT-Neo; similar refactor pattern but not the same change." + }, + { + "left": "pull_request:44018", + "right": "pull_request:44066", + "accept": false, + "reason": "GPT-Neo and GPT-J/CodeGen are separate model-specific refactors." + }, + { + "left": "pull_request:44019", + "right": "pull_request:44722", + "accept": false, + "reason": "ResNet output tracing and GPT-J/CodeGen are unrelated changes." + }, + { + "left": "pull_request:44086", + "right": "pull_request:44722", + "accept": false, + "reason": "MGP-STR and GPT-J/CodeGen are different models and different concrete fixes." + }, + { + "left": "pull_request:44019", + "right": "pull_request:44085", + "accept": false, + "reason": "ResNet versus GPT-J; same issue tag is not enough." + }, + { + "left": "pull_request:44018", + "right": "pull_request:44019", + "accept": false, + "reason": "GPT-Neo and ResNet are different code paths." + }, + { + "left": "pull_request:44019", + "right": "pull_request:44071", + "accept": false, + "reason": "ResNet versus MPT are unrelated implementations." + }, + { + "left": "pull_request:44068", + "right": "pull_request:44722", + "accept": false, + "reason": "GPT-Neo and GPT-J/CodeGen are different model families." + }, + { + "left": "pull_request:44068", + "right": "pull_request:44116", + "accept": false, + "reason": "GPT-Neo and Flaubert are unrelated model-specific refactors." + }, + { + "left": "pull_request:44018", + "right": "pull_request:44116", + "accept": false, + "reason": "GPT-Neo and Flaubert do not share the same concrete bug/change." + }, + { + "left": "pull_request:44071", + "right": "pull_request:44722", + "accept": false, + "reason": "MPT and GPT-J/CodeGen are different code paths." + }, + { + "left": "pull_request:44018", + "right": "pull_request:44085", + "accept": false, + "reason": "GPT-Neo versus GPT-J; same pattern, different model fix." + }, + { + "left": "pull_request:44018", + "right": "pull_request:44086", + "accept": false, + "reason": "GPT-Neo and MGP-STR are unrelated." + }, + { + "left": "pull_request:44068", + "right": "pull_request:44086", + "accept": false, + "reason": "GPT-Neo and MGP-STR are unrelated." + }, + { + "left": "pull_request:44071", + "right": "pull_request:44085", + "accept": false, + "reason": "MPT and GPT-J are different model internals." + }, + { + "left": "pull_request:44085", + "right": "pull_request:44086", + "accept": false, + "reason": "GPT-J and MGP-STR are different changes." + }, + { + "left": "pull_request:44059", + "right": "pull_request:44086", + "accept": false, + "reason": "GPT-2 and MGP-STR are unrelated model-specific refactors." + }, + { + "left": "pull_request:44098", + "right": "pull_request:44101", + "accept": false, + "reason": "ViLT versus Flaubert/XLM are different model families." + }, + { + "left": "pull_request:44059", + "right": "pull_request:44071", + "accept": false, + "reason": "GPT-2 and MPT are separate implementations." + }, + { + "left": "pull_request:43997", + "right": "pull_request:44017", + "accept": false, + "reason": "RegNet and Segformer are unrelated model code paths." + }, + { + "left": "pull_request:43997", + "right": "pull_request:44056", + "accept": false, + "reason": "RegNet and MPNet are different model implementations." + }, + { + "left": "pull_request:44071", + "right": "pull_request:44076", + "accept": false, + "reason": "MPT and ImageGPT are unrelated changes." + }, + { + "left": "pull_request:44074", + "right": "pull_request:44101", + "accept": false, + "reason": "TextNet and Flaubert/XLM are different model-specific refactors." + }, + { + "left": "pull_request:43997", + "right": "pull_request:44018", + "accept": false, + "reason": "RegNet and GPT-Neo are different code paths." + }, + { + "left": "pull_request:44056", + "right": "pull_request:44161", + "accept": false, + "reason": "MPNet and LongT5 are unrelated model fixes." + }, + { + "left": "pull_request:43997", + "right": "pull_request:44161", + "accept": false, + "reason": "RegNet and LongT5 are different model families." + }, + { + "left": "pull_request:44074", + "right": "pull_request:44161", + "accept": false, + "reason": "TextNet and LongT5 are unrelated." + }, + { + "left": "pull_request:44018", + "right": "pull_request:44161", + "accept": false, + "reason": "GPT-Neo and LongT5 are different model-specific changes." + }, + { + "left": "pull_request:44010", + "right": "pull_request:44071", + "accept": false, + "reason": "SqueezeBert and MPT are unrelated implementations." + }, + { + "left": "pull_request:44010", + "right": "pull_request:44073", + "accept": false, + "reason": "SqueezeBert and VisualBert are different model code paths." + }, + { + "left": "pull_request:44018", + "right": "pull_request:44129", + "accept": false, + "reason": "GPT-Neo and SpeechT5 are unrelated model families." + }, + { + "left": "pull_request:43997", + "right": "pull_request:44074", + "accept": false, + "reason": "RegNet and TextNet are different model-specific refactors." + }, + { + "left": "pull_request:44056", + "right": "pull_request:44101", + "accept": false, + "reason": "MPNet and Flaubert/XLM are unrelated." + }, + { + "left": "pull_request:44071", + "right": "pull_request:44076", + "accept": false, + "reason": "MPT and ImageGPT are unrelated changes." + }, + { + "left": "pull_request:44018", + "right": "pull_request:44059", + "accept": false, + "reason": "GPT-Neo and GPT-2 are different model code paths." + }, + { + "left": "pull_request:44059", + "right": "pull_request:44068", + "accept": false, + "reason": "GPT-2 and GPT-Neo are separate refactors." + }, + { + "left": "pull_request:44068", + "right": "pull_request:44722", + "accept": false, + "reason": "GPT-Neo and GPT-J/CodeGen are not the same concrete change." + }, + { + "left": "pull_request:44071", + "right": "pull_request:44086", + "accept": false, + "reason": "MPT and MGP-STR are unrelated." + }, + { + "left": "pull_request:44017", + "right": "pull_request:44073", + "accept": false, + "reason": "Segformer and VisualBert are different model implementations." + }, + { + "left": "pull_request:44010", + "right": "pull_request:44018", + "accept": false, + "reason": "SqueezeBert and GPT-Neo are unrelated." + }, + { + "left": "pull_request:43997", + "right": "pull_request:44129", + "accept": false, + "reason": "RegNet and SpeechT5 are different model code paths." + }, + { + "left": "pull_request:44071", + "right": "pull_request:44129", + "accept": false, + "reason": "MPT and SpeechT5 are unrelated changes." + }, + { + "left": "pull_request:43996", + "right": "pull_request:44161", + "accept": false, + "reason": "CVT/FNet and LongT5 are unrelated model-specific refactors." + }, + { + "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:44044", + "accept": false, + "reason": "SqueezeBert and DeBERTa v2 are unrelated." + }, + { + "left": "pull_request:44044", + "right": "pull_request:44074", + "accept": false, + "reason": "DeBERTa v2 and TextNet are different code paths." + }, + { + "left": "pull_request:44024", + "right": "pull_request:44076", + "accept": false, + "reason": "FocalNet and ImageGPT are unrelated model implementations." + }, + { + "left": "pull_request:44085", + "right": "pull_request:44722", + "accept": true, + "reason": "Same GPT-J output-tracing change; 44722 is the broader later consolidation that subsumes this file-level fix." + }, + { + "left": "pull_request:44085", + "right": "pull_request:44129", + "accept": false, + "reason": "GPT-J and SpeechT5 are unrelated model-specific refactors." + }, + { + "left": "pull_request:44044", + "right": "pull_request:44129", + "accept": false, + "reason": "DeBERTa v2 and SpeechT5 are different code paths." + }, + { + "left": "pull_request:44044", + "right": "pull_request:44154", + "accept": false, + "reason": "DeBERTa v2 and Vits are unrelated changes." + }, + { + "left": "pull_request:44024", + "right": "pull_request:44076", + "accept": false, + "reason": "FocalNet and ImageGPT are different model families." + }, + { + "left": "pull_request:44007", + "right": "pull_request:44019", + "accept": true, + "reason": "Both include the ResNet output-tracing refactor in the same concrete file; 44007 is the broader superset." + }, + { + "left": "pull_request:44101", + "right": "pull_request:44116", + "accept": true, + "reason": "Both cover the same Flaubert output-tracing change in the same file; 44101 additionally includes XLM." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays conservative about cross-model overlaps. The accepted soft pairs are all supported by either the same concrete file or a clear broader PR that subsumes the narrower file-level change, so they look mergeable into a single bugfix PR. The rejected cross-model pairs are handled cautiously and are consistent with the evidence." + }, + "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": "06ad026ccb0114dbbf145150e503fdc2df6607ea7f4cd1319e30f78f11f50ad2", + "component_budget": { + "node_count": 28, + "item_count": 28, + "soft_pair_count": 77, + "serialized_chars": 84505, + "estimated_input_tokens": 21127, + "estimated_eval_tokens": 42510 + }, + "budget": { + "node_count": 6, + "item_count": 6, + "soft_pair_count": 5, + "serialized_chars": 13069, + "estimated_input_tokens": 3268, + "estimated_eval_tokens": 6792 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "All six PRs are model-specific refactors around standardized output tracing/capture, but they affect different architectures and code paths, so they are not duplicates of the same change.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 44129 is the strongest representative because it is the only one tied to an explicit issue and makes a concrete SpeechT5 tracing refactor rather than a purely generic migration.", + "best_issue_reason": null, + "best_pr_reason": "PR 44129 has the clearest scope and traceability, with an explicit issue link and a concrete code-path change in SpeechT5.", + "soft_edge_verdicts": [ + { + "left": "pull_request:44073", + "right": "pull_request:44074", + "accept": false, + "reason": "Both are output-tracing refactors, but one is for VisualBert and the other for TextNet; different model code paths mean they are not the same change." + }, + { + "left": "pull_request:44056", + "right": "pull_request:44073", + "accept": false, + "reason": "These touch different models (MPNet vs VisualBert) and different tracing implementations, so they are not mergeable as one duplicate fix." + }, + { + "left": "pull_request:44056", + "right": "pull_request:44074", + "accept": false, + "reason": "MPNet and TextNet are separate model implementations; this is the same refactor theme, not the same underlying bug or code change." + }, + { + "left": "pull_request:43997", + "right": "pull_request:44010", + "accept": false, + "reason": "RegNet and SqueezeBert are unrelated model files with independent output-handling updates, so this is not a duplicate PR pair." + }, + { + "left": "pull_request:44010", + "right": "pull_request:44129", + "accept": false, + "reason": "SqueezeBert output collection and SpeechT5 output tracing/caching are different concrete code paths; shared tooling does not make them duplicates." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 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": "5a20ed1d0ec6145ebdde55438fc49d476b5a425ec49115ef9e0ba92692bf18e5", + "component_budget": { + "node_count": 13, + "item_count": 13, + "soft_pair_count": 40, + "serialized_chars": 40181, + "estimated_input_tokens": 10046, + "estimated_eval_tokens": 20348 + }, + "budget": { + "node_count": 13, + "item_count": 13, + "soft_pair_count": 40, + "serialized_chars": 40181, + "estimated_input_tokens": 10046, + "estimated_eval_tokens": 20348 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These PRs are highly similar textually, but they are independent, model-specific output-tracing refactors across different architectures. They should not be deduplicated as the same change unless they touch the same concrete code path, which they do not.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": "If one PR must represent the cluster, 44025 is the best anchor: it is a self-contained refactor across two closely related depth-anything models and includes tests, making it the most complete example of the pattern.", + "best_issue_reason": null, + "best_pr_reason": "44025 is the strongest representative of the family because it is broader than the single-file refactors and has accompanying test coverage, while still staying within one coherent model family.", + "soft_edge_verdicts": [ + { + "left": "pull_request:43998", + "right": "pull_request:44002", + "accept": false, + "reason": "Different model families and code paths; same output-tracing refactor theme is too broad." + }, + { + "left": "pull_request:43998", + "right": "pull_request:44000", + "accept": false, + "reason": "Timm backbone vs vision-text dual encoder are unrelated implementations." + }, + { + "left": "pull_request:43998", + "right": "pull_request:44001", + "accept": false, + "reason": "Different architectures; no shared concrete bug or mergeable code path." + }, + { + "left": "pull_request:43998", + "right": "pull_request:44003", + "accept": false, + "reason": "Timm backbone refactor is unrelated to Falcon/Mamba state/output handling." + }, + { + "left": "pull_request:43998", + "right": "pull_request:44029", + "accept": false, + "reason": "Different model code paths; only a broad tracing-refactor similarity." + }, + { + "left": "pull_request:43999", + "right": "pull_request:44001", + "accept": false, + "reason": "MobileNetV1 and UnivNet are separate models with separate output semantics." + }, + { + "left": "pull_request:43999", + "right": "pull_request:44002", + "accept": false, + "reason": "MobileNetV1 and UperNet are unrelated code paths." + }, + { + "left": "pull_request:43999", + "right": "pull_request:44025", + "accept": false, + "reason": "Different model families; no same underlying bug or change." + }, + { + "left": "pull_request:43999", + "right": "pull_request:44026", + "accept": false, + "reason": "MobileNetV1 refactor is unrelated to vision encoder-decoder wrapper changes." + }, + { + "left": "pull_request:43999", + "right": "pull_request:44027", + "accept": false, + "reason": "Different architectures and concrete implementations." + }, + { + "left": "pull_request:44000", + "right": "pull_request:44001", + "accept": false, + "reason": "Vision-text dual encoder and UnivNet are unrelated model paths." + }, + { + "left": "pull_request:44000", + "right": "pull_request:44002", + "accept": false, + "reason": "Different models; same refactor motif is not enough to merge." + }, + { + "left": "pull_request:44000", + "right": "pull_request:44003", + "accept": false, + "reason": "Vision-text dual encoder vs Falcon/Mamba output tracing are different changes." + }, + { + "left": "pull_request:44000", + "right": "pull_request:44004", + "accept": false, + "reason": "Different architectures and code paths; not the same fix." + }, + { + "left": "pull_request:44000", + "right": "pull_request:44025", + "accept": false, + "reason": "Vision-text dual encoder is unrelated to depth-anything model tracing." + }, + { + "left": "pull_request:44001", + "right": "pull_request:44002", + "accept": false, + "reason": "UnivNet and UperNet are separate model families." + }, + { + "left": "pull_request:44001", + "right": "pull_request:44003", + "accept": false, + "reason": "Different concrete output/caching logic in different models." + }, + { + "left": "pull_request:44001", + "right": "pull_request:44025", + "accept": false, + "reason": "UnivNet refactor is unrelated to depth-anything refactor." + }, + { + "left": "pull_request:44002", + "right": "pull_request:44003", + "accept": false, + "reason": "UperNet segmentation and Falcon/Mamba generation paths are unrelated." + }, + { + "left": "pull_request:44002", + "right": "pull_request:44004", + "accept": false, + "reason": "Different models and output contracts; not mergeable as one fix." + }, + { + "left": "pull_request:44003", + "right": "pull_request:44028", + "accept": false, + "reason": "Falcon/Mamba state tracing and SuperPoint feature extraction are unrelated." + }, + { + "left": "pull_request:44003", + "right": "pull_request:44004", + "accept": false, + "reason": "Different concrete code paths: Mamba/Falcon state handling vs CodeGen cache handling." + }, + { + "left": "pull_request:43998", + "right": "pull_request:44004", + "accept": false, + "reason": "Timm backbone and CodeGen are unrelated model implementations." + }, + { + "left": "pull_request:44001", + "right": "pull_request:44004", + "accept": false, + "reason": "UnivNet and CodeGen do not share the same bug or change." + }, + { + "left": "pull_request:44002", + "right": "pull_request:44004", + "accept": false, + "reason": "UperNet and CodeGen are different code paths." + }, + { + "left": "pull_request:44025", + "right": "pull_request:44026", + "accept": false, + "reason": "Depth-anything models vs vision encoder-decoder wrappers are different concrete changes." + }, + { + "left": "pull_request:44025", + "right": "pull_request:44027", + "accept": false, + "reason": "Depth-anything and speech encoder-decoder are unrelated model families." + }, + { + "left": "pull_request:44025", + "right": "pull_request:44028", + "accept": false, + "reason": "Depth-anything refactor is unrelated to SuperPoint output tracing." + }, + { + "left": "pull_request:44025", + "right": "pull_request:44029", + "accept": false, + "reason": "Depth-anything and RWKV are different architectures and output contracts." + }, + { + "left": "pull_request:44025", + "right": "pull_request:44030", + "accept": false, + "reason": "Depth-anything refactor is unrelated to DPR encoder output handling." + }, + { + "left": "pull_request:44026", + "right": "pull_request:44027", + "accept": false, + "reason": "Vision encoder-decoder and speech encoder-decoder are separate wrapper implementations." + }, + { + "left": "pull_request:44026", + "right": "pull_request:44028", + "accept": false, + "reason": "Different model families; not the same underlying bug or change." + }, + { + "left": "pull_request:44026", + "right": "pull_request:44029", + "accept": false, + "reason": "Vision encoder-decoder wrapper and RWKV generation path are unrelated." + }, + { + "left": "pull_request:44026", + "right": "pull_request:44030", + "accept": false, + "reason": "Different models and code paths." + }, + { + "left": "pull_request:44027", + "right": "pull_request:44028", + "accept": false, + "reason": "Speech encoder-decoder and SuperPoint are unrelated implementations." + }, + { + "left": "pull_request:44027", + "right": "pull_request:44029", + "accept": false, + "reason": "Speech encoder-decoder wrapper changes do not match RWKV output tracing." + }, + { + "left": "pull_request:44027", + "right": "pull_request:44030", + "accept": false, + "reason": "Different architectures and concrete fixes." + }, + { + "left": "pull_request:44028", + "right": "pull_request:44029", + "accept": false, + "reason": "SuperPoint feature tracing and RWKV sequence modeling are unrelated." + }, + { + "left": "pull_request:44028", + "right": "pull_request:44030", + "accept": false, + "reason": "Different model families and output structures." + }, + { + "left": "pull_request:44029", + "right": "pull_request:44030", + "accept": false, + "reason": "RWKV and DPR are separate code paths with different output behavior." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 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": "c9358f835f79bd06fb8233b38d0a62a9cbe91c1937b4fbe0c0460159464c1825", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 6552, + "estimated_input_tokens": 1638, + "estimated_eval_tokens": 3532 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 6552, + "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 unrelated model-addition PRs: one adds PenguinVL, the other adds Voxtral TTS. They share some generic integration files, but they target different models, different model codepaths, and only one links a specific issue. No duplicate relationship is present.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": "No canonical PR for deduplication: the PRs implement different models and should not be collapsed into one change.", + "best_issue_reason": null, + "best_pr_reason": "PR #44662 is the better representative artifact only by maturity and review activity, but it is not a duplicate of #45401.", + "soft_edge_verdicts": [ + { + "left": "pull_request:44662", + "right": "pull_request:45401", + "accept": false, + "reason": "Reject: PenguinVL and Voxtral TTS are distinct model implementations with different titles, filenames, and likely different codepaths. Overlap is limited to shared Transformers registration/docs plumbing, which 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": 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": "8866b115fbc0df34b9b0278ac2586f8ad455be36235585fcab6bf425c4750c64", + "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": "These two pull requests are not duplicates: one is a workflow change to route PR CI to a smaller runner/branch target, while the other is an unrelated attempt to convert PRs to draft via a documentation workflow. They do not target the same code path or same underlying fix.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 46845 is the more coherent and concrete change set, but it is not a duplicate of PR 44965; the two PRs address different automation paths and should remain separate.", + "best_issue_reason": null, + "best_pr_reason": "PR 46845 is the better representative of the cluster because it has a clearer, self-contained CI-related purpose and an explicit branch/workflow adjustment, whereas PR 44965 is a different draft-conversion workflow change.", + "soft_edge_verdicts": [ + { + "left": "pull_request:44965", + "right": "pull_request:46845", + "accept": false, + "reason": "Different underlying automation changes: 44965 touches PR documentation/draft conversion workflow, while 46845 changes PR CI routing and adds unrelated code/test churn. They are not the same bug or mergeable as one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 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": "55c4f258eca2e1b525148b1d9d7a30eb13b24b8e6db87fe21fcb35c42c094f24", + "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 are related to kernel/module fusion infrastructure, but they implement different layers of the stack and do not look like the same mergeable change. PR #44979 adds a standalone Module Fusion API and tests; PR #45363 adds n-to-1 kernel fusion wiring via KernelConfig across integrations/modeling utilities.", + "confidence": 0.94, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #45363 is the better canonical representative for this cluster because it is the broader, end-user-facing fusion change and touches the runtime integration path rather than just introducing a lower-level API scaffold.", + "best_issue_reason": null, + "best_pr_reason": "PR #45363 is the best overall fit: it has the clearest concrete behavior change (KernelConfig-driven fusion in model loading/execution) and is more likely to anchor follow-up discussion than the smaller API-only foundation PR.", + "soft_edge_verdicts": [ + { + "left": "pull_request:44979", + "right": "pull_request:45363", + "accept": false, + "reason": "Different code paths and scopes: #44979 introduces a general Module Fusion API in a new module, while #45363 implements n-to-1 kernel fusion through KernelConfig and integration hooks. Related, but not the same underlying change and not plausibly one merged PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 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": "b56b14de3cfd1956a4b112c7dfaa72472769978f8a25a25342556eefe297eed7", + "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 PRs patch the same AttributeError in `_patch_mistral_regex` by switching from `tokenizer.backend_tokenizer.pre_tokenizer` to `tokenizer.pre_tokenizer` and updating the sequence assignment accordingly; one PR just adds tests and a slightly broader diff around the same fix.", + "confidence": 0.99, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #45317 is the better canonical PR because it contains the same core fix as #45086 plus test coverage, making it the more complete representation of the change.", + "best_issue_reason": null, + "best_pr_reason": "PR #45317 is the best representative because it fixes the same concrete code-path problem as #45086 and includes validation via tests, so it is the strongest standalone record of the change.", + "soft_edge_verdicts": [ + { + "left": "pull_request:45086", + "right": "pull_request:45317", + "accept": true, + "reason": "Accept: these PRs make the same exact code change in the same file/function to fix the same AttributeError, and the right PR simply adds tests around 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": 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": "f0e6332fac5f2e9bbab1fc838db6bde1d843efec31d5488637067b433f01637e", + "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": "Both items are NemotronH-related PRs, but they address different changes: one refactors conversion/mapping compatibility, while the other splits NemotronH into dense/sparse model variants and updates docs/auto-mapping.", + "confidence": 0.91, + "canonical_issue_reason": null, + "canonical_pr_reason": "No clear canonical PR for duplication: #45115 is a conversion-mapping refactor, while #45569 is a model-architecture split and documentation/API update.", + "best_issue_reason": null, + "best_pr_reason": "#45569 is the stronger representative of the main product change because it introduces the new dense/sparse NemotronH model structure and corresponding integration work.", + "soft_edge_verdicts": [ + { + "left": "pull_request:45115", + "right": "pull_request:45569", + "accept": false, + "reason": "They share the NemotronH subsystem, but not the same underlying fix. #45115 only adds conversion-key renames for checkpoint loading, whereas #45569 reworks the model/package layout into dense and sparse variants plus docs/tests. These could not plausibly be merged as one PR without mixing distinct work." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 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": "c6bc5fe1c029dde9faaf12eb3df4567fbeeb2378b17d5727a4995521a961a653", + "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 feature PRs for heterogeneous model support, but they are not duplicates: #45333 is a config-only slice, while #45332 includes config plus broader modeling/runtime integration. I would keep #45332 as the representative PR for the cluster and reject the soft-duplicate link.", + "confidence": 0.91, + "canonical_issue_reason": null, + "canonical_pr_reason": "#45332 is the broader, more complete implementation: it covers per-layer config plus modeling/runtime changes across cache, masking, mxfp4, and heterogeneous modeling utilities, so it best represents the underlying feature work.", + "best_issue_reason": null, + "best_pr_reason": "#45332 is the best PR to anchor the cluster because it subsumes the config work and extends into the core code paths needed for actual heterogeneous model support.", + "soft_edge_verdicts": [ + { + "left": "pull_request:45332", + "right": "pull_request:45333", + "accept": false, + "reason": "They share the same feature area and some files, but #45333 is only per-layer configuration while #45332 also adds modeling/runtime support. That makes #45333 a subset, not the same concrete PR-level 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": "bb1ce2ec1b2af41f1f415c3778ed90eac49c748e6290c7c41765f4065b02ff99", + "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 both typing-focused PRs in transformers, but they address different scopes: #45415 is a broad type-checking sweep across many files, while #45425 adds/adjusts specific typing protocols and annotations around modeling_utils and related helpers. Shared typing infrastructure is not enough to make them the same change.", + "confidence": 0.86, + "canonical_issue_reason": null, + "canonical_pr_reason": "#45425 is the more focused, concrete typing change set; #45415 is a larger repo-wide type-checking cleanup and not the same underlying change.", + "best_issue_reason": null, + "best_pr_reason": "#45425 is the better representative PR for this cluster because it makes a narrower, self-contained typing fix rather than a broad linting/type-check sweep.", + "soft_edge_verdicts": [ + { + "left": "pull_request:45415", + "right": "pull_request:45425", + "accept": false, + "reason": "Both are typing-related, but they fix different problems at different scopes. #45415 is a wide sweep to add type checking across many modules and silence type-check issues, while #45425 introduces specific typing protocols/aliases for modeling_utils and related integrations. They do not look mergeable as one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 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": "50adef346135bb2fb9bd94d598c43001bf14ee390cb6ed197af01f39c891c4c2", + "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": "Two PRs touch expert-parallel loading paths, but they fix different failure modes: one is about preserving EP-related config and launch-time loading for DeepSpeed ZeRO-3, while the other prevents EP experts from being overwritten during FSDP2/rank-0 broadcast. They are related by subsystem, not the same underlying change.", + "confidence": 0.94, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #45662 is the better canonical PR for this cluster because it targets a concrete EP loading corruption bug in the weight-loading path and spans the relevant post-shard / distributed tensor handling. It is the more substantive and behaviorally specific fix.", + "best_issue_reason": null, + "best_pr_reason": "PR #45662 is the best representative of the cluster because it addresses a clearer end-to-end correctness problem in the model-loading code path, whereas PR #45548 is a narrower config/launch integration fix.", + "soft_edge_verdicts": [ + { + "left": "pull_request:45548", + "right": "pull_request:45662", + "accept": false, + "reason": "Both involve expert parallelism, but they fix different bugs and code paths: #45548 preserves distributed config and adds an EP flag for accelerate/ZeRO-3 loading, while #45662 fixes experts being overwritten after shard wrapping in FSDP2. Not mergeable as one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 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": "c4eff0bb07013c6542e403b08445f3e8da8df8b54921f871c4c9ebb30fbde358", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4673, + "estimated_input_tokens": 1169, + "estimated_eval_tokens": 2594 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4673, + "estimated_input_tokens": 1169, + "estimated_eval_tokens": 2594 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both PRs target the same xdist-related captured_info logging problem in transformers testing utilities; PR 45645 is the broader, more complete version because it includes the core xdist-safe file handling plus the CI workflow change to aggregate multiple captured_info files.", + "confidence": 0.88, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 45645 is the better canonical PR because it covers the underlying fix in testing_utils and also updates the CI workflow to read all xdist-split captured_info artifacts, making the end-to-end solution more complete.", + "best_issue_reason": null, + "best_pr_reason": "PR 45645 is the strongest representative of the change set: it fixes the xdist-safe output path and preserves CI debug log collection across multiple worker files, whereas 45639 appears to cover only the code-side portion.", + "soft_edge_verdicts": [ + { + "left": "pull_request:45639", + "right": "pull_request:45645", + "accept": true, + "reason": "Same underlying bug: captured_info logging breaks under xdist. 45645 looks like a superset of 45639 by including the same testing_utils fix plus the workflow-side aggregation of split log files, so they could plausibly be merged into one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 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": "5ba90fed84929d18bddad0ac449ae14298bfeed26c270e5403b55b031364e63f", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 7234, + "estimated_input_tokens": 1809, + "estimated_eval_tokens": 3874 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 7234, + "estimated_input_tokens": 1809, + "estimated_eval_tokens": 3874 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "The two PRs are both linter-rule changes, but they implement different rules on different code paths (TRF003 vs TRF018). They should not be treated as duplicates.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #46259 is the better cluster representative because it adds the more substantial modeling-rule implementation across many model files, while #45720 is a narrower partial TRF003 change.", + "best_issue_reason": null, + "best_pr_reason": "PR #46259 is the strongest standalone candidate: it has a broader, clearer scope and a more complete rule implementation, making it more suitable as the representative PR for this cluster.", + "soft_edge_verdicts": [ + { + "left": "pull_request:45720", + "right": "pull_request:46259", + "accept": false, + "reason": "Reject: these are different lint rules with different intended behavior. #45720 implements part of TRF003, while #46259 adds TRF018; they do not appear to fix the same concrete bug or change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 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": "74cbde91c9343bcab1d2ba59e55a9ab7b4ca9a8704db2853d15f74f99cd10ea3", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 7883, + "estimated_input_tokens": 1971, + "estimated_eval_tokens": 4198 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 7883, + "estimated_input_tokens": 1971, + "estimated_eval_tokens": 4198 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are three unrelated open test/CI-oriented PRs that only overlap superficially in labels and workflow/testing areas. They do not appear to fix the same underlying problem or belong in one duplicate cluster.", + "confidence": 0.92, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #45932 is the broadest and most central of the three, with the largest surface area and explicit CI/workflow changes, but it is not a true duplicate of the others.", + "best_issue_reason": null, + "best_pr_reason": "#45932 is the best representative PR in this set because it is the most substantial and most centrally about CI/test plumbing, whereas #46100 is a test migration and #46891 is a separate PEFT integration CI experiment.", + "soft_edge_verdicts": [ + { + "left": "pull_request:45932", + "right": "pull_request:46100", + "accept": false, + "reason": "Superficial overlap only: both are test-related, but #45932 changes model code and CI secrets, while #46100 adds kernel/model tests and a DDP YAML. Different concrete changes and no shared bug path." + }, + { + "left": "pull_request:45932", + "right": "pull_request:46891", + "accept": false, + "reason": "Both touch the CI caller workflow, but #45932 is a grafana/testing PR with model mapping changes, while #46891 is a PEFT integration test/CI branch update. Same subsystem, different change intent and code path." + }, + { + "left": "pull_request:46100", + "right": "pull_request:46891", + "accept": false, + "reason": "Only shared theme is testing. #46100 is a broad test migration across vision models; #46891 is specifically about PEFT integration CI and related integration code. 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": 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": "db55ce1e3880d9c4c7b2baae9c0df521c6a1beb5b018d0aa9f6e2ec1cf1f92fc", + "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, while PR 46269 is a broad distributed-branch base that happens to touch overlapping distributed files but represents a much larger, different change set.", + "confidence": 0.92, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 46126 is the more canonical artifact for this cluster because it describes a specific, coherent feature/change with a clear scope; PR 46269 is a broad base branch with many unrelated distributed updates and CI/config churn.", + "best_issue_reason": null, + "best_pr_reason": "PR 46126 is the best representative PR here: it targets one concrete code path and could be reviewed/merged on its own, unlike PR 46269 which is too sprawling to serve as the canonical duplicate target.", + "soft_edge_verdicts": [ + { + "left": "pull_request:46126", + "right": "pull_request:46269", + "accept": false, + "reason": "Reject: the overlap is only at the level of shared distributed filenames. PR 46126 implements EP + Trainer integration, whereas PR 46269 is a wide distributed branch base touching CI, FSDP, modeling, and multiple integration paths; they are not the same underlying fix/change and would not plausibly be merged as one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 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": "8f152e3bcec53dc722aa5106a0a0c4c6688499307cc2b0c643fdef4a16010dcb", + "component_budget": { + "node_count": 10, + "item_count": 10, + "soft_pair_count": 29, + "serialized_chars": 25931, + "estimated_input_tokens": 6483, + "estimated_eval_tokens": 13222 + }, + "budget": { + "node_count": 10, + "item_count": 10, + "soft_pair_count": 29, + "serialized_chars": 25931, + "estimated_input_tokens": 6483, + "estimated_eval_tokens": 13222 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are mostly unrelated workflow-maintenance PRs that only look similar at the title level. Even the repeated pr-ci-security-gate.yml changes address different concerns, so there is no safe duplicate merge here.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": "No canonical duplicate PR stands out because the cluster is heterogeneous. If one representative is needed, PR 46554 is the most substantive and central security-gate workflow change, but it is not a duplicate of the others.", + "best_issue_reason": null, + "best_pr_reason": "PR 46554 is the best single representative because it makes the core pr-ci-security-gate workflow logic change and has the strongest overlap with the other security-gate variants, though the group is not a true duplicate set.", + "soft_edge_verdicts": [ + { + "left": "pull_request:46138", + "right": "pull_request:46554", + "accept": false, + "reason": "Different workflow files and different changes: self-comment CI stdin handling vs security-gate checks." + }, + { + "left": "pull_request:46138", + "right": "pull_request:46558", + "accept": false, + "reason": "Different workflow files; one updates comment parsing, the other changes PR CI gate permissions/approval flow." + }, + { + "left": "pull_request:46138", + "right": "pull_request:46563", + "accept": false, + "reason": "Unrelated workflows: self-comment CI command plumbing vs adding safety notes to the security gate reusable workflow." + }, + { + "left": "pull_request:46138", + "right": "pull_request:46716", + "accept": false, + "reason": "Different workflow files and unrelated changes; not the same underlying bug or fix." + }, + { + "left": "pull_request:46138", + "right": "pull_request:46504", + "accept": false, + "reason": "Different workflow files and different behavior; only a broad CI-workflow similarity." + }, + { + "left": "pull_request:46432", + "right": "pull_request:46503", + "accept": false, + "reason": "SSH runner hardening vs benchmark trigger logic; same subsystem is too broad to merge." + }, + { + "left": "pull_request:46432", + "right": "pull_request:46501", + "accept": false, + "reason": "Unrelated workflow changes in different files: SSH runner hardening vs anti-slop permissions layout." + }, + { + "left": "pull_request:46432", + "right": "pull_request:46502", + "accept": false, + "reason": "Different workflows and problems; SSH runner secret handling is not the same as reviewer assignment checkout behavior." + }, + { + "left": "pull_request:46432", + "right": "pull_request:46554", + "accept": false, + "reason": "Different files and distinct fixes; SSH runner environment cleanup is unrelated to security-gate logic." + }, + { + "left": "pull_request:46432", + "right": "pull_request:46558", + "accept": false, + "reason": "SSH runner hardening is unrelated to PR CI security-gate permissions/approval changes." + }, + { + "left": "pull_request:46501", + "right": "pull_request:46503", + "accept": false, + "reason": "Different workflow files and different triggers/permissions; similar title only." + }, + { + "left": "pull_request:46501", + "right": "pull_request:46502", + "accept": false, + "reason": "Anti-slop permissions and reviewer checkout are separate workflow concerns, not one bug." + }, + { + "left": "pull_request:46501", + "right": "pull_request:46504", + "accept": false, + "reason": "Different workflow files and distinct behavior changes; not mergeable as one fix." + }, + { + "left": "pull_request:46501", + "right": "pull_request:46554", + "accept": false, + "reason": "Different workflow files; anti-slop permissions vs security-gate code-path changes." + }, + { + "left": "pull_request:46502", + "right": "pull_request:46503", + "accept": false, + "reason": "Reviewer checkout ref and benchmark trigger conditions are unrelated." + }, + { + "left": "pull_request:46502", + "right": "pull_request:46504", + "accept": false, + "reason": "Different workflow files and unrelated changes; only superficial CI similarity." + }, + { + "left": "pull_request:46502", + "right": "pull_request:46554", + "accept": false, + "reason": "Assign-reviewers checkout behavior is unrelated to security-gate logic." + }, + { + "left": "pull_request:46503", + "right": "pull_request:46504", + "accept": false, + "reason": "Benchmark dispatch trigger and build-ci docker commit parsing are distinct workflow fixes." + }, + { + "left": "pull_request:46503", + "right": "pull_request:46554", + "accept": false, + "reason": "Benchmark runner trigger logic is unrelated to PR CI security gate changes." + }, + { + "left": "pull_request:46504", + "right": "pull_request:46554", + "accept": false, + "reason": "Build-ci tag selection from commit messages is unrelated to security-gate workflow hardening." + }, + { + "left": "pull_request:46504", + "right": "pull_request:46558", + "accept": false, + "reason": "Different workflow files and fixes; no shared underlying defect." + }, + { + "left": "pull_request:46554", + "right": "pull_request:46558", + "accept": false, + "reason": "Same file, but different concrete changes: one adds checkout and file-change checks, the other adjusts permissions and approval gating." + }, + { + "left": "pull_request:46554", + "right": "pull_request:46563", + "accept": false, + "reason": "Same workflow file, but one is security logic and the other is explanatory comments only." + }, + { + "left": "pull_request:46554", + "right": "pull_request:46716", + "accept": false, + "reason": "Different security-gate edits: one adds checks, the other changes secrets inheritance; not the same change." + }, + { + "left": "pull_request:46558", + "right": "pull_request:46563", + "accept": false, + "reason": "Permissions/approval logic vs comment-only update; not the same underlying fix." + }, + { + "left": "pull_request:46558", + "right": "pull_request:46716", + "accept": false, + "reason": "Different changes in the same workflow file: approval/permissions vs secrets handling." + }, + { + "left": "pull_request:46563", + "right": "pull_request:46716", + "accept": false, + "reason": "Comment-only documentation change vs functional secrets scoping change." + }, + { + "left": "pull_request:46504", + "right": "pull_request:46563", + "accept": false, + "reason": "Different workflow files and unrelated changes." + }, + { + "left": "pull_request:46504", + "right": "pull_request:46716", + "accept": false, + "reason": "Different workflow files and unrelated fixes; not duplicate PRs." + }, + { + "left": "pull_request:46558", + "right": "pull_request:46563", + "accept": false, + "reason": "Different functional area in the same file; not the same concrete problem." + }, + { + "left": "pull_request:46558", + "right": "pull_request:46716", + "accept": false, + "reason": "Permissions/approval gating vs secrets scoping are distinct 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": 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": "8924788f6a3cf7debd8f6938790d69f7ffcd4ae233d2678bcc306843be5a83f4", + "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-file inspection to avoid direct open() calls, but they address different code paths in different modules and are not duplicates.", + "confidence": 0.91, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #46152 is the broader representative of the shared theme: it introduces a reusable source-path resolution helper in import_utils to replace file-open based self-inspection. PR #46207 is a separate, narrower fix in modeling_utils.", + "best_issue_reason": null, + "best_pr_reason": "PR #46152 is the best cluster representative because it captures the more general refactor pattern and the core change in the shared code-reading approach, while PR #46207 is a specific follow-up in a different subsystem.", + "soft_edge_verdicts": [ + { + "left": "pull_request:46152", + "right": "pull_request:46207", + "accept": false, + "reason": "Same high-level goal, but different concrete functions and files with different implementation paths; they do not look mergeable into one PR or like the same underlying bug." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "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": "983f6eb82adc9cde687cb700fc7cc81aac57f8d404c3e3451bf20c605bd6dbeb", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4271, + "estimated_input_tokens": 1068, + "estimated_eval_tokens": 2392 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4271, + "estimated_input_tokens": 1068, + "estimated_eval_tokens": 2392 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both items are security-focused PRs that harden path handling against traversal, but they target different untrusted inputs and different code paths: sharded checkpoint weight_map vs. transformers_weights config. Similar theme, not the same underlying fix.", + "confidence": 0.92, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #46281 is the more concrete and self-contained fix: it addresses a specific loader path traversal in sharded-checkpoint resolution from the index weight_map, with a clear containment check and targeted test coverage.", + "best_issue_reason": null, + "best_pr_reason": "PR #46281 is the best representative of the cluster because it has the clearest, most explicit bug surface and a direct fix for a specific code path; #46890 is related security hardening but for a distinct input path.", + "soft_edge_verdicts": [ + { + "left": "pull_request:46281", + "right": "pull_request:46890", + "accept": false, + "reason": "Both are path-traversal mitigations, but they fix different sources of untrusted filenames in different loaders. They do not appear to be the same bug or a single change that 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": 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": "16bbd7e8b1dac8d6148a7dec9a38a54f938fda6e5d43dc199242550340da243e", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4851, + "estimated_input_tokens": 1213, + "estimated_eval_tokens": 2682 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4851, + "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 in the right side of the sliding-window mask (`-dist <= right_window_size` instead of `<`); PR 46546 is the more complete change because it also updates Gemma4 and adds tests.", + "confidence": 0.93, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 46546 is the best canonical PR because it contains the full fix across all affected implementations and includes test coverage, while PR 46310 is a narrower partial fix.", + "best_issue_reason": null, + "best_pr_reason": "PR 46546 is the strongest representative of the change: broader coverage, same core code fix, and explicit tests make it the more complete/merge-ready patch.", + "soft_edge_verdicts": [ + { + "left": "pull_request:46310", + "right": "pull_request:46546", + "accept": true, + "reason": "Same underlying boundary bug in the sliding-window right mask, with identical fix in the shared Moonshine streaming code path; 46546 is a superset 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": 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": "3c005043df5839de116ea965e0c35267380c4722ace45042be3c834038934941", + "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 all distinct: three separate model additions, one Gemma sliding-window masking bugfix, one continuous-batching multimodal feature, and one deprecation cleanup. The soft links are broad-topic similarity only, not duplicate code changes.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 46425 is the best anchor for the cluster because it is the most central and broadly model-integration oriented PR, but it still addresses a different model family than the others.", + "best_issue_reason": null, + "best_pr_reason": "PR 46425 is the best overall representative of the cluster\u2019s loose similarity pattern, since it is the hub with the most soft connections and the broadest feature-addition scope.", + "soft_edge_verdicts": [ + { + "left": "pull_request:46376", + "right": "pull_request:46417", + "accept": false, + "reason": "Different features for different model families: continuous batching multimodal support vs. HunYuan VL model support." + }, + { + "left": "pull_request:46376", + "right": "pull_request:46425", + "accept": false, + "reason": "Unrelated PRs: continuous batching multimodal support is a runtime feature change; RADIO is a model implementation/docs addition." + }, + { + "left": "pull_request:46425", + "right": "pull_request:46509", + "accept": false, + "reason": "Both add model support, but for different architectures and code paths (RADIO vs. Nemotron Omni), so they are not the same change." + }, + { + "left": "pull_request:46425", + "right": "pull_request:46917", + "accept": false, + "reason": "RADIO model support and deprecated-args cleanup are unrelated changes touching different subsystems." + }, + { + "left": "pull_request:46425", + "right": "pull_request:46850", + "accept": false, + "reason": "Model addition vs. Gemma masking bugfix; they do not fix the same concrete problem or change the same code path." + }, + { + "left": "pull_request:46850", + "right": "pull_request:46917", + "accept": false, + "reason": "Gemma attention-mask boundary fix and deprecated training/image-processor args cleanup are unrelated." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 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": "916d065f7e80368b21f4ac9eb2733ff15fc216c2250c5735967cb41a42c7268c", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4410, + "estimated_input_tokens": 1103, + "estimated_eval_tokens": 2462 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4410, + "estimated_input_tokens": 1103, + "estimated_eval_tokens": 2462 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Two unrelated open PRs: one fixes CodeLlama decode stripping of real leading spaces, the other fixes DeepSeek-Coder/DeepSeek-LLM auto-tokenizer class selection. They should not be treated as duplicates.", + "confidence": 0.99, + "canonical_issue_reason": null, + "canonical_pr_reason": "No canonical PR: the PRs fix different bugs in different tokenizer pathways (CodeLlama decode behavior vs. auto-tokenizer checkpoint routing).", + "best_issue_reason": null, + "best_pr_reason": "46580 is the broader tokenizer-routing fix, but it is still unrelated to 46574 and should not be merged with it as a duplicate.", + "soft_edge_verdicts": [ + { + "left": "pull_request:46574", + "right": "pull_request:46580", + "accept": false, + "reason": "Reject: 46574 changes CodeLlama decoder setup to preserve leading spaces on decode, while 46580 changes auto-tokenizer checkpoint-prefix logic for DeepSeek models. Different model families, files, and code paths." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 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": "566a339282b9105a15a9cb0ba86dc86ac6750ee494b66b7c69a371fff3c21513", + "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 open Mistral PRs share files but address different changes: one refactors/moves the Mistral converter into an integrations package, while the other adds tekken tokenizer detection/conversion/save utilities. They are related maintenance around the same subsystem, but not the same concrete change.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 46604 is the broader functional change focused on tokenizer detection/conversion/save support; PR 46603 is mainly a code organization refactor.", + "best_issue_reason": null, + "best_pr_reason": "PR 46604 is the best representative of the cluster because it introduces the main user-facing tokenizer capability change, whereas 46603 is an internal move/refactor.", + "soft_edge_verdicts": [ + { + "left": "pull_request:46603", + "right": "pull_request:46604", + "accept": false, + "reason": "Different underlying changes: 46603 relocates MistralConverter into the integrations package, while 46604 implements tekken tokenizer detection/conversion/save utilities. Shared files/subsystem are not enough to merge them as duplicates." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 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-46741-3", + "nodes": [ + "pull_request:46741", + "pull_request:46773", + "pull_request:46781" + ], + "soft_pairs": [ + "pull_request:46741|pull_request:46773", + "pull_request:46741|pull_request:46781" + ], + "prepared_review_unit_hash": "f601a5a2ccc1087797963cafe9e386982baa67b5ba775adcb1c9c7da68eb7a55", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 8456, + "estimated_input_tokens": 2114, + "estimated_eval_tokens": 4484 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 8456, + "estimated_input_tokens": 2114, + "estimated_eval_tokens": 4484 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are three distinct PRs around linear-attention/cache behavior, but they fix different bugs: one for chunked-prefill/speculative decoding state reuse, one for padding-mask application at batch size 1, and one for QuantizedCache handling on hybrid models. None are true duplicates of each other.", + "confidence": 0.95, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #46741 is the best canonical PR because it is the broadest concrete fix in the shared model-cache area and the title/diff clearly center on a single bug family: incorrect Mamba2 chunked-prefill/speculative-decoding behavior across several affected models.", + "best_issue_reason": null, + "best_pr_reason": "PR #46741 is the strongest representative of the cluster since it addresses a specific forward/cache-path bug with clear model-level impact and the most substantial shared code changes.", + "soft_edge_verdicts": [ + { + "left": "pull_request:46741", + "right": "pull_request:46773", + "accept": false, + "reason": "Both touch similar linear-attention model files, but the bugs are different: #46741 changes cache/state reuse for chunked-prefill/speculative decoding, while #46773 fixes padding-mask application being skipped for batch size 1. Different code paths and failure modes." + }, + { + "left": "pull_request:46741", + "right": "pull_request:46781", + "accept": false, + "reason": "#46741 is a model forward/cache bug fix across several architectures, whereas #46781 fixes QuantizedCache construction for hybrid SSM+attention models in cache_utils. They are related to caches broadly but not the same underlying change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 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": "1b4af156e3127d63c0efe87c8cbbe60194522d25ba57d9e4c093528e2c16cb0d", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 7713, + "estimated_input_tokens": 1929, + "estimated_eval_tokens": 4114 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 7713, + "estimated_input_tokens": 1929, + "estimated_eval_tokens": 4114 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Two of the PRs are model-specific variants of the same training-loss double-shift fix, while the third is a bundled PR that contains both fixes. The only clear non-duplicate pair is CohereASR vs Florence2.", + "confidence": 0.76, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #46916 is the broadest representative: it contains the same concrete loss/shifting fix for CohereASR and Florence2, so it best captures the cluster\u2019s combined change.", + "best_issue_reason": null, + "best_pr_reason": "PR #46916 is the best single representative because it subsumes the narrower CohereASR-only and Florence2-only fixes and points to both affected code paths.", + "soft_edge_verdicts": [ + { + "left": "pull_request:46895", + "right": "pull_request:46898", + "accept": false, + "reason": "Different models and different issue targets; they share the symptom pattern but not the same concrete code path or change." + }, + { + "left": "pull_request:46895", + "right": "pull_request:46916", + "accept": true, + "reason": "46916 includes the same CohereASR training-loss double-shift fix in the same files; it is a superset PR, but this edge still matches the same underlying CohereASR bug/change." + }, + { + "left": "pull_request:46898", + "right": "pull_request:46916", + "accept": true, + "reason": "46916 includes the same Florence2 training-loss double-shift fix in the same files; despite bundling CohereASR too, this is the same underlying Florence2 bug/change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 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": "c668f32b49e267942925b11311db62ebd69e17449d695d494e85c96d8fa14584", + "component_budget": { + "node_count": 5, + "item_count": 5, + "soft_pair_count": 10, + "serialized_chars": 11925, + "estimated_input_tokens": 2982, + "estimated_eval_tokens": 6220 + }, + "budget": { + "node_count": 5, + "item_count": 5, + "soft_pair_count": 10, + "serialized_chars": 11925, + "estimated_input_tokens": 2982, + "estimated_eval_tokens": 6220 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These PRs are all model-specific fixes with different root causes: one real code-path bug in EdgeTam, two offload/load-test workarounds for different models, one dtype-cast fix in GLM46V, and one tokenizer registry update for BLIP-2. They are not duplicates of the same underlying change.", + "confidence": 0.95, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 46907 is the strongest representative because it fixes an actual runtime bug in the model forward path, while the others are unrelated model-specific test or registration fixes.", + "best_issue_reason": null, + "best_pr_reason": "PR 46907 is the best single PR to represent the cluster since it contains a concrete code fix rather than a test-only workaround and is clearly distinct from the other PRs.", + "soft_edge_verdicts": [ + { + "left": "pull_request:46907", + "right": "pull_request:46927", + "accept": false, + "reason": "Different bugs in different models: EdgeTam drops processor-only kwargs, while GLM46V casts input_ids to long." + }, + { + "left": "pull_request:46907", + "right": "pull_request:46926", + "accept": false, + "reason": "EdgeTam forward-path fix vs MiniMax-M3-VL test offload_folder workaround; not the same underlying issue." + }, + { + "left": "pull_request:46907", + "right": "pull_request:46908", + "accept": false, + "reason": "Different models and fixes: EdgeTam code-path cleanup vs DeepSeek-V3.2 test offload_folder workaround." + }, + { + "left": "pull_request:46907", + "right": "pull_request:46928", + "accept": false, + "reason": "EdgeTam runtime fix vs BLIP-2 tokenizer auto-registration update; unrelated changes." + }, + { + "left": "pull_request:46908", + "right": "pull_request:46926", + "accept": false, + "reason": "Both are load-related test changes, but they target different models and different load failures, so they are not the same bug." + }, + { + "left": "pull_request:46908", + "right": "pull_request:46927", + "accept": false, + "reason": "DeepSeek test offload workaround and GLM46V input dtype fix are unrelated." + }, + { + "left": "pull_request:46926", + "right": "pull_request:46927", + "accept": false, + "reason": "MiniMax offload_folder test workaround vs GLM46V forward input cast; different code paths and failures." + }, + { + "left": "pull_request:46926", + "right": "pull_request:46928", + "accept": false, + "reason": "MiniMax load workaround and BLIP-2 tokenizer registry change are not the same underlying change." + }, + { + "left": "pull_request:46927", + "right": "pull_request:46928", + "accept": false, + "reason": "GLM46V input_ids casting fix vs BLIP-2 tokenizer registration update; unrelated model fixes." + }, + { + "left": "pull_request:46908", + "right": "pull_request:46928", + "accept": false, + "reason": "DeepSeek load_error workaround vs BLIP-2 tokenizer registry update; different models and different bugs." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + } + ] +} diff --git a/snapshots/20260627T060016Z/analysis-runs/hybrid-model-20260627t060016z/manifest.json b/snapshots/20260627T060016Z/analysis-runs/hybrid-model-20260627t060016z/manifest.json new file mode 100644 index 0000000000000000000000000000000000000000..33d2b27c8ff16899b5ff218c7d403384a1dd3c05 --- /dev/null +++ b/snapshots/20260627T060016Z/analysis-runs/hybrid-model-20260627t060016z/manifest.json @@ -0,0 +1,14 @@ +{ + "analysis_id": "hybrid-model-20260627t060016z", + "artifacts": { + "hybrid": "snapshots/20260627T060016Z/analysis-runs/hybrid-model-20260627t060016z/analysis-report-hybrid.json", + "hybrid_reviews": "snapshots/20260627T060016Z/analysis-runs/hybrid-model-20260627t060016z/analysis-report-hybrid.llm-reviews.json" + }, + "channel": "canonical", + "model": null, + "published_at": "2026-06-27T06:09:30Z", + "repo": "huggingface/transformers", + "schema_version": 1, + "snapshot_id": "20260627T060016Z", + "variant": "hybrid" +} diff --git a/snapshots/20260627T060016Z/manifest.json b/snapshots/20260627T060016Z/manifest.json index 932e01aa58c6e6a054d06c6113f3a53e9a217df3..0eb048dbd1af40df65831c96bbd9bfbc8ec829f8 100644 --- a/snapshots/20260627T060016Z/manifest.json +++ b/snapshots/20260627T060016Z/manifest.json @@ -32,6 +32,24 @@ "timeline_events": 61 }, "extracted_at": "2026-06-27T06:00:16Z", + "published_analysis": { + "canonical_analysis_id": "hybrid-model-20260627t060016z", + "runs": { + "hybrid-model-20260627t060016z": { + "analysis_id": "hybrid-model-20260627t060016z", + "artifacts": { + "hybrid": "snapshots/20260627T060016Z/analysis-runs/hybrid-model-20260627t060016z/analysis-report-hybrid.json", + "hybrid_reviews": "snapshots/20260627T060016Z/analysis-runs/hybrid-model-20260627t060016z/analysis-report-hybrid.llm-reviews.json" + }, + "channel": "canonical", + "manifest_path": "snapshots/20260627T060016Z/analysis-runs/hybrid-model-20260627t060016z/manifest.json", + "model": null, + "published_at": "2026-06-27T06:09:30Z", + "variant": "hybrid" + } + }, + "schema_version": 1 + }, "repo": "huggingface/transformers", "snapshot_id": "20260627T060016Z", "watermark": {