diff --git a/analysis/current/analysis-report-hybrid.json b/analysis/current/analysis-report-hybrid.json index 9b582ad5d99137487dcf63c7c1cc9a971e73d349..859fda2246fb29c427c38b731e0c8273a3d23f82 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": "20260616T060022Z", - "generated_at": "2026-06-16T06:09:50Z", + "snapshot_id": "20260616T120046Z", + "generated_at": "2026-06-16T12:10:41Z", "evidence_quality": "full", "llm_enrichment": true, "meta_bugs": [ @@ -1740,31 +1740,53 @@ ] }, { - "cluster_id": "cluster-46562-2", + "cluster_id": "cluster-46554-2", "summary": "Cluster of 2 related pull requests linked by soft_similarity.", "status": "open", "confidence": 0.5, "canonical_issue_number": null, - "canonical_pr_number": 46562, + "canonical_pr_number": 46558, "issue_numbers": [], "pr_numbers": [ - 46562, - 46658 + 46554, + 46558 ], "evidence_types": [ "soft_similarity" ], "pr_comparisons": [ { - "left_pr_number": 46562, - "right_pr_number": 46658, - "code_similarity": 0.127, - "size_similarity": 0.556, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.107, - "shared_filenames": [], - "shared_file_areas": [] + "left_pr_number": 46554, + "right_pr_number": 46558, + "code_similarity": 0.433, + "size_similarity": 0.429, + "file_overlap": 1.0, + "area_overlap": 0.1, + "patch_similarity": 0.081, + "shared_filenames": [ + ".github/workflows/pr-ci-security-gate.yml" + ], + "shared_file_areas": [ + { + "filename": ".github/workflows/pr-ci-security-gate.yml", + "left_ranges": [ + [ + 16, + 45 + ] + ], + "right_ranges": [ + [ + 13, + 19 + ], + [ + 57, + 63 + ] + ] + } + ] } ] }, @@ -2154,13 +2176,13 @@ "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-46562-2", - "canonical_pr_number": 46562, + "cluster_id": "cluster-46554-2", + "canonical_pr_number": 46558, "duplicate_pr_numbers": [ - 46658 + 46554 ], "target_issue_number": null, - "reason": "PRs in cluster-46562-2 are treated as duplicates because they share soft_similarity evidence." + "reason": "PRs in cluster-46554-2 are treated as duplicates because they share soft_similarity evidence." } ], "best_issue": { diff --git a/analysis/current/analysis-report-hybrid.llm-reviews.json b/analysis/current/analysis-report-hybrid.llm-reviews.json index 9c6003bff47abbb8495e49ba057bd54544ef1bf3..b1632b8116de79f556308e968f5f9834e9067490 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": "20260616T060022Z", - "generated_at": "2026-06-16T06:09:50Z", + "snapshot_id": "20260616T120046Z", + "generated_at": "2026-06-16T12:10:41Z", "model": "gpt-5.4-mini?service_tier=flex", "reviews": [ { @@ -130,427 +130,367 @@ "issue:45237|issue:46015", "issue:43039|issue:44805" ], - "prepared_review_unit_hash": "4cf30975fd20e026d89b1298ee0d9c274aea9673c1e589e66f370d35c70b8773", + "prepared_review_unit_hash": "88e6572aee9f47660b4bbf1c0e5799069bce465f5790a2a11f0e63a28694b8f0", "component_budget": { "node_count": 473, "item_count": 473, "soft_pair_count": 1738, - "serialized_chars": 901383, - "estimated_input_tokens": 225346, - "estimated_eval_tokens": 450948 + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 62, - "serialized_chars": 45726, - "estimated_input_tokens": 11432, - "estimated_eval_tokens": 23120 + "serialized_chars": 45736, + "estimated_input_tokens": 11434, + "estimated_eval_tokens": 23124 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster still breaks into a few real duplicate subclusters, mainly the DeepSpeed ZeRO-3/BERT index-error pair and the torch 2.4.1 compatibility pair. The Qwen3.5 MoE rotary/offload issues are related, but not established as the same bug, so they should stay separate.", - "confidence": 0.86, - "canonical_issue_reason": "issue:43596 is the best canonical issue because it cleanly represents the DeepSpeed ZeRO-3 BERT out-of-bounds index failure and matches the strongest duplicate pair in the cluster.", + "summary": "This cluster is mostly heterogeneous. I only see a few true duplicate mini-clusters: the DeepSpeed ZeRO-3 Bert IndexError pair, the Qwen3.5Moe rotary-embedding/offload pair, and the torch 2.4.1 incompatibility pair. Most other soft edges connect broadly similar but clearly different bugs, models, or docs issues.", + "confidence": 0.81, + "canonical_issue_reason": "issue:43596 is the cleanest concrete bug report among the clearly duplicated subgroup: it describes the same ZeRO-3 init IndexError (size 0 / index 0 out of bounds) as issue:43638, with the older issue being the better representative.", "canonical_pr_reason": null, - "best_issue_reason": "issue:43596 is the best overall issue to anchor triage: it is concrete, reproducible, and has the clearest true duplicate relationship to issue:43638.", + "best_issue_reason": "issue:43596 is the best issue representative for the only strong duplicate subgroup in this set because it is specific, reproducible, and directly matches the paired ZeRO-3/Bert failure. The rest of the cluster is too mixed to justify a single global canonical across all items.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:44322", "right": "issue:45464", "accept": false, - "reason": "Different Qwen3.5 failures: missing config attribute vs streaming chat/completions breakage." + "reason": "Different problems: missing Qwen3.5 config field vs streaming chat/completions failure." }, { "left": "issue:45237", "right": "issue:45685", "accept": false, - "reason": "Different concrete bugs: AMD GPU failure vs MPS histogram op missing for Int." + "reason": "Different backends and failures: AMD GPU support vs MPS histogram not implemented for Int." }, { "left": "issue:43716", "right": "issue:45237", "accept": false, - "reason": "Different model/runtime problems: image preprocessor dtype mismatch vs GPT-OSS on AMD GPUs." + "reason": "Different root causes: dtype mismatch in image preprocessing vs AMD GPU execution failure." }, { "left": "issue:43828", "right": "issue:45237", "accept": false, - "reason": "Both are dtype-related, but they affect different models and code paths." + "reason": "Different issues: autocast dtype mismatch vs AMD GPU-specific model failure." }, { "left": "issue:45237", "right": "issue:46468", "accept": false, - "reason": "Unrelated failures: GPT-OSS on AMD GPUs vs Nemotron example breakage." + "reason": "Different artifacts: AMD GPU incompatibility vs broken Nemotron example." }, { "left": "issue:45237", "right": "issue:45507", "accept": false, - "reason": "Different concrete bugs: AMD GPU failure vs GraniteMoEHybrid invalid method call." + "reason": "Different bugs: AMD GPU failure vs invalid method call in GraniteMoEHybrid." }, { "left": "issue:30277", "right": "issue:35545", "accept": false, - "reason": "Different subsystems: DeepSpeed collective mismatch vs ONNX export error." + "reason": "Different code paths: DeepSpeed collective mismatch vs ONNX export error." }, { "left": "issue:35545", "right": "issue:42915", "accept": false, - "reason": "Different models and failures: ModernBERT export vs Qwen3Moe FineGrainedFP8Config." + "reason": "Different model/export issue vs FP8 config failure." }, { "left": "issue:35545", "right": "issue:41720", "accept": false, - "reason": "Different issues: ONNX export failure vs Qwen3 auto device-map CUDA assert." + "reason": "Different failures: ONNX export vs auto device mapping cuda assert." }, { "left": "issue:45901", "right": "issue:46015", "accept": false, - "reason": "Unrelated: table QA crash vs outdated RAG docs." + "reason": "Different topics: table-question-answering crash vs outdated RAG examples." }, { "left": "issue:44509", "right": "issue:44512", "accept": false, - "reason": "Both are stale-doc issues for v5 removals, but they target different removed items and pages." + "reason": "Both are docs cleanup, but they target different removed items and pages." }, { "left": "issue:35545", "right": "issue:36010", "accept": false, - "reason": "Different failures: export to ONNX vs GenerationMixin import breakage." + "reason": "Different import/export issues with no shared concrete bug." }, { "left": "issue:44918", "right": "issue:46165", "accept": false, - "reason": "Different code paths: TRL SFT embedding unpacking vs resume/trust-remote-code training issue." + "reason": "Different failures: Qwen3.5 input-embedding unpacking vs resume/trust_remote_code issue." }, { "left": "issue:44805", "right": "issue:45823", "accept": false, - "reason": "Different failures: mask shape mismatch vs Gemma4 device mismatch in forward." + "reason": "Different problems: mask shape IndexError vs Gemma4 PLE device mismatch." }, { "left": "issue:43638", "right": "issue:44805", "accept": false, - "reason": "Same broad error class, but different concrete bugs and models." + "reason": "Different error signatures and model paths: ZeRO-3 Bert init vs generic mask shape mismatch." }, { "left": "issue:43638", "right": "issue:46165", "accept": false, - "reason": "Unrelated: ZeRO-3 BERT index error vs resuming local model training with remote code." + "reason": "Different training/resume issue vs ZeRO-3 Bert IndexError." }, { "left": "issue:43596", "right": "issue:44805", "accept": false, - "reason": "Different concrete error messages and model paths." + "reason": "Different bugs: ZeRO-3 Bert IndexError vs mask shape mismatch." }, { "left": "issue:44805", "right": "issue:46294", "accept": false, - "reason": "Both involve shape mismatches, but the models and mechanisms differ." + "reason": "Different multimodal/video failure vs generic mask shape mismatch." }, { "left": "issue:44805", "right": "issue:45161", "accept": false, - "reason": "Different failures: tensor mask mismatch vs GPT-OSS MoE TP issue." + "reason": "Different problems: mask shape IndexError vs GPT-OSS MoE tensor-parallel issue." }, { "left": "issue:46165", "right": "issue:46468", "accept": false, - "reason": "Different issues: remote-code resume failure vs Nemotron example breakage." + "reason": "Different artifacts: resume/trust_remote_code issue vs broken Nemotron example." }, { "left": "issue:44560", "right": "issue:46165", "accept": false, - "reason": "Different failures: Qwen3-VL video StopIteration vs training resume/remote-code issue." + "reason": "Different failures: video StopIteration vs training resume problem." }, { "left": "issue:43596", "right": "issue:46165", "accept": false, - "reason": "Unrelated bugs." + "reason": "Different failures: ZeRO-3 Bert IndexError vs resume/trust_remote_code issue." }, { "left": "issue:34634", "right": "issue:35545", "accept": false, - "reason": "Different models and domains: BarkProcessor preset handling vs ModernBERT export." + "reason": "Different model subsystem bugs: BarkProcessor voice preset vs ONNX export." }, { "left": "issue:30277", "right": "issue:43638", "accept": false, - "reason": "Both involve DeepSpeed, but the concrete failures and models are different." + "reason": "Both involve DeepSpeed/ZeRO, but the concrete failures and models differ." }, { "left": "issue:43596", "right": "issue:43638", "accept": true, - "reason": "Same DeepSpeed ZeRO-3 BERT index-out-of-bounds bug; one title is just a more specific variant." + "reason": "Same underlying DeepSpeed ZeRO-3 Bert IndexError; only the model variant differs." }, { "left": "issue:43039", "right": "issue:45161", "accept": false, - "reason": "Different failures: Liger kernel cross-entropy call vs GPT-OSS MoE TP issue." + "reason": "Different issues: Liger Kernel cross-entropy path vs GPT-OSS tensor-parallel failure." }, { "left": "issue:43975", "right": "issue:44457", "accept": false, - "reason": "Different bugs: detokenization regression vs LoRA merge/save/reload mismatch." + "reason": "Different bugs: detokenization regression vs LoRA merge/save/load mismatch." }, { "left": "issue:44292", "right": "issue:44457", "accept": false, - "reason": "Unrelated model/runtime errors." + "reason": "Different model/runtime error vs LoRA result inconsistency." }, { "left": "issue:44841", "right": "issue:46661", "accept": false, - "reason": "Different areas: processor failure for Voxtral vs remote-code config_class import error." + "reason": "Different failures: Voxtral processor bug vs remote-code AutoModel config_class error." }, { "left": "issue:34928", "right": "issue:44805", "accept": false, - "reason": "Different failures: FSDP checkpointing recompute mismatch vs tensor mask shape mismatch." + "reason": "Both are tensor/shape errors, but they come from different mechanisms and code paths." }, { "left": "issue:24643", "right": "issue:30277", "accept": false, - "reason": "Both are training/DeepSpeed issues, but not the same failure mode or code path." + "reason": "Different DeepSpeed errors: weight dimensionality vs collective mismatch." }, { "left": "issue:45859", "right": "issue:45902", - "accept": false, - "reason": "Related Qwen3.5 MoE rotary/offload symptoms, but the packet does not establish the same underlying bug." - }, - { - "left": "issue:43638", - "right": "issue:45161", - "accept": false, - "reason": "Different bugs: BERT ZeRO-3 index error vs GPT-OSS only-TP failure." - }, - { - "left": "issue:43975", - "right": "issue:44457", - "accept": false, - "reason": "Different model behaviors: detokenization bug vs LoRA save/reload mismatch." - }, - { - "left": "issue:44292", - "right": "issue:45237", - "accept": false, - "reason": "Different bugs: NVFP4 runtime error vs AMD GPU support failure." - }, - { - "left": "issue:43039", - "right": "issue:43828", - "accept": false, - "reason": "Different models and code paths despite both involving training-time computation concerns." - }, - { - "left": "issue:41762", - "right": "issue:43638", - "accept": false, - "reason": "Both are ZeRO-3 index errors, but they affect different models and are not clearly the same bug." + "accept": true, + "reason": "Same Qwen3.5Moe rotary-embedding/materialization bug under CPU offload/meta-to-empty handling." }, { "left": "issue:43827", "right": "issue:44512", "accept": false, - "reason": "Different stale-doc references: removed pipeline tasks vs removed transformers run command." - }, - { - "left": "issue:43828", - "right": "issue:45685", - "accept": false, - "reason": "Different backend/model issues: autocast dtype mismatch vs MPS Int histogram op missing." - }, - { - "left": "issue:44918", - "right": "issue:45685", - "accept": false, - "reason": "Unrelated: TRL SFT embedding unpacking vs MPS histogram failure." - }, - { - "left": "issue:43039", - "right": "issue:44918", - "accept": false, - "reason": "Different failures and subsystems." - }, - { - "left": "issue:43596", - "right": "issue:45161", - "accept": false, - "reason": "Different bugs: BERT ZeRO-3 index error vs GPT-OSS MoE TP issue." - }, - { - "left": "issue:39290", - "right": "issue:45081", - "accept": false, - "reason": "Different models and failure modes: Gemma3 vLLM config error vs Mistral tokenizer regex patch crash." - }, - { - "left": "issue:45823", - "right": "issue:45874", - "accept": false, - "reason": "Same family (Gemma4), but different concrete bugs: device-mismatch forward vs expensive reverse embedding lookup." + "reason": "Both are docs removals, but they reference different removed commands/tasks." }, { "left": "issue:45800", "right": "issue:46515", "accept": true, - "reason": "Same torch 2.4.1 compatibility problem; the latter is just a less specific restatement." + "reason": "Near-identical torch 2.4.1 incompatibility reports." }, { "left": "issue:43425", "right": "issue:44292", "accept": false, - "reason": "Unrelated: generic torch version incompatibility vs Qwen-3 NVFP4 runtime error." + "reason": "Different issues: Torch version incompatibility vs Qwen-3 NVFP4 runtime failure." }, { "left": "issue:34928", "right": "issue:41720", "accept": false, - "reason": "Different failures: activation-checkpointing shape issue vs Qwen3 auto device-map assert." + "reason": "Different bugs: activation-checkpointing/FSDP recompute mismatch vs Qwen3 auto device-map assert." }, { "left": "issue:43828", "right": "issue:45161", "accept": false, - "reason": "Different problems: autocast dtype mismatch vs GPT-OSS MoE TP failure." + "reason": "Different model/runtime problems: autocast dtype mismatch vs GPT-OSS MoE tensor-parallel failure." }, { "left": "issue:43541", "right": "issue:43716", "accept": false, - "reason": "Different models and failures: torch dynamo grouped_mm error vs image-preprocessor dtype mismatch." + "reason": "Different failures: torch dynamo/grouped_mm tracing vs image preprocessor/model dtype mismatch." }, { "left": "issue:43716", "right": "issue:46468", "accept": false, - "reason": "Unrelated issues." + "reason": "Different bugs: Mistral dtype mismatch vs Nemotron example breakage." }, { "left": "issue:43716", "right": "issue:45685", "accept": false, - "reason": "Different model/runtime problems." + "reason": "Different issues: image/model dtype mismatch vs MPS histogram kernel limitation." }, { "left": "issue:44292", "right": "issue:44918", "accept": false, - "reason": "Different bugs: NVFP4 runtime error vs TRL SFT embedding unpacking." + "reason": "Different bugs: Qwen-3 NVFP4 runtime failure vs TRL SFT embedding unpacking failure." }, { "left": "issue:43716", "right": "issue:44292", "accept": false, - "reason": "Different models and failure modes." + "reason": "Different models and failure modes; not the same bug." }, { "left": "issue:45507", "right": "issue:45685", "accept": false, - "reason": "Different bugs: invalid method call in GraniteMoEHybrid vs MPS histogram implementation gap." + "reason": "Different code paths: invalid method call vs missing MPS kernel support." }, { "left": "issue:44918", "right": "issue:45161", "accept": false, - "reason": "Different areas: TRL SFT embedding unpacking vs GPT-OSS MoE tensor-parallel issue." + "reason": "Different problems: TRL input embedding handling vs GPT-OSS MoE tensor-parallel issue." + }, + { + "left": "issue:43039", + "right": "issue:44918", + "accept": false, + "reason": "Different issues with no shared concrete failure path." }, { "left": "issue:41762", "right": "issue:43638", "accept": false, - "reason": "Different models and not clearly the same ZeRO-3 bug." + "reason": "Both are ZeRO-3 IndexErrors, but the models and triggering paths differ enough to avoid merging." }, { "left": "issue:43827", "right": "issue:44512", "accept": false, - "reason": "Different documentation cleanup targets." + "reason": "Different docs omissions; not the same underlying change." }, { "left": "issue:43828", "right": "issue:45685", "accept": false, - "reason": "Different underlying problems." + "reason": "Different failures: autocast dtype mismatch vs MPS Int histogram limitation." }, { "left": "issue:44918", "right": "issue:45685", "accept": false, - "reason": "Unrelated failures." + "reason": "Different bugs: SFT embedding unpacking vs MPS kernel limitation." }, { "left": "issue:43039", "right": "issue:43828", "accept": false, - "reason": "Different bugs despite both touching training-time computation." + "reason": "Different failures in different models/subsystems." }, { "left": "issue:30277", "right": "issue:36331", "accept": false, - "reason": "Different failures: collective mismatch in ZeRO-3 vs custom trainer loss signature mismatch." + "reason": "Different problems: DeepSpeed collective mismatch vs trainer API signature change." }, { "left": "issue:34928", "right": "issue:36010", "accept": false, - "reason": "Unrelated: checkpointing/FSDP issue vs import error." + "reason": "Different issues: FSDP/checkpointing size mismatch vs GenerationMixin import error." }, { "left": "issue:45237", "right": "issue:46015", "accept": false, - "reason": "Different areas: GPU support bug vs outdated RAG docs." + "reason": "Different artifacts: AMD GPU model failure vs docs/examples issue." }, { - "left": "issue:45823", - "right": "issue:45874", + "left": "issue:43039", + "right": "issue:44805", "accept": false, - "reason": "Same model family, but different concrete forward-path bugs." + "reason": "Different code paths: Liger Kernel cross-entropy vs generic mask shape mismatch." } ] }, "evaluator_result": { - "accept": false, - "feedback": "Mostly grounded, but one soft-edge acceptance is too strong: issue:45859 vs issue:45902 are clearly related to Qwen3.5 MoE rotary-embedding/materialization/offload behavior, yet the packet does not establish they are the same underlying bug. The wording \"same underlying ... problem\" overstates the evidence. The other accepted pairs are well supported, but this one makes the overall triage too aggressive." + "accept": true, + "feedback": "The summary is grounded in the packet and stays conservative. It correctly treats the cluster as mostly heterogeneous while identifying the three plausible duplicate mini-clusters (43596/43638, 45859/45902, 45800/46515). The soft-edge rejections are appropriately cautious, and the accepted pairs look mergeable as the same bugfix." }, "evaluator_used": true, - "retried": true, + "retried": false, "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null @@ -591,13 +531,13 @@ "issue:43901", "issue:43975", "issue:44292", - "issue:44387", "issue:44512", "issue:44560", "issue:44589", + "issue:44805", "issue:44841", "issue:44863", - "issue:45005", + "issue:44918", "issue:45020", "issue:45070", "issue:45084", @@ -658,291 +598,291 @@ "issue:45399|issue:46661", "issue:43572|issue:45070", "issue:45800|issue:45823", - "issue:44387|issue:45005" + "issue:44805|issue:44918" ], - "prepared_review_unit_hash": "a863bca78f5f6f140b485663b93e6fb2ce33c48325897bef058a6a3f26910ea6", + "prepared_review_unit_hash": "60435c75a71799a810b1478b5782fe4db6503a1900e99f765f97c39a310f0463", "component_budget": { "node_count": 473, "item_count": 473, "soft_pair_count": 1738, - "serialized_chars": 901383, - "estimated_input_tokens": 225346, - "estimated_eval_tokens": 450948 + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 41, - "serialized_chars": 39788, - "estimated_input_tokens": 9947, - "estimated_eval_tokens": 20150 + "serialized_chars": 39813, + "estimated_input_tokens": 9954, + "estimated_eval_tokens": 20164 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a mix of unrelated Transformers issues. I only see two clear duplicate families: a DeepSpeed ZeRO-3 empty-tensor IndexError and a v5 AutoTokenizer wrong-token-IDs regression. The rest are too model-/subsystem-specific to merge as duplicates.", - "confidence": 0.91, - "canonical_issue_reason": "issue:45920 is the best issue canonical: it describes the broader tokenizer regression, covers multiple affected models, and is the clearest representative of the underlying bug family.", + "summary": "The cluster is mostly heterogeneous. There are two accepted duplicate subclusters: the ZeRO-3 load/init out-of-bounds reports (41762\u219443596) and the v5 AutoTokenizer token-ID regression reports (45812\u219445920). So 45920 is a strong representative for one cluster, but not the only clear duplicate anchor in the packet.", + "confidence": 0.86, + "canonical_issue_reason": "Use 45920 as the canonical issue for the broad v5 AutoTokenizer token-ID regression cluster: it is open and spans multiple model families. The packet also supports a separate accepted duplicate subcluster around 41762\u219443596, so 45920 should be treated as one representative cluster anchor, not the only one.", "canonical_pr_reason": null, - "best_issue_reason": "issue:45920 is the strongest global issue candidate because it is open, broad in scope, and captures the shared v5 AutoTokenizer regression better than the Granite-only report.", + "best_issue_reason": "45920 is the best overall issue representative because it is open, broad, and central to one of the two clear duplicate clusters. 41762 is also a valid duplicate-cluster anchor, but it is narrower and closed.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43866", "right": "issue:46402", "accept": false, - "reason": "Both are loading/weight issues, but they affect different models and failure modes; not the same bug." + "reason": "Different failures: corrupted checkpoint / missing weights loading, not the same code path." }, { "left": "issue:45800", "right": "issue:46304", "accept": false, - "reason": "Torch/version incompatibility and USE_DISTRIBUTED=0 support breakage are different problems." + "reason": "Both are compatibility complaints, but they point to different breakages and fixes." }, { "left": "issue:43296", "right": "issue:46304", "accept": false, - "reason": "PaddleOCR-VL/vLLM load failure is unrelated to distributed build support." + "reason": "Unrelated: model load failure in vLLM vs a PyTorch distributed-support regression." }, { "left": "issue:41762", "right": "issue:43596", "accept": true, - "reason": "Same underlying ZeRO-3 empty-tensor IndexError during model initialization/loading; model names differ but the code-path failure matches." + "reason": "Same underlying ZeRO-3 loading bug: both report an out-of-bounds index 0 on size 0 during model init/load, with different model classes only." }, { "left": "issue:39290", "right": "issue:41720", "accept": false, - "reason": "Different model/config problems: missing Gemma3 config attribute vs Qwen3 auto device-map CUDA assert." + "reason": "Different model/config and different runtime error; only both mention vLLM/auto mapping loosely." }, { "left": "issue:43901", "right": "issue:44512", "accept": false, - "reason": "One is a docs mismatch, the other is obsolete run-command documentation; not the same issue." + "reason": "Both are docs issues, but they concern different API/documentation changes." }, { "left": "issue:44589", "right": "issue:46304", "accept": false, - "reason": "Float8 storage lookup failure is unrelated to the distributed-support regression." + "reason": "Completely different failures: missing storage type vs distributed build/runtime support." }, { "left": "issue:45237", "right": "issue:45901", "accept": false, - "reason": "Different surfaces entirely: AMD GPU runtime support vs table-question-answering crash." + "reason": "Different subsystems and error classes; no shared bug path." }, { "left": "issue:43039", "right": "issue:44560", "accept": false, - "reason": "Liger Kernel cross-entropy fallback and Qwen3-vl video StopIteration are different code paths." + "reason": "Liger-kernel cross-entropy dispatch vs video chat-template StopIteration are unrelated." }, { "left": "issue:45823", "right": "issue:46294", "accept": false, - "reason": "Both involve multimodal models, but device-mismatch forward failure is not the same as chat-template video token mismatch." + "reason": "Different models and different device/template failures; not the same bug." }, { "left": "issue:43975", "right": "issue:46015", "accept": false, - "reason": "Incorrect detokenization and outdated RAG examples are unrelated." + "reason": "Tokenizer detokenization bug vs RAG docs drift are unrelated." }, { "left": "issue:44841", "right": "issue:45823", "accept": false, - "reason": "Processor failure for Voxtral and Gemma4 PLE device mismatch are different bugs." + "reason": "Different models and distinct forward/device issues." }, { "left": "issue:44863", "right": "issue:46015", "accept": false, - "reason": "NemotronH checkpoint loading and outdated RAG docs do not match." + "reason": "Checkpoint-loading failure vs outdated docs; unrelated." }, { "left": "issue:45020", "right": "issue:46304", "accept": false, - "reason": "Remote-code regressions and USE_DISTRIBUTED=0 support are separate compatibility issues." + "reason": "Remote-code regressions and PyTorch build support are different problem areas." }, { "left": "issue:45812", "right": "issue:45920", "accept": true, - "reason": "Same tokenizer regression: both report wrong token IDs from AutoTokenizer in v5; 45920 is the broader representative report." + "reason": "Same v5 AutoTokenizer regression: wrong token IDs across Granite and other model families, so they describe the same underlying bug." }, { "left": "issue:43866", "right": "issue:44863", "accept": false, - "reason": "Different model checkpoint-loading failures with no shared concrete code path." + "reason": "Both mention loading, but one is a corrupted checkpoint and the other is a model implementation/load issue." }, { "left": "issue:44292", "right": "issue:46015", "accept": false, - "reason": "Qwen-3 NVFP4 runtime error is unrelated to RAG example documentation." + "reason": "Qwen NVFP4 runtime error and RAG docs issues are unrelated." }, { "left": "issue:43872", "right": "issue:46304", "accept": false, - "reason": "bitsandbytes Int8Params constructor mismatch is not the same as the PyTorch distributed support break." + "reason": "bitsandbytes parameter-construction failure is unrelated to distributed build support." }, { "left": "issue:39290", "right": "issue:43531", "accept": false, - "reason": "Both mention sliding window, but the model/config failures are different and not clearly the same bug." + "reason": "Both mention sliding_window, but the models and concrete errors differ." }, { "left": "issue:41093", "right": "issue:43638", "accept": false, - "reason": "Both are IndexErrors, but the tensor-mask mismatch and ZeRO-3 size-0 initialization failure are distinct." + "reason": "Both are IndexErrors, but one is a mask/tensor length mismatch and the other is a zero-size out-of-bounds load error." }, { "left": "issue:43792", "right": "issue:45901", "accept": false, - "reason": "Whisper load failure and table-question-answering crash are unrelated." + "reason": "Whisper load failure vs table-question-answering crash are unrelated." }, { "left": "issue:43854", "right": "issue:43866", "accept": false, - "reason": "GLM-4.7-Flash test loading failure and Ovis2 checkpoint corruption are different model issues." + "reason": "Different model load problems with no shared failure mechanism." }, { "left": "issue:43425", "right": "issue:45070", "accept": false, - "reason": "Torch 2.10 compatibility and pydantic PretrainedConfig regression are different compatibility layers." + "reason": "Torch-version incompatibility vs pydantic PretrainedConfig field break are separate regressions." }, { "left": "issue:43975", "right": "issue:45901", "accept": false, - "reason": "Incorrect detokenization and table-question-answering crashes do not share a concrete bug." + "reason": "Different tasks and error modes; no shared underlying defect." }, { "left": "issue:43596", "right": "issue:45800", "accept": false, - "reason": "ZeRO-3 IndexError and torch/version incompatibility are unrelated." + "reason": "ZeRO-3 loading bug and torch/transformers compatibility issue are unrelated." }, { "left": "issue:43792", "right": "issue:45070", "accept": false, - "reason": "Whisper model loading and PretrainedConfig pydantic breakage are unrelated." + "reason": "Whisper runtime failure is unrelated to the config/pydantic regression." }, { "left": "issue:45362", "right": "issue:45464", "accept": false, - "reason": "Both mention Qwen3.5 and chat, but one is a crash and the other is a streaming API failure; not clearly the same bug." + "reason": "Both are Qwen3.5 chat-related, but the reported failures and call paths differ." }, { "left": "issue:44292", "right": "issue:45901", "accept": false, - "reason": "Different model/runtime failure and task crash; no shared concrete code path." + "reason": "Different model/runtime issue vs table QA crash; no duplicate evidence." }, { "left": "issue:39290", "right": "issue:45823", "accept": false, - "reason": "Different model families and different failure types." + "reason": "Different models and different failure surfaces." }, { "left": "issue:45464", "right": "issue:45685", "accept": false, - "reason": "Streaming chat API failure and MPS histogram operator error are unrelated." + "reason": "Chat streaming failure and MPS histogram kernel error are unrelated." }, { "left": "issue:43541", "right": "issue:45685", "accept": false, - "reason": "Torch dynamo grouped_mm tracing failure is unrelated to the MPS int histogram issue." + "reason": "Different op/runtime failures; no sign of the same bug." }, { "left": "issue:43541", "right": "issue:43828", "accept": false, - "reason": "Grouped_mm tracing failure and autocast dtype mismatch are different runtime bugs." + "reason": "Both involve numerical/tracing issues, but the concrete failures and affected models differ." }, { "left": "issue:36010", "right": "issue:39290", "accept": false, - "reason": "ImportError for GenerationMixin and Gemma3 sliding_window_pattern are unrelated." + "reason": "ImportError for GenerationMixin and Gemma config attribute errors are unrelated." }, { "left": "issue:43404", "right": "issue:43866", "accept": false, - "reason": "Tied lm_head weights in Mistral3 and Ovis2 checkpoint corruption are separate problems." + "reason": "Weight-tying bug is unrelated to a corrupted checkpoint/load failure." }, { "left": "issue:44841", "right": "issue:45084", "accept": false, - "reason": "Processor failure and template-node compilation error are unrelated." + "reason": "Processor failure for Voxtral and template compilation errors do not share a concrete bug." }, { "left": "issue:41093", "right": "issue:41762", "accept": false, - "reason": "Different IndexError shapes and different loading scenarios; not the same bug." + "reason": "Different IndexErrors from different code paths: mask-shape mismatch vs ZeRO-3 load-time out-of-bounds." }, { "left": "issue:43296", "right": "issue:43366", "accept": false, - "reason": "PaddleOCR-VL load failure and GGUF gpt-oss support are unrelated." + "reason": "Different model-support problems and different runtimes." }, { "left": "issue:45399", "right": "issue:46661", "accept": false, - "reason": "Flash-attn fallback logic and remote-code model_class/config_class handling are different issues." + "reason": "Flash-attn fallback gating and remote-code config_class errors are unrelated." }, { "left": "issue:43572", "right": "issue:45070", "accept": false, - "reason": "Missing pad_token_idx in StableLmConfig is not the same as the PretrainedConfig pydantic regression." + "reason": "Missing pad_token_idx and pydantic PretrainedConfig-field regression are different config issues." }, { "left": "issue:45800", "right": "issue:45823", "accept": false, - "reason": "Torch/transformers version incompatibility and Gemma4 device mismatch are different bugs." + "reason": "Torch/version compatibility vs device-map mismatch are unrelated." }, { - "left": "issue:44387", - "right": "issue:45005", + "left": "issue:44805", + "right": "issue:44918", "accept": false, - "reason": "Int4 CUDA reserved-memory/OOM regression and tied-weights translation-model issue are unrelated." + "reason": "Same broad area of model inputs, but different errors and different code paths." } ] }, "evaluator_result": { - "accept": true, - "feedback": "The summary is grounded in the packet and stays conservative. The canonical issue choice is supported by the broader v5 AutoTokenizer regression in issue:45920, and the two accepted duplicate families (issue:41762/issue:43596 and issue:45812/issue:45920) are plausible from the titles and descriptions. The remaining non-duplicates are appropriately treated as model- or subsystem-specific rather than overmerged." + "accept": false, + "feedback": "The 45920 anchor is plausible, but the write-up overstates uniqueness. The packet supports two accepted duplicate subclusters: 41762\u219443596 and 45812\u219445920. So saying 45920 anchors the \"only clear duplicate subcluster\" is not grounded. Please revise the summary to acknowledge both clusters and keep the duplicate claims more conservative." }, "evaluator_used": true, - "retried": false, + "retried": true, "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null @@ -979,12 +919,12 @@ "issue:44291", "issue:44292", "issue:44360", + "issue:44387", "issue:44479", "issue:44492", "issue:44512", "issue:44530", "issue:44589", - "issue:44805", "issue:44863", "issue:44918", "issue:44960", @@ -1009,7 +949,7 @@ "issue:46661" ], "soft_pairs": [ - "issue:44805|issue:44918", + "issue:44387|issue:45005", "issue:43541|issue:45161", "issue:43828|issue:45464", "issue:43366|issue:45464", @@ -1041,226 +981,226 @@ "issue:43975|issue:44292", "issue:44292|issue:45507" ], - "prepared_review_unit_hash": "c822c6849464fd22d1acabd7161119693e01d403d086fa9fe479dc33803de570", + "prepared_review_unit_hash": "9061d7c7ce21e8e9fb4d9b7d5543dec825dff52b3e47925708c87c7312e239d8", "component_budget": { "node_count": 473, "item_count": 473, "soft_pair_count": 1738, - "serialized_chars": 901383, - "estimated_input_tokens": 225346, - "estimated_eval_tokens": 450948 + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 31, - "serialized_chars": 36261, - "estimated_input_tokens": 9066, - "estimated_eval_tokens": 18388 + "serialized_chars": 36232, + "estimated_input_tokens": 9058, + "estimated_eval_tokens": 18372 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly heterogeneous: validation errors, multimodal regressions, tokenizer issues, and model-loading bugs. Only the remote_code/custom-module loading pair looks like a likely duplicate; the rest are distinct root causes.", - "confidence": 0.61, - "canonical_issue_reason": "issue:45698 is the broader, more representative report: it describes the end-user failure mode (`from_pretrained` loading the wrong custom module), while issue:45632 looks like a specific cache-path collision symptom of the same remote-code loading path.", + "summary": "This cluster is mostly a heterogeneous mix of unrelated Transformers bug reports. The only plausible duplicate pair is the remote-code/custom-module path-collision reports (45632/45698); the rest should stay separate.", + "confidence": 0.74, + "canonical_issue_reason": "issue:45632 is the clearest root-cause report: it explicitly describes a trust_remote_code cache-path collision for local models sharing a leaf directory name, which matches the symptom in 45698.", "canonical_pr_reason": null, - "best_issue_reason": "issue:45698 is the best representative issue because it captures the general custom-module loading bug in a way that can subsume the narrower cache-collision symptom.", + "best_issue_reason": "issue:45632 is the best anchor for this cluster because it states the underlying loading/cache collision directly and is the most likely canonical duplicate target for 45698.", "best_pr_reason": null, "soft_edge_verdicts": [ { - "left": "issue:44805", - "right": "issue:44918", + "left": "issue:44387", + "right": "issue:45005", "accept": false, - "reason": "Both mention mask/shape failures, but they appear to involve different code paths and models (generic tensor masking vs Qwen3.5 input embeddings with TRL)." + "reason": "Both concern training/quantization quirks, but one is int4 OOM memory behavior and the other is tied-weights handling; different bugs." }, { "left": "issue:43541", "right": "issue:45161", "accept": false, - "reason": "Different failures: torch dynamo/grouped_mm tracing in Mixtral vs TP not working for GPT-OSS MoE. Same broad MoE area, not the same bug." + "reason": "Grouped_mm tracing errors in Mixtral are unrelated to GPT-OSS tensor-parallel support." }, { "left": "issue:43828", "right": "issue:45464", "accept": false, - "reason": "Autocast dtype mismatch on Phi-tiny-MoE is unrelated to Qwen3.5 streaming chat/completions failure." + "reason": "Phi MoE autocast dtype mismatch is unrelated to Qwen3.5 streaming/chat-completions video inference." }, { "left": "issue:43366", "right": "issue:45464", "accept": false, - "reason": "GPT-OSS GGUF architecture support is a loader/model-compatibility request, not the same issue as Qwen3.5 streaming inference failure." + "reason": "GGUF gpt-oss architecture support is a different feature gap from Qwen3.5 streaming/video failures." }, { "left": "issue:44291", "right": "issue:45070", "accept": false, - "reason": "One is an `_is_hf_initialized` constructor argument error, the other is a pydantic `PretrainedConfig` field break. Different regressions." + "reason": "Unexpected _is_hf_initialized argument in init_empty_weights is a different loading bug than the pydantic PretrainedConfig field regression." }, { "left": "issue:43531", "right": "issue:45070", "accept": false, - "reason": "Qwen3-MoE sliding_window behavior and pydantic `PretrainedConfig` break are unrelated changes." + "reason": "Sliding-window behavior in Qwen3-MoE is unrelated to the PretrainedConfig/pydantic breakage." }, { "left": "issue:43716", "right": "issue:46294", "accept": false, - "reason": "Both are multimodal/video-related, but one is an image preprocessor dtype mismatch and the other is a video chat-template token/feature count mismatch." + "reason": "Both involve multimodal dtype/input issues, but one is image-preprocessor dtype mismatch and the other is video token/feature mismatch; not the same bug." }, { "left": "issue:43541", "right": "issue:44589", "accept": false, - "reason": "Grouped_mm tracing failure and missing Float8 storage are different runtime/kernel issues." + "reason": "Float32 grouped_mm tracing failure and missing Float8 storage type are distinct runtime issues." }, { "left": "issue:45020", "right": "issue:45800", "accept": false, - "reason": "Remote-code breakage in recent transformers and torch/transformers version incompatibility are different problems." + "reason": "Generic remote_code breakage is broader than the specific torch 2.4.1 vs transformers 5.8.0 incompatibility report." }, { "left": "issue:41093", "right": "issue:43596", "accept": false, - "reason": "Both are IndexErrors, but one is a mask/tensor length mismatch and the other is a DeepSpeed ZeRO-3 init out-of-bounds error." + "reason": "Mask-length mismatch and zero-sized tensor index error are different failure modes with no clear shared code path." }, { "left": "issue:44479", "right": "issue:46294", "accept": false, - "reason": "Both involve Qwen video input, but one is a broad v5.3.0 regression across several model families and the other is a specific apply_chat_template mismatch on Qwen3.5-9B." + "reason": "Both mention Qwen video input regressions, but 44479 is a broader v5.3.0 regression across several models while 46294 is a specific chat-template token/feature mismatch." }, { "left": "issue:43054", "right": "issue:43994", "accept": false, - "reason": "SigLIP2 text embedding quality and SigLIP2 AutoModel/pipeline nonsense outputs are related model-family symptoms, but not the same concrete bug." + "reason": "Siglip2 embedding quality and nonsensical AutoModel/pipeline output are related in topic, but not clearly the same concrete bug." }, { "left": "issue:45759", "right": "issue:45902", "accept": false, - "reason": "`from_config` not unwrapping `text_config` and uninitialized `inv_freq` after `to_empty` are separate Qwen multimodal issues." + "reason": "Text_config unwrapping in from_config and uninitialized inv_freq after meta\u2192to_empty(cuda) are separate multimodal initialization issues." }, { "left": "issue:46640", "right": "issue:46661", "accept": false, - "reason": "Flash Attention 2 issue and remote-code `config_class` AttributeError are unrelated loader/back-end failures." + "reason": "Flash Attention 2 integration failure is unrelated to remote-code models missing config_class." }, { "left": "issue:39186", "right": "issue:44918", "accept": false, - "reason": "FSDP 2-D weight constraint and Qwen3.5 input embedding unpacking are different bugs." + "reason": "FSDP 2-D weight error and Qwen3.5 embedding unpacking in TRL are different model/training issues." }, { "left": "issue:43716", "right": "issue:44918", "accept": false, - "reason": "Mistral-3 image preprocessor dtype mismatch and Qwen3.5/TR L input embedding unpacking do not share the same failure path." + "reason": "Image-preprocessor dtype mismatch is not the same as TRL SFT unpacking of Qwen3.5 input embeddings." }, { "left": "issue:40990", "right": "issue:44960", "accept": false, - "reason": "High perplexity on GPT-OSS and GLM5 are model-specific quality/behavior reports, not the same issue." + "reason": "High perplexity on gpt-oss-20b and GLM5 are unrelated model-specific reports." }, { "left": "issue:45632", "right": "issue:45698", "accept": true, - "reason": "Both describe the same remote_code/custom-module loading problem: a cache/path resolution bug causing `from_pretrained` to pick the wrong module, with 45632 presenting a specific collision symptom." + "reason": "Both point to the same remote-code/custom-module loading problem: path collisions cause the wrong module to be loaded after save/load or when models share a leaf directory name." }, { "left": "issue:43645", "right": "issue:45399", "accept": false, - "reason": "Jupyter custom-model init breakage and flash-attn fallback blocking are unrelated compatibility issues." + "reason": "Jupyter custom-model initialization regression is unrelated to flash-attn2 fallback gating." }, { "left": "issue:43404", "right": "issue:45005", "accept": false, - "reason": "Untied lm_head in Mistral3 image-text generation and translation-model tied-weights issues are different model families and code paths." + "reason": "A specific untied lm_head in Mistral3ForConditionalGeneration is not the same bug as translation-model tied-weights issues in v5." }, { "left": "issue:42915", "right": "issue:45399", "accept": false, - "reason": "Qwen3MoE FP8 failure and flash-attn fallback checks are unrelated." + "reason": "FineGrainedFP8Config failure on Qwen3Moe is unrelated to flash-attn2 fallback checks." }, { "left": "issue:44530", "right": "issue:45005", "accept": false, - "reason": "PagedAttentionCache crash on Qwen3.5 and tied-weights translation-model bug are unrelated." + "reason": "PagedAttentionCache group-type crash on Qwen3.5 is unrelated to tied-weight regressions." }, { "left": "issue:43873", "right": "issue:45005", "accept": false, - "reason": "Quantization/offloading behavior and tied-weight translation models are different concerns." + "reason": "Quantization/offloading memory behavior is a different issue from tied weights." }, { "left": "issue:44360", "right": "issue:44512", "accept": false, - "reason": "DSA indexer discussion and docs mentioning removed run command are unrelated." + "reason": "A DSA indexer ReLU discussion is unrelated to a docs typo." }, { "left": "issue:44863", "right": "issue:45005", "accept": false, - "reason": "NemotronH checkpoint loading and translation-model tied weights are different bugs." + "reason": "NemotronH checkpoint loading failure is unrelated to translation-model tied weights." }, { "left": "issue:45183", "right": "issue:46463", "accept": false, - "reason": "Both are validation/error-message issues, but one is generic `input_ids`/`inputs_embeds` XOR logic and the other is T5 decoder-input validation; not enough to treat as the same bug." + "reason": "Generic XOR validation errors across models are broader than the T5 decoder-input error message bug; not clearly the same code-path defect." }, { "left": "issue:44492", "right": "issue:44512", "accept": false, - "reason": "A typo in cache strategies and docs mentioning a removed command are unrelated." + "reason": "Cache-strategy typo and removed run-command docs are unrelated documentation issues." }, { "left": "issue:46489", "right": "issue:46491", "accept": false, - "reason": "Both are tokenizer regressions, but they affect different tokenizers and symptoms (DeepSeek-Coder output vs CodeLlama leading-space round-trip)." + "reason": "Both are tokenizer regressions, but they affect different model families and likely different code paths." }, { "left": "issue:44918", "right": "issue:45507", "accept": false, - "reason": "Qwen3.5 embedding unpacking and GraniteMoEHybrid invalid method calls are different model-integration failures." + "reason": "Qwen3.5 SFT embedding unpacking and GraniteMoEHybrid invalid method calls are different model-specific bugs." }, { "left": "issue:43975", "right": "issue:44292", "accept": false, - "reason": "DeepSeek-Coder detokenization and Qwen-3-8B-NVFP4 runtime error are unrelated." + "reason": "DeepSeek-Coder tokenizer detokenization regression is unrelated to init_empty_weights argument handling." }, { "left": "issue:44292", "right": "issue:45507", "accept": false, - "reason": "Qwen-3-8B-NVFP4 runtime error and GraniteMoEHybrid invalid method calls are distinct model-specific failures." + "reason": "Unexpected _is_hf_initialized handling and GraniteMoEHybrid invalid method calls are unrelated." } ] }, "evaluator_result": { "accept": true, - "feedback": "The cluster summary is well grounded in the packet and stays conservative. The only accepted soft pair (45632/45698) is a plausible duplicate based on the titles alone, while the other pairs are reasonably rejected as different bugs or code paths." + "feedback": "The cluster summary is grounded in the packet: most items are clearly unrelated, and the 45632/45698 pair is a plausible canonical duplicate around trust_remote_code/custom-module loading path collisions. The soft-edge rejections are conservative and do not overstate evidence." }, "evaluator_used": true, "retried": false, @@ -1350,8 +1290,8 @@ "issue:41553|issue:43881", "issue:44403|issue:44451", "issue:43975|issue:45520", - "issue:42947|issue:43856", "issue:24643|issue:28530", + "issue:42947|issue:43856", "issue:44403|issue:44488", "issue:43643|issue:45020", "issue:41762|issue:42915", @@ -1366,247 +1306,247 @@ "issue:42491|issue:44368", "issue:43828|issue:45823" ], - "prepared_review_unit_hash": "77df2676d467009f09fdf7415025ddd968257076420b4d51df23339b14225d6d", + "prepared_review_unit_hash": "e81250a261248f6fe31b197efe0932332f48daff0cd262373d4fc0848c75de96", "component_budget": { "node_count": 473, "item_count": 473, "soft_pair_count": 1738, - "serialized_chars": 901383, - "estimated_input_tokens": 225346, - "estimated_eval_tokens": 450948 + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 34, - "serialized_chars": 38110, - "estimated_input_tokens": 9528, - "estimated_eval_tokens": 19312 + "serialized_chars": 38115, + "estimated_input_tokens": 9529, + "estimated_eval_tokens": 19314 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster is mostly heterogeneous, but there is a clear duplicate thread around the Transformers v5 non-persistent buffer corruption regression in `from_pretrained` / `to_empty`. Most other pairs only share a broad model-loading or tokenizer theme and should not be merged.", - "confidence": 0.68, - "canonical_issue_reason": "`issue:43950` is the clearest concrete bug in the set and the strongest duplicate hub: it precisely describes the v5 non-persistent-buffer corruption regression, with close overlap to `issue:44534` and `issue:45902`.", + "summary": "Most pairs are unrelated. The strongest duplicate clusters are the non-persistent-buffer corruption regression, the SigLIP2 implementation/discrepancy reports, and the remote_code loading regression. No pull requests were present.", + "confidence": 0.79, + "canonical_issue_reason": "issue:45920 is the best umbrella issue: it is open, broad, and explicitly covers a multi-model tokenizer regression affecting several model families, so it is the most useful canonical anchor for deduping nearby reports.", "canonical_pr_reason": null, - "best_issue_reason": "`issue:43950` is the best representative issue overall because it is specific, actionable, and central to the only clearly duplicated bug thread in this cluster.", + "best_issue_reason": "issue:45920 has the broadest scope and strongest canonical value for grouping related tokenizer-regression reports, and it remains open with inbound references.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43950", "right": "issue:44534", "accept": true, - "reason": "Same underlying v5 regression: non-persistent buffers are mishandled/corrupted during load/materialization." + "reason": "Same non-persistent-buffer corruption regression in Transformers v5; the titles describe the same defect with different wording." }, { "left": "issue:39692", "right": "issue:43054", "accept": false, - "reason": "Both mention SigLIP2, but one is a doc example/model mismatch and the other is embedding quality; not the same bug." + "reason": "Docs/example errors and quantization mismatch are not the same bug as SigLIP2 embedding quality degradation." }, { "left": "issue:43644", "right": "issue:43950", - "accept": false, - "reason": "`trust_remote_code` missing fields is a different config-return bug than buffer corruption in `from_pretrained`." + "accept": true, + "reason": "Both report the same v5 regression where non-persistent buffers are corrupted or junk-filled on load." }, { "left": "issue:36331", "right": "issue:41762", "accept": false, - "reason": "Different code paths: custom trainer API signature vs DeepSpeed ZeRO-3 loading failure." + "reason": "Unexpected compute_loss kwarg and ZeRO-3 loading IndexError are unrelated bugs." }, { "left": "issue:43828", "right": "issue:44928", "accept": false, - "reason": "Different symptoms and mechanisms: autocast dtype mismatch vs Qwen3.5 3D `position_ids` / flash-attn crash." + "reason": "Different Qwen/Phi failure modes and different root causes; not the same code-path bug." }, { "left": "issue:43872", "right": "issue:44589", "accept": false, - "reason": "Both are quantization-related, but the `Int8Params` kwarg breakage and missing `Float8` storage are separate incompatibilities." + "reason": "Both are low-level quantization issues, but they affect different storage/parameter paths." }, { "left": "issue:45920", "right": "issue:46489", - "accept": true, - "reason": "Both are tokenizer-output regressions in v5; `46489` looks like a model-specific gap in the broader wrong-token-IDs bug." + "accept": false, + "reason": "Related tokenizer regressions, but 46489 is a DeepSeek-Coder-specific detokenization gap rather than the same concrete issue." }, { "left": "issue:43054", "right": "issue:45701", "accept": false, - "reason": "SigLIP2 embedding quality issue is not the same as a general tokenization-version change." + "reason": "SigLIP2 embedding quality and a generic tokenization-version change are not the same bug." }, { "left": "issue:43866", "right": "issue:44403", "accept": false, - "reason": "Checkpoint corruption and load-time noise are unrelated." + "reason": "Checkpoint corruption and loading noise are different problems." }, { "left": "issue:43950", "right": "issue:45902", - "accept": true, - "reason": "Same family of bug: non-persistent buffer/materialization path leaves buffers uninitialized or junk-filled." + "accept": false, + "reason": "43950 is a generic non-persistent-buffer corruption regression, while 45902 is a specific Qwen3_5MoeTextRotaryEmbedding uninitialized-memory/materialization bug; the evidence does not show the same underlying defect." }, { "left": "issue:38175", "right": "issue:43054", "accept": false, - "reason": "Both concern SigLIP2 quality, but zero probabilities and worse embeddings are not clearly the same defect." + "reason": "Both involve SigLIP2, but zero probabilities and degraded text embeddings are not clearly the same underlying defect." }, { "left": "issue:41553", "right": "issue:43883", "accept": false, - "reason": "Different models and failures: bad AutoTokenizer error message vs missing attribute on `MolmoForCausalLM`." + "reason": "Bad AutoTokenizer error messaging and a missing attribute on Molmo are unrelated." }, { "left": "issue:44910", "right": "issue:45859", "accept": false, - "reason": "Flash-attn packed-sequence crash and CPU-offload incompatibility in Qwen3.5 MoE are distinct issues." + "reason": "Flash-Attn packed-sequence crash and CPU-offload incompatibility are different Qwen3.5 issues." }, { "left": "issue:42947", "right": "issue:44387", "accept": false, - "reason": "LoRA gradient-checkpointing behavior and int4 memory/OOM regression are different problems." + "reason": "Gradient checkpointing ineffective and int4 quantization OOM are distinct memory bugs." }, { "left": "issue:43054", "right": "issue:43493", - "accept": false, - "reason": "Performance regression in SigLIP2 embeddings is not the same as a general HF-vs-JAX implementation discrepancy." + "accept": true, + "reason": "Both point to the same SigLIP2 implementation discrepancy, which plausibly explains the degraded embeddings." }, { "left": "issue:45663", "right": "issue:45902", "accept": false, - "reason": "Gemma4 FSDP2 `KeyError` from shared state rebuilding is unrelated to buffer initialization corruption." + "reason": "Gemma4 FSDP2 KeyError is unrelated to non-persistent-buffer materialization corruption." }, { "left": "issue:41553", "right": "issue:43881", "accept": false, - "reason": "Different model-loading failures with no clear shared code-path." + "reason": "Different models and different loading failures; no shared underlying bug." }, { "left": "issue:44403", "right": "issue:44451", "accept": false, - "reason": "Generic loading noise and a specific ScandiBERT load failure are not the same bug." + "reason": "Generic load noise vs inability to load a specific model are not the same defect." }, { "left": "issue:43975", "right": "issue:45520", "accept": false, - "reason": "Tokenizer detokenization regression and Python 3.13 `flash_attn` import failure are unrelated." + "reason": "Tokenizer detokenization bug and a Python 3.13 flash_attn import error are unrelated." }, { - "left": "issue:42947", - "right": "issue:43856", + "left": "issue:24643", + "right": "issue:28530", "accept": false, - "reason": "Gradient checkpointing/LoRA issue is separate from Qwen3 MoE memory inefficiency." + "reason": "DeepSpeed training RuntimeError and early-stopping metric handling are unrelated." }, { - "left": "issue:24643", - "right": "issue:28530", + "left": "issue:42947", + "right": "issue:43856", "accept": false, - "reason": "DeepSpeed weight-shape error and early-stopping metric handling are unrelated training issues." + "reason": "LoRA checkpointing and Qwen3 MoE memory usage are different performance/memory issues." }, { "left": "issue:44403", "right": "issue:44488", "accept": false, - "reason": "Generic load noise and `cjvt/sleng-bert` load failure are not the same concrete bug." + "reason": "Load noise and a specific model load failure are not the same bug." }, { "left": "issue:43643", "right": "issue:45020", - "accept": false, - "reason": "`trust_remote_code` returning incomplete config fields is broader-adjacent but not the same as remote-code model breakage." + "accept": true, + "reason": "Same remote_code regression surface: AutoConfig missing fields is a concrete manifestation of models using remote_code breaking in recent versions." }, { "left": "issue:41762", "right": "issue:42915", "accept": false, - "reason": "Gemma3 ZeRO-3 load failure and Qwen3 MoE FP8 failure are different model/setup issues." + "reason": "Gemma3 ZeRO-3 load failure and Qwen3Moe FineGrainedFP8 failure are unrelated." }, { "left": "issue:44991", "right": "issue:45399", "accept": false, - "reason": "Tokenizer loading failure and flash-attn2 fallback gating are unrelated." + "reason": "Tokenizer loading failure and flash-attn fallback gating are unrelated." }, { "left": "issue:45685", "right": "issue:45823", "accept": false, - "reason": "MPS histogram/int support error and Gemma4 PLE device mismatch are distinct." + "reason": "MPS histogram-on-int error and Gemma4 device mismatch are different device-placement bugs." }, { "left": "issue:43782", "right": "issue:44960", "accept": false, - "reason": "Qwen3VL weight-only load error and GLM5 issue are different model-specific failures." + "reason": "Qwen3VL weight_only loading and GLM5 are unrelated." }, { "left": "issue:43824", "right": "issue:43872", "accept": false, - "reason": "Importing Qwen2.5-VL and bitsandbytes `Int8Params` incompatibility are unrelated." + "reason": "Importing a missing model class and bitsandbytes Int8Params construction are unrelated." }, { "left": "issue:43881", "right": "issue:43883", "accept": false, - "reason": "GLM-4V loading failure and missing `all_tied_weights_keys` on Molmo are different bugs." + "reason": "Different model load failures with different symptoms and causes." }, { "left": "issue:45020", "right": "issue:45823", "accept": false, - "reason": "Remote-code breakage is not the same as Gemma4 device-map forwarding mismatch." + "reason": "Remote_code regression and Gemma4 PLE device mismatch are unrelated." }, { "left": "issue:44661", "right": "issue:45020", "accept": false, - "reason": "Tokenizer mapping-name edge case and remote-code regression are different code paths." + "reason": "Tokenizer-mapping edge case and remote_code regression are different code paths." }, { "left": "issue:45081", "right": "issue:45399", "accept": false, - "reason": "Mistral regex patch crash and flash-attn2 fallback gating are unrelated." + "reason": "Mistral regex patch crash and flash-attn fallback gating are unrelated." }, { "left": "issue:42491", "right": "issue:44368", "accept": false, - "reason": "LoRA compatibility break between versions and a tie-word-embeddings warning are not the same bug." + "reason": "LoRA model incompatibility across branches and a tied-embeddings warning are not the same bug." }, { "left": "issue:43828", "right": "issue:45823", "accept": false, - "reason": "Autocast dtype mismatch and device-map forward device mismatch are different failures." + "reason": "Autocast dtype mismatch and Gemma4 device mismatch are different forward-pass failures." } ] }, "evaluator_result": { - "accept": true, - "feedback": "The cluster summary is grounded in the packet and stays conservative overall. The 43950/44534/45902 thread is clearly the strongest duplicate family from the titles, and the other pairwise rejections are appropriately cautious. The accepted tokenizer pair (45920/46489) is also plausibly the same v5 regression family from the titles, not an overclaim." + "accept": false, + "feedback": "The overall grouping is mostly grounded, but at least one soft-edge accept is too strong for the evidence in the packet. In particular, issue:43950 \u2194 issue:45902 is not clearly the same bug from the titles alone: 43950 is a generic non-persistent-buffer corruption regression, while 45902 is a specific Qwen3_5MoeTextRotaryEmbedding uninitialized-memory/materialization issue. The SigLIP2 and remote_code links are more plausible, but the verdict set is not uniformly conservative enough to accept." }, "evaluator_used": true, - "retried": false, + "retried": true, "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null @@ -1674,8 +1614,8 @@ "issue:46661" ], "soft_pairs": [ - "issue:45715|issue:46315", "issue:45020|issue:45356", + "issue:45715|issue:46315", "issue:45488|issue:46489", "issue:43827|issue:44509", "issue:43404|issue:43950", @@ -1694,8 +1634,8 @@ "issue:44479|issue:44560", "issue:41093|issue:44805", "issue:41553|issue:44488", - "issue:44863|issue:46402", "issue:24643|issue:30064", + "issue:44863|issue:46402", "issue:44530|issue:44863", "issue:43366|issue:43828", "issue:43596|issue:46515", @@ -1703,20 +1643,20 @@ "issue:43366|issue:43531", "issue:45663|issue:45823" ], - "prepared_review_unit_hash": "da55aa20d941cd43cdbbbc847a0848d0bdd2176149056bad86d56dfc2ef647a2", + "prepared_review_unit_hash": "4061509026ecae0360822f0e21a6a92947f2cd5ed1f7a384810c8548c9ecad62", "component_budget": { "node_count": 473, "item_count": 473, "soft_pair_count": 1738, - "serialized_chars": 901383, - "estimated_input_tokens": 225346, - "estimated_eval_tokens": 450948 + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 28, - "serialized_chars": 36626, + "serialized_chars": 36625, "estimated_input_tokens": 9157, "estimated_eval_tokens": 18570 }, @@ -1725,186 +1665,186 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a loose set of unrelated Transformers issues, with a small subset of true duplicates around docs cleanup, `_is_hf_initialized` init-time incompatibilities, and the same mask/index shape-mismatch error pattern. Most other pairs are only thematically similar (tokenizers, DeepSpeed, Qwen/Gemma loading) but not the same bug.", - "confidence": 0.83, - "canonical_issue_reason": "Issue 45020 is the broadest umbrella in the set: it frames recent-version regressions breaking remote_code / model loading, which best matches the overall cluster theme.", + "summary": "This cluster is mostly heterogeneous; there is one clear duplicate pair (docs mentioning removed pipeline tasks) and the rest look like separate regressions across different model/tokenizer/loading paths.", + "confidence": 0.79, + "canonical_issue_reason": "Issue 45020 is the broadest umbrella in the set: it frames a general v5 regression affecting remote-code model loading, which is the closest match to several of the nearby loading/tokenizer breakages.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 46661 is the most concrete, actionable representative of the loading-regression subset: a specific remote-code `from_pretrained` failure with a clear stack-level symptom.", + "best_issue_reason": "Issue 45020 is the most representative cluster anchor because it is broad enough to subsume related v5 loading regressions, even though most other items are still distinct.", "best_pr_reason": null, "soft_edge_verdicts": [ { - "left": "issue:45715", - "right": "issue:46315", + "left": "issue:45020", + "right": "issue:45356", "accept": false, - "reason": "Both are tokenizer performance regressions, but they hit different methods (`convert_ids_to_tokens` vs `convert_tokens_to_ids`) and look like separate hotspots." + "reason": "Both are version-related loading/tokenizer regressions, but they target different concrete failures and model paths." }, { - "left": "issue:45020", - "right": "issue:45356", + "left": "issue:45715", + "right": "issue:46315", "accept": false, - "reason": "Broad remote_code loading regressions vs a specific Kimi-K2.5 tokenizer codec-handling regression; related theme, not the same bug." + "reason": "Both are tokenizer performance regressions, but they affect different conversion directions and different hot paths." }, { "left": "issue:45488", "right": "issue:46489", "accept": false, - "reason": "Both mention DeepSeek tokenization, but one is a LlamaTokenizer pre-tokenizer override issue and the other is a DeepSeek-Coder output regression; different root causes." + "reason": "Both involve DeepSeek tokenization, but the reported failures and affected model families are not the same concrete bug." }, { "left": "issue:43827", "right": "issue:44509", "accept": true, - "reason": "Both report stale docs still referencing removed v5 pipeline tasks; this is the same documentation cleanup issue." + "reason": "Same docs regression: both report stale pipeline-task documentation for removed v5 pipelines." }, { "left": "issue:43404", "right": "issue:43950", "accept": false, - "reason": "`lm_head` tying in Mistral3 and non-persistent buffer corruption in `from_pretrained()` are different loading bugs." + "reason": "Different failures: untied lm_head weights versus corrupted non-persistent buffers." }, { "left": "issue:45759", "right": "issue:46661", "accept": false, - "reason": "Both are composite/remote-code model-loading issues, but one is `from_config` unwrapping `text_config` and the other is a `from_pretrained` `config_class` AttributeError." + "reason": "Both involve model construction/loading, but they are different API failures and not the same code-path bug." }, { "left": "issue:43872", "right": "issue:44291", - "accept": true, - "reason": "Both describe the same `_is_hf_initialized` kwarg incompatibility during initialization/loading of model weights." + "accept": false, + "reason": "Shared '_is_hf_initialized' wording, but the failures arise from different loading mechanisms and are not clearly one bug." }, { "left": "issue:42915", "right": "issue:43872", "accept": false, - "reason": "FineGrainedFP8Config failure in Qwen3 MoE is unrelated to the bitsandbytes `_is_hf_initialized` constructor error." + "reason": "Different incompatibilities: FP8 config handling versus bitsandbytes parameter construction." }, { "left": "issue:45020", "right": "issue:46661", "accept": false, - "reason": "Issue 45020 is a broad umbrella; 46661 is one specific remote-code failure mode, so they are not the same concrete bug." + "reason": "Both are remote-code loading problems, but one is a broad regression and the other is a specific missing config_class attribute issue." }, { "left": "issue:43792", "right": "issue:44220", "accept": false, - "reason": "Both involve Whisper/audio, but the titles point to a generic \"can't run\" report versus a specific `_torch_extract_fbank_features()` problem; not enough to treat as the same bug." + "reason": "Both are whisper/audio related, but the titles do not indicate the same concrete failure." }, { "left": "issue:43257", "right": "issue:45305", "accept": false, - "reason": "One is a DeepSpeed weight-conversion/load failure, the other is a gradient-averaging bug during training; different code paths." + "reason": "Both mention DeepSpeed, but one is weight conversion/loading and the other is gradient averaging behavior." }, { "left": "issue:42947", "right": "issue:45923", "accept": false, - "reason": "Gradient checkpointing ineffective with PEFT LoRA is not the same as a missing `supports_gradient_checkpointing` flag on a trust_remote_code model." + "reason": "Gradient-checkpointing problems, but one is ineffective LoRA checkpointing and the other is a missing capability flag." }, { "left": "issue:41553", "right": "issue:44991", "accept": false, - "reason": "Both are tokenizer-loading problems, but they concern different models and different failure modes (bad error message vs outright load failure)." + "reason": "Both are tokenizer-loading issues, but they concern different models and different failure modes." }, { "left": "issue:36331", "right": "issue:41093", "accept": false, - "reason": "`compute_loss()` kwarg incompatibility and a mask/tensor shape mismatch are unrelated training/runtime errors." + "reason": "Both are trainer/indexing errors, but the specific kwargs/mask mismatch bugs are different." }, { "left": "issue:43940", "right": "issue:45313", "accept": false, - "reason": "Both are ZeRO-3 load failures for Qwen models, but they target different model variants and different missing-weight symptoms." + "reason": "Both are DeepSpeed ZeRO-3 loading problems, but one is missing all weights for Qwen3-Next and the other is language_model weight loading for Qwen3.5." }, { "left": "issue:43854", "right": "issue:44863", "accept": false, - "reason": "Different architectures and different load failures: GLM-4.7-Flash unit-test loading vs NemotronH checkpoint loading." + "reason": "Different model-loading failures for different architectures." }, { "left": "issue:44928", "right": "issue:45588", "accept": false, - "reason": "RLHF NaN explosion from Qwen3.5 position IDs is a numerical/training issue, not the flash_attention `s_aux=None` crash." + "reason": "Both touch attention/cache paths, but one is RLHF position-id instability and the other is a flash-attention auxiliary-state crash." }, { "left": "issue:44479", "right": "issue:44560", "accept": false, - "reason": "Both concern video inputs in Qwen VL-family models, but one is a broad v5.3.0 regression and the other is a specific `StopIteration` in Qwen3-VL-Embedding; not clearly the same bug." + "reason": "Both are Qwen video regressions, but they are distinct symptoms and model variants." }, { "left": "issue:41093", "right": "issue:44805", - "accept": true, - "reason": "These are the same mask/index shape-mismatch error pattern, just with different concrete dimensions shown in the error message." + "accept": false, + "reason": "Similar error template, but the mismatched shapes are different and there is no evidence of one shared root cause." }, { "left": "issue:41553", "right": "issue:44488", "accept": false, - "reason": "Different tokenizer-loading failures on different models; only superficially similar." + "reason": "Both concern tokenizer loading, but they are different models and different error reports." }, { - "left": "issue:44863", - "right": "issue:46402", + "left": "issue:24643", + "right": "issue:30064", "accept": false, - "reason": "Both are model weight-loading issues, but they involve different model families and different missing parameters." + "reason": "Training DeepSpeed weight-shape error versus image-processor segmentation-map handling; unrelated bugs." }, { - "left": "issue:24643", - "right": "issue:30064", + "left": "issue:44863", + "right": "issue:46402", "accept": false, - "reason": "DeepSpeed training `weight must be 2-D` and void segmentation-map processing are unrelated bugs." + "reason": "Different loading failures for different model families." }, { "left": "issue:44530", "right": "issue:44863", "accept": false, - "reason": "PagedAttentionCache `linear_attention` crash and NemotronH checkpoint loading are different problems." + "reason": "Different Qwen/GGUF loading issues with unrelated failure mechanisms." }, { "left": "issue:43366", "right": "issue:43828", "accept": false, - "reason": "GGUF/gpt-oss architecture support and a Phi-tiny-MoE autocast dtype mismatch are unrelated." + "reason": "Unrelated: GGUF architecture support versus autocast dtype mismatch." }, { "left": "issue:43596", "right": "issue:46515", "accept": false, - "reason": "Zero3/BertModel empty-index failure and torch 2.4.1 incompatibility are not the same bug." + "reason": "Different incompatibilities: ZeRO-3/Bert index error versus torch version incompatibility." }, { "left": "issue:45325", "right": "issue:45902", "accept": false, - "reason": "Qwen2.5-VL rope scaling for still images and Qwen3.5-MoE uninitialized `inv_freq` are different rotary/position-id issues." + "reason": "Both concern Qwen position/rotary internals, but the reported bugs are different and not the same code path." }, { "left": "issue:43366", "right": "issue:43531", "accept": false, - "reason": "GPT-OSS GGUF support and Qwen3-MoE `sliding_window` behavior are unrelated." + "reason": "Different Qwen-related bugs: GGUF architecture support versus sliding-window handling." }, { "left": "issue:45663", "right": "issue:45823", "accept": false, - "reason": "Both are Gemma4 issues, but one is a FSDP2 KeyError in attention and the other is a device-map mismatch during forward; distinct bugs." + "reason": "Same model family, but one is an FSDP2 KeyError and the other is a device-map mismatch." } ] }, "evaluator_result": { "accept": true, - "feedback": "The summary is grounded in the packet and stays appropriately conservative on the soft-edge pairs. The accepted duplicates are well-supported (docs cleanup, `_is_hf_initialized`, and the mask/index shape mismatch), and the rejected pairs are phrased as related-but-distinct rather than overclaimed duplicates. Canonical/best issue choices are also defensible from the cluster." + "feedback": "The cluster summary is grounded in the packet and stays conservative. The one accepted soft pair (43827/44509) is well supported by the titles, and the remaining false verdicts avoid overclaiming shared root cause. I don\u2019t see evidence of overstated links or an unmergeable PR-pair issue here." }, "evaluator_used": true, "retried": false, @@ -2000,8 +1940,8 @@ "issue:43596|issue:44661", "issue:43317|issue:44756", "issue:44661|issue:44805", - "issue:38175|issue:39692", "issue:42491|issue:43720", + "issue:38175|issue:39692", "issue:43644|issue:44534", "issue:41720|issue:42915", "issue:42491|issue:45276", @@ -2010,20 +1950,20 @@ "issue:46459|issue:46512", "issue:44960|issue:44977" ], - "prepared_review_unit_hash": "09b6ffdf12886ea1f4282f536d0d025cd7cbdabc9ad8b56e81ebca5857b5674a", + "prepared_review_unit_hash": "071d11b9cc084d3556931ef8072de4e7244037addda7ce4319806b28ee1659a4", "component_budget": { "node_count": 473, "item_count": 473, "soft_pair_count": 1738, - "serialized_chars": 901383, - "estimated_input_tokens": 225346, - "estimated_eval_tokens": 450948 + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 34, - "serialized_chars": 37775, + "serialized_chars": 37773, "estimated_input_tokens": 9444, "estimated_eval_tokens": 19144 }, @@ -2032,222 +1972,222 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Mostly a heterogeneous set of unrelated Transformers issues, with only a few likely duplicate pairs. The clearest duplicates are the non-persistent buffer regression, the SigLIP2 output regression, and the BERT model-loading failures for two specific checkpoints. The rest are different models, code paths, or failure modes.", - "confidence": 0.93, - "canonical_issue_reason": null, + "summary": "Mostly a noisy cluster of unrelated Transformers issues. The only clear duplicate pair is the Transformers v5 non-persistent-buffer corruption regression; there is also a likely duplicate pair about custom model initialization in v5.", + "confidence": 0.79, + "canonical_issue_reason": "Issue 43644 is the clearest and earliest report of the shared Transformers 5.0.0 non-persistent-buffer corruption bug; 44534 is the same regression restated later.", "canonical_pr_reason": null, - "best_issue_reason": null, + "best_issue_reason": "Issue 43644 is the best anchor for the cluster because it describes the core bug most explicitly and is the earliest representative of the only strong duplicate group.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43638", "right": "issue:46515", "accept": false, - "reason": "Different bugs: DeepSpeed ZeRO3/BertModel indexing vs a generic torch 2.4.1 incompatibility." + "reason": "Different failures: Zero3/Bert index error vs a torch 2.4.1 compatibility complaint." }, { "left": "issue:44155", "right": "issue:45584", "accept": false, - "reason": "Different models and failure modes: AudioFlamingo3 track leakage vs Whisper empty-transcription handling." + "reason": "Both are audio/generation problems, but the symptoms and code paths are different." }, { "left": "issue:43716", "right": "issue:44560", "accept": false, - "reason": "Both are multimodal, but the concrete bugs differ: image preprocessor dtype mismatch vs video StopIteration." + "reason": "Different model families and failure modes; only both are multimodal regressions." }, { "left": "issue:43950", "right": "issue:45834", "accept": false, - "reason": "Different code paths: non-persistent buffer corruption vs long-OCR index error." + "reason": "Both mention loading/model behavior, but the bugs are not the same concrete failure." }, { "left": "issue:43366", "right": "issue:44589", "accept": false, - "reason": "Unrelated issues: GGUF gpt-oss architecture support vs missing Float8 storage handling." + "reason": "Unrelated issues: GGUF gpt-oss support vs a Float8 storage/load error." }, { "left": "issue:43824", "right": "issue:43957", "accept": false, - "reason": "Different loading problems: missing Qwen2.5-VL export vs meta-device loading failure." + "reason": "Different model-loading regressions; one is an import failure, the other is meta-device loading breakage." }, { "left": "issue:44560", "right": "issue:45685", "accept": false, - "reason": "Different subsystems: Qwen3-vl video embedding failure vs MPS histogram support in MoE." + "reason": "Both involve model internals, but they describe different bugs and platforms." }, { "left": "issue:42915", "right": "issue:43957", "accept": false, - "reason": "Different failures: Qwen3Moe FP8 config vs meta-device model loading." + "reason": "Shared broad v5 regression theme, but the concrete failures differ." }, { "left": "issue:43645", "right": "issue:43646", - "accept": false, - "reason": "Related theme, but not clearly the same concrete bug: notebook-specific custom model init vs general custom model initialization." + "accept": true, + "reason": "Same Transformers v5 custom-model initialization breakage; the Jupyter detail is just a usage context." }, { "left": "issue:43550", "right": "issue:44910", "accept": false, - "reason": "Different attention-path bugs in different models: Bamba torch.compile/SDPA vs Qwen3.5 flash-attn illegal access." + "reason": "Different model-specific attention/compile failures with no clear shared root cause." }, { "left": "issue:44560", "right": "issue:44918", "accept": false, - "reason": "Different failures: Qwen3-vl video handling vs TRL SFT embedding unpacking." + "reason": "Both are Qwen3.5-related, but one is video embedding StopIteration and the other is SFT input-embedding unpacking." }, { "left": "issue:44683", "right": "issue:46515", "accept": false, - "reason": "Different version incompatibilities: compiled flex_attention on torch>=2.9 vs a torch 2.4.1 issue." + "reason": "Different issues: compiled flex_attention vs a torch version incompatibility report." }, { "left": "issue:42617", "right": "issue:43366", "accept": false, - "reason": "Unrelated: a 3d_parallel runtime failure vs GGUF gpt-oss support." + "reason": "Not the same bug; one is a script/runtime failure, the other is a model-architecture support request." }, { "left": "issue:45812", "right": "issue:46489", "accept": false, - "reason": "Both are tokenizer regressions, but for different models and different root causes." + "reason": "Related tokenizer regressions, but different models and different concrete failures." }, { "left": "issue:36010", "right": "issue:42915", "accept": false, - "reason": "Completely different issues: GenerationMixin import error vs Qwen3Moe FP8 failure." + "reason": "ImportError for GenerationMixin is unrelated to the Qwen3Moe FP8 failure." }, { "left": "issue:45823", "right": "issue:46515", "accept": false, - "reason": "Different problems: Gemma4 PLE device mismatch vs a generic torch version incompatibility." + "reason": "Device-mapping forward mismatch is not the same as a torch 2.4.1 incompatibility issue." }, { "left": "issue:44661", "right": "issue:45823", "accept": false, - "reason": "Different bugs: add-new-model-like validation vs Gemma4 forward device mismatch." + "reason": "Different bug classes: tokenizer mapping-name validation vs Gemma4 PLE device mismatch." }, { "left": "issue:44560", "right": "issue:44805", "accept": false, - "reason": "Different failure modes: video StopIteration vs mask/tensor shape mismatch." + "reason": "Both are index/error reports, but the models and code paths differ." }, { "left": "issue:38175", "right": "issue:43994", - "accept": true, - "reason": "Same underlying SigLIP2-base-patch16-224 regression: both report nonsensical/zero outputs from the same model family." + "accept": false, + "reason": "Same SigLIP2 model family, but the reports are too different to treat as the same bug confidently." }, { "left": "issue:43632", "right": "issue:45834", "accept": false, - "reason": "Different issues: _is_hf_initialized breakage vs Kosmos2.5 long-input index error." + "reason": "Both have index-related symptoms, but the underlying failures are not clearly the same." }, { "left": "issue:44877", "right": "issue:45030", "accept": false, - "reason": "Both involve config validation, but the affected models and specific validation failures are different." + "reason": "Different config-validation failures with no clear shared root cause." }, { "left": "issue:43638", "right": "issue:44661", "accept": false, - "reason": "Different paths: DeepSpeed ZeRO3 BertModel initialization vs add-new-model-like handling of tokenizer mappings." + "reason": "Zero3/Bert index error is unrelated to tokenizer mapping-name handling." }, { "left": "issue:43596", "right": "issue:44661", "accept": false, - "reason": "Different code paths: empty-tensor/BertModel ZeRO3 init vs add-new-model-like validation." + "reason": "Different failures: Zero3/Bert initialization vs tokenizer mapping validation." }, { "left": "issue:43317", "right": "issue:44756", "accept": false, - "reason": "Different bugs: gpu+cpu offload loading vs Strix Halo mmap/OOM behavior." + "reason": "Both are loading/offload related, but one is dequantized model offload and the other is mmap/OOM on Strix Halo." }, { "left": "issue:44661", "right": "issue:44805", "accept": false, - "reason": "Different issues: tokenizer-mapping validation vs tensor mask shape mismatch." + "reason": "No evidence they are the same bug; just both are loader/config issues." }, { - "left": "issue:38175", - "right": "issue:39692", + "left": "issue:42491", + "right": "issue:43720", "accept": false, - "reason": "Same model family, but different concrete reports: output regression vs docs example/model-processor mismatch and quantization errors." + "reason": "Different model- and framework-specific loading/training issues." }, { - "left": "issue:42491", - "right": "issue:43720", + "left": "issue:38175", + "right": "issue:39692", "accept": false, - "reason": "Different issues: Qwen3 MoE LoRA compatibility across versions vs BitNet packed-weight loading." + "reason": "Both involve SigLIP2, but one is wrong outputs and the other is a documentation example with multiple errors." }, { "left": "issue:43644", "right": "issue:44534", "accept": true, - "reason": "Same regression: Transformers v5 fills non-persistent buffers with junk; the titles describe the same bug." + "reason": "Exact same Transformers v5 regression: non-persistent buffers are filled with junk on load." }, { "left": "issue:41720", "right": "issue:42915", "accept": false, - "reason": "Different model-specific failures: Qwen3 auto device mapping CUDA assert vs Qwen3Moe FP8 config." + "reason": "Qwen3 auto-device-map CUDA assert is unrelated to Qwen3Moe FP8 config failure." }, { "left": "issue:42491", "right": "issue:45276", "accept": false, - "reason": "Different bugs in different model families: Qwen3 MoE LoRA loading vs Gemma4 token embedding resize behavior." + "reason": "Different bugs: qwen3_moe LoRA compatibility vs Gemma4 embedding resize behavior." }, { "left": "issue:42915", "right": "issue:44756", "accept": false, - "reason": "Unrelated: Qwen3Moe FP8 failure vs mmap/OOM mitigation on Strix Halo." + "reason": "Different subsystems and failure modes; only a broad loading-performance resemblance." }, { "left": "issue:44451", "right": "issue:44488", - "accept": true, - "reason": "Both report the same latest-version inability to load specific BERT checkpoints; likely the same loader/config regression." + "accept": false, + "reason": "Similar symptom of loading failure, but the affected models differ and no shared concrete bug is shown." }, { "left": "issue:46459", "right": "issue:46512", "accept": false, - "reason": "Related dtype theme, but not the same bug: composite-checkpoint loading ignores dtype vs from_config leaves config.dtype stale." + "reason": "Both mention dtype, but one is silent dtype ignore on checkpoint load and the other is stale config dtype after from_config." }, { "left": "issue:44960", "right": "issue:44977", "accept": false, - "reason": "Different model families and failures: GLM5 issue vs Qwen3.5 flash-attention generation problem." + "reason": "No clear duplicate relation; GLM5 and Qwen3.5 flash-attention issues are different." } ] }, "evaluator_result": { "accept": true, - "feedback": "The summary is grounded in the packet and stays conservative. The accepted duplicate pairs are well supported by the titles, and the non-accepted soft edges mostly explain plausible but distinct bug surfaces without overclaiming." + "feedback": "The cluster summary is grounded in the packet: 43644/44534 are clearly the same non-persistent-buffer regression, and 43645/43646 are a plausible duplicate pair on custom model initialization. The rejected soft edges are generally conservative and do not overclaim duplicates from weak title similarity." }, "evaluator_used": true, "retried": false, @@ -2336,236 +2276,233 @@ "issue:44164|issue:44291", "issue:44910|issue:45588", "issue:44402|issue:44488", - "issue:42947|issue:45704", - "issue:43873|issue:44387", "issue:43404|issue:43454", + "issue:42947|issue:45704", "issue:41720|issue:44155", + "issue:43873|issue:44387", "issue:39692|issue:44960", "issue:43425|issue:43957", "issue:45663|issue:45759", - "issue:36010|issue:39404", "issue:44291|issue:44756", + "issue:36010|issue:39404", "issue:45663|issue:45859", "issue:44387|issue:44756" ], - "prepared_review_unit_hash": "c8ef710b0c0140de2373b8e5c234b613279e1fdefbd5097c54bf8a7ad9adb7ac", + "prepared_review_unit_hash": "ed56ad1fb12d9f41b0d9cd1eff3c8ff0c070f5dd81112d6d76be3f6a9a750de4", "component_budget": { "node_count": 473, "item_count": 473, "soft_pair_count": 1738, - "serialized_chars": 901383, - "estimated_input_tokens": 225346, - "estimated_eval_tokens": 450948 + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 30, - "serialized_chars": 36203, - "estimated_input_tokens": 9051, - "estimated_eval_tokens": 18358 + "serialized_chars": 36207, + "estimated_input_tokens": 9052, + "estimated_eval_tokens": 18360 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly heterogeneous. I only found a couple of true duplicate-like pairs: the two resize_token_embeddings regressions and the two lm_head-tie regressions; the rest look like separate bugs that happen to live in similar model/loading areas.", - "confidence": 0.78, - "canonical_issue_reason": "issue:45276 is the clearest issue anchor: it describes a concrete, model-facing resize_token_embeddings regression and is the broader of the two matching embedding issues, so it best represents the duplicate pair with 45335.", + "summary": "This set is mostly a noisy cluster of Transformers issues that share broad themes (model loading, generation heads, tokenizer regressions, Qwen/Gemma variants) but usually point to different concrete code paths. I did not find a safe cross-issue duplicate merge; the closest pair is the resize_token_embeddings regression, but it still looks model-specific.", + "confidence": 0.64, + "canonical_issue_reason": "issue:45276 is the clearest representative of the only near-duplicate subtopic in the set: a resize_token_embeddings propagation bug affecting output-side embeddings. It is more general than its paired Gemma/T5Gemma variant, but the cluster overall is too heterogeneous to treat it as a true umbrella duplicate.", "canonical_pr_reason": null, - "best_issue_reason": "issue:45276 is the strongest global issue candidate among the genuinely related items because it is specific, reproducible, and broad enough to cover the shared resize_token_embeddings failure mode.", + "best_issue_reason": "issue:45488 is the strongest standalone issue in the list: it is active, has discussion and inbound references, and describes a concrete user-facing tokenizer regression with broad impact on DeepSeek V3/R1-style models.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:45313", "right": "issue:45357", "accept": false, - "reason": "Different failure modes: ZeRO-3 weight loading for language_model vs incorrect visual-encoder keys saved by save_pretrained." + "reason": "Both are Qwen3.5 loading/save regressions, but one is DeepSpeed ZeRO-3 loading and the other is save_pretrained visual-encoder key serialization; different failure modes and code paths." }, { "left": "issue:45507", "right": "issue:46468", "accept": false, - "reason": "Unrelated models and symptoms; one is an invalid method call in GraniteMoEHybrid, the other is a broken Nemotron example." + "reason": "Generic model call failure vs a broken official Nemotron example; too little evidence they share the same root cause." }, { "left": "issue:43856", "right": "issue:43873", "accept": false, - "reason": "Both mention MoE/quantization themes, but one is training memory inefficiency and the other is offloading behavior with quantization." + "reason": "Both mention Qwen3 MoE/offloading/memory, but one is training memory usage and the other is quantization offloading behavior; not the same bug." }, { "left": "issue:44488", "right": "issue:44991", "accept": false, - "reason": "Both are tokenizer/loading regressions, but they involve different models and different concrete breakages." + "reason": "Both are tokenizer load failures on different models, but the affected models and reported symptoms differ enough that this looks like a shared regression class, not one duplicate bug." }, { "left": "issue:43644", "right": "issue:43645", "accept": false, - "reason": "Non-persistent buffer junk vs Jupyter custom-model initialization are distinct transformer-5 regressions." + "reason": "Both are Transformers 5.0 breakages, but one is non-persistent buffers and the other is custom model initialization in Jupyter; separate issues." }, { "left": "issue:43708", "right": "issue:44075", "accept": false, - "reason": "Trainer checkpoint step calculation and SGD argument handling are unrelated bugs." + "reason": "Trainer checkpoint resume/max_steps logic is unrelated to optimizer SGD arguments being ignored." }, { "left": "issue:39692", "right": "issue:43994", "accept": false, - "reason": "Both involve SigLIP2, but one is a docs example with multiple issues while the other is a model/pipeline output problem." + "reason": "Both involve SigLIP2-ish model behavior, but one is documentation example errors and the other is bad outputs from AutoModel/pipeline; different artifacts and likely different fixes." }, { "left": "issue:43856", "right": "issue:44387", "accept": false, - "reason": "Different scope: Qwen3 MoE training memory usage vs generic int4 quantization OOM in Transformers 5.x." + "reason": "Both concern memory, but one is Qwen3 MoE training memory usage and the other is int4 quantization reserved-memory/OOM regression; distinct underlying problems." }, { "left": "issue:43257", "right": "issue:43854", "accept": false, - "reason": "Different model families and code paths: Qwen3 MoE accelerate/deepspeed conversion vs GLM-4.7-Flash load failure." + "reason": "Qwen3 MoE weight conversion under accelerate+deepspeed is not the same as a GLM-4.7 model-loading failure in unit tests." }, { "left": "issue:43854", "right": "issue:43931", "accept": false, - "reason": "Both are loading issues, but they concern different models and different failure mechanisms." + "reason": "Both are model loading errors, but for different models and different mismatches (GLM-4.7 test load vs Qwen3-VL weight shapes)." }, { "left": "issue:42491", "right": "issue:45292", "accept": false, - "reason": "LoRA compatibility across versions is unrelated to a generic resize_token_embeddings/output_embeddings bug." + "reason": "LoRA qwen3_moe hf4.x/hf5.x compatibility is unrelated to a generic resize_token_embeddings/output_embeddings issue." }, { "left": "issue:42491", "right": "issue:43931", "accept": false, - "reason": "Different problems: Qwen3 MoE LoRA portability vs a Qwen3-VL shape-mismatch load error." + "reason": "The LoRA hf4.x vs hf5.x incompatibility is a training/compatibility regression, not a Qwen3-VL shape-mismatch loading bug." }, { "left": "issue:44960", "right": "issue:46050", "accept": false, - "reason": "GLM5 loading/setup issue is not the same as save_pretrained/from_pretrained corruption after sinq quantization." + "reason": "GLM5 is too vague to merge with a sinq quantization save/load corruption report; no clear shared code path." }, { "left": "issue:45488", "right": "issue:45812", "accept": false, - "reason": "Both are tokenizer regressions, but they affect different model families and likely different code paths." + "reason": "Both are tokenizer regressions, but one is LlamaTokenizer pre-tokenizer override and the other is Granite token IDs; different tokenizer logic and likely separate fixes." }, { "left": "issue:45362", "right": "issue:45406", "accept": false, - "reason": "One is a chat crash on Qwen3.5-35B, the other is a Gemma4Processor attribute error in serve." + "reason": "Qwen3.5 chat crash and Gemma4Processor missing _tokenizer are different runtime failures in different components." }, { "left": "issue:45276", "right": "issue:45335", - "accept": true, - "reason": "Same underlying bug: resize_token_embeddings does not propagate to tied/nested output or decoder embeddings across related composite model classes." + "accept": false, + "reason": "Both are resize_token_embeddings propagation bugs, but they hit different model families and different internal targets (Gemma4 output/embed_tokens_per_layer vs T5Gemma decoder.embed_tokens); not safe to merge as one duplicate." }, { "left": "issue:44164", "right": "issue:44291", "accept": false, - "reason": "Serialization extra_state handling and init_empty_weights _is_hf_initialized argument handling are separate loading/saving issues." + "reason": "save/from_pretrained extra_state handling and init_empty_weights _is_hf_initialized TypeError are separate serialization/loading issues." }, { "left": "issue:44910", "right": "issue:45588", "accept": false, - "reason": "Both are in flash_attention code, but they fail on different conditions and different stack paths." + "reason": "Flash attention illegal memory access on Qwen3.5 position_ids is unrelated to a sink-less models AttributeError in integrations/flash_attention.py." }, { "left": "issue:44402", "right": "issue:44488", "accept": false, - "reason": "Tokenizer vocab-size mismatch in one model is not the same as a separate model-loading failure in another." + "reason": "Tokenizer vocab-size mismatch for a specific model is not the same underlying bug as a separate model load failure in cjvt/sleng-bert." }, { - "left": "issue:42947", - "right": "issue:45704", + "left": "issue:43404", + "right": "issue:43454", "accept": false, - "reason": "Gradient checkpointing with PEFT LoRA and apex RMSNorm memory leak are unrelated." + "reason": "Both involve untied lm_head weights and bad generation, but they are different model classes (Mistral3 multimodal vs AyaVision); likely model-specific omissions rather than one duplicate bug." }, { - "left": "issue:43873", - "right": "issue:44387", + "left": "issue:42947", + "right": "issue:45704", "accept": false, - "reason": "Both are memory-related, but one is offloading behavior and the other is CUDA reserved memory growth under int4 quantization." - }, - { - "left": "issue:43404", - "right": "issue:43454", - "accept": true, - "reason": "Both describe the same concrete bug class: lm_head weights are not tied, causing bad generation in related multimodal model implementations." + "reason": "Gradient checkpointing with PEFT LoRA is unrelated to T5 silently using apex.FusedRMSNorm and its memory leak." }, { "left": "issue:41720", "right": "issue:44155", "accept": false, - "reason": "Qwen3 auto device mapping CUDA assert and AudioFlamingo3 batched inference token/embedding leakage are unrelated." + "reason": "Qwen3 auto device mapping cudaErrorAssert and AudioFlamingo3 batched inference leakage are different subsystems and failure modes." + }, + { + "left": "issue:43873", + "right": "issue:44387", + "accept": false, + "reason": "Quantization/offloading not working as expected and int4 reserved-memory OOM are both memory-related, but not the same concrete bug." }, { "left": "issue:39692", "right": "issue:44960", "accept": false, - "reason": "SigLIP2 docs/example problems are unrelated to GLM5 model-loading issues." + "reason": "SigLIP2 docs errors and GLM5 are not clearly the same underlying issue; the titles do not point to one shared code path." }, { "left": "issue:43425", "right": "issue:43957", "accept": false, - "reason": "Torch 2.10 compatibility and meta-device loading breakage are different incompatibility classes." + "reason": "Torch 2.10 incompatibility and meta-device model loading breakage are separate compatibility regressions." }, { "left": "issue:45663", "right": "issue:45759", "accept": false, - "reason": "Gemma-4 FSDP2 attention KeyError and Qwen 3.5 multimodal config unwrapping are separate model-specific bugs." + "reason": "Gemma-4 FSDP2 shared_kv_states KeyError and Qwen 3.5/3.6 multimodal config unwrapping are different model/config bugs." }, { - "left": "issue:36010", - "right": "issue:39404", + "left": "issue:44291", + "right": "issue:44756", "accept": false, - "reason": "ImportError for GenerationMixin and Whisper pipeline return_language regression are unrelated." + "reason": "init_empty_weights/_is_hf_initialized TypeError and mmap OOM on Strix Halo are unrelated loading/runtime issues." }, { - "left": "issue:44291", - "right": "issue:44756", + "left": "issue:36010", + "right": "issue:39404", "accept": false, - "reason": "init_empty_weights loading and mmap-on-Strix-Halo OOM are unrelated." + "reason": "An ImportError for GenerationMixin and Whisper pipeline return_language regression are completely different problems." }, { "left": "issue:45663", "right": "issue:45859", "accept": false, - "reason": "Gemma-4 FSDP2 attention state handling and Qwen3.5 MoE CPU offload compatibility are different bugs." + "reason": "Gemma-4 FSDP2 attention state bug and Qwen3.5MoE CPU offload rotary embedding incompatibility are distinct code paths." }, { "left": "issue:44387", "right": "issue:44756", "accept": false, - "reason": "Int4 quantization OOM and disabling mmap on Strix Halo address different memory-pressure sources." + "reason": "Int4 quantization reserved-memory growth and mmap OOM on Strix Halo are not the same bug." } ] }, - "evaluator_result": { - "accept": true, - "feedback": "The summary is grounded in the packet and the soft-edge verdicts stay mostly conservative. The two accepted pairs are well supported by the titles, and the rejected pairs are generally explained with appropriately narrow reasoning rather than overclaiming 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 }, @@ -2644,8 +2581,8 @@ "issue:44164|issue:44756", "issue:43381|issue:44928", "issue:43854|issue:45216", - "issue:45684|issue:45698", "issue:45588|issue:45663", + "issue:45684|issue:45698", "issue:42796|issue:46326", "issue:45292|issue:45335", "issue:43646|issue:43708", @@ -2658,8 +2595,8 @@ "issue:46315|issue:46491", "issue:39404|issue:45701", "issue:43611|issue:43645", - "issue:39404|issue:46665", "issue:44683|issue:45823", + "issue:39404|issue:46665", "issue:44530|issue:45084", "issue:43425|issue:45823", "issue:43425|issue:43716", @@ -2667,226 +2604,244 @@ "issue:43906|issue:44530", "issue:44589|issue:45084" ], - "prepared_review_unit_hash": "ce69eb365f80a40eee85e80abc721e9f1c837d05ed75ea758541df9cc34c0991", + "prepared_review_unit_hash": "b4fa6ca8ecd1046c008a9631a85d861f4eef3ae6e813cd37de3c0335c1cd7186", "component_budget": { "node_count": 473, "item_count": 473, "soft_pair_count": 1738, - "serialized_chars": 901383, - "estimated_input_tokens": 225346, - "estimated_eval_tokens": 450948 + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 34, - "serialized_chars": 37883, - "estimated_input_tokens": 9471, - "estimated_eval_tokens": 19198 + "serialized_chars": 37880, + "estimated_input_tokens": 9470, + "estimated_eval_tokens": 19196 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly heterogeneous, but there is one clear duplicate family around `resize_token_embeddings` failing to propagate updates to tied/output embeddings across model variants.", - "confidence": 0.81, - "canonical_issue_reason": "`issue:45292` is the broadest statement of the shared bug (`resize_token_embeddings` not updating `output_embeddings`), and it cleanly subsumes the Gemma4- and T5Gemma-specific reports.", + "summary": "Most pairs are related only by broad theme, not the same bug. The only clear duplicate cluster is the `resize_token_embeddings` regression across related model variants (`45276`, `45292`, `45335`).", + "confidence": 0.93, + "canonical_issue_reason": "`issue:45292` is the cleanest core report of the shared regression: `resize_token_embeddings` fails to update output embeddings. The other accepted reports are model-specific manifestations of the same missing propagation behavior.", "canonical_pr_reason": null, - "best_issue_reason": "`issue:45292` is the best cluster representative because it captures the common code-path problem at the right abstraction level and is the most likely umbrella issue for the duplicate family.", + "best_issue_reason": "`issue:45292` is the best global representative because it states the underlying API bug in the most generic, reusable way and avoids model-specific detail.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:44292", "right": "issue:44912", "accept": false, - "reason": "Both are quantization/model-loading regressions, but they involve different models and different failure modes (`NVFP4` load error vs `MXFP4` fallback to bf16)." + "reason": "Both are quantization/loading problems, but they affect different models and describe different failure modes." }, { "left": "issue:44155", "right": "issue:45823", "accept": false, - "reason": "Different models and different bugs: track-to-track embedding leak in AudioFlamingo3 vs device mismatch in Gemma4 PLE." + "reason": "Different subsystems and symptoms: batched audio track leakage vs device-mapping mismatch." }, { "left": "issue:43335", "right": "issue:45663", "accept": false, - "reason": "Both touch model internals, but one is a sparse-layer config bug and the other is an FSDP2 `shared_kv_states` rebuild issue; not the same underlying defect." + "reason": "Both mention model configuration/training failures, but the actual bugs are unrelated." }, { "left": "issue:42947", "right": "issue:43381", "accept": false, - "reason": "Both mention gradient checkpointing, but one is about ineffective checkpointing with LoRA and the other about eval-mode restrictions; distinct code paths." + "reason": "Different gradient-checkpointing issues: LoRA effectiveness vs eval-mode prohibition." }, { "left": "issue:43957", "right": "issue:44756", "accept": false, - "reason": "Meta-device loading failures and Strix Halo mmap OOM are unrelated regressions." + "reason": "Unrelated loading/OOM problems despite both involving model initialization." }, { "left": "issue:45276", "right": "issue:45292", "accept": true, - "reason": "Same underlying bug: `resize_token_embeddings` does not propagate to output/tied embeddings; 45276 is the Gemma4-specific manifestation of 45292's broader report." + "reason": "Same underlying bug: `resize_token_embeddings` does not propagate updated embeddings correctly; `45276` is a Gemma4-specific manifestation." }, { "left": "issue:44863", "right": "issue:46468", "accept": false, - "reason": "Both are Nemotron-related, but one is a checkpoint loading failure and the other is a broken official example; insufficient evidence they are the same bug." + "reason": "Both are Nemotron-related, but one is checkpoint loading and the other is an example failure; not the same bug." }, { "left": "issue:45685", "right": "issue:46468", "accept": false, - "reason": "MPS histogram error in a MoE path is not the same as the Nemotron example breakage." + "reason": "MPS integer histogram support and Nemotron example breakage are unrelated." }, { "left": "issue:44479", "right": "issue:44877", "accept": false, - "reason": "Video-input regression and strict-config loading failure are different issues." + "reason": "Different regressions in different model families and code paths." }, { "left": "issue:44164", "right": "issue:44756", "accept": false, - "reason": "`extra_state` save/load handling and mmap/OOM are unrelated." + "reason": "Save/load extra_state handling is unrelated to the Strix Halo mmap OOM issue." }, { "left": "issue:43381", "right": "issue:44928", "accept": false, - "reason": "Eval-mode checkpointing prohibition and RLHF gradient explosion from SDPA fallback are not the same defect." + "reason": "One is about checkpointing in eval mode; the other is an RLHF NaN/SDPA fallback regression." }, { "left": "issue:43854", "right": "issue:45216", "accept": false, - "reason": "GLM-4.7-Flash loading failure and Qwen3.5 `save_pretrained` regression are separate model-loading/save issues." + "reason": "Different model-loading regressions with different failure points and triggers." }, { - "left": "issue:45684", - "right": "issue:45698", + "left": "issue:45588", + "right": "issue:45663", "accept": false, - "reason": "Both are custom-model save/load topics, but one is a file-permissions problem and the other is wrong-module loading; different failure modes." + "reason": "Flash-attention sinkless-model crash and Gemma4 FSDP KeyError are distinct issues." }, { - "left": "issue:45588", - "right": "issue:45663", + "left": "issue:45684", + "right": "issue:45698", "accept": false, - "reason": "Flash-attention `s_aux=None` crash and Gemma4 FSDP2 `KeyError` are unrelated." + "reason": "Both concern save/load custom models, but one is permissions propagation and the other is loading the wrong module." }, { "left": "issue:42796", "right": "issue:46326", "accept": false, - "reason": "Tokenizer/BPE-marker behavior and `continue_final_message` behavior are different tokenizer bugs." + "reason": "Different tokenizer bugs with different APIs and symptoms." }, { "left": "issue:45292", "right": "issue:45335", "accept": true, - "reason": "Same root bug family: `resize_token_embeddings` failing to update downstream embedding modules; `45335` is another model-specific manifestation." + "reason": "Same resize-token-embeddings regression: the updated embeddings are not reflected in model output/decoder embeddings." }, { "left": "issue:43646", "right": "issue:43708", "accept": false, - "reason": "Custom model initialization in notebooks and checkpoint-resume `max_steps` calculation are unrelated." + "reason": "Both are Transformers 5.0 regressions, but one is custom model initialization and the other is checkpoint-resume step calculation." }, { "left": "issue:43828", "right": "issue:43957", "accept": false, - "reason": "Autocast dtype mismatch and meta-device loading failures are different code paths." + "reason": "Different dtype/device-loading regressions with different repros and code paths." }, { "left": "issue:44402", "right": "issue:44451", "accept": false, - "reason": "Tokenizer vocab-size mismatch and ScandiBERT loading failure may be adjacent, but there is not enough evidence they are the same bug." + "reason": "Both are model/tokenizer loading issues, but they concern different checkpoints and different root causes." }, { "left": "issue:43645", "right": "issue:44164", "accept": false, - "reason": "Custom-model initialization in notebooks and `extra_state` save/load handling are different issues." + "reason": "Notebook custom-model initialization and extra_state save/load handling are not the same bug." }, { "left": "issue:44589", "right": "issue:44756", "accept": false, - "reason": "Missing `Float8_e4m3fnStorage` and Strix Halo mmap OOM are unrelated runtime failures." + "reason": "Float8 storage lookup and mmap OOM are unrelated failures." }, { "left": "issue:43425", "right": "issue:43828", "accept": false, - "reason": "Torch-version incompatibility and an autocast dtype mismatch are different problems." + "reason": "Torch version incompatibility is broader and different from the autocast dtype mismatch bug." }, { "left": "issue:45812", "right": "issue:46315", "accept": false, - "reason": "Tokenizer ID regression and `convert_tokens_to_ids` performance regression are distinct issues." + "reason": "Tokenizer ID regression and slow `convert_tokens_to_ids` are distinct tokenizer issues." }, { "left": "issue:46315", "right": "issue:46491", "accept": false, - "reason": "Performance regression and CodeLlama leading-space round-trip regression are unrelated." + "reason": "Performance regression in token lookup is unrelated to CodeLlama leading-space round-trip behavior." }, { "left": "issue:39404", "right": "issue:45701", "accept": false, - "reason": "Whisper `return_language` pipeline breakage and a tokenizer-version tokenization change are not the same underlying bug." + "reason": "Whisper pipeline `return_language` regression and generic tokenization changes are not the same issue." }, { "left": "issue:43611", "right": "issue:43645", "accept": false, - "reason": "Both involve custom models, but one is `base_model_prefix` loading and the other is notebook-side initialization; not the same code-path failure." + "reason": "Both involve custom models in Transformers 5.0, but one is base-model-prefix loading and the other is notebook initialization; too different to merge." }, { - "left": "issue:39404", - "right": "issue:46665", + "left": "issue:44683", + "right": "issue:45823", "accept": false, - "reason": "Whisper pipeline behavior and `from_pretrained` inside Python classes are unrelated." + "reason": "Compiled flex_attention on torch>=2.9 and Gemma4 device mismatch are unrelated." }, { - "left": "issue:44683", - "right": "issue:45823", + "left": "issue:39404", + "right": "issue:46665", "accept": false, - "reason": "Compiled flex_attention on torch>=2.9 and Gemma4 device mismatch are different regressions." + "reason": "Whisper pipeline language return and `from_pretrained` inside Python classes are unrelated behaviors." }, { "left": "issue:44530", "right": "issue:45084", "accept": false, - "reason": "PagedAttentionCache `linear_attention` group-type crash and compile-time non-template-node error are distinct problems." + "reason": "PagedAttentionCache group-type crash and template-node compilation error are different code paths." }, { "left": "issue:43425", "right": "issue:45823", "accept": false, - "reason": "Torch 2.10 incompatibility and Gemma4 device mismatch are unrelated." + "reason": "Torch compatibility issue vs Gemma4 device-mapping forward failure." }, { "left": "issue:43425", "right": "issue:43716", "accept": false, - "reason": "Torch 2.10 incompatibility and Mistral-3 image-preprocessor dtype mismatch are different issues." + "reason": "Torch version incompatibility is not the same as the Mistral image dtype mismatch." + }, + { + "left": "issue:44479", + "right": "issue:44805", + "accept": false, + "reason": "Video-input regression and tensor mask shape mismatch are unrelated." + }, + { + "left": "issue:43906", + "right": "issue:44530", + "accept": false, + "reason": "Isolated reproduction of another issue and Qwen3.5 cache crash are different reports." + }, + { + "left": "issue:44589", + "right": "issue:45084", + "accept": false, + "reason": "Different failures: missing Float8 storage vs compilation of non-template nodes." } ] }, "evaluator_result": { "accept": true, - "feedback": "The analysis is grounded in the packet: the 45292/45276/45335 cluster is a coherent resize_token_embeddings family, and the rest are handled conservatively as non-duplicates. No soft-edge verdict overstates the evidence." + "feedback": "Grounded and conservative overall. The only merged cluster that is clearly supported by the packet is the `resize_token_embeddings` regression across `45276`, `45292`, and `45335`; the other pairwise rejections stay appropriately narrow and do not overclaim duplicates." }, "evaluator_used": true, "retried": false, @@ -2964,17 +2919,17 @@ "issue:44671|issue:44991", "issue:36032|issue:43452", "issue:46050|issue:46326", - "issue:43299|issue:43931", - "issue:44164|issue:45759", "issue:43906|issue:46326", + "issue:44164|issue:45759", + "issue:43299|issue:43931", "issue:44991|issue:45081", "issue:43856|issue:44756", "issue:43596|issue:44683", "issue:44164|issue:44479", "issue:44488|issue:45081", "issue:44462|issue:46489", - "issue:42796|issue:43404", "issue:44488|issue:44987", + "issue:42796|issue:43404", "issue:43611|issue:43644", "issue:39404|issue:44671", "issue:44737|issue:45859", @@ -2985,209 +2940,212 @@ "issue:43582|issue:43632", "issue:43643|issue:44530" ], - "prepared_review_unit_hash": "63ec51d1a829e7aa4aca4c44405d31d21aa51fdc759508136f34f947bb75d325", + "prepared_review_unit_hash": "ab598da95210c25570707dac98d055422afb1764c223fd87cf927992e1e67e94", "component_budget": { "node_count": 473, "item_count": 473, "soft_pair_count": 1738, - "serialized_chars": 901383, - "estimated_input_tokens": 225346, - "estimated_eval_tokens": 450948 + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 28, - "serialized_chars": 35587, - "estimated_input_tokens": 8897, - "estimated_eval_tokens": 18050 + "serialized_chars": 35589, + "estimated_input_tokens": 8898, + "estimated_eval_tokens": 18052 }, "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 (tokenizer loading, model loading, quantization, docs, and a few training/runtime bugs). I don\u2019t see any soft pair that looks like the same underlying bug, so none should be merged as duplicates. If one issue has to act as an umbrella representative, issue 43611 is the broadest loading-regression report here.", - "confidence": 0.79, - "canonical_issue_reason": "Issue 43611 is the broadest and most central v5 regression in the set: it describes a generic model-loading breakage tied to `base_model_prefix`, which is closer to the common theme than the more model-specific or docs-only reports.", + "summary": "This cluster is a loose grouping of mostly distinct Transformers v5 regression reports, centered on model/tokenizer loading and behavior changes. They read as related by theme, not as true duplicates, so the soft edges should mostly be rejected.", + "confidence": 0.82, + "canonical_issue_reason": "issue:43611 is the broadest representative of the cluster\u2019s main theme: v5 breaking model loading behavior. It is generic enough to stand in for the larger set of loading regressions without implying the specific tokenizer/model bugs are duplicates.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 43611 is the best single representative because it is the most general Transformers v5 loading issue in the cluster and could plausibly anchor discussion of related v5 loading regressions.", + "best_issue_reason": "issue:43611 is the best single issue to anchor this cluster because it is a general, high-level regression report rather than one narrow model-specific symptom.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43901", "right": "issue:44509", "accept": false, - "reason": "Both are docs issues, but they cover different APIs/tasks (`return_all_scores` vs removed pipeline tasks); not the same bug." + "reason": "Both are docs issues, but they cover different pipeline/documentation regressions: stale return_all_scores wording vs removed text2text/summarization/translation tasks." }, { "left": "issue:44265", "right": "issue:44898", "accept": false, - "reason": "Different failure modes and code paths: `torch.export`/`torch_compilable_check` versus Perceiver interpolation handling." + "reason": "Both involve model/export behavior, but one is about torch.export and torch_compilable_check, while the other is a Perceiver image-classification resolution/interpolation bug." }, { "left": "issue:39404", "right": "issue:44912", "accept": false, - "reason": "Whisper pipeline language return regression and MXFP4 quantization fallback are unrelated." + "reason": "Whisper pipeline return_language and MXFP4 quantization fallback on git-oss are unrelated bugs in different subsystems." }, { "left": "issue:43994", "right": "issue:44079", "accept": false, - "reason": "One is a model/pipeline output quality issue; the other is a `ModelOutput` key assignment bug." + "reason": "One is a SigLIP2 output quality/loading regression, the other is a ModelOutput key assignment bug; same broad area, different code-path problem." }, { "left": "issue:43425", "right": "issue:44918", "accept": false, - "reason": "Torch version incompatibility and Qwen3.5 embedding unpacking in TRL are different problems." + "reason": "Torch 2.10 compatibility and Qwen3.5 embedding unpacking in TRL are unrelated failures." }, { "left": "issue:44671", "right": "issue:44991", "accept": false, - "reason": "Different models and root causes: CamemBERT masked-LM prediction regression vs tokenizer loading for EMBEDDIA/est-roberta." + "reason": "Both are tokenizer/model regressions in v5, but they affect different models and likely different loading paths." }, { "left": "issue:36032", "right": "issue:43452", "accept": false, - "reason": "Both involve loading, but the specific tokenizer conflict and the `gguf_file` from_pretrained breakage are distinct." + "reason": "T5 tokenizer special-token conflict and gguf_file loading breakage are different tokenizer-loading issues." }, { "left": "issue:46050", "right": "issue:46326", "accept": false, - "reason": "Quantized save/load corruption and `continue_final_message` behavior for mistral_common tokenizers are unrelated." + "reason": "Quantized save/load corruption and mistral_common continue_final_message behavior are not the same bug." }, { - "left": "issue:43299", - "right": "issue:43931", + "left": "issue:43906", + "right": "issue:46326", "accept": false, - "reason": "Both mention Qwen3-VL, but one is a v5 dev0 loading breakage and the other is a weight-shape mismatch; not the same concrete bug." + "reason": "An isolated reproduction of an older issue is not the same underlying tokenizer behavior as mistral_common continue_final_message." }, { "left": "issue:44164", "right": "issue:45759", "accept": false, - "reason": "`extra_state` serialization handling and `text_config` unwrapping in composite Qwen configs are different loading paths." + "reason": "extra_state save/load handling and text_config unwrapping in composite configs are distinct serialization/configuration bugs." }, { - "left": "issue:43906", - "right": "issue:46326", - "accept": false, - "reason": "The reproduction issue is too generic and there is no evidence it is the same `continue_final_message` tokenizer bug." + "left": "issue:43299", + "right": "issue:43931", + "accept": true, + "reason": "Both point to the same Qwen3-VL MoE model-loading failure in v5, with the second looking like a concrete manifestation of the same shape-mismatch/load regression." }, { "left": "issue:44991", "right": "issue:45081", "accept": false, - "reason": "Both are tokenizer-loading regressions, but they concern different models and different internal failures (`backend_tokenizer` vs tokenizer selection)." + "reason": "Both are tokenizer regressions, but one is a generic tokenizer load failure for a specific model and the other is a mistral regex patch crash." }, { "left": "issue:43856", "right": "issue:44756", "accept": false, - "reason": "Both are memory-related, but one is Qwen3 MoE training efficiency and the other is Strix Halo mmap OOM avoidance." + "reason": "Qwen3 MoE training memory usage and disabling mmap on Strix Halo are unrelated performance/oom reports." }, { "left": "issue:43596", "right": "issue:44683", "accept": false, - "reason": "Deepspeed Zero3/BertModel indexing and compiled flex attention on newer torch versions are separate runtime issues." + "reason": "DeepSpeed ZeRO-3/BertModel indexing and compiled flex_attention on torch>=2.9 are different failure modes." }, { "left": "issue:44164", "right": "issue:44479", "accept": false, - "reason": "Saving/loading `extra_state` is unrelated to the v5 video-input regression for Qwen VL models." + "reason": "save/from_pretrained extra_state handling is unrelated to the v5 video-input regression on Qwen VL models." }, { "left": "issue:44488", "right": "issue:45081", "accept": false, - "reason": "Both are tokenizer load failures, but the models and failure mechanisms differ substantially." + "reason": "Both involve tokenizer loading, but one is a missing tokenizer load for a model repo and the other is a specific mistral regex patch backend_tokenizer crash." }, { "left": "issue:44462", "right": "issue:46489", "accept": false, - "reason": "`AutoTokenizer` ignoring `tokenizer.json` and DeepSeek-Coder v1 output corruption are not the same bug, even if both are tokenizer-related." + "reason": "AutoTokenizer ignoring tokenizer.json and DeepSeek-Coder v1 incorrect output after a prior fix are not clearly the same underlying bug." }, { - "left": "issue:42796", - "right": "issue:43404", + "left": "issue:44488", + "right": "issue:44987", "accept": false, - "reason": "Tokenizer BPE-marker handling and Mistral3 `lm_head` weight tying are unrelated issues." + "reason": "Both mention model/tokenizer loading failures, but they concern different repositories and different root causes." }, { - "left": "issue:44488", - "right": "issue:44987", + "left": "issue:42796", + "right": "issue:43404", "accept": false, - "reason": "Different failing models and different loading regressions; no shared concrete code path." + "reason": "Ministral tokenizer BPE-marker handling and Mistral3 lm_head weight tying are separate model/tokenizer issues." }, { "left": "issue:43611", "right": "issue:43644", "accept": false, - "reason": "Both are v5 regressions, but one is a `base_model_prefix` loading issue and the other is non-persistent buffer initialization corruption." + "reason": "Both are v5 loading regressions, but one is about base_model_prefix handling and the other about non-persistent buffers being initialized with junk." }, { "left": "issue:39404", "right": "issue:44671", "accept": false, - "reason": "Whisper pipeline language return and CamemBERT masked-LM prediction problems are unrelated." + "reason": "Whisper pipeline language return and CamemBERT masked-LM prediction regressions are unrelated model/pipeline bugs." }, { "left": "issue:44737", "right": "issue:45859", "accept": false, - "reason": "XLNet CPU placement bug and Qwen3.5 MoE rotary embedding CPU-offload incompatibility are different model-specific device issues." + "reason": "Both are device/offload related, but XLNet CPU arange placement and Qwen3.5 MoE rotary embedding CPU offload incompatibility are different issues." }, { "left": "issue:46285", "right": "issue:46402", "accept": false, - "reason": "CLIP structure inconsistency and Sam3 text encoder weights not loading are separate model-loading regressions." + "reason": "CLIP structure inconsistency and Sam3Model text-encoder weights not loading are distinct architectural/load regressions." }, { "left": "issue:43716", "right": "issue:45084", "accept": false, - "reason": "Mistral image-preprocessor dtype mismatch and template-node compile error are unrelated." + "reason": "Mistral-3 image preprocessor dtype mismatch and non-template-node compilation failure are unrelated." }, { "left": "issue:36032", "right": "issue:42915", "accept": false, - "reason": "Tokenizer special-tokens conflict and Qwen3 MoE FP8 config failure are different bug classes." + "reason": "T5 tokenizer special-token conflict and Qwen3 MoE FP8 config failure are different model-loading problems." }, { "left": "issue:44661", "right": "issue:44683", "accept": false, - "reason": "`add-new-model-like` mapping logic and flex attention compilation failures do not share the same underlying defect." + "reason": "add-new-model-like / TOKENIZER_MAPPING_NAMES and compiled flex_attention are unrelated." }, { "left": "issue:43582", "right": "issue:43632", "accept": false, - "reason": "Apple Silicon warmup TypeError and `_is_hf_initialized` flag breakage are unrelated runtime issues." + "reason": "Apple Silicon caching_allocator_warmup TypeError and the _is_hf_initialized flag regression are different bugs." }, { "left": "issue:43643", "right": "issue:44530", "accept": false, - "reason": "`trust_remote_code` config field loss and PagedAttentionCache linear-attention crash are different code paths." + "reason": "AutoConfig trust_remote_code field loss and PagedAttentionCache linear_attention crashes are unrelated." } ] }, - "evaluator_result": null, - "evaluator_used": false, + "evaluator_result": { + "accept": true, + "feedback": "The summary stays grounded in the packet and is appropriately conservative about the cluster being a loose theme rather than a duplicate set. The soft-edge verdicts generally avoid overclaiming, and the one accepted pair is plausibly the same Qwen3-VL family regression." + }, + "evaluator_used": true, "retried": false, - "accepted_nontrivial_soft_edge": false, + "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null }, @@ -3257,10 +3215,10 @@ "issue:40990|issue:43421", "issue:44451|issue:44987", "issue:42915|issue:44530", - "issue:34567|issue:40444", "issue:43541|issue:45084", - "issue:42175|issue:46178", + "issue:34567|issue:40444", "issue:43644|issue:43646", + "issue:42175|issue:46178", "issue:44811|issue:45584", "issue:44206|issue:44220", "issue:43883|issue:44488", @@ -3272,12 +3230,12 @@ "issue:44987|issue:44991", "issue:43262|issue:43377", "issue:43746|issue:46326", - "issue:44464|issue:45987", "issue:43883|issue:44991", + "issue:44464|issue:45987", "issue:43646|issue:44534", "issue:44403|issue:45081", - "issue:43645|issue:44534", "issue:42915|issue:45084", + "issue:43645|issue:44534", "issue:43404|issue:44466", "issue:43377|issue:43819", "issue:43994|issue:44610", @@ -3289,190 +3247,238 @@ "issue:43322|issue:43540", "issue:44265|issue:44610" ], - "prepared_review_unit_hash": "73bd89098d7cfbc4db47f059876780ed39a36b1b51eadc66ac3f44fa62e07f44", + "prepared_review_unit_hash": "00732948046e6ba93428485b2f85e9b96edcf5091da0e81a5e70df484fcfe318", "component_budget": { "node_count": 473, "item_count": 473, "soft_pair_count": 1738, - "serialized_chars": 901383, - "estimated_input_tokens": 225346, - "estimated_eval_tokens": 450948 + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 34, - "serialized_chars": 38071, - "estimated_input_tokens": 9518, - "estimated_eval_tokens": 19292 + "serialized_chars": 38077, + "estimated_input_tokens": 9520, + "estimated_eval_tokens": 19296 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster is mostly a loose set of unrelated transformer/model-loading regressions that only share broad surface similarity (v5, tokenizer, audio, loading, tracing). I would not merge any of the soft pairs as duplicates based on the available evidence.", - "confidence": 0.64, - "canonical_issue_reason": "issue:43646 is the broadest, most general regression in the set and best represents the cluster\u2019s recurring theme of Transformers v5 breakage, even though the nearby issues are not true duplicates.", + "summary": "This cluster is mostly a grab bag of unrelated Transformers issues. The only clear duplicate pair is the v5 non-persistent-buffer regression; the rest are superficially similar but point to different models, code paths, or failure modes.", + "confidence": 0.72, + "canonical_issue_reason": "issue:43644 is the cleanest canonical because it is the earlier, broader report of the same v5 non-persistent-buffer corruption bug, and issue:44534 reads like a near-verbatim follow-up.", "canonical_pr_reason": null, - "best_issue_reason": "issue:43646 has the clearest generalizable title and issue framing among this heterogeneous set, so it is the best single issue to anchor the cluster.", + "best_issue_reason": "issue:43644 is the best representative of the cluster because it captures the only strong duplicate relationship and states the underlying regression clearly.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:40990", "right": "issue:43421", "accept": false, - "reason": "Perplexity on a model/dataset issue vs tokenizer post-processor runtime updates; different bug classes and code paths." + "reason": "Both concern training/Tokenizer behavior, but one is early stopping metric selection and the other is runtime special-token post-processing; different bugs." }, { "left": "issue:44451", "right": "issue:44987", "accept": false, - "reason": "Both are loading failures, but they involve different models and likely different breakpoints; no evidence of the same underlying bug." + "reason": "Both are model-loading failures, but they affect different models and likely different config/tokenizer paths; too broad to merge." }, { "left": "issue:42915", "right": "issue:44530", "accept": false, - "reason": "One is FP8 config failure on Qwen3Moe, the other is PagedAttentionCache group-type parsing on Qwen3.5; distinct failures." - }, - { - "left": "issue:34567", - "right": "issue:40444", - "accept": false, - "reason": "Trainer token-count tracking vs multimodal IterableDataset loading; unrelated symptoms and subsystems." + "reason": "Both are multimodal-model failures, but one is FP8 config on Qwen3Moe and the other is PagedAttention cache group-type handling on Qwen3.5; different code paths." }, { "left": "issue:43541", "right": "issue:45084", "accept": false, - "reason": "Torch Dynamo grouped_mm tracing failure vs template compilation TypeError; both involve compilation, but not the same concrete bug." + "reason": "A torch dynamo grouped_mm tracing error and a template-node compilation error are unrelated failures." }, { - "left": "issue:42175", - "right": "issue:46178", + "left": "issue:34567", + "right": "issue:40444", "accept": false, - "reason": "Packaging/backends issues, but one is missing tensorflow in an extra and the other is an unmapped tensorflow_text backend; not the same defect." + "reason": "Num-input-token accounting and Qwen2.5-VL iterable-dataset multi-image loading are unrelated issues." }, { "left": "issue:43644", "right": "issue:43646", "accept": false, - "reason": "Non-persistent buffer initialization bug vs custom model initialization breakage; related version window, different issues." + "reason": "Both are v5 regressions, but one is buffer serialization/junk state and the other is custom model initialization; not the same bug." + }, + { + "left": "issue:42175", + "right": "issue:46178", + "accept": false, + "reason": "One is missing tensorflow in an extra install, the other is an unknown backend mapping error for tensorflow_text; different packaging/runtime issues." }, { "left": "issue:44811", "right": "issue:45584", "accept": false, - "reason": "Whisper batch_decode skip_special_tokens handling vs empty transcription after align_special_tokens; same area, different concrete bug." + "reason": "Both involve Whisper token handling, but batch_decode skip_special_tokens and empty-transcription handling after align_special_tokens are distinct defects." }, { "left": "issue:44206", "right": "issue:44220", "accept": false, - "reason": "Both are audio feature-extraction related, but one is a specific unsupported center arg regression and the other is a different fbank helper issue." + "reason": "One is a LASRFeatureExtractor API regression, the other is a low-level fbank feature extraction issue; not the same code path." }, { "left": "issue:43883", "right": "issue:44488", "accept": false, - "reason": "Molmo tied-weights attribute error vs model loading failure for a specific BERT checkpoint; unrelated." + "reason": "Both mention loading models, but one is an all_tied_weights_keys attribute error and the other is a specific model-loading regression; different failures." }, { "left": "issue:40444", "right": "issue:46326", "accept": false, - "reason": "Multimodal iterable dataset failure vs mistral_common continue_final_message behavior; different feature areas." + "reason": "Multiple-image iterable dataset handling and mistral_common continue_final_message behavior are unrelated." }, { "left": "issue:36032", "right": "issue:43645", "accept": false, - "reason": "Tokenizer method conflict vs Jupyter custom model initialization regression; not the same bug." + "reason": "Tokenizer class-method conflict and custom model initialization in notebooks are different problems." }, { "left": "issue:43452", "right": "issue:43482", "accept": false, - "reason": "Both touch GGUF loading, but one is a generic from_pretrained breakage and the other is a specific Qwen2.5-GGUF failure; insufficiently identical." + "reason": "GGUF-related loader breakage and a specific Qwen2.5-GGUF v5 loading failure are not the same underlying issue." }, { "left": "issue:43317", "right": "issue:43482", "accept": false, - "reason": "Device-map/offload dequantized-model loading vs GGUF loading in v5; different loading paths and failure modes." + "reason": "Dequantized model offload with device_map=auto and Qwen2.5-GGUF loading failure are separate loader paths." }, { "left": "issue:45084", "right": "issue:45507", "accept": false, - "reason": "Template compilation error vs GraniteMoEHybrid calling an invalid method; no shared root cause." + "reason": "Template compilation failure and GraniteMoEHybrid invalid method call are unrelated." }, { "left": "issue:44987", "right": "issue:44991", "accept": false, - "reason": "Both are latest-version loading regressions, but on different models and likely different code paths." + "reason": "Both are model/tokenizer loading regressions on v5, but for different models and failure sites; not enough to unify." }, { "left": "issue:43262", "right": "issue:43377", "accept": false, - "reason": "Audio chat-template sampling-rate default vs MIMI batched/single padding-mask mismatch; different bugs." + "reason": "Audio processor sampling-rate defaulting and MIMI encoder batching/padding-mask behavior are unrelated." }, { "left": "issue:43746", "right": "issue:46326", "accept": false, - "reason": "PEFT checkpoint loading for GraniteSpeech vs mistral_common continue_final_message; unrelated." + "reason": "PEFT local-checkpoint loading for GraniteSpeech and mistral_common continue_final_message behavior are different bugs." }, { - "left": "issue:44464", - "right": "issue:45987", + "left": "issue:43883", + "right": "issue:44991", "accept": false, - "reason": "Chunked generation inconsistency under compiled forward vs cache length contract mismatch; related to generation internals but not the same bug." + "reason": "An attribute error during weight tying and a tokenizer loading regression for a different model are not the same issue." }, { - "left": "issue:43883", - "right": "issue:44991", + "left": "issue:44464", + "right": "issue:45987", "accept": false, - "reason": "Different affected models and symptoms; one is a missing attribute, the other is a tokenizer loading regression." + "reason": "Chunked generation inconsistency under compiled forward and StaticCache seq-length type mismatch are different runtime behaviors." }, { "left": "issue:43646", "right": "issue:44534", "accept": false, - "reason": "These are the same broad v5 theme but not the same issue: one is custom model init, the other is non-persistent buffers being corrupted." + "reason": "Custom model initialization and non-persistent-buffer junk are different regressions, even if both are v5-related." }, { "left": "issue:44403", "right": "issue:45081", "accept": false, - "reason": "Loading-noise regression vs Mistral regex patch crashing on backend_tokenizer; not the same underlying defect." + "reason": "Noise on loading a transformer and a Mistral regex patch crash are unrelated." + }, + { + "left": "issue:42915", + "right": "issue:45084", + "accept": false, + "reason": "FP8 config failure on Qwen3Moe and template-node compilation failure are different code paths." }, { "left": "issue:43645", "right": "issue:44534", "accept": false, - "reason": "Jupyter notebook custom model initialization breakage vs buffer junk during v5 load; different failure modes." + "reason": "Custom notebook model initialization and non-persistent-buffer corruption are distinct v5 problems." + }, + { + "left": "issue:43404", + "right": "issue:44466", + "accept": false, + "reason": "lm_head tying in Mistral3ForConditionalGeneration and device-dependent lm_head serialization are related only at a high level; the concrete bugs differ." + }, + { + "left": "issue:43377", + "right": "issue:43819", + "accept": false, + "reason": "Batched-vs-single MIMI encoder outputs and DAC.from_latents missing STE are separate model-specific correctness bugs." + }, + { + "left": "issue:43994", + "right": "issue:44610", + "accept": false, + "reason": "Siglip2 nonsensical outputs and OmDet-Turbo processor size mismatch are different vision-model issues." + }, + { + "left": "issue:44849", + "right": "issue:45325", + "accept": false, + "reason": "Qwen3.5 hidden_states behavior and Qwen2.5-VL rope index scaling are separate model bugs." + }, + { + "left": "issue:28530", + "right": "issue:34567", + "accept": false, + "reason": "Early stopping metric lookup and token-count tracking are unrelated trainer issues." + }, + { + "left": "issue:44451", + "right": "issue:45081", + "accept": false, + "reason": "A model load failure for one checkpoint and a Mistral tokenizer regex crash are not the same underlying bug." + }, + { + "left": "issue:44987", + "right": "issue:45081", + "accept": false, + "reason": "Both involve loading regressions, but one is a model-load failure and the other is a tokenizer regex patch failure; different code paths." }, { "left": "issue:42915", "right": "issue:44291", "accept": false, - "reason": "FP8/Qwen3Moe failure vs init_empty_weights unexpected argument error; distinct initialization paths." + "reason": "FineGrainedFP8Config on Qwen3Moe and init_empty_weights/_is_hf_initialized incompatibility are unrelated." }, { "left": "issue:43322", "right": "issue:43540", "accept": false, - "reason": "Llava Next loading segfault vs Qwen3OmniMoe video-input ValueError; different multimodal bugs." + "reason": "Llava Next loading segfault and Qwen3OmniMoe video-processing ValueError are distinct multimodal issues." }, { "left": "issue:44265", "right": "issue:44610", "accept": false, - "reason": "torch.export/exportability issue vs processor image-size mismatch; unrelated." + "reason": "torch.export exportability with torch_compilable_check and an image processor input-size mismatch are unrelated." } ] }, @@ -3501,24 +3507,24 @@ "issue:36032", "issue:40444", "issue:40990", + "issue:41553", "issue:41720", "issue:42796", "issue:42915", "issue:43089", "issue:43296", - "issue:43317", "issue:43381", "issue:43421", "issue:43425", "issue:43482", "issue:43525", "issue:43531", + "issue:43582", "issue:43611", "issue:43632", "issue:43742", "issue:43746", "issue:43792", - "issue:43856", "issue:44062", "issue:44079", "issue:44220", @@ -3556,235 +3562,235 @@ "issue:36032|issue:41720", "issue:43531|issue:44877", "issue:44220|issue:44683", - "issue:40444|issue:45325", - "issue:43611|issue:44534", "issue:43425|issue:44403", "issue:43296|issue:43531", + "issue:40444|issue:45325", + "issue:43611|issue:44534", "issue:43632|issue:44756", "issue:45812|issue:46491", "issue:42796|issue:43746", "issue:44403|issue:44991", - "issue:33666|issue:40444", "issue:43381|issue:45923", + "issue:33666|issue:40444", "issue:43792|issue:44589", - "issue:44079|issue:45663", "issue:40990|issue:44811", + "issue:44079|issue:45663", "issue:44291|issue:44589", "issue:43632|issue:44534", "issue:43421|issue:44568", "issue:43482|issue:43525", "issue:35707|issue:40444", - "issue:43317|issue:43856", - "issue:43482|issue:45538" + "issue:43482|issue:45538", + "issue:41553|issue:43582" ], - "prepared_review_unit_hash": "a31eadbb56df82fb6c1331217fd447c9a192ce4560f05728bc27fdc12c6c6e11", + "prepared_review_unit_hash": "cc636a9ea0aae7f34a809876fb4091556f9005e5a5a3697f5b2a01a8b34c83b2", "component_budget": { "node_count": 473, "item_count": 473, "soft_pair_count": 1738, - "serialized_chars": 901383, - "estimated_input_tokens": 225346, - "estimated_eval_tokens": 450948 + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 30, - "serialized_chars": 36546, - "estimated_input_tokens": 9137, - "estimated_eval_tokens": 18530 + "serialized_chars": 36533, + "estimated_input_tokens": 9134, + "estimated_eval_tokens": 18524 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This is a heterogeneous cluster of mostly unrelated Hugging Face/transformers issues; the soft pairs only share broad subsystem vocabulary, not the same underlying bug or change.", - "confidence": 0.96, - "canonical_issue_reason": "If forced to anchor the cluster, issue:43089 is the broadest and most general problem statement here (generation performance/overhead) rather than a narrow model-specific regression.", + "summary": "This cluster is heterogeneous: most items are unrelated model-specific bugs, tokenizer regressions, or environment/version incompatibilities. I don\u2019t see any soft pair that is clearly the same underlying bug/change, so I would not merge any of the candidate edges.", + "confidence": 0.91, + "canonical_issue_reason": "issue:45812 is the strongest anchor if one must be chosen: it is a broad, concrete v4\u2192v5 regression affecting all Granite models and has multiple inbound references. The rest of the cluster is mostly narrow, unrelated model/version-specific issues.", "canonical_pr_reason": null, - "best_issue_reason": "issue:43089 is the most plausible representative because it is system-level and generic, but the cluster is too mixed to treat it as a true duplicate hub.", + "best_issue_reason": "issue:45812 is the best representative issue overall because it has the broadest impact and clearest regression framing among an otherwise mixed set of mostly unrelated bugs.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43089", "right": "issue:45987", "accept": false, - "reason": "Different problems: generation dispatch/GPU sync overhead vs. StaticCache return type contract." + "reason": "Both mention GPU-side generation behavior, but one is a generation performance overhead report and the other is a cache API contract bug; different code paths and symptoms." }, { "left": "issue:36032", "right": "issue:43742", "accept": false, - "reason": "Both are loading-time errors, but one is a T5 tokenizer method conflict and the other is a MobileLLM key error." + "reason": "A tokenizer class-name conflict is unrelated to a KeyError while loading a specific model; no shared underlying bug." }, { "left": "issue:44062", "right": "issue:44987", "accept": false, - "reason": "Both involve tokenizer/model loading, but the specific failure modes and affected models are unrelated." + "reason": "One is an AddedToken keyword collision, the other is a model load failure on transformers>=5.1.0; too vague to treat as the same bug." }, { "left": "issue:42915", "right": "issue:44589", "accept": false, - "reason": "Different regressions: Qwen3MoE with FineGrainedFP8Config vs. missing Float8 storage type." + "reason": "Both are FP8-related, but one is a Qwen3MoE FineGrainedFP8Config failure and the other is a missing Float8 storage type; different failure mechanisms." }, { "left": "issue:44479", "right": "issue:45325", "accept": false, - "reason": "Both are Qwen2.5-VL-related, but one is a broad video-input regression and the other is a rope index scaling bug for still images." + "reason": "Both involve Qwen VL models, but one is a video-input regression and the other is still-image temporal position-id scaling; different bugs." }, { "left": "issue:45216", "right": "issue:45357", "accept": false, - "reason": "Same model family, but different save_pretrained bugs affecting different saved keys and versions." + "reason": "Both are Qwen3.5 save_pretrained regressions, but one reports incorrect saved checkpoints broadly while the other is specifically about visual encoder keys; not clearly the same defect." }, { "left": "issue:44991", "right": "issue:46315", "accept": false, - "reason": "Tokenizer loading regression vs. convert_tokens_to_ids performance regression; different code paths." + "reason": "Tokenizer loading failure vs performance regression in convert_tokens_to_ids; unrelated symptoms and code paths." }, { "left": "issue:36032", "right": "issue:41720", "accept": false, - "reason": "Unrelated bugs: tokenizer attribute conflict vs. device-mapping cuda assert." + "reason": "A T5 tokenizer method-conflict bug is unrelated to a Qwen3 auto-device-map CUDA assert." }, { "left": "issue:43531", "right": "issue:44877", "accept": false, - "reason": "Both mention configuration/loading, but sliding_window behavior and strict config parsing are separate issues." + "reason": "Sliding-window behavior in Qwen3-MoE and strict config loading for granite_speech are different issues." }, { "left": "issue:44220", "right": "issue:44683", "accept": false, - "reason": "Different feature failures: audio fbank extraction vs. compiled flex_attention on newer torch." + "reason": "Audio fbank extraction and compiled flex_attention are unrelated subsystems." }, { - "left": "issue:40444", - "right": "issue:45325", + "left": "issue:43425", + "right": "issue:44403", "accept": false, - "reason": "One is finetuning with multiple images per prompt; the other is a still-image position_ids scaling regression." + "reason": "Torch version incompatibility and noisy load logging are not the same underlying bug." }, { - "left": "issue:43611", - "right": "issue:44534", + "left": "issue:43296", + "right": "issue:43531", "accept": false, - "reason": "Both are Transformers v5 regressions, but the base_model_prefix loading bug is distinct from non-persistent buffer initialization." + "reason": "PaddleOCR-VL model loading in vLLM and Qwen3-MoE sliding_window handling are unrelated." }, { - "left": "issue:43425", - "right": "issue:44403", + "left": "issue:40444", + "right": "issue:45325", "accept": false, - "reason": "Torch version incompatibility is not the same as noisy model-loading behavior." + "reason": "Both are Qwen2.5-VL-related, but one is finetuning with multiple images per prompt and the other is rope/position-id scaling for still images; different code paths." }, { - "left": "issue:43296", - "right": "issue:43531", + "left": "issue:43611", + "right": "issue:44534", "accept": false, - "reason": "Different model/config issues: PaddleOCR-VL load failure vs. Qwen3-MoE sliding_window behavior." + "reason": "Base-model-prefix loading and non-persistent buffer initialization are distinct v5 regressions." }, { "left": "issue:43632", "right": "issue:44756", "accept": false, - "reason": "Both are v5 regressions, but one is _is_hf_initialized handling and the other is mmap/OOM on Strix Halo." + "reason": "An initialization flag regression and an mmap/OOM workaround are unrelated." }, { "left": "issue:45812", "right": "issue:46491", "accept": false, - "reason": "Both are tokenizer regressions, but one is Granite token ID assignment and the other is CodeLlama space-preservation round-trip behavior." + "reason": "Both are tokenizer regressions, but one is wrong token IDs for Granite and the other is a leading-space round-trip issue in CodeLlama; not the same bug." }, { "left": "issue:42796", "right": "issue:43746", "accept": false, - "reason": "Tokenizer BPE-marker handling and GraniteSpeech PEFT checkpoint loading are different bugs." + "reason": "Tokenizer BPE-marker handling and GraniteSpeech PEFT checkpoint loading are unrelated." }, { "left": "issue:44403", "right": "issue:44991", "accept": false, - "reason": "Noise during loading and tokenizer loading failures are too generic to be the same bug." + "reason": "General load-noise logging is unrelated to a tokenizer loading failure for EMBEDDIA/est-roberta." }, { - "left": "issue:33666", - "right": "issue:40444", + "left": "issue:43381", + "right": "issue:45923", "accept": false, - "reason": "Both concern multi-image/multi-GPU Qwen VL workflows, but one is training setup and the other is iterable-dataset finetuning failure." + "reason": "Eval-mode gradient-checkpointing restriction and a missing supports_gradient_checkpointing flag are related in theme only, not the same concrete defect." }, { - "left": "issue:43381", - "right": "issue:45923", + "left": "issue:33666", + "right": "issue:40444", "accept": false, - "reason": "Gradient checkpointing in eval mode is unrelated to a missing supports_gradient_checkpointing flag." + "reason": "Both are Qwen VL training issues, but one is multi-GPU training and the other is iterable-dataset/multi-image prompt failure; different bugs." }, { "left": "issue:43792", "right": "issue:44589", "accept": false, - "reason": "Whisper runtime failure vs. Float8 storage lookup error; no shared code path." + "reason": "Whisper runtime failure and Float8 storage lookup error are unrelated." }, { - "left": "issue:44079", - "right": "issue:45663", + "left": "issue:40990", + "right": "issue:44811", "accept": false, - "reason": "ModelOutput key assignment bug is unrelated to Gemma4 FSDP2 shared_kv_states rebuild failure." + "reason": "High perplexity on a language-model benchmark and Whisper batch_decode behavior are unrelated." }, { - "left": "issue:40990", - "right": "issue:44811", + "left": "issue:44079", + "right": "issue:45663", "accept": false, - "reason": "Perplexity anomaly on gpt-oss and Whisper batch_decode special-token handling are different symptoms and subsystems." + "reason": "ModelOutput key assignment and Gemma4 FSDP2 shared_kv_states rebuilding are different code-path problems." }, { "left": "issue:44291", "right": "issue:44589", "accept": false, - "reason": "Both mention Transformers v5 and initialization/typing, but the unexpected _is_hf_initialized argument bug is separate from Float8 storage resolution." + "reason": "init_empty_weights argument handling and Float8 storage lookup are unrelated regressions." }, { "left": "issue:43632", "right": "issue:44534", "accept": false, - "reason": "Distinct regressions: _is_hf_initialized init logic vs. non-persistent buffer junk filling." + "reason": "_is_hf_initialized handling and non-persistent buffer junk values are different initialization bugs." }, { "left": "issue:43421", "right": "issue:44568", "accept": false, - "reason": "Runtime special-token update behavior is different from mdeberta add_special_tokens not adding BOS/EOS." + "reason": "Runtime post-processor updates for special tokens and add_special_tokens BOS/EOS behavior in mdeberta-v3 are not the same issue." }, { "left": "issue:43482", "right": "issue:43525", "accept": false, - "reason": "Qwen2.5-GGUF loading failure and Llama4Config missing pad_token_id are separate model-specific issues." + "reason": "Qwen2.5-GGUF loading failure and missing pad_token_id on Llama4Config are unrelated." }, { "left": "issue:35707", "right": "issue:40444", "accept": false, - "reason": "Both are generation/training adjacent, but progressive generation with inputs_embeds is unrelated to iterable-dataset multi-image finetuning." + "reason": "Progressive generation with inputs_embeds/past_key_values is unrelated to Qwen2.5-VL multi-image finetuning failure." }, { - "left": "issue:43317", - "right": "issue:43856", + "left": "issue:43482", + "right": "issue:45538", "accept": false, - "reason": "Both are model loading/training issues, but dequantized device_map offload is unrelated to Qwen3 MoE memory usage." + "reason": "GGUF loading in transformers v5 and CLIPTokenizer model_max_length are unrelated tokenization/loading issues." }, { - "left": "issue:43482", - "right": "issue:45538", + "left": "issue:41553", + "right": "issue:43582", "accept": false, - "reason": "Qwen2.5-GGUF loading and CLIPTokenizer model_max_length are unrelated tokenizer/model-loading problems." + "reason": "A bad AutoTokenizer error message for Voxtral and an AppleSilicon allocator TypeError are unrelated." } ] }, @@ -3809,17 +3815,16 @@ "cluster_id": "cluster-10000-473", "nodes": [ "issue:40444", - "issue:41553", "issue:41720", "issue:42175", "issue:42796", "issue:42915", "issue:43296", + "issue:43317", "issue:43377", "issue:43454", "issue:43482", "issue:43531", - "issue:43582", "issue:43606", "issue:43611", "issue:43632", @@ -3829,6 +3834,7 @@ "issue:43749", "issue:43792", "issue:43818", + "issue:43856", "issue:43873", "issue:43957", "issue:43994", @@ -3858,22 +3864,22 @@ "issue:46402" ], "soft_pairs": [ - "issue:41553|issue:43582", + "issue:43317|issue:43856", "issue:42175|issue:45542", "issue:43531|issue:44291", "issue:44829|issue:45479", "issue:44079|issue:44534", "issue:44462|issue:45812", - "issue:43818|issue:46402", "issue:41720|issue:45081", + "issue:43818|issue:46402", "issue:43377|issue:45491", "issue:43818|issue:45357", "issue:43611|issue:43646", "issue:43644|issue:44079", "issue:44265|issue:45072", "issue:43749|issue:44079", - "issue:40444|issue:45381", "issue:42175|issue:43296", + "issue:40444|issue:45381", "issue:45325|issue:45381", "issue:43994|issue:44898", "issue:45325|issue:45834", @@ -3889,224 +3895,224 @@ "issue:43606|issue:44265", "issue:44479|issue:45823" ], - "prepared_review_unit_hash": "1112532a6a79dfb76292b22783c4612b81d7524dd9d152138fa2d0d7663c93ba", + "prepared_review_unit_hash": "345a2181e9233e070ac21b38a7777d93e70991ac9ac774b01ad40d272ab80288", "component_budget": { "node_count": 473, "item_count": 473, "soft_pair_count": 1738, - "serialized_chars": 901383, - "estimated_input_tokens": 225346, - "estimated_eval_tokens": 450948 + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 30, - "serialized_chars": 36841, - "estimated_input_tokens": 9211, - "estimated_eval_tokens": 18678 + "serialized_chars": 36853, + "estimated_input_tokens": 9214, + "estimated_eval_tokens": 18684 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Mostly a grab-bag of unrelated model/regression reports. The only strong duplicate-like clusters are around Transformers v5 model initialization/loading and a Qwen2.5-VL positional-index regression; most other pairs are too far apart to merge.", - "confidence": 0.52, - "canonical_issue_reason": "Issue 43611 is the clearest representative of the shared Transformers v5 model-loading/initialization regression, and it is the best anchor for the only convincing duplicate subcluster in this set.", + "summary": "This cluster is still mostly a set of distinct Transformers v5 regression reports rather than a tight duplicate cluster. The `issue:43611` \u2194 `issue:43646` link should be treated as related but not duplicated: both concern v5 load/init behavior, but the packet does not show the same concrete code path or fix.", + "confidence": 0.71, + "canonical_issue_reason": "Issue 43646 is the broadest and most representative anchor for the cluster because it captures the general Transformers v5 custom-model initialization regression theme, but it should not be used to imply that nearby load/init reports are definite duplicates without stronger evidence.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 43611 is the broadest, most central report of a concrete v5 initialization/loading breakage and aligns best with the strongest near-duplicate candidate (43646).", + "best_issue_reason": "Issue 43646 is the best single issue anchor because it is broad, actionable, and closest to the shared v5 loading/initialization regression pattern.", "best_pr_reason": null, "soft_edge_verdicts": [ { - "left": "issue:41553", - "right": "issue:43582", + "left": "issue:43317", + "right": "issue:43856", "accept": false, - "reason": "Different subsystems and symptoms: tokenizer loading error message vs Apple Silicon warmup TypeError." + "reason": "Different problems: dequantized model offload/device mapping vs MoE training memory usage." }, { "left": "issue:42175", "right": "issue:45542", "accept": false, - "reason": "Both touch TensorFlow-related packaging, but one is missing TF in the torch extra and the other is a tensorboard-only backend error; not clearly the same bug." + "reason": "Both mention TensorFlow packaging, but one is missing TF in the `.[torch]` backend and the other is a tensorboard-only undefined-backend error." }, { "left": "issue:43531", "right": "issue:44291", "accept": false, - "reason": "Unrelated failures: Qwen3-MoE sliding_window behavior vs init_empty_weights/_is_hf_initialized argument handling." + "reason": "Unrelated bugs: Qwen3-MoE sliding-window behavior vs `init_empty_weights`/`_is_hf_initialized` argument handling." }, { "left": "issue:44829", "right": "issue:45479", "accept": false, - "reason": "Different training bugs in different code paths: flash_attention_3 degeneration vs num_labels=1 zero-loss classification behavior." + "reason": "Different training regressions with different triggers and failure modes." }, { "left": "issue:44079", "right": "issue:44534", "accept": false, - "reason": "ModelOutput key assignment bug vs non-persistent buffer initialization bug are distinct." + "reason": "`ModelOutput` key assignment bug vs non-persistent buffer initialization bug." }, { "left": "issue:44462", "right": "issue:45812", "accept": false, - "reason": "Both are tokenizer regressions, but one is repository tokenizer.json loading and the other is Granite token ID mapping; different underlying problems." + "reason": "Both are tokenizer-related, but one ignores repo `tokenizer.json` while the other miscomputes Granite token IDs." }, { - "left": "issue:43818", - "right": "issue:46402", + "left": "issue:41720", + "right": "issue:45081", "accept": false, - "reason": "Different models and failure modes: Video-LLaVA tower architecture issue vs Sam3 text-encoder weights not loading." + "reason": "Different subsystems: Qwen3 device mapping crash vs Mistral tokenizer regex patch crash." }, { - "left": "issue:41720", - "right": "issue:45081", + "left": "issue:43818", + "right": "issue:46402", "accept": false, - "reason": "CUDA/device-map crash on Qwen3 vs Mistral regex/tokenizer attribute error; not the same issue." + "reason": "Unrelated models and failures: Video-LLaVA tower/weights issue vs Sam3Model text-encoder loading issue." }, { "left": "issue:43377", "right": "issue:45491", "accept": false, - "reason": "Both involve batching/masking, but they affect different models and different mechanisms (padding mask vs sliding-window all-padding windows)." + "reason": "Both involve batching, but the underlying bugs differ: missing padding-mask support vs sliding-window/all-padding NaNs." }, { "left": "issue:43818", "right": "issue:45357", "accept": false, - "reason": "Video-LLaVA tower bug is unrelated to Qwen3.5 visual-encoder save_pretrained key corruption." + "reason": "Different model families and issues: Video-LLaVA architecture bug vs Qwen3.5 save_pretrained key regression." }, { "left": "issue:43611", "right": "issue:43646", - "accept": true, - "reason": "Both report Transformers v5 breaking model initialization/loading for custom architectures; same underlying init path regression and plausibly one fix." + "accept": false, + "reason": "Related v5 initialization/loading regressions, but the packet only shows broad thematic overlap (`base_model_prefix` loading vs custom model init); not enough evidence for the same underlying bug or code path." }, { "left": "issue:43644", "right": "issue:44079", "accept": false, - "reason": "Non-persistent buffer serialization/init bug is unrelated to ModelOutput key assignment." + "reason": "Different core bugs: non-persistent buffer corruption vs ModelOutput key assignment." }, { "left": "issue:44265", "right": "issue:45072", "accept": false, - "reason": "torch.export exportability issue vs dtype mismatch in inference; different code paths." + "reason": "Different code paths: `torch.export`/`torch_compilable_check` vs dtype mismatch in inference." }, { "left": "issue:43749", "right": "issue:44079", "accept": false, - "reason": "FSDP CPU RAM efficient loading regression is unrelated to ModelOutput key handling." + "reason": "FSDP CPU-efficient loading breakage is unrelated to ModelOutput key assignment." }, { - "left": "issue:40444", - "right": "issue:45381", + "left": "issue:42175", + "right": "issue:43296", "accept": false, - "reason": "Both are Qwen2.5-VL-related, but one is multi-image iterable finetuning failure and the other is video vision_position_ids regression; not enough evidence of the same bug." + "reason": "Different packaging/model-loading issues with no clear shared code-path." }, { - "left": "issue:42175", - "right": "issue:43296", + "left": "issue:40444", + "right": "issue:45381", "accept": false, - "reason": "Packaging/backend availability issue vs vLLM loading failure for a specific multimodal model." + "reason": "Both are Qwen2.5-VL vision-position related, but one is finetuning with multiple images and the other is a video-input regression; not clearly the same bug." }, { "left": "issue:45325", "right": "issue:45381", - "accept": true, - "reason": "Both are Qwen2.5-VL positional-index regressions in the same family of code; still-image temporal position_ids and video vision_position_ids look like the same underlying rope/positioning bug." + "accept": false, + "reason": "Closely related area, but one reports still-image temporal scaling and the other video `vision_position_ids`; not enough to treat as the same defect." }, { "left": "issue:43994", "right": "issue:44898", "accept": false, - "reason": "Different vision-model inference failures: SigLIP2 nonsensical outputs vs Perceiver non-default-resolution interpolation failure." + "reason": "Different vision-model failures on different architectures and inputs." }, { "left": "issue:45325", "right": "issue:45834", "accept": false, - "reason": "Qwen2.5-VL positional-index bug vs Kosmos2.5 long OCR index error; different models and failures." + "reason": "Both concern Qwen2.5-VL position logic, but the reported failures are different: temporal scaling vs OCR long-input index error." }, { "left": "issue:43454", "right": "issue:44466", "accept": false, - "reason": "Both mention lm_head/tied weights, but one is a weight-tying failure in config/init and the other is save_pretrained serialization inconsistency; not clearly the same code-path bug." + "reason": "Different symptoms around weight tying/serialization vs inconsistent `lm_head.weight` serialization." }, { "left": "issue:43792", "right": "issue:43957", "accept": false, - "reason": "Whisper runtime failure vs meta-device model loading breakage; too different." + "reason": "Whisper runtime failure vs generic `torch.device('meta')` model-loading regression." }, { "left": "issue:42915", "right": "issue:43482", "accept": false, - "reason": "Different model families and different loading/config problems." + "reason": "Unrelated: Qwen3-MoE FP8 config failure vs Qwen2.5 GGUF loading under v5." }, { "left": "issue:43632", "right": "issue:45072", "accept": false, - "reason": "_is_hf_initialized regression is about model init state, while the other is a bfloat16 dtype mismatch." + "reason": "Different regressions: `_is_hf_initialized` flag handling vs bfloat16 dtype mismatches." }, { "left": "issue:42796", "right": "issue:45381", "accept": false, - "reason": "Tokenizer BPE marker handling issue is unrelated to Qwen2.5-VL video position IDs." + "reason": "Tokenizer BPE-marker handling is unrelated to Qwen2.5-VL video position IDs." }, { "left": "issue:43701", "right": "issue:45325", "accept": false, - "reason": "Checkpoint resume key mismatch is unrelated to Qwen2.5-VL rope/position-id scaling." + "reason": "Checkpoint key mismatch is unrelated to Qwen2.5-VL position-id scaling." }, { "left": "issue:44291", "right": "issue:44877", "accept": false, - "reason": "init_empty_weights argument error vs strict config rejection are different loading bugs." + "reason": "`init_empty_weights` argument handling and strict config loading are different issues." }, { "left": "issue:43873", "right": "issue:45701", "accept": false, - "reason": "Quantization/offloading behavior vs tokenization changes are unrelated." + "reason": "Quantized offloading behavior and tokenization changes are unrelated." }, { "left": "issue:43296", "right": "issue:45800", "accept": false, - "reason": "PaddleOCR-VL loading issue vs torch/transformers incompatibility; different root causes." + "reason": "Different compatibility problems: PaddleOCR-VL/vLLM loading vs torch 2.4.1 and Transformers 5.8.0." }, { "left": "issue:43606", "right": "issue:44265", "accept": false, - "reason": "CPU offload device mismatch is unrelated to torch.export export failure." + "reason": "CPU offload device mismatch is unrelated to `torch.export` failures." }, { "left": "issue:44479", "right": "issue:45823", "accept": false, - "reason": "Both are multimodal/device-map issues, but they concern different model families and distinct failure modes." + "reason": "Different model families and failures: Qwen video-input regression vs Gemma4 PLE device-mapping mismatch." } ] }, "evaluator_result": { - "accept": true, - "feedback": "The summary stays grounded in the packet and the soft-edge calls are mostly conservative. The two accepted pairs are plausibly mergeable into single bugfix clusters (Transformers v5 init/loading regression; Qwen2.5-VL position-id regression), while the rest are reasonably rejected as distinct failures." + "accept": false, + "feedback": "Most soft-edge rejections are appropriately conservative, but the accepted pair `issue:43611` \u2194 `issue:43646` is overstated. From the packet we only know one is about `base_model_prefix` loading and the other about custom model initialization; both are v5 load/init regressions, but there is not enough evidence here to call them the same underlying bug or a clear duplicate-like pair. The cluster summary should keep that linkage tentative unless the issue bodies show the same code path or fix." }, "evaluator_used": true, - "retried": false, - "accepted_nontrivial_soft_edge": true, + "retried": true, + "accepted_nontrivial_soft_edge": false, "error_kind": null, "error_message": null }, @@ -4176,23 +4182,23 @@ "issue:43262|issue:43746", "issue:40990|issue:45584", "issue:35707|issue:41720", - "issue:34634|issue:35707", "issue:43881|issue:44488", - "issue:39692|issue:40444", + "issue:34634|issue:35707", "issue:43454|issue:43550", "issue:43606|issue:43819", + "issue:39692|issue:40444", "issue:44186|issue:44265", - "issue:46305|issue:46315", "issue:41720|issue:45823", + "issue:46305|issue:46315", "issue:43377|issue:44610", "issue:42796|issue:46209", - "issue:43257|issue:45923", "issue:44568|issue:45920", + "issue:43257|issue:45923", "issue:43454|issue:45072", "issue:43317|issue:43873", "issue:43377|issue:44898", - "issue:43454|issue:43720", "issue:42796|issue:44977", + "issue:43454|issue:43720", "issue:40990|issue:42947", "issue:43322|issue:46050", "issue:42617|issue:42915", @@ -4200,218 +4206,218 @@ "issue:43572|issue:45759", "issue:41720|issue:43322", "issue:43526|issue:43883", - "issue:44387|issue:45704", "issue:43881|issue:44991", + "issue:44387|issue:45704", "issue:41720|issue:44977" ], - "prepared_review_unit_hash": "444041268b89d6d342bbf89a3c9285d90e7609b14843f562cf9a7683ceda4e75", + "prepared_review_unit_hash": "c7f4fa6103f192407af89a1fe222a6bceb2bd9615f3af22b37e95ade36f02e80", "component_budget": { "node_count": 473, "item_count": 473, "soft_pair_count": 1738, - "serialized_chars": 901383, - "estimated_input_tokens": 225346, - "estimated_eval_tokens": 450948 + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 30, - "serialized_chars": 36793, - "estimated_input_tokens": 9199, - "estimated_eval_tokens": 18654 + "serialized_chars": 36797, + "estimated_input_tokens": 9200, + "estimated_eval_tokens": 18656 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is not a true duplicate set; it mixes many unrelated bug reports that only share broad themes like loading, offloading, tokenization, and quantization. No PRs are present.", - "confidence": 0.58, - "canonical_issue_reason": "issue:43873 is the broadest open issue in the set and the closest thing to a cluster anchor because it covers offloading plus quantization, which recurs across several nearby reports.", + "summary": "This cluster is highly heterogeneous and does not contain any clear duplicate pairs. Several items share broad themes like tokenizer regressions, loading/offload problems, or device-map issues, but they point to different models and different failure modes.", + "confidence": 0.88, + "canonical_issue_reason": "No true duplicate cluster stands out; if a representative issue must be chosen, #45920 is the broadest open regression affecting multiple models and has the strongest evidence of being a general issue rather than a niche case.", "canonical_pr_reason": null, - "best_issue_reason": "issue:43873 is the best representative issue because it is open, topical, and general enough to sit at the center of the loading/offloading/quantization subset without being overly model-specific.", + "best_issue_reason": "#45920 is the best single issue to represent this set because it is open, cross-model, and more broadly scoped than the other candidates, which are mostly model-specific or narrowly scoped bugs.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43262", "right": "issue:43746", "accept": false, - "reason": "Different bugs: audio processor chat-template sampling-rate default vs PEFT adapter local checkpoint loading." + "reason": "Different subsystems: audio chat-template sampling rate vs PEFT checkpoint loading." }, { "left": "issue:40990", "right": "issue:45584", "accept": false, - "reason": "Both are generation-quality problems, but they involve different models and different failure modes (perplexity vs empty transcription handling)." + "reason": "Unrelated failures: perplexity evaluation on GPT-OSS vs Whisper empty-transcription generation." }, { "left": "issue:35707", "right": "issue:41720", "accept": false, - "reason": "Generation cache/input-embed handling is unrelated to the Qwen3 device-mapping CUDA assert." - }, - { - "left": "issue:34634", - "right": "issue:35707", - "accept": false, - "reason": "Voice preset handling in BarkProcessor is a different subsystem from progressive generation with inputs_embeds/past_key_values." + "reason": "Different code paths: inputs_embeds/past_key_values generation vs auto device mapping on Qwen3." }, { "left": "issue:43881", "right": "issue:44488", "accept": false, - "reason": "Both fail to load, but they are different models with no clear shared code-path or root cause." + "reason": "Both are loading failures, but for different models and likely different root causes." }, { - "left": "issue:39692", - "right": "issue:40444", + "left": "issue:34634", + "right": "issue:35707", "accept": false, - "reason": "Documentation/model-processor mismatch is not the same bug as iterable-dataset finetuning with multiple images." + "reason": "Bark voice_preset handling is unrelated to generation with inputs_embeds and past_key_values." }, { "left": "issue:43454", "right": "issue:43550", "accept": false, - "reason": "Weight-tying failure in AyaVisionConfig is unrelated to the Bamba torch.compile SDPA failure." + "reason": "Different problems: weight tying/generation corruption vs torch.compile SDPA failure." }, { "left": "issue:43606", "right": "issue:43819", "accept": false, - "reason": "CPU offload device mismatch for Bark is a different problem from DAC.from_latents not matching forward due to missing STE." + "reason": "CPU offload device mismatch is unrelated to DAC latent reconstruction parity." }, { - "left": "issue:44186", - "right": "issue:44265", + "left": "issue:39692", + "right": "issue:40444", "accept": false, - "reason": "Tokenizer crashes on NER padding/truncation and torch.export failures with torch_compilable_check are different code paths." + "reason": "Documentation example errors are unrelated to Qwen2.5-VL fine-tuning with iterable datasets." }, { - "left": "issue:46305", - "right": "issue:46315", + "left": "issue:44186", + "right": "issue:44265", "accept": false, - "reason": "One is an off-by-one masking bug; the other is a tokenizer performance regression." + "reason": "Tokenizer crash on NER/padding is unrelated to torch.export compatibility checks." }, { "left": "issue:41720", "right": "issue:45823", "accept": false, - "reason": "Both mention device_map=auto, but the models and concrete failures differ, so they are not the same bug." + "reason": "Both mention auto device mapping, but they involve different models and distinct runtime failures." + }, + { + "left": "issue:46305", + "right": "issue:46315", + "accept": false, + "reason": "Sliding-window attention off-by-one is unrelated to tokenizer conversion performance." }, { "left": "issue:43377", "right": "issue:44610", "accept": false, - "reason": "A padding-mask batching bug in MIMI is unrelated to OmDet-Turbo's processor/model input-size mismatch." + "reason": "MIMI padding-mask batching bug is unrelated to OmDet-Turbo image-size mismatch." }, { "left": "issue:42796", "right": "issue:46209", "accept": false, - "reason": "Tokenizer BPE-marker handling is unrelated to FP8 inference failing on B200 hardware." + "reason": "Tokenizer BPE marker handling and FP8 inference on B200 are different bugs." }, { - "left": "issue:43257", - "right": "issue:45923", + "left": "issue:44568", + "right": "issue:45920", "accept": false, - "reason": "MOE weight conversion during accelerate+deepspeed loading is a different issue from a missing gradient-checkpointing flag." + "reason": "Specific BOS/EOS insertion failure for one tokenizer is not the same as broad wrong-token-id regressions." }, { - "left": "issue:44568", - "right": "issue:45920", + "left": "issue:43257", + "right": "issue:45923", "accept": false, - "reason": "Special-token insertion for one tokenizer is not the same as the broader convert-to-IDs regression across multiple models." + "reason": "MOE weight conversion under accelerate+deepspeed is unrelated to a missing supports_gradient_checkpointing flag." }, { "left": "issue:43454", "right": "issue:45072", "accept": false, - "reason": "Weight tying and bfloat16 dtype mismatches are different bugs." + "reason": "Weight tying issue and bfloat16 dtype mismatch are separate model/runtime bugs." }, { "left": "issue:43317", "right": "issue:43873", "accept": false, - "reason": "These are related only broadly; one is a dequantized-model load failure under gpu+cpu offload, the other is a more general quantized offloading regression." + "reason": "Dequantized model loading with cpu offload and generic quantization offloading are related only at a broad level, not the same bug." }, { "left": "issue:43377", "right": "issue:44898", "accept": false, - "reason": "Different models and failure modes: batched-vs-single encoder outputs vs non-default-resolution image classification." + "reason": "Batching/padding-mask output mismatch is unrelated to Perceiver non-default resolution failures." }, { - "left": "issue:43454", - "right": "issue:43720", + "left": "issue:42796", + "right": "issue:44977", "accept": false, - "reason": "AyaVision weight tying and BitNet packed-weight unpacking are unrelated loading bugs." + "reason": "Tokenizer BPE marker bug is unrelated to flash-attention generation issues." }, { - "left": "issue:42796", - "right": "issue:44977", + "left": "issue:43454", + "right": "issue:43720", "accept": false, - "reason": "Tokenizer BPE-marker handling is unrelated to Qwen3.5 flash-attention generation issues." + "reason": "Weight-tying failure is unrelated to packed-weight unpacking during accelerate loading." }, { "left": "issue:40990", "right": "issue:42947", "accept": false, - "reason": "High perplexity on a model is not the same problem as gradient checkpointing being ineffective with LoRA." + "reason": "Perplexity anomaly and gradient-checkpointing ineffectiveness are different issues." }, { "left": "issue:43322", "right": "issue:46050", "accept": false, - "reason": "Segfault loading Llava Next models is different from save/load corruption of sinq-quantized models." + "reason": "Llava Next loading segfault and sinq save/load breakage are distinct model-loading bugs." }, { "left": "issue:42617", "right": "issue:42915", "accept": false, - "reason": "Running a 3d_parallel script is unrelated to FineGrainedFP8Config failures." + "reason": "3d_parallel.py execution failure is unrelated to Qwen3Moe FP8 config loading." }, { "left": "issue:43749", "right": "issue:44534", "accept": false, - "reason": "FSDP efficient loading and non-persistent buffer initialization are different state-loading problems." + "reason": "FSDP CPU RAM efficient loading and non-persistent buffer junk are not the same code-path problem." }, { "left": "issue:43572", "right": "issue:45759", "accept": false, - "reason": "Missing pad_token_idx in one config is not the same as failing to unwrap text_config in composite multimodal configs." + "reason": "Missing pad_token_idx and text_config unwrapping are different config bugs." }, { "left": "issue:41720", "right": "issue:43322", "accept": false, - "reason": "Different models and different crash types: Qwen3 device-mapping assert vs Llava Next segmentation fault." + "reason": "Auto device-map assert on Qwen3 and Llava Next segfault are unrelated failures." }, { "left": "issue:43526", "right": "issue:43883", "accept": false, - "reason": "BeitImageProcessorFast label reduction is unrelated to the Molmo tied-weights attribute error." + "reason": "BEiT label reduction bug is unrelated to missing all_tied_weights_keys on Molmo." }, { - "left": "issue:44387", - "right": "issue:45704", + "left": "issue:43881", + "right": "issue:44991", "accept": false, - "reason": "Both are memory-related, but int4 quantization OOM and apex FusedRMSNorm memory leak are not the same underlying bug." + "reason": "Model loading failure for glm-4v-9b is unrelated to tokenizer loading regression for EMBEDDIA/est-roberta." }, { - "left": "issue:43881", - "right": "issue:44991", + "left": "issue:44387", + "right": "issue:45704", "accept": false, - "reason": "A model loading failure and a tokenizer loading regression are distinct issues." + "reason": "Int4 quantization memory growth and apex RMSNorm memory leak are different resource issues." }, { "left": "issue:41720", "right": "issue:44977", "accept": false, - "reason": "Same general Qwen family, but the actual bugs differ: device-mapping crash versus flash-attention generation behavior." + "reason": "Both involve Qwen variants, but one is a device-map crash and the other is a flash-attention generation bug." } ] }, @@ -4493,8 +4499,8 @@ "issue:44877|issue:44991", "issue:40990|issue:45701", "issue:44186|issue:45072", - "issue:43873|issue:44756", "issue:43065|issue:46402", + "issue:43873|issue:44756", "issue:43550|issue:43720", "issue:41720|issue:42617", "issue:43577|issue:43708", @@ -4505,8 +4511,8 @@ "issue:44186|issue:44610", "issue:44008|issue:45823", "issue:44610|issue:44898", - "issue:44423|issue:45406", "issue:42915|issue:43317", + "issue:44423|issue:45406", "issue:36010|issue:42617", "issue:45632|issue:45684", "issue:41720|issue:45161", @@ -4518,226 +4524,208 @@ "issue:36331|issue:41720", "issue:43526|issue:43881" ], - "prepared_review_unit_hash": "c0c62d24f97aa263837458cff363f3aa8025ffbdb23670fd1ae8256b840efea8", + "prepared_review_unit_hash": "ff3c64188044eb1ba97f249d5f3d487db05a308a1ca8a4778a1ba13a2f5dfc36", "component_budget": { "node_count": 473, "item_count": 473, "soft_pair_count": 1738, - "serialized_chars": 901383, - "estimated_input_tokens": 225346, - "estimated_eval_tokens": 450948 + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 32, - "serialized_chars": 37367, - "estimated_input_tokens": 9342, - "estimated_eval_tokens": 18940 + "serialized_chars": 37369, + "estimated_input_tokens": 9343, + "estimated_eval_tokens": 18942 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Mostly unrelated issue reports with a few superficial theme overlaps (loading/tokenization, multimodal serving, offload), but no pair is clearly the same underlying bug. Best representative is the broad custom-model initialization regression.", - "confidence": 0.62, - "canonical_issue_reason": "issue:43646 is the broadest, most framework-level regression in the set and best fits a representative duplicate bucket around custom model loading/init behavior.", + "summary": "This set is mostly a loose bag of unrelated issue reports; the soft pairs share broad subsystem vocabulary but not the same concrete bug or change. I would not merge any of the proposed pairs as duplicates.", + "confidence": 0.87, + "canonical_issue_reason": "issue:43873 is the broadest and most active issue in the set, centered on offloading/quantization behavior, so it is the best representative anchor if one must pick a canonical issue. It is not a duplicate of the other reports, just the most central among several adjacent device/offload problems.", "canonical_pr_reason": null, - "best_issue_reason": "issue:43646 is the most suitable single issue to keep as the cluster representative because its scope is broader and closer to the only other near-match in the set.", + "best_issue_reason": "issue:43873 has the strongest triage signal (discussion + inbound refs) and sits closest to the recurring device/offload theme seen across a few nearby reports, making it the best issue candidate overall.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:44466", "right": "issue:45357", "accept": false, - "reason": "Both mention model saving/tied weights, but they concern different models and different serialization failures." + "reason": "Different failures: lm_head serialization/device dependence vs incorrect visual encoder keys in save_pretrained." }, { "left": "issue:36331", "right": "issue:42617", "accept": false, - "reason": "Unexpected compute_loss kwargs vs a 3d_parallel.py runtime failure are different code paths and symptoms." + "reason": "Unexpected compute_loss kwarg in CustomTrainer is unrelated to a 3d_parallel.py execution problem." }, { "left": "issue:26413", "right": "issue:28530", "accept": false, - "reason": "Checkpoint resume epoch-iterator failure and early-stopping metric discovery are unrelated Trainer behaviors." + "reason": "Checkpoint resume/epoch_iterator failure is distinct from early-stopping metric resolution." }, { "left": "issue:36331", "right": "issue:42915", "accept": false, - "reason": "CustomTrainer API breakage and Qwen3Moe FP8 loading failure are separate bugs." + "reason": "Trainer signature regression and Qwen3Moe FP8 config failure are different code paths." }, { "left": "issue:43450", "right": "issue:45701", "accept": false, - "reason": "Video processor shape handling and tokenization differences are not the same problem." + "reason": "Video batch-shape bug is unrelated to tokenization behavior changes across versions." }, { "left": "issue:44877", "right": "issue:44991", "accept": false, - "reason": "Strict config loading for granite_speech and tokenizer loading for est-roberta are unrelated loader failures." + "reason": "Strict config loading for granite_speech is not the same as tokenizer loading failure for EMBEDDIA/est-roberta." }, { "left": "issue:40990", "right": "issue:45701", "accept": false, - "reason": "High perplexity on gpt-oss-20b is not clearly the same as a version-dependent tokenization change." + "reason": "High perplexity on gpt-oss is a model/eval-quality issue, not the tokenizer-version regression described in 45701." }, { "left": "issue:44186", "right": "issue:45072", "accept": false, - "reason": "Tokenizer crashes on NER/batched padding and dtype mismatches in inference are different bugs." + "reason": "Tokenizer crash on NER/padding is unrelated to dtype mismatch in bfloat16 inference." }, { - "left": "issue:43873", - "right": "issue:44756", + "left": "issue:43065", + "right": "issue:46402", "accept": false, - "reason": "Quantization offload behavior and Strix Halo mmap OOM are only loosely related by memory pressure, not the same code path." + "reason": "Dummy Conv2d in Sam3PixelDecoder and missing Sam3Model text encoder weights are different SAM3 bugs." }, { - "left": "issue:43065", - "right": "issue:46402", + "left": "issue:43873", + "right": "issue:44756", "accept": false, - "reason": "Sam3 pixel decoder dummy Conv2d and Sam3 text-encoder weights not loading affect different model components." + "reason": "Quantization/offload malfunction is not the same as disabling mmap to avoid OOM on Strix Halo." }, { "left": "issue:43550", "right": "issue:43720", "accept": false, - "reason": "torch.compile+SDPA failure in Bamba and packed-weight unpacking in BitNet are different implementation issues." + "reason": "torch.compile+SDPA failure for Bamba is unrelated to packed-weight unpacking during accelerate loading." }, { "left": "issue:41720", "right": "issue:42617", "accept": false, - "reason": "Qwen3-VL auto device mapping cudaErrorAssert and a 3d_parallel.py failure are unrelated." + "reason": "Qwen3 auto device_map cuda assert is a different failure from the 3d_parallel.py runtime issue." }, { "left": "issue:43577", "right": "issue:43708", "accept": false, - "reason": "BLIP2 dtype propagation and checkpoint max_steps recomputation are separate Trainer/model-loading issues." + "reason": "Blip2 dtype staying float32 is unrelated to resume_from_checkpoint max_steps calculation." }, { "left": "issue:43404", "right": "issue:45357", "accept": false, - "reason": "Both involve saved weights, but one is tied lm_head handling and the other is wrong visual encoder keys." + "reason": "Tied lm_head weights in Mistral3 are not the same bug as incorrect visual encoder keys in Qwen3.5 save_pretrained." }, { "left": "issue:43550", "right": "issue:43606", "accept": false, - "reason": "These are different runtime failures in different models and execution modes." + "reason": "Both are runtime/model-loading bugs, but they affect different models and code paths; not one concrete duplicate." }, { "left": "issue:43646", "right": "issue:46665", "accept": false, - "reason": "Both are about custom loading patterns, but one is a broad 5.0 regression and the other is a specific from_pretrained-in-class behavior; not clearly the same bug." + "reason": "Both concern custom-model loading, but the titles point to different symptoms and not a clearly shared concrete bug." }, { "left": "issue:43295", "right": "issue:45701", "accept": false, - "reason": "Processor.tokenizer regression and tokenizer differences across versions are too broad to treat as the same concrete bug." + "reason": "Processor.tokenizer/API regression is broader custom-code breakage, not the same as tokenization changes reported in 45701." }, { "left": "issue:44186", "right": "issue:44610", "accept": false, - "reason": "A tokenizer crash on NER inputs is not the same as a processor producing the wrong image size." + "reason": "LayoutLMv2Tokenizer crashing and OmDet-Turbo image-size mismatch are unrelated model-specific processor bugs." }, { "left": "issue:44008", "right": "issue:45823", "accept": false, - "reason": "Gemma3n variable-name collision in forward and Gemma4 device-mapping mismatch are unrelated." + "reason": "Gemma3n variable collision in forward is unrelated to Gemma4 PLE device mismatch under device_map=\"auto\"." }, { "left": "issue:44610", "right": "issue:44898", "accept": false, - "reason": "Both touch image resolution, but they involve different models and different failure mechanisms." + "reason": "Both involve image sizing, but they affect different models and different mechanisms (processor output vs positional encoding handling)." }, { - "left": "issue:44423", - "right": "issue:45406", + "left": "issue:42915", + "right": "issue:43317", "accept": false, - "reason": "Both are serve/multimodal crashes, but the attribute errors point to different broken internals and code paths." + "reason": "Qwen3Moe FP8 config failure is different from dequantized model loading with gpu+cpu offload." }, { - "left": "issue:42915", - "right": "issue:43317", + "left": "issue:44423", + "right": "issue:45406", "accept": false, - "reason": "Qwen3Moe FP8 failure and dequantized model offload loading failure are different quantization/offload bugs." + "reason": "Continuous batching crash in multimodal serving and Gemma4Processor missing _tokenizer are separate server/processor issues." }, { "left": "issue:36010", "right": "issue:42617", "accept": false, - "reason": "GenerationMixin import error and 3d_parallel.py runtime failure are unrelated." + "reason": "GenerationMixin import failure is unrelated to the 3d_parallel.py issue." }, { "left": "issue:45632", "right": "issue:45684", "accept": false, - "reason": "Cache-path collision for trust_remote_code and read-only permission propagation in save_pretrained are distinct filesystem issues." + "reason": "Cache-path collision for local models is unrelated to read-only permission propagation in save_pretrained." }, { "left": "issue:41720", "right": "issue:45161", "accept": false, - "reason": "Both mention parallelism, but Qwen3 auto device mapping and GPT-OSS tensor parallel behavior are not the same bug." + "reason": "Both mention parallelism/device setup, but they concern different models and distinct failures." }, { "left": "issue:28530", "right": "issue:33666", "accept": false, - "reason": "Early stopping configuration and Qwen2-VL multi-GPU training are unrelated." + "reason": "Early stopping metric resolution and Qwen2-VL multi-GPU training are not the same bug." }, { "left": "issue:43531", "right": "issue:43572", "accept": false, - "reason": "Sliding-window behavior in Qwen3-MoE and missing pad_token_idx in StableLmConfig are separate issues." + "reason": "Qwen3-MoE sliding_window behavior and missing pad_token_idx in StableLmConfig are separate config/model issues." }, { "left": "issue:44671", "right": "issue:44743", "accept": false, - "reason": "CamemBERT masked LM output issues and Qwen3.5 recurrent-state reset behavior are unrelated model bugs." - }, - { - "left": "issue:42915", - "right": "issue:46291", - "accept": false, - "reason": "Qwen3Moe FP8 loading and pipelines import failure due to kernels version/revision requirements are unrelated." - }, - { - "left": "issue:44186", - "right": "issue:44361", - "accept": false, - "reason": "Tokenizer crashes in different model families do not indicate the same underlying defect." - }, - { - "left": "issue:36331", - "right": "issue:41720", - "accept": false, - "reason": "Trainer loss-signature breakage and Qwen3-VL device mapping assert are unrelated." + "reason": "CamemBERT masked-LM predictions and Qwen3.5 recurrent-state reset are unrelated model behaviors." }, { "left": "issue:43526", "right": "issue:43881", "accept": false, - "reason": "BeitImageProcessorFast label reduction and glm-4v-9b loading failure are different problems." + "reason": "BeitImageProcessorFast label reduction bug is unrelated to glm-4v-9b loading failure." } ] }, @@ -4811,13 +4799,13 @@ "issue:46665" ], "soft_pairs": [ - "issue:42915|issue:43322", "issue:43316|issue:44913", + "issue:42915|issue:43322", "issue:28530|issue:30064", "issue:43540|issue:43782", "issue:33357|issue:35707", - "issue:43873|issue:45704", "issue:43550|issue:43653", + "issue:43873|issue:45704", "issue:43606|issue:43720", "issue:29159|issue:36032", "issue:30064|issue:34634", @@ -4831,212 +4819,218 @@ "issue:43606|issue:43653", "issue:45406|issue:45464", "issue:34567|issue:35707", - "issue:45357|issue:46402", "issue:34634|issue:41720", + "issue:45357|issue:46402", "issue:44451|issue:44991", "issue:45759|issue:45859", "issue:43531|issue:44589", - "issue:44077|issue:44625", "issue:44625|issue:45479", + "issue:44077|issue:44625", "issue:43644|issue:46665", "issue:44898|issue:45072" ], - "prepared_review_unit_hash": "62276c13512563372968dc410b22f8e4182d9da18b11806cfc55f0b8a71452c5", + "prepared_review_unit_hash": "e98a17af1a0fd3e7c0b296ee7616b5578fa91d6a22389e6a2b3cc9ed80269505", "component_budget": { "node_count": 473, "item_count": 473, "soft_pair_count": 1738, - "serialized_chars": 901383, - "estimated_input_tokens": 225346, - "estimated_eval_tokens": 450948 + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 29, - "serialized_chars": 36432, - "estimated_input_tokens": 9108, - "estimated_eval_tokens": 18472 + "serialized_chars": 36440, + "estimated_input_tokens": 9110, + "estimated_eval_tokens": 18476 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is heterogeneous: it spans unrelated tokenizer, multimodal, offload, config, docs, and training bugs. None of the soft pairs look like the same underlying issue, so all are rejected.", - "confidence": 0.96, - "canonical_issue_reason": "No single canonical issue stands out because the items do not form one duplicate theme; they are a mix of unrelated bugs across different subsystems and models.", + "summary": "These issues span unrelated bugs across configs, tokenizers, multimodal loading, quantization/offload, and docs. None of the soft pairs look like safe duplicates or the same underlying change.", + "confidence": 0.91, + "canonical_issue_reason": "\"Transformers v5 fills non-persistent buffers with junk\" is a concrete, regression-style bug with a clear runtime symptom and broad impact, making it a strong anchor issue.", "canonical_pr_reason": null, - "best_issue_reason": "No best issue for deduplication: the cluster is too diverse for one issue to serve as a representative duplicate target.", + "best_issue_reason": "This is the most representative standalone bug report in the set: specific, actionable, and broad enough to serve as a good canonical issue.", "best_pr_reason": null, "soft_edge_verdicts": [ { - "left": "issue:42915", - "right": "issue:43322", + "left": "issue:43316", + "right": "issue:44913", "accept": false, - "reason": "Different failure modes in different model families: Qwen3Moe FP8 config vs Llava Next load-time segfault." + "reason": "Both are config/serialization-related, but they affect different models and different fields (`Gemma3TextConfig` API vs `rotary_pct` reload behavior)." }, { - "left": "issue:43316", - "right": "issue:44913", + "left": "issue:42915", + "right": "issue:43322", "accept": false, - "reason": "Both are config-related, but one is Gemma3TextConfig API mismatch and the other is GPTNeoX rotary_pct reload behavior." + "reason": "Different failure classes and code paths: Qwen3Moe FP8 config handling versus Llava Next loading segfault." }, { "left": "issue:28530", "right": "issue:30064", "accept": false, - "reason": "Training early-stopping metric lookup vs image processor void segmentation-map handling are unrelated." + "reason": "Early-stopping metric selection and void segmentation-map processing are unrelated subsystems and symptoms." }, { "left": "issue:43540", "right": "issue:43782", "accept": false, - "reason": "Both mention Qwen3 multimodal loading, but one is video input processing and the other is from_pretrained weight_only loading; different code paths." + "reason": "Different Qwen multimodal bugs: video input processing error versus `weight_only=True` loading error." }, { "left": "issue:33357", "right": "issue:35707", "accept": false, - "reason": "MacOS CLIP bus error and progressive generation with inputs_embeds/past_key_values are unrelated bugs." + "reason": "MacOS bus error while loading a CLIP model is unrelated to progressive generation with `inputs_embeds` and `past_key_values`." }, { - "left": "issue:43873", - "right": "issue:45704", + "left": "issue:43550", + "right": "issue:43653", "accept": false, - "reason": "Quantization/offloading behavior and a T5 RMSNorm memory leak are separate issues." + "reason": "Torch.compile/SDPA failure in Bamba and BigBird tokenizer special-token registration are different bugs." }, { - "left": "issue:43550", - "right": "issue:43653", + "left": "issue:43873", + "right": "issue:45704", "accept": false, - "reason": "torch.compile/SDPA failure in Bamba is unrelated to BigBirdTokenizer special-token decode behavior." + "reason": "Quantization/offload behavior and a T5 RMSNorm memory leak are distinct problems with different fixes." }, { "left": "issue:43606", "right": "issue:43720", "accept": false, - "reason": "Both involve loading/offload infrastructure, but the concrete bugs differ: Bark CPU offload device mismatch vs BitNet packed-weight unpacking." + "reason": "Both involve loading/offload, but one is a CPU offload device mismatch and the other is packed-weight unpacking during accelerate loading; not the same defect." }, { "left": "issue:29159", "right": "issue:36032", "accept": false, - "reason": "General slow-vs-fast tokenizer inconsistency is not the same as the T5 add_special_tokens method-name conflict." + "reason": "Slow/fast tokenizer inconsistency and a T5 tokenizer loading name conflict are separate tokenizer issues." }, { "left": "issue:30064", "right": "issue:34634", "accept": false, - "reason": "Image processor segmentation-map failure and BarkProcessor voice_preset behavior are unrelated." + "reason": "Image processor segmentation-map handling is unrelated to BarkProcessor `voice_preset` behavior." }, { "left": "issue:44387", "right": "issue:45701", "accept": false, - "reason": "Int4 quantization memory/OOM regression does not match a tokenization-version change issue." + "reason": "CUDA memory growth under int4 quantization is not the same as a version-dependent tokenization change." }, { "left": "issue:44534", "right": "issue:46665", "accept": false, - "reason": "Non-persistent buffer initialization junk and from_pretrained inside Python classes are different loading bugs." + "reason": "Non-persistent buffer initialization junk and inconsistent `from_pretrained` behavior inside Python classes are different bugs." }, { "left": "issue:44492", "right": "issue:44509", "accept": false, - "reason": "Cache-strategy typo and outdated docs mentioning removed pipeline tasks are unrelated." + "reason": "A cache-strategy typo and outdated docs for removed pipeline tasks are documentation issues of different scope." }, { "left": "issue:43388", "right": "issue:45059", "accept": false, - "reason": "Metric gathering label truncation is unrelated to SAM3 text/bounding-box behavior." + "reason": "Metric gathering drops tuple labels in the last batch; SAM3 text/box behavior is a different multimodal-input bug." }, { "left": "issue:41720", "right": "issue:43746", "accept": false, - "reason": "Qwen3 auto device-mapping CUDA assert and GraniteSpeech PEFT local-checkpoint loading are different problems." + "reason": "Qwen3 auto device mapping CUDA assert and GraniteSpeech PEFT local-checkpoint loading are unrelated loading/runtime failures." }, { "left": "issue:43531", "right": "issue:45759", "accept": false, - "reason": "Qwen3-MoE sliding_window behavior and composite Qwen text_config unwrapping are distinct config/model issues." + "reason": "Sliding-window behavior in Qwen3-MoE and `from_config` not unwrapping `text_config` are different configuration bugs." }, { "left": "issue:42490", "right": "issue:43329", "accept": false, - "reason": "Trailing-slash local path warning is unrelated to the multimodal token-count video-branch undefined-name bug." + "reason": "Local-path import warning and an undefined-variable bug in multimodal token counting are unrelated." }, { "left": "issue:43606", "right": "issue:43653", "accept": false, - "reason": "Bark CPU offload mismatch and BigBirdTokenizer mask-token decode failure are unrelated." + "reason": "Device-mismatch offload bug versus tokenizer special-token registration bug; no shared root cause." }, { "left": "issue:45406", "right": "issue:45464", "accept": false, - "reason": "Gemma4Processor serve crash and Qwen3.5 streaming chat/completions failure are different serving bugs." + "reason": "Gemma4Processor missing `_tokenizer` in serve and Qwen3.5 streaming inference failure are different API/runtime issues." }, { "left": "issue:34567", "right": "issue:35707", "accept": false, - "reason": "TrainerState token counter update and progressive generation cache behavior do not share a code-path." + "reason": "Trainer state token accounting and progressive generation are unrelated trainer/generation paths." }, { - "left": "issue:45357", - "right": "issue:46402", + "left": "issue:34634", + "right": "issue:41720", "accept": false, - "reason": "Both are multimodal weight-loading issues, but they affect different models and different saved/loaded components." + "reason": "BarkProcessor voice-preset handling and Qwen3 device-mapping CUDA asserts do not share a code path." }, { - "left": "issue:34634", - "right": "issue:41720", + "left": "issue:45357", + "right": "issue:46402", "accept": false, - "reason": "BarkProcessor voice_preset and Qwen3 auto device mapping are unrelated." + "reason": "Incorrect save_pretrained visual-encoder keys and SAM3 text-encoder weights not loading are different multimodal serialization/load bugs." }, { "left": "issue:44451", "right": "issue:44991", "accept": false, - "reason": "Model/tokenizer loading failures for two different repos are not enough to treat as the same bug." + "reason": "Both are tokenizer-load failures, but they affect different models and likely different tokenizer regressions." }, { "left": "issue:45759", "right": "issue:45859", "accept": false, - "reason": "from_config text_config unwrapping and CPU-offload incompatibility in rotary embedding are separate issues." + "reason": "Composite Qwen config unwrapping and CPU-offload rotary embedding compatibility are unrelated fixes." }, { "left": "issue:43531", "right": "issue:44589", "accept": false, - "reason": "Qwen3-MoE sliding window support and missing Float8 storage are unrelated." + "reason": "Sliding-window handling and missing Float8 storage lookup are completely different bug classes." + }, + { + "left": "issue:44625", + "right": "issue:45479", + "accept": false, + "reason": "`num_labels` propagation and zero-loss single-label classification are both classification-related but stem from different logic errors." }, { "left": "issue:44077", "right": "issue:44625", "accept": false, - "reason": "Optional post_init policy for patchtsmixer and Qwen3.5 num_labels propagation are distinct API/config problems." + "reason": "PatchTSMixer `post_init` policy and Qwen3.5 `num_labels` propagation are unrelated API/config issues." }, { "left": "issue:43644", "right": "issue:46665", "accept": false, - "reason": "Non-persistent buffer junk in v5 and inconsistent from_pretrained behavior inside Python classes are not the same bug." + "reason": "Non-persistent buffer junk and inconsistent `from_pretrained` in Python classes are different failure modes." }, { "left": "issue:44898", "right": "issue:45072", "accept": false, - "reason": "Perceiver resolution handling and bfloat16 dtype mismatches in SwitchTransformers/TimmWrapperModel are unrelated." + "reason": "Perceiver positional encoding interpolation and bf16 dtype mismatch in SwitchTransformers/TimmWrapperModel are unrelated." } ] }, @@ -5116,11 +5110,11 @@ "issue:44987|issue:45042", "issue:39692|issue:44610", "issue:44514|issue:44545", - "issue:43421|issue:43927", "issue:44918|issue:44936", + "issue:43421|issue:43927", "issue:45704|issue:45923", - "issue:41720|issue:45704", "issue:43746|issue:46050", + "issue:41720|issue:45704", "issue:43381|issue:45704", "issue:43531|issue:43645", "issue:42886|issue:44843", @@ -5128,8 +5122,8 @@ "issue:45759|issue:45923", "issue:43782|issue:45381", "issue:43421|issue:43653", - "issue:39692|issue:43873", "issue:43994|issue:44671", + "issue:39692|issue:43873", "issue:43746|issue:43873", "issue:43421|issue:44186", "issue:45081|issue:45356", @@ -5141,224 +5135,218 @@ "issue:43295|issue:45834", "issue:43550|issue:45072" ], - "prepared_review_unit_hash": "2ad5488cdb4997f2963540dd6fb4d34c0302760f54635eb6882354c42195f0c7", + "prepared_review_unit_hash": "392866393daa0b82ec856dd429d39a25d6f3f1cc5267f4d6b0349c3845f3f7b2", "component_budget": { "node_count": 473, "item_count": 473, "soft_pair_count": 1738, - "serialized_chars": 901383, - "estimated_input_tokens": 225346, - "estimated_eval_tokens": 450948 + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 30, - "serialized_chars": 37056, - "estimated_input_tokens": 9264, - "estimated_eval_tokens": 18784 + "serialized_chars": 37057, + "estimated_input_tokens": 9265, + "estimated_eval_tokens": 18786 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Mostly a loose semantic cluster of related but distinct issue reports. The offline-tokenizer pair and the Qwen2_5_VL batched chat-template pair are similar, but the packet only gives soft similarity and marks them non-deterministic, so they should stay tentative rather than be treated as confirmed duplicates.", - "confidence": 0.31, - "canonical_issue_reason": "issue:44843 is the strongest anchor for the offline-tokenizer regression because it names the concrete _patch_mistral_regex/model_info() path; issue:42886 is the broader symptom report, but the packet does not provide direct duplicate-level evidence.", + "summary": "This cluster is mostly noisy and heterogeneous: most items are unrelated Transformers bugs spanning tokenizers, multimodal processors, quantization, and training/runtime regressions. Only one soft edge looks like a true duplicate pair (`Qwen2_5_VLProcessor.apply_chat_template` with `padding=False`), while the rest should stay separate.", + "confidence": 0.87, + "canonical_issue_reason": "Issue 43421 is the strongest representative of the cluster\u2019s main tokenizer/runtime theme: it describes a concrete core-library bug around special-token changes not propagating to the post-processor, and it sits near several other tokenizer-related reports.", "canonical_pr_reason": null, - "best_issue_reason": "issue:44843 is the best representative issue because it pinpoints the likely root cause and makes the offline-cache failure easiest to triage.", + "best_issue_reason": "Issue 43421 is the best overall issue to keep because it is the most general and reusable bug report in the set, with the clearest cross-cutting impact compared with the many narrower model-specific regressions.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:44062", "right": "issue:44451", "accept": false, - "reason": "Both are tokenizer/load failures, but the error modes and code paths look different." + "reason": "Different problems: `AddedToken(..., special=...)` constructor conflict vs a model-loading failure for `vesteinn/ScandiBERT`." }, { "left": "issue:44368", "right": "issue:45357", "accept": false, - "reason": "Different bugs: a training warning about tie_word_embeddings vs a save/load regression for visual encoder keys." + "reason": "Both are warnings/regressions in multimodal setups, but one is about `tie_word_embeddings` and the other about incorrect visual encoder keys on save/load." }, { "left": "issue:43381", "right": "issue:43844", "accept": false, - "reason": "Eval-mode gradient checkpointing and ZeRO-3 gradient growth are separate problems." + "reason": "Eval-mode gradient checkpointing behavior and ZeRO-3 gradient growth are distinct training-path issues." }, { "left": "issue:44987", "right": "issue:45042", "accept": false, - "reason": "Model-loading breakage and PIL/torchvision dependency handling are unrelated." + "reason": "Unrelated failures: loading `physical-intelligence/fast` vs PIL image processors incorrectly requiring torchvision." }, { "left": "issue:39692", "right": "issue:44610", "accept": false, - "reason": "Both touch vision processing, but they involve different models and different failures." + "reason": "Different model/documentation bugs: SigLIP2 example errors vs OmDet-Turbo image-size mismatch." }, { "left": "issue:44514", "right": "issue:44545", - "accept": false, - "reason": "The titles match closely, but the packet only shows soft similarity and no direct duplicate evidence." + "accept": true, + "reason": "Same concrete bug: `Qwen2_5_VLProcessor.apply_chat_template` crashes on batched input when `padding=False`." }, { - "left": "issue:43421", - "right": "issue:43927", + "left": "issue:44918", + "right": "issue:44936", "accept": false, - "reason": "Related tokenizer-surface area, but runtime post-processor updates and save/load token-ID loss are different bugs." + "reason": "Different code paths: Qwen3.5 embedding unpacking in TRL SFT vs `trainer.evaluate()` failing after `trainer.train()`." }, { - "left": "issue:44918", - "right": "issue:44936", + "left": "issue:43421", + "right": "issue:43927", "accept": false, - "reason": "Different trainer failures: embedding unpacking vs evaluate-after-train behavior." + "reason": "Related tokenizer state, but one is runtime special-token/post-processor updates and the other is config persistence for custom token IDs." }, { "left": "issue:45704", "right": "issue:45923", "accept": false, - "reason": "T5 Apex RMSNorm memory leak and missing gradient-checkpointing flag are unrelated." + "reason": "Different subsystems: T5/apex RMSNorm memory leak vs missing `supports_gradient_checkpointing` on a trust-remote-code model." }, { - "left": "issue:41720", - "right": "issue:45704", + "left": "issue:43746", + "right": "issue:46050", "accept": false, - "reason": "CUDA assert on Qwen3 auto mapping is a different failure from the T5 memory-leak report." + "reason": "Adapter-loading failure for local checkpoints vs quantized-model save/load breakage are not the same bug." }, { - "left": "issue:43746", - "right": "issue:46050", + "left": "issue:41720", + "right": "issue:45704", "accept": false, - "reason": "Both involve model loading/saving, but the concrete breakages differ." + "reason": "Qwen3 auto device-mapping CUDA assert is unrelated to the T5/apex RMSNorm leak." }, { "left": "issue:43381", "right": "issue:45704", "accept": false, - "reason": "Eval-mode checkpointing and T5 memory leak are different issues." + "reason": "Gradient checkpointing in eval mode and the T5 apex memory leak are separate issues." }, { "left": "issue:43531", "right": "issue:43645", "accept": false, - "reason": "Sliding-window behavior and Jupyter custom-model initialization are separate bugs." + "reason": "Sliding-window handling for Qwen3-MoE is unrelated to notebook-defined custom model initialization in Transformers 5.0.0." }, { "left": "issue:42886", "right": "issue:44843", "accept": false, - "reason": "They are likely related offline-tokenizer reports, but the packet does not provide enough direct evidence to call them the same bug." + "reason": "Both concern offline/tokenizer loading, but one is cache resolution for tokenizers and the other is an unconditional `model_info()` call in `_patch_mistral_regex`." }, { "left": "issue:46032", "right": "issue:46463", "accept": false, - "reason": "Mamba2Mixer cache correctness and T5 decoder-input error messaging are unrelated." + "reason": "Mamba2Mixer cache correctness with `seq_len > 1` is unrelated to T5 decoder-input validation messaging." }, { "left": "issue:45759", "right": "issue:45923", "accept": false, - "reason": "Composite config unwrapping and missing gradient-checkpointing support are different problems." + "reason": "`AutoModelForCausalLM.from_config` unwrapping `text_config` is a different composite-config bug from a missing gradient-checkpointing flag." }, { "left": "issue:43782", "right": "issue:45381", "accept": false, - "reason": "Different Qwen VL regressions: weight_only loading vs video position IDs." + "reason": "Qwen3VL weight-only loading and Qwen2.5-VL video `vision_position_ids` are different multimodal failures." }, { - "left": "issue:43421", - "right": "issue:43653", + "left": "issue:43994", + "right": "issue:44671", "accept": false, - "reason": "Both mention tokenizer internals, but they are different bugs with different symptoms." + "reason": "Different symptoms and models: incorrect SigLIP2 results vs CamemBERT masked-LM predictions." }, { "left": "issue:39692", "right": "issue:43873", "accept": false, - "reason": "SigLIP2 example failures and offloading-with-quantization behavior are not the same bug." - }, - { - "left": "issue:43994", - "right": "issue:44671", - "accept": false, - "reason": "Different model-specific correctness regressions." + "reason": "Documentation/example issues for SigLIP2 are not the same as quantization/offloading behavior." }, { "left": "issue:43746", "right": "issue:43873", "accept": false, - "reason": "Local-checkpoint loading with PEFT adapters is unrelated to quantized offloading behavior." + "reason": "PEFT adapter checkpoint loading and quantization offloading are separate model-loading paths." }, { "left": "issue:43421", - "right": "issue:44186", + "right": "issue:43653", "accept": false, - "reason": "Special-token post-processor updates and LayoutLMv2 tokenizer crashes are different issues." + "reason": "Both touch tokenizer special-token handling, but one is post-processor refresh on runtime edits and the other is BigBird mask-token registration causing empty decode output." }, { "left": "issue:45081", "right": "issue:45356", "accept": false, - "reason": "Both are Mistral-regex-related tokenizer regressions, but the concrete failures are not the same." + "reason": "Both involve Mistral/Kimi tokenizer regex handling, but the reported failures are not the same concrete bug." }, { "left": "issue:30064", "right": "issue:36010", "accept": false, - "reason": "Image segmentation-map processing is unrelated to the GenerationMixin import error." + "reason": "Image processor void-segmentation failure is unrelated to the `GenerationMixin` import regression." }, { "left": "issue:43720", "right": "issue:43819", "accept": false, - "reason": "Packed-weight loading in BitNet and DAC STE/forward mismatch are unrelated." + "reason": "BitNet packed-weight unpacking during accelerate loading is unrelated to DAC `from_latents` vs forward-pass mismatch." }, { "left": "issue:45356", "right": "issue:45907", "accept": false, - "reason": "Tokenizer regression and list_repo_files revision forwarding are unrelated." + "reason": "Tokenizer codec/regex regression is unrelated to `list_repo_files` failing to forward `revision`." }, { "left": "issue:43653", "right": "issue:43720", "accept": false, - "reason": "BigBirdTokenizer special-token registration and BitNet packed-weight loading are different issues." + "reason": "Both are CI-tagged issues, but BigBird special-token decoding and BitNet packed-weight loading are different bugs." }, { "left": "issue:43540", "right": "issue:46294", "accept": false, - "reason": "Both involve video inputs, but the affected models and failure stages differ." + "reason": "Both are Qwen multimodal video issues, but one is a processing `ValueError` in Qwen3OmniMoe and the other is a video-feature/token mismatch in Qwen3.5-9B." }, { "left": "issue:43295", "right": "issue:45834", "accept": false, - "reason": "Different model-loading/processor regressions with no shared concrete code path." + "reason": "Custom model processor/tokenizer access regression is unrelated to Kosmos2.5 long-OCR index errors." }, { "left": "issue:43550", "right": "issue:45072", "accept": false, - "reason": "Different inference failures in different models and execution modes." + "reason": "Different runtime paths: Bamba-9B-v2 torch.compile with SDPA vs dtype mismatches in SwitchTransformers/TimmWrapperModel." } ] }, "evaluator_result": { - "accept": false, - "feedback": "The summary is directionally reasonable, but it overstates evidence for the two accepted soft edges. In the packet, both `issue:42886`\u2013`issue:44843` and `issue:44514`\u2013`issue:44545` are marked `deterministic_accept: false`, so treating them as true duplicates is too strong. The rest of the rejects are conservative enough, but the duplicate calls need to stay at most tentative unless the packet gives stronger direct evidence." + "accept": true, + "feedback": "The summary is grounded in the packet, and the soft-edge decisions are conservative overall. The only accepted pair, issue:44514 / issue:44545, is an evident duplicate with the same Qwen2_5_VLProcessor.apply_chat_template padding=False crash. The remaining pairs are plausibly related by theme but not the same concrete bug, so rejecting them is safe." }, "evaluator_used": true, - "retried": true, - "accepted_nontrivial_soft_edge": false, + "retried": false, + "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null }, @@ -5426,8 +5414,8 @@ "soft_pairs": [ "issue:43596|issue:44220", "issue:43550|issue:43927", - "issue:43618|issue:43697", "issue:44062|issue:44488", + "issue:43618|issue:43697", "issue:43611|issue:46665", "issue:44186|issue:44898", "issue:44220|issue:44661", @@ -5435,213 +5423,207 @@ "issue:43531|issue:43643", "issue:43335|issue:43756", "issue:43550|issue:43819", - "issue:43643|issue:44877", "issue:43540|issue:44008", "issue:44843|issue:45081", "issue:41720|issue:44991", + "issue:43643|issue:44877", "issue:30064|issue:36331", - "issue:39692|issue:45834", "issue:44821|issue:45084", + "issue:39692|issue:45834", "issue:44186|issue:44442", "issue:43322|issue:45685", "issue:44448|issue:44671", "issue:45584|issue:45834", + "issue:45183|issue:45479", "issue:43388|issue:46032", "issue:43716|issue:44008", - "issue:45183|issue:45479", - "issue:43644|issue:43749", - "issue:44485|issue:44509" + "issue:44485|issue:44509", + "issue:43644|issue:43749" ], - "prepared_review_unit_hash": "4efafb91e39cc5f1a7e5e119e18166c54826607e00df709554ce39581c5eb1ee", + "prepared_review_unit_hash": "ccca756ad017e74c71b72a71f8b12adc5d727fb9fae4a6585076010ba5a0d397", "component_budget": { "node_count": 473, "item_count": 473, "soft_pair_count": 1738, - "serialized_chars": 901383, - "estimated_input_tokens": 225346, - "estimated_eval_tokens": 450948 + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 27, - "serialized_chars": 35390, - "estimated_input_tokens": 8848, - "estimated_eval_tokens": 17952 + "serialized_chars": 35388, + "estimated_input_tokens": 8847, + "estimated_eval_tokens": 17950 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is noisy: many items are broadly about Transformers v5 loading/tokenizer regressions, but the soft pairs mostly describe different concrete failures, so I would not merge them as duplicates.", - "confidence": 0.29, - "canonical_issue_reason": "issue:43611 is the broadest, most central loading-regression report in the set; it cleanly represents the cluster\u2019s general v5 model-loading breakage better than the narrower, more specialized bug reports.", + "summary": "This cluster is not a duplicate set; it mixes many unrelated Transformers bugs across tokenizers, multimodal models, loading/config paths, and v5 regressions. None of the soft edges look like the same concrete underlying defect.", + "confidence": 0.94, + "canonical_issue_reason": "No clear canonical issue: the items are heterogeneous and do not collapse to one shared bug family.", "canonical_pr_reason": null, - "best_issue_reason": "issue:43611 is the best single issue to anchor the cluster because it is general, user-facing, and closest to the shared theme of loading regressions across v5.", + "best_issue_reason": "No best issue candidate stands out as representing the rest; the cluster lacks a single common code path or symptom.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43596", "right": "issue:44220", "accept": false, - "reason": "Different subsystems and symptoms: Deepspeed/zero3 BertModel init versus audio feature extraction. No shared underlying bug." + "reason": "Different subsystems and failures: Deepspeed/BERT init index error vs audio fbank extraction." }, { "left": "issue:43550", "right": "issue:43927", "accept": false, - "reason": "One is a torch.compile/SDPA model execution bug; the other is config save/load losing custom token IDs. Different code paths." + "reason": "Unrelated bugs: torch.compile/SDPA failure vs DiaConfig losing custom token IDs on save/load." }, { - "left": "issue:43618", - "right": "issue:43697", + "left": "issue:44062", + "right": "issue:44488", "accept": false, - "reason": "CLIP attentions missing and RTDetrV2 output drift are separate model regressions with different mechanisms." + "reason": "Both tokenizer-related, but one is an AddedToken kwarg conflict and the other is a model loading failure for a specific tokenizer." }, { - "left": "issue:44062", - "right": "issue:44488", + "left": "issue:43618", + "right": "issue:43697", "accept": false, - "reason": "Both touch tokenizer loading, but one is an AddedToken API error and the other is a specific tokenizer model load failure. Not the same bug." + "reason": "Both are v5 regressions, but they affect different models and code paths, so they are not the same bug." }, { "left": "issue:43611", "right": "issue:46665", "accept": false, - "reason": "Both involve from_pretrained/loading, but 43611 is a base_model_prefix regression while 46665 is about behavior inside Python wrapper classes. Different failure modes." + "reason": "Both concern loading, but one is base_model_prefix behavior and the other is from_pretrained inside Python classes; distinct issues." }, { "left": "issue:44186", "right": "issue:44898", "accept": false, - "reason": "LayoutLMv2 tokenization/padding crash and Perceiver image-size interpolation failure are unrelated." + "reason": "Different domains and symptoms: LayoutLMv2 tokenizer/NER padding crash vs Perceiver image classification interpolation failure." }, { "left": "issue:44220", "right": "issue:44661", "accept": false, - "reason": "Audio feature extraction bug versus add-new-model-like failing in TOKENIZER_MAPPING_NAMES. Different underlying issues." + "reason": "Feature extraction bug vs add-new-model-like failure in TOKENIZER_MAPPING_NAMES; no shared concrete defect." }, { "left": "issue:43653", "right": "issue:43819", "accept": false, - "reason": "BigBirdTokenizer special-token handling and DAC.from_latents vs forward mismatch are unrelated." + "reason": "Tokenizer special-token registration issue vs DAC from_latents mismatch; unrelated code paths." }, { "left": "issue:43531", "right": "issue:43643", "accept": false, - "reason": "Sliding-window MoE behavior and AutoConfig trust_remote_code field loss are different parts of the stack." + "reason": "Sliding-window behavior in Qwen3-MoE is unrelated to trust_remote_code config field loss." }, { "left": "issue:43335", "right": "issue:43756", "accept": false, - "reason": "SwitchTransformers sparse-layer config bug is unrelated to Smollm3 RoPE-layer dropping." + "reason": "Config creates sparse layers incorrectly vs RoPE layer drop in Smollm3; different models and bugs." }, { "left": "issue:43550", "right": "issue:43819", "accept": false, - "reason": "torch.compile SDPA failure and DAC latent conversion mismatch are different model bugs." - }, - { - "left": "issue:43643", - "right": "issue:44877", - "accept": false, - "reason": "Both concern config loading, but one is missing fields from trust_remote_code and the other is overly strict config blocking granite_speech. Not the same issue." + "reason": "torch.compile/SDPA failure is unrelated to DAC latent decoding mismatch." }, { "left": "issue:43540", "right": "issue:44008", "accept": false, - "reason": "Video-input processing in Qwen3OmniMoe and a Gemma3n variable-name collision are unrelated." + "reason": "Different multimodal failures: video input processing in Qwen3OmniMoe vs Gemma3n variable collision." }, { "left": "issue:44843", "right": "issue:45081", "accept": false, - "reason": "Both mention _patch_mistral_regex, but one is an offline-mode model_info() call and the other is a backend_tokenizer AttributeError. Distinct bugs in the same area." + "reason": "Same helper function area, but different failures: offline model_info call vs backend_tokenizer attribute error." }, { "left": "issue:41720", "right": "issue:44991", "accept": false, - "reason": "Qwen3 auto device mapping CUDA assert and tokenizer load failure for EMBEDDIA/est-roberta are different issues." + "reason": "Qwen3 auto device mapping cuda assert is unrelated to tokenizer loading for EMBEDDIA/est-roberta." }, { - "left": "issue:30064", - "right": "issue:36331", + "left": "issue:43643", + "right": "issue:44877", "accept": false, - "reason": "Image processor segmentation-map handling and CustomTrainer.compute_loss signature regression are unrelated." + "reason": "Both are config-loading problems, but the root causes and affected objects differ." }, { - "left": "issue:39692", - "right": "issue:45834", + "left": "issue:30064", + "right": "issue:36331", "accept": false, - "reason": "SigLIP2 docs/model mismatch with quantization and Kosmos2.5 long OCR input index error are separate problems." + "reason": "Image segmentation map processing bug vs CustomTrainer compute_loss signature change; unrelated." }, { "left": "issue:44821", "right": "issue:45084", "accept": false, - "reason": "AutoImageProcessor URL loading and template compilation errors are different components and failure modes." + "reason": "AutoImageProcessor URL loading failure vs template-node compilation error; no shared bug." }, { - "left": "issue:44186", - "right": "issue:44442", + "left": "issue:39692", + "right": "issue:45834", "accept": false, - "reason": "Tokenizer crash on NER/batched padding versus AutoTokenizer failing to load FastSpeech2ConformerTokenizer are not the same bug." + "reason": "Doc example/model mismatch and quantization errors are unrelated to Kosmos2.5 long-ocr index errors." }, { - "left": "issue:43322", - "right": "issue:45685", + "left": "issue:44186", + "right": "issue:44442", "accept": false, - "reason": "Llava Next segmentation fault and MPS histogram support error are unrelated runtime issues." + "reason": "Tokenizer crash on NER/padding is not the same as AutoTokenizer failing to load a FastSpeech2Conformer tokenizer." }, { "left": "issue:44448", "right": "issue:44671", "accept": false, - "reason": "Pegasus v4/v5 output drift and CamemBERT masked-LM prediction regression are different model-specific behaviors." + "reason": "Both are v5 output regressions, but they affect different models and likely different causes." }, { "left": "issue:45584", "right": "issue:45834", "accept": false, - "reason": "Whisper empty transcription after token alignment and Kosmos2.5 long OCR input index error are unrelated." + "reason": "Whisper empty transcription failure after token alignment is unrelated to Kosmos2.5 long-input indexing." + }, + { + "left": "issue:45183", + "right": "issue:45479", + "accept": false, + "reason": "Input validation/XOR messaging bug vs degenerate zero-loss classification behavior; different underlying defects." }, { "left": "issue:43388", "right": "issue:46032", "accept": false, - "reason": "gather_for_metrics label truncation and Mamba2Mixer use_cache correctness are different training/inference issues." + "reason": "Metrics gathering last-batch bug is unrelated to Mamba2Mixer use_cache incorrect results." }, { "left": "issue:43716", "right": "issue:44008", "accept": false, - "reason": "Mistral-3 image preprocessor dtype mismatch and Gemma3n audio tensor attribute collision are unrelated." + "reason": "Different multimodal model families and failures: dtype mismatch in Mistral-3 vs Gemma3n attribute collision." }, { - "left": "issue:45183", - "right": "issue:45479", + "left": "issue:44485", + "right": "issue:44509", "accept": false, - "reason": "Input validation XOR error messaging and degenerate single-label classification loss are different bugs." + "reason": "RoPE implementation discussion is unrelated to docs still mentioning removed pipeline tasks." }, { "left": "issue:43644", "right": "issue:43749", "accept": false, - "reason": "Non-persistent buffer junk values and FSDP CPU RAM efficient loading breakage are separate initialization/loading regressions." - }, - { - "left": "issue:44485", - "right": "issue:44509", - "accept": false, - "reason": "GLM-5 RoPE implementation discussion and docs mentioning removed pipeline tasks are not duplicates." + "reason": "Both are v5/loading related, but one is non-persistent buffers filled with junk and the other is broken FSDP CPU RAM efficient loading." } ] }, @@ -5671,6 +5653,7 @@ "issue:36010", "issue:36331", "issue:42175", + "issue:43262", "issue:43421", "issue:43540", "issue:43577", @@ -5681,6 +5664,7 @@ "issue:43749", "issue:43782", "issue:43818", + "issue:43819", "issue:43825", "issue:43827", "issue:43950", @@ -5700,7 +5684,6 @@ "issue:45020", "issue:45072", "issue:45081", - "issue:45325", "issue:45381", "issue:45701", "issue:45834", @@ -5715,8 +5698,8 @@ ], "soft_pairs": [ "issue:43606|issue:45072", - "issue:43950|issue:45987", "issue:43582|issue:46512", + "issue:43950|issue:45987", "issue:34634|issue:36010", "issue:44479|issue:45834", "issue:43825|issue:43994", @@ -5733,218 +5716,218 @@ "issue:28218|issue:29159", "issue:43825|issue:43827", "issue:43577|issue:43582", - "issue:45987|issue:46192", "issue:43749|issue:44291", + "issue:45987|issue:46192", "issue:43645|issue:46665", "issue:44671|issue:45701", "issue:42175|issue:45081", "issue:34634|issue:36331", "issue:43582|issue:44857", "issue:45003|issue:46285", - "issue:43782|issue:45325" + "issue:43262|issue:43819" ], - "prepared_review_unit_hash": "8dc84860e73b8e4b29ee7822f1b9571754b8c7b4624c34e8bfc2e85d84b74a9a", + "prepared_review_unit_hash": "4fb6955d9525ba4762d8687fb97df70f8b6cd246dd25cf63fefaca75df43b34f", "component_budget": { "node_count": 473, "item_count": 473, "soft_pair_count": 1738, - "serialized_chars": 901383, - "estimated_input_tokens": 225346, - "estimated_eval_tokens": 450948 + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 }, "budget": { - "node_count": 47, - "item_count": 47, + "node_count": 48, + "item_count": 48, "soft_pair_count": 28, - "serialized_chars": 35840, - "estimated_input_tokens": 8960, - "estimated_eval_tokens": 18176 + "serialized_chars": 36429, + "estimated_input_tokens": 9108, + "estimated_eval_tokens": 18472 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Most pairs are not true duplicates: they describe different models, different code paths, or one is just a broader umbrella report. The only clear duplicate-like match is the tokenizer added-token spacing regression and the slow/fast tokenizer inconsistency report.", - "confidence": 0.81, - "canonical_issue_reason": "Issue 28218 is the clearest open, concrete report for the tokenizer added-token spacing regression, making it the best canonical issue for the one accepted duplicate family.", + "summary": "This cluster is mostly a grab bag of unrelated regression reports. The only likely duplicate pair is the tokenizer spacing/inconsistency reports; the rest are too model- or feature-specific to safely merge.", + "confidence": 0.38, + "canonical_issue_reason": "issue:45701 is the broadest, most umbrella-like issue in the set, and it best represents the general tokenizer/regression theme among the items.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 28218 is the best overall representative because it is specific, actionable, and matches the only pair that looks like the same underlying bug.", + "best_issue_reason": "issue:45701 has the widest scope and strongest chance of serving as the representative issue for the cluster's tokenizer-regression subset.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43606", "right": "issue:45072", "accept": false, - "reason": "Different failures: CPU offload device mismatch vs bfloat16 dtype mismatch." + "reason": "Both are inference mismatches, but they describe different models and different symptoms (device offload vs bfloat16 dtype mismatch)." }, { - "left": "issue:43950", - "right": "issue:45987", + "left": "issue:43582", + "right": "issue:46512", "accept": false, - "reason": "Unrelated bugs: non-persistent buffer corruption vs StaticCache seq-length return type." + "reason": "Unrelated: AppleSilicon warmup TypeError vs stale config dtype after from_config." }, { - "left": "issue:43582", - "right": "issue:46512", + "left": "issue:43950", + "right": "issue:45987", "accept": false, - "reason": "Different symptoms and code paths: Apple Silicon warmup TypeError vs stale config dtype after from_config()." + "reason": "Different regressions: non-persistent buffer corruption vs StaticCache length return type." }, { "left": "issue:34634", "right": "issue:36010", "accept": false, - "reason": "BarkProcessor voice_preset issue is unrelated to GenerationMixin import failure." + "reason": "BarkProcessor voice preset failure and GenerationMixin import failure are unrelated." }, { "left": "issue:44479", "right": "issue:45834", "accept": false, - "reason": "Different model-specific regressions: Qwen video-input failure vs Kosmos2.5 long-OCR index error." + "reason": "Different vision/video failures in different models; no clear shared bug." }, { "left": "issue:43825", "right": "issue:43994", "accept": false, - "reason": "One is an error-message wording issue; the other is a model output regression in SigLIP2." + "reason": "One is an error-message/doc issue around pipeline deprecation; the other is a model output regression." }, { "left": "issue:43577", "right": "issue:43818", "accept": false, - "reason": "BLIP2 dtype loading bug and Video-LLaVA architecture/weights bug are not the same issue." + "reason": "Different BLIP2 dtype propagation issue vs Video-LLaVA architecture/weight issue." }, { "left": "issue:43421", "right": "issue:44811", "accept": false, - "reason": "Tokenizer post-processor update bug is separate from Whisper batch_decode() ignoring skip_special_tokens." + "reason": "Both involve tokenizers, but one is runtime special-token post-processing and the other is batch_decode skip_special_tokens behavior." }, { "left": "issue:43782", "right": "issue:44977", "accept": false, - "reason": "Loading error with weight_only=True is different from flash-attention generation behavior." + "reason": "Different Qwen generation/loading failures with different triggers and code paths." }, { "left": "issue:46294", "right": "issue:46463", "accept": false, - "reason": "Qwen3.5 video feature/token mismatch is unrelated to T5 decoder input validation messaging." + "reason": "Qwen video feature/token mismatch vs T5 decoder-input validation messaging are unrelated." }, { "left": "issue:43540", "right": "issue:45381", "accept": false, - "reason": "Both are video-input issues, but they affect different Qwen models and likely different code paths." + "reason": "Both are video-input bugs, but they affect different Qwen families and likely different preprocessing paths." }, { "left": "issue:44805", "right": "issue:44936", "accept": false, - "reason": "Mask-shape IndexError is not the same as trainer.evaluate() failing after train()." + "reason": "Mask shape IndexError and trainer.evaluate() failing after train are not the same underlying bug." }, { "left": "issue:44991", "right": "issue:46291", "accept": false, - "reason": "Tokenizer loading regression and kernels import/version requirement failure are separate problems." + "reason": "Tokenizer load failure for a specific model and kernels import/version constraint are unrelated." }, { "left": "issue:44561", "right": "issue:45020", "accept": false, - "reason": "45020 is a broad umbrella for remote_code breakages, not the same concrete symbol-removal bug as 44561." + "reason": "Both touch remote_code compatibility, but the titles point to different failure modes and 45020 is a broad umbrella rather than the same concrete bug." }, { "left": "issue:44610", "right": "issue:45072", "accept": false, - "reason": "Processor output size mismatch is unrelated to bfloat16 dtype mismatch in inference." + "reason": "OmDet input-size mismatch and dtype mismatch in inference are different issues." }, { "left": "issue:43720", "right": "issue:45072", "accept": false, - "reason": "Packed-weights loading failure is a different issue from dtype mismatch." + "reason": "Packed-weight loading failure and bfloat16 dtype mismatch are separate accelerate/loading regressions." }, { "left": "issue:28218", "right": "issue:29159", "accept": true, - "reason": "Both point to the same tokenizer behavior regression around added tokens; 29159 frames it as slow/fast tokenizer inconsistency." + "reason": "Both describe the same tokenizer inconsistency around added tokens/spacing; 29159 is the broader version of the same symptom." }, { "left": "issue:43825", "right": "issue:43827", "accept": false, - "reason": "Related topic, but one is a runtime error message and the other is stale documentation." + "reason": "One is an error-message regression, the other is stale docs; related topic, but not the same underlying bug/change." }, { "left": "issue:43577", "right": "issue:43582", "accept": false, - "reason": "These are unrelated failures in different subsystems: dtype handling vs allocator warmup." + "reason": "Different dtype handling bug vs AppleSilicon warmup TypeError." }, { - "left": "issue:45987", - "right": "issue:46192", + "left": "issue:43749", + "right": "issue:44291", "accept": false, - "reason": "Non-persistent buffer corruption and return_tensors precedence are unrelated bugs." + "reason": "Both are loading-related, but they describe different breakages in different code paths." }, { - "left": "issue:43749", - "right": "issue:44291", + "left": "issue:45987", + "right": "issue:46192", "accept": false, - "reason": "Both concern model loading, but the FSDP loading break and init_empty_weights TypeError are different failure modes." + "reason": "Both are processor/return-shape issues, but they affect different parameters and code paths." }, { "left": "issue:43645", "right": "issue:46665", "accept": false, - "reason": "Notebook custom-model initialization and from_pretrained inside classes are distinct issues." + "reason": "Both mention from_pretrained in class-like contexts, but the titles are too broad and point to different scenarios to treat as the same bug." }, { "left": "issue:44671", "right": "issue:45701", "accept": false, - "reason": "CamemBERT MLM prediction regression is not the same as general tokenization changes." + "reason": "CamemBERT masked-LM regression is model-specific; 45701 is a broad tokenization-change report, not the same concrete bug." }, { "left": "issue:42175", "right": "issue:45081", "accept": false, - "reason": "Package extra/backend installation issue is unrelated to the Mistral regex patch crash." + "reason": "TensorFlow backend packaging and Mistral tokenizer patch crash are unrelated." }, { "left": "issue:34634", "right": "issue:36331", "accept": false, - "reason": "BarkProcessor voice_preset and CustomTrainer.compute_loss signature mismatch are different bugs." + "reason": "BarkProcessor voice preset and CustomTrainer.compute_loss signature mismatch are unrelated." }, { "left": "issue:43582", "right": "issue:44857", "accept": false, - "reason": "Apple Silicon warmup TypeError is separate from CUDA AMP crash in LwDetrImageLoss." + "reason": "Different numerical/device issues: AppleSilicon warmup TypeError vs CUDA AMP crash." }, { "left": "issue:45003", "right": "issue:46285", "accept": false, - "reason": "Unsafe sys.modules access and CLIPTextModel structure inconsistency are unrelated." + "reason": "sys.modules access bug and CLIP text-model structure inconsistency are unrelated." }, { - "left": "issue:43782", - "right": "issue:45325", + "left": "issue:43262", + "right": "issue:43819", "accept": false, - "reason": "Qwen3VL weight_only loading error and Qwen2.5-VL RoPE scaling bug do not share the same root cause." + "reason": "Sampling-rate defaulting in audio chat templates and DAC latent/forward mismatch are different bugs." } ] }, "evaluator_result": { "accept": true, - "feedback": "The summary is grounded in the packet and stays mostly conservative. The accepted pair (28218/29159) is plausibly the same tokenizer-family regression from the titles alone, and the rejected pairs are generally justified by distinct models, subsystems, or failure modes. No mergeability issues apply here because this is not a PR-pair cluster." + "feedback": "The analysis is grounded in the packet and stays conservative. The umbrella choice of issue:45701 is reasonable given its broad tokenizer-change title, and the only accepted soft-edge pair (28218/29159) is well supported by the titles. The rejected pairs mostly look safely non-mergeable as single bugfixes based on the information provided." }, "evaluator_used": true, "retried": false, @@ -5972,7 +5955,6 @@ "issue:42175", "issue:42490", "issue:43257", - "issue:43262", "issue:43329", "issue:43452", "issue:43526", @@ -5985,6 +5967,7 @@ "issue:43720", "issue:43742", "issue:43746", + "issue:43782", "issue:43819", "issue:43854", "issue:43906", @@ -6005,6 +5988,7 @@ "issue:45042", "issue:45081", "issue:45084", + "issue:45325", "issue:45372", "issue:45406", "issue:45538", @@ -6014,18 +5998,18 @@ "issue:46326" ], "soft_pairs": [ - "issue:43262|issue:43819", "issue:43653|issue:44912", - "issue:43526|issue:43697", + "issue:43782|issue:45325", "issue:45081|issue:45406", + "issue:43526|issue:43697", "issue:45372|issue:46326", "issue:44060|issue:44368", "issue:43329|issue:45600", "issue:44589|issue:45042", "issue:43742|issue:45020", - "issue:43550|issue:44912", - "issue:43526|issue:44554", "issue:45084|issue:45538", + "issue:43526|issue:44554", + "issue:43550|issue:44912", "issue:43906|issue:46209", "issue:43452|issue:44991", "issue:43638|issue:44936", @@ -6035,229 +6019,210 @@ "issue:36010|issue:36331", "issue:43720|issue:44610", "issue:43531|issue:45042", - "issue:38175|issue:44610", "issue:42175|issue:46291", + "issue:38175|issue:44610", "issue:39692|issue:44912", - "issue:43746|issue:44811", "issue:44387|issue:44617", + "issue:43746|issue:44811", "issue:42490|issue:43257", - "issue:43854|issue:44977", - "issue:43262|issue:44811" + "issue:43854|issue:44977" ], - "prepared_review_unit_hash": "19f65f9f0160afa9674ada3898d99213422bcba32597599582e69b800ac40d7f", + "prepared_review_unit_hash": "df735a76fb0a470a220a1f8afdfc12e6080a8cf4d4ad56916003b7416dec58c3", "component_budget": { "node_count": 473, "item_count": 473, "soft_pair_count": 1738, - "serialized_chars": 901383, - "estimated_input_tokens": 225346, - "estimated_eval_tokens": 450948 + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 }, "budget": { - "node_count": 47, - "item_count": 47, - "soft_pair_count": 29, - "serialized_chars": 36177, - "estimated_input_tokens": 9045, - "estimated_eval_tokens": 18346 + "node_count": 48, + "item_count": 48, + "soft_pair_count": 28, + "serialized_chars": 36420, + "estimated_input_tokens": 9105, + "estimated_eval_tokens": 18466 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster is heterogeneous: the soft pairs mostly share broad Transformers themes (loading, tokenizers, multimodal processors, quantization, training) but not the same concrete defect. I would not merge any of the proposed soft edges.", + "summary": "This is not a true duplicate cluster: the items cover unrelated Transformers regressions across tokenizers, multimodal processors, quantization, training, packaging, and backend-specific behavior. Issue 45020 is the broadest umbrella, but it is still too general to absorb the rest as duplicates.", "confidence": 0.93, - "canonical_issue_reason": "No single canonical issue stands out; the items span unrelated bugs across different models, processors, tokenizers, training paths, and docs.", + "canonical_issue_reason": "issue:45020 is the broadest, highest-level regression report in the set (recent Transformers versions breaking remote_code models), so it is the closest thing to an umbrella issue, though not a true duplicate of the others.", "canonical_pr_reason": null, - "best_issue_reason": "No strong global representative: the cluster is too mixed for one issue to serve as a reliable canonical duplicate target.", + "best_issue_reason": "issue:45020 is the best representative of the cluster because it captures the general theme of version-induced loading breakages; the remaining issues are too specific and materially different to serve as one canonical duplicate target.", "best_pr_reason": null, "soft_edge_verdicts": [ - { - "left": "issue:43262", - "right": "issue:43819", - "accept": false, - "reason": "Different subsystems and failures: audio processor sampling-rate default vs DAC latent reconstruction mismatch." - }, { "left": "issue:43653", "right": "issue:44912", "accept": false, - "reason": "Tokenizer special-token registration bug is unrelated to MXFP4 quantization loading/fallback." + "reason": "Both involve model/loading behavior, but one is a tokenizer special-token decode bug and the other is MXFP4 loading fallback; different code paths and symptoms." }, { - "left": "issue:43526", - "right": "issue:43697", + "left": "issue:43782", + "right": "issue:45325", "accept": false, - "reason": "BEiT label reduction bug is unrelated to RTDetrV2 output drift." + "reason": "Qwen3VL weight_only loading failure is unrelated to Qwen2.5-VL rope index scaling; different models and bugs." }, { "left": "issue:45081", "right": "issue:45406", "accept": false, - "reason": "Both are tokenizer/processor load crashes, but they affect different models and different code paths." + "reason": "Tokenizer regex patch crash in Mistral loading is not the same as Gemma4Processor missing _tokenizer in transformers serve." + }, + { + "left": "issue:43526", + "right": "issue:43697", + "accept": false, + "reason": "BEiT reduce_labels returning one label and RTDetrV2 differing outputs are distinct model bugs with no shared failure mechanism." }, { "left": "issue:45372", "right": "issue:46326", "accept": false, - "reason": "Both involve mistral_common, but one is an import failure and the other is a continue_final_message behavior bug." + "reason": "Both touch mistral_common, but one is a missing import breaking Gemma 4 processor loading and the other is continue_final_message behavior; not the same defect." }, { "left": "issue:44060", "right": "issue:44368", "accept": false, - "reason": "Related warning wording, but the underlying model/config issues are not the same concrete bug." + "reason": "Both mention tied-weights warnings in Qwen3 variants, but one is an internal incorrect tying bug and the other is a user-facing LoRA warning; not one concrete issue." }, { "left": "issue:43329", "right": "issue:45600", "accept": false, - "reason": "Multimodal token counting video-path bug is unrelated to auto_mappings referencing removed configs." + "reason": "Undefined video-token helper variables and removed Sam3LiteText mappings are unrelated regressions despite both affecting multimodal internals." }, { "left": "issue:44589", "right": "issue:45042", "accept": false, - "reason": "Float8 storage lookup failure is unrelated to PIL backend requiring torchvision." + "reason": "Missing Float8 storage support and PIL backend requiring torchvision are unrelated backend/environment failures." }, { "left": "issue:43742", "right": "issue:45020", "accept": false, - "reason": "A specific MobileLLM load KeyError is not the same as generic remote_code breakage." + "reason": "MobileLLM key error and remote_code breakage are different loading problems affecting different code paths." }, { - "left": "issue:43550", - "right": "issue:44912", + "left": "issue:45084", + "right": "issue:45538", "accept": false, - "reason": "torch.compile + SDPA failure is unrelated to MXFP4 quantization loading behavior." + "reason": "Template compilation failure and CLIPTokenizer model_max_length are unrelated tokenizer/template issues." }, { "left": "issue:43526", "right": "issue:44554", "accept": false, - "reason": "Different bugs: label reduction in an image processor vs MPS attention correctness." + "reason": "A label reduction bug in BeitImageProcessorFast is unrelated to the MPS attention correctness issue." }, { - "left": "issue:45084", - "right": "issue:45538", + "left": "issue:43550", + "right": "issue:44912", "accept": false, - "reason": "Template compilation error is unrelated to CLIPTokenizer max length." + "reason": "torch.compile SDPA failure in Bamba and MXFP4 quantization fallback in git-oss-20b are different model/runtime bugs." }, { "left": "issue:43906", "right": "issue:46209", "accept": false, - "reason": "A reproduction of an older issue is not the same as FP8 inference failure on B200." + "reason": "An isolated reproduction of issue 38071 is not the same as FP8 inference failing on B200 GPUs." }, { "left": "issue:43452", "right": "issue:44991", "accept": false, - "reason": "Both are loading problems, but one is gguf_file handling and the other is a specific tokenizer regression." + "reason": "GGUF file handling breakage and tokenizer loading failure for a specific model are separate loading regressions." }, { "left": "issue:43638", "right": "issue:44936", "accept": false, - "reason": "Zero3/Bert index error is unrelated to evaluate-after-train failure." + "reason": "DeepSpeed ZeRO-3 index error during training and trainer.evaluate() failing after train are distinct trainer/runtime issues." }, { "left": "issue:43257", "right": "issue:43329", "accept": false, - "reason": "Qwen3 MOE weight conversion/loading is unrelated to multimodal token counting." + "reason": "Qwen3 MOE weight conversion under accelerate+deepspeed is unrelated to the multimodal token counting video-branch bug." }, { "left": "issue:43819", "right": "issue:43927", "accept": false, - "reason": "DAC latent/STE mismatch is unrelated to DiaConfig losing custom token IDs." + "reason": "DAC from_latents mismatch and DiaConfig custom token IDs lost after save/load are different serialization/inference bugs." }, { "left": "issue:36010", "right": "issue:43645", "accept": false, - "reason": "ImportError for GenerationMixin is unrelated to Jupyter custom model initialization regression." + "reason": "GenerationMixin import failure and Jupyter custom model initialization breakage do not share the same failure mode." }, { "left": "issue:36010", "right": "issue:36331", "accept": false, - "reason": "Import path breakage is unrelated to compute_loss signature change." + "reason": "GenerationMixin import failure is unrelated to CustomTrainer.compute_loss receiving an unexpected keyword argument." }, { "left": "issue:43720", "right": "issue:44610", "accept": false, - "reason": "Packed-weight loading in BitNet is unrelated to OmDet-Turbo image-size mismatch." + "reason": "BitNet packed-weight unpacking during accelerate loading and OmDet-Turbo processor size mismatch are unrelated defects." }, { "left": "issue:43531", "right": "issue:45042", "accept": false, - "reason": "Qwen3-MoE sliding_window issue is unrelated to PIL backend torchvision dependency." - }, - { - "left": "issue:38175", - "right": "issue:44610", - "accept": false, - "reason": "SigLIP2 zero-probability behavior is unrelated to OmDet-Turbo preprocessing size mismatch." + "reason": "Qwen3-MoE sliding_window behavior and PIL backend torchvision dependency are different issues in different subsystems." }, { "left": "issue:42175", "right": "issue:46291", "accept": false, - "reason": "TensorFlow extra packaging issue is unrelated to kernels v0.15 import requirements." + "reason": "Missing tensorflow in the torch extra and kernels v0.15 import-time version/revision validation are unrelated packaging/import failures." }, { - "left": "issue:39692", - "right": "issue:44912", + "left": "issue:38175", + "right": "issue:44610", "accept": false, - "reason": "Docs/example errors for SigLIP2 are unrelated to git-oss-20b MXFP4 loading." + "reason": "Unexpected zero probabilities in SigLIP2 and OmDet-Turbo input-size mismatch are unrelated model/processor bugs." }, { - "left": "issue:43746", - "right": "issue:44811", + "left": "issue:39692", + "right": "issue:44912", "accept": false, - "reason": "PEFT checkpoint loading for GraniteSpeech is unrelated to Whisper batch_decode special-token handling." + "reason": "Broken SigLIP2 docs example and MXFP4 quantization fallback are different problems; one is documentation/example correctness, the other runtime loading." }, { "left": "issue:44387", "right": "issue:44617", "accept": false, - "reason": "Both mention OOM, but the causes are different: quantization memory regression vs model-specific video OOM." + "reason": "Both mention memory, but int4 quantization reserved-memory growth and Sam3Video CUDA OOM are not the same underlying bug." }, { - "left": "issue:42490", - "right": "issue:43257", - "accept": false, - "reason": "Local-path slash import warning is unrelated to Qwen3 MOE weight conversion." - }, - { - "left": "issue:43854", - "right": "issue:44977", + "left": "issue:43746", + "right": "issue:44811", "accept": false, - "reason": "GLM-4.7 loading issue is unrelated to Qwen3.5 flash-attention generation behavior." + "reason": "PEFT local-checkpoint loading for GraniteSpeech and Whisper batch_decode ignoring skip_special_tokens are unrelated issues." }, { - "left": "issue:43906", - "right": "issue:46209", + "left": "issue:42490", + "right": "issue:43257", "accept": false, - "reason": "These are not the same bug; one is an isolated reproduction, the other is an FP8 hardware-specific inference failure." + "reason": "Local-path trailing-slash import warning and Qwen3 MOE weight conversion are unrelated." }, { - "left": "issue:45020", + "left": "issue:43854", "right": "issue:44977", "accept": false, - "reason": "Generic remote_code breakage is unrelated to Qwen3.5 flash-attention generation." - }, - { - "left": "issue:43262", - "right": "issue:44811", - "accept": false, - "reason": "Audio chat-template sampling-rate default is unrelated to Whisper batch decode skip_special_tokens." + "reason": "GLM-4.7-Flash loading in tests and Qwen3.5 flash-attention generation are separate model-specific regressions." } ] }, @@ -6287,9 +6252,9 @@ "issue:34567", "issue:34689", "issue:40990", - "issue:42175", "issue:42915", "issue:43257", + "issue:43262", "issue:43329", "issue:43404", "issue:43425", @@ -6304,7 +6269,6 @@ "issue:43697", "issue:43742", "issue:43749", - "issue:43824", "issue:43906", "issue:43927", "issue:43931", @@ -6312,6 +6276,7 @@ "issue:44479", "issue:44568", "issue:44756", + "issue:44811", "issue:44821", "issue:44912", "issue:44987", @@ -6331,213 +6296,231 @@ "issue:46424" ], "soft_pairs": [ + "issue:43262|issue:44811", "issue:43643|issue:45759", "issue:43257|issue:43696", - "issue:45459|issue:45803", "issue:46421|issue:46424", + "issue:45459|issue:45803", "issue:43329|issue:44479", "issue:43650|issue:43906", "issue:43257|issue:43931", "issue:43329|issue:43450", "issue:42915|issue:43742", - "issue:43749|issue:44756", "issue:43606|issue:45859", + "issue:43749|issue:44756", "issue:34567|issue:40990", "issue:44291|issue:45042", "issue:43454|issue:43577", "issue:42915|issue:44821", "issue:29159|issue:34567", - "issue:43927|issue:44912", "issue:44568|issue:45356", + "issue:43927|issue:44912", "issue:43525|issue:46291", "issue:44987|issue:45538", "issue:43697|issue:45357", "issue:43404|issue:45127", "issue:28282|issue:34689", "issue:42915|issue:45538", - "issue:45137|issue:45405", - "issue:33357|issue:34567", "issue:43425|issue:46291", - "issue:42175|issue:43824" + "issue:45137|issue:45405", + "issue:33357|issue:34567" ], - "prepared_review_unit_hash": "51133d500aec757b4d7591da794b2c97926d9e04216fd9251b5482f2c0cb0895", + "prepared_review_unit_hash": "1dbb12db71b1f3ec8707869dd24b76d4484c11218e6c2ef5ba5f0fc9d47ca03f", "component_budget": { "node_count": 473, "item_count": 473, "soft_pair_count": 1738, - "serialized_chars": 901383, - "estimated_input_tokens": 225346, - "estimated_eval_tokens": 450948 + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 28, - "serialized_chars": 36227, - "estimated_input_tokens": 9057, - "estimated_eval_tokens": 18370 + "serialized_chars": 36276, + "estimated_input_tokens": 9069, + "estimated_eval_tokens": 18394 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a loose embedding grouping of unrelated Transformers issues. The proposed soft pairs share broad vocabulary at best, but not the same underlying bug or change, so they should all be kept separate.", - "confidence": 0.18, - "canonical_issue_reason": "issue:43257 is the most concrete and representative of the cluster\u2019s recurring theme of model-loading/weight-conversion failures, with a specific reproducible path (Qwen3 MoE + accelerate/deepspeed) rather than a vague symptom.", + "summary": "This cluster is highly heterogeneous: tokenizer, multimodal, cache, loading, training, and offload issues are all mixed together. No soft pair appears to describe the same underlying bug or change.", + "confidence": 0.87, + "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "issue:43257 has the clearest, most actionable failure mode and the strongest fit as a representative issue among the mostly unrelated model-loading regressions in this set.", + "best_issue_reason": "issue:46424 is the most suitable representative here: it is open, concrete, and already has multiple inbound references, so it is the clearest actionable issue in an otherwise mixed set.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:43262", + "right": "issue:44811", + "accept": false, + "reason": "Both are audio-related, but they target different APIs and bugs: chat template sampling-rate defaulting vs batch_decode skip_special_tokens handling." + }, { "left": "issue:43643", "right": "issue:45759", "accept": false, - "reason": "Both concern configuration/model construction, but they hit different code paths: one is trust_remote_code field loss in AutoConfig, the other is unwrapping text_config in AutoModelForCausalLM.from_config for composite Qwen configs." + "reason": "Both involve config/model loading, but one is about missing fields with trust_remote_code and the other about not unwrapping text_config in from_config." }, { "left": "issue:43257", "right": "issue:43696", "accept": false, - "reason": "Different bugs and symptoms: Qwen3 MoE weight conversion/loading vs a GPT-oss CUDA OOM. No shared underlying defect." + "reason": "Qwen3 MoE weight conversion with accelerate/deepspeed is unrelated to a GPT-oss CUDA OOM failure." }, { - "left": "issue:45459", - "right": "issue:45803", + "left": "issue:46421", + "right": "issue:46424", "accept": false, - "reason": "Both involve exception handling, but the failures are unrelated: tokenizer decode-error masking vs bare except swallowing KeyboardInterrupt in Fuyu tensor conversion." + "reason": "Both mention cache, but one is about spurious recompilation from lazy init and the other about max_cache_length only applying to Quantized Cache." }, { - "left": "issue:46421", - "right": "issue:46424", + "left": "issue:45459", + "right": "issue:45803", "accept": false, - "reason": "Same cache subsystem, but different bugs: lazy initialization causing recompilation vs max_cache_length being ignored except for Quantized Cache." + "reason": "Both are exception-handling complaints, but they affect different tokenizer/processor code paths and different failure modes." }, { "left": "issue:43329", "right": "issue:44479", "accept": false, - "reason": "Both mention video VLM behavior, but one is an undefined-variable bug in _get_num_multimodal_tokens and the other is a broader regression across several models; not the same concrete fix." + "reason": "Both mention video regressions, but one is a specific undefined-variable bug in token counting and the other is a broader multi-model v5.3 regression." }, { "left": "issue:43650", "right": "issue:43906", "accept": false, - "reason": "No meaningful overlap; these are unrelated issue reports with no shared code path or failure mode." + "reason": "No shared bug is evident; one title is a placeholder and the other is an isolated reproduction of a different issue." }, { "left": "issue:43257", "right": "issue:43931", "accept": false, - "reason": "Both are Qwen-related loading issues, but one is MoE weight conversion with accelerate/deepspeed while the other is a shape mismatch for Qwen3-VL-30B. Different defects." + "reason": "Both are Qwen3-related load problems, but one is about MoE weight conversion and the other is a shape mismatch for Qwen3-VL." }, { "left": "issue:43329", "right": "issue:43450", "accept": false, - "reason": "Different video-processing failures: undefined multimodal token helper variables vs batched video processor shape handling." + "reason": "Different video bugs: undefined variables in multimodal token counting versus incorrect batched video processor output shapes." }, { "left": "issue:42915", "right": "issue:43742", "accept": false, - "reason": "Different models and failures: FineGrainedFP8Config support for Qwen3Moe vs a key error loading MobileLLM-125M." + "reason": "Different loading failures with different models and causes: FineGrainedFP8Config vs a KeyError on MobileLLM-125M." }, { - "left": "issue:43749", - "right": "issue:44756", + "left": "issue:43606", + "right": "issue:45859", "accept": false, - "reason": "Both are performance/memory-related, but one is broken FSDP CPU RAM efficient loading and the other is mmap-triggered OOM on Strix Halo. Not the same bug." + "reason": "Both involve CPU offload, but they are distinct model-specific incompatibilities with different underlying code paths." }, { - "left": "issue:43606", - "right": "issue:45859", + "left": "issue:43749", + "right": "issue:44756", "accept": false, - "reason": "Both mention CPU offload, but they are distinct model-specific bugs with different failing components and symptoms." + "reason": "One is FSDP CPU RAM-efficient loading being broken; the other is a Strix Halo mmap/OOM workaround. Not the same bug." }, { "left": "issue:34567", "right": "issue:40990", "accept": false, - "reason": "Unrelated issues: TrainerState token counting vs high perplexity on GPT-oss-20b." + "reason": "Training-state token counting is unrelated to perplexity being high on gpt-oss-20b." }, { "left": "issue:44291", "right": "issue:45042", "accept": false, - "reason": "Completely different areas: init_empty_weights argument handling vs image processor backend dependency requirements." + "reason": "Different problems entirely: init_empty_weights argument handling vs image processor torchvision dependency requirements." }, { "left": "issue:43454", "right": "issue:43577", "accept": false, - "reason": "Different subsystems and failures: tied lm_head weights in AyaVision vs dtype propagation in Blip2." + "reason": "Different model-loading bugs: untied lm_head weights in AyaVision versus dtype propagation in Blip2." }, { "left": "issue:42915", "right": "issue:44821", "accept": false, - "reason": "No shared underlying bug; Qwen3Moe FP8 loading and AutoImageProcessor URL loading are unrelated." + "reason": "No underlying bug match: Qwen3 MoE FP8 config failure is unrelated to loading an AutoImageProcessor from URL." }, { "left": "issue:29159", "right": "issue:34567", "accept": false, - "reason": "Tokenizer inconsistency vs training token-count bookkeeping; not the same problem." + "reason": "Tokenizer slow/fast inconsistency is unrelated to TrainerState token-count updating." }, { - "left": "issue:43927", - "right": "issue:44912", + "left": "issue:44568", + "right": "issue:45356", "accept": false, - "reason": "Different bugs: DiaConfig save/load losing token IDs vs MXFP4 quantization falling back to bf16." + "reason": "Both are tokenizer regressions, but they affect different models and behaviors; no evidence they share one fix." }, { - "left": "issue:44568", - "right": "issue:45356", + "left": "issue:43927", + "right": "issue:44912", "accept": false, - "reason": "Both are tokenizer regressions, but they affect different tokenizers and failure modes, so they should not be merged." + "reason": "DiaConfig save/load token IDs and MXFP4 quantization fallback are unrelated load-path bugs." }, { "left": "issue:43525", "right": "issue:46291", "accept": false, - "reason": "Unrelated issues: missing pad_token_id on Llama4Config vs kernels dependency requirements during transformers.pipelines import." + "reason": "A missing pad_token_id on Llama4Config is unrelated to the kernels v0.15 import failure in transformers.pipelines." }, { "left": "issue:44987", "right": "issue:45538", "accept": false, - "reason": "Different tokenizer/model-loading problems with no common code-path: loading physical-intelligence/fast vs CLIPTokenizer max_length handling." + "reason": "Different tokenizer issues: loading physical-intelligence/fast under v5.1.0 versus CLIPTokenizer's absurd max length." }, { "left": "issue:43697", "right": "issue:45357", "accept": false, - "reason": "One is an RTDetrV2 output regression, the other is incorrect visual encoder keys during Qwen3.5 save_pretrained. Different defects." + "reason": "Different regressions: RTDetrV2 output changes in v5 versus Qwen3.5 save_pretrained key mapping." }, { "left": "issue:43404", "right": "issue:45127", "accept": false, - "reason": "Both involve weight tying, but one is a missing tie in Mistral3 multimodal generation while the other is LoRA merge collapse with extended vocab. Not the same fix." + "reason": "Both involve weight tying in a broad sense, but they are different models and different bugs: tie failure vs LoRA merge collapse." }, { "left": "issue:28282", "right": "issue:34689", "accept": false, - "reason": "Different failure modes: missing PyTorch import vs a specific model-loading regression for Llama 3.2 Vision." + "reason": "ImportError from missing PyTorch is unrelated to a model-loading regression in Transformers 4.46.2." + }, + { + "left": "issue:42915", + "right": "issue:45538", + "accept": false, + "reason": "No shared underlying issue: Qwen3 FP8 config failure versus CLIPTokenizer max-length behavior." + }, + { + "left": "issue:43425", + "right": "issue:46291", + "accept": false, + "reason": "Torch version incompatibility is not the same as the kernels v0.15 package requirement failure." }, { "left": "issue:45137", "right": "issue:45405", "accept": false, - "reason": "Completely unrelated: DeepSpeed ZeRO3 deque error vs a version pin bump for PEFT." + "reason": "DeepSpeed ZeRO3 deque error is unrelated to the PEFT version bump/release issue." }, { "left": "issue:33357", "right": "issue:34567", "accept": false, - "reason": "Bus error loading a CLIP model on macOS vs TrainerState token counting not updating; no shared bug." + "reason": "MacOS bus error with CLIP is unrelated to TrainerState token-count not updating." } ] }, @@ -6582,6 +6565,7 @@ "issue:43742", "issue:43749", "issue:43756", + "issue:43824", "issue:43994", "issue:44062", "issue:44291", @@ -6607,10 +6591,10 @@ "issue:45907", "issue:46399", "issue:46463", - "issue:46491", "issue:46512" ], "soft_pairs": [ + "issue:42175|issue:43824", "issue:43577|issue:46512", "issue:43452|issue:43582", "issue:28282|issue:29127", @@ -6618,238 +6602,234 @@ "issue:44062|issue:44991", "issue:43317|issue:45538", "issue:44849|issue:45701", - "issue:42915|issue:43749", "issue:41720|issue:43540", + "issue:42915|issue:43749", "issue:43540|issue:44560", "issue:44479|issue:45381", "issue:43540|issue:44479", "issue:24643|issue:29159", - "issue:38175|issue:44898", "issue:44568|issue:45488", - "issue:27957|issue:28282", + "issue:38175|issue:44898", "issue:44987|issue:45907", + "issue:27957|issue:28282", "issue:42175|issue:44488", "issue:43493|issue:43994", "issue:43257|issue:43756", - "issue:34567|issue:36331", "issue:44291|issue:44821", - "issue:44737|issue:46463", + "issue:34567|issue:36331", "issue:44792|issue:44964", + "issue:44737|issue:46463", "issue:44062|issue:45081", "issue:43577|issue:45072", "issue:43756|issue:46399", - "issue:43742|issue:44589", - "issue:44568|issue:46491" + "issue:43742|issue:44589" ], - "prepared_review_unit_hash": "52033fc88088bb27c47cb078fb91f48f453bbfcd169ccd80d33106d1c6b7c3a3", + "prepared_review_unit_hash": "486907808cbbc0762b0b718fc7b3b3fe3e9cbcb34e251cd78d2606654bdc8f16", "component_budget": { "node_count": 473, "item_count": 473, "soft_pair_count": 1738, - "serialized_chars": 901383, - "estimated_input_tokens": 225346, - "estimated_eval_tokens": 450948 + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 29, - "serialized_chars": 36520, - "estimated_input_tokens": 9130, - "estimated_eval_tokens": 18516 + "serialized_chars": 36522, + "estimated_input_tokens": 9131, + "estimated_eval_tokens": 18518 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Heterogeneous issue cluster spanning tokenizer regressions, model-loading failures, dtype/config mismatches, and multimodal/video bugs; no PRs were provided. The soft pairs look like topical similarity rather than same underlying defects, so none should be merged as duplicates.", - "confidence": 0.38, - "canonical_issue_reason": "issue:45488 is the strongest concrete anchor: it describes a specific, high-impact tokenizer regression with clear reproduction and broader impact than the other candidates.", + "summary": "Mostly unrelated transformer issues, with two clear duplicate-like pairs: one around Qwen video-input regressions and one around SigLIP2 discrepancy. The rest are too different to merge.", + "confidence": 0.91, + "canonical_issue_reason": "issue:45488 is the strongest standalone report in this set: high discussion, inbound references, and a clear, user-facing regression in core tokenizer behavior affecting a widely used model family.", "canonical_pr_reason": null, - "best_issue_reason": "issue:45488 is the most useful representative issue in this set because it is specific, actionable, and closest to a reusable regression pattern; the rest are largely unrelated one-off bugs.", + "best_issue_reason": "issue:45488 has the best mix of activity, impact, and diagnostic value, making it the best anchor issue for triage even though the cluster is not a single duplicate group.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:42175", + "right": "issue:43824", + "accept": false, + "reason": "Different problems: backend extras packaging vs a missing model class import." + }, { "left": "issue:43577", "right": "issue:46512", "accept": false, - "reason": "Both are dtype-related, but one is Blip2 loading behavior while the other is from_config leaving config.dtype stale; different code paths and failures." + "reason": "Both involve dtype handling, but one is a BLIP2 loading bug and the other is stale config metadata from from_config; not the same code-path bug." }, { "left": "issue:43452", "right": "issue:43582", "accept": false, - "reason": "Tokenizer/AutoModel loading via gguf_file is unrelated to an Apple Silicon caching_allocator_warmup TypeError." + "reason": "Unrelated: GGUF loading breakage vs Apple Silicon allocator warmup TypeError." }, { "left": "issue:28282", "right": "issue:29127", "accept": false, - "reason": "AutoModel missing PyTorch is a backend/import problem, while LayoutLMv3 is about clearer validation errors for bad boxes." + "reason": "Completely different: missing PyTorch import vs LayoutLMv3 box-validation error messaging." }, { "left": "issue:43742", "right": "issue:44291", "accept": false, - "reason": "MobileLLM key errors and init_empty_weights argument mismatches are separate loading failures with different triggers." + "reason": "Different loader failures with different causes: model config/key loading vs init_empty_weights argument handling." }, { "left": "issue:44062", "right": "issue:44991", "accept": false, - "reason": "AddedToken duplicate-special-argument error is a token construction bug; EMBEDDIA tokenizer loading is a different model-specific regression." + "reason": "Both are tokenizer-load failures, but they point to different regressions and different model-specific paths." }, { "left": "issue:43317", "right": "issue:45538", "accept": false, - "reason": "device_map/offload dequantized-model loading is unrelated to CLIPTokenizer's oversized model_max_length." + "reason": "Device-map/offload loading bug vs CLIP tokenizer max_length misconfiguration." }, { "left": "issue:44849", "right": "issue:45701", "accept": false, - "reason": "Qwen3.5 hidden-state output bug and general tokenization-version changes are not the same code-path or symptom." + "reason": "Different issues: Qwen3.5 hidden-state handling vs a broader tokenization-change complaint." }, { - "left": "issue:42915", - "right": "issue:43749", + "left": "issue:41720", + "right": "issue:43540", "accept": false, - "reason": "Qwen3MoE FP8 config failure is model/config specific; FSDP CPU RAM efficient loading is a separate training/loading feature bug." + "reason": "Both mention Qwen models, but one is auto device mapping cuda assert and the other is video input processing failure." }, { - "left": "issue:41720", - "right": "issue:43540", + "left": "issue:42915", + "right": "issue:43749", "accept": false, - "reason": "Qwen3 auto device-map cuda assert on A800 differs from Qwen3OmniMoe video input processing error." + "reason": "Distinct training/runtime paths: Qwen3 MoE FP8 config vs FSDP CPU RAM efficient loading." }, { "left": "issue:43540", "right": "issue:44560", "accept": false, - "reason": "Both involve Qwen3 video, but one is input processing failure and the other is a StopIteration in embedding/video handling; not clearly the same bug." + "reason": "Both are video-input failures, but they affect different Qwen model families and appear to be separate regressions." }, { "left": "issue:44479", "right": "issue:45381", - "accept": false, - "reason": "These are both video regressions, but one spans multiple Qwen vl families and the other is specifically wrong vision_position_ids in qwen2.5-vl." + "accept": true, + "reason": "These describe the same Qwen/VL video-input regression; 45381 is a specific manifestation of the broader regression described in 44479." }, { - "left": "issue:43540", - "right": "issue:44479", + "left": "issue:24643", + "right": "issue:29159", "accept": false, - "reason": "Qwen3OmniMoe video processing and the broader v5.3.0 video regression are related only at subsystem level, not the same defect." + "reason": "Training DeepSpeed weight-shape error is unrelated to tokenizer inconsistency." }, { - "left": "issue:24643", - "right": "issue:29159", + "left": "issue:44568", + "right": "issue:45488", "accept": false, - "reason": "DeepSpeed weight-shape RuntimeError and tokenizer slow/fast inconsistency are completely different classes of bugs." + "reason": "Both are tokenizer regressions, but they affect different tokenizers and different behavior; not one concrete bug." }, { "left": "issue:38175", "right": "issue:44898", "accept": false, - "reason": "SigLIP2 zero probabilities and Perceiver image-classification resize/interpolation failure affect different models and mechanisms." + "reason": "Different vision-model failures with different symptoms and model families." }, { - "left": "issue:44568", - "right": "issue:45488", + "left": "issue:44987", + "right": "issue:45907", "accept": false, - "reason": "mdeberta add_special_tokens not adding BOS/EOS and LlamaTokenizer overriding ByteLevel with Metaspace are both tokenizer regressions, but not the same model or implementation bug." + "reason": "One is a model-loading failure, the other is a repo-files API forwarding bug." }, { "left": "issue:27957", "right": "issue:28282", "accept": false, - "reason": "XLMRoberta Flash Attention 2 and missing PyTorch import are unrelated model/backend issues." - }, - { - "left": "issue:44987", - "right": "issue:45907", - "accept": false, - "reason": "Loading fast model on transformers>=5.1.0 and list_repo_files not forwarding revision are different API regressions with no shared code-path." + "reason": "Unrelated: Flash Attention 2 on XLM-Roberta vs missing PyTorch import." }, { "left": "issue:42175", "right": "issue:44488", "accept": false, - "reason": "TensorFlow backend packaging and failure to load cjvt/sleng-bert are separate concerns." + "reason": "Packaging/backend dependency issue vs a specific model-loading problem." }, { "left": "issue:43493", "right": "issue:43994", - "accept": false, - "reason": "SigLIP2 discrepancy vs JAX is an implementation parity issue; nonsensical AutoModel/pipeline output is broader and not clearly the same bug." + "accept": true, + "reason": "Same underlying SigLIP2 implementation discrepancy: one reports mismatch with JAX, the other shows incorrect outputs in HF AutoModel/pipeline." }, { "left": "issue:43257", "right": "issue:43756", "accept": false, - "reason": "Qwen3 MoE loading with accelerate+deepspeed and Smollm3 RoPE-layer count mismatch are different model-specific defects." - }, - { - "left": "issue:34567", - "right": "issue:36331", - "accept": false, - "reason": "TrainerState token counting and CustomTrainer compute_loss signature mismatch are unrelated trainer API issues." + "reason": "Different model/config bugs: Qwen3 MoE weight conversion vs SmolLM3 RoPE layer drop." }, { "left": "issue:44291", "right": "issue:44821", "accept": false, - "reason": "init_empty_weights unexpected argument and AutoImageProcessor URL loading are separate loading mechanisms." + "reason": "Init-empty-weights argument handling is unrelated to loading AutoImageProcessor from URL." }, { - "left": "issue:44737", - "right": "issue:46463", + "left": "issue:34567", + "right": "issue:36331", "accept": false, - "reason": "XLNet CPU arange placement bug and T5 decoder-input validation messaging are unrelated." + "reason": "TrainerState token counting and CustomTrainer compute_loss signature are separate trainer API issues." }, { "left": "issue:44792", "right": "issue:44964", "accept": false, - "reason": "Janus image-generation test failure and Phi-4 multimodal loading failure are different multimodal problems with different symptoms." + "reason": "Both involve multimodal models, but the failures and code paths are different." + }, + { + "left": "issue:44737", + "right": "issue:46463", + "accept": false, + "reason": "XLNet CPU device placement bug is unrelated to T5 decoder-input validation messaging." }, { "left": "issue:44062", "right": "issue:45081", "accept": false, - "reason": "AddedToken special-argument conflict and Mistral regex patch crash are distinct tokenizer internals bugs." + "reason": "Different tokenizer-regression surfaces and different failing code paths." }, { "left": "issue:43577", "right": "issue:45072", "accept": false, - "reason": "Blip2 dtype persistence and SwitchTransformers/TimmWrapperModel bfloat16 mismatches are different dtype problems in different models." + "reason": "Different dtype bugs in different model wrappers; not the same concrete defect." }, { "left": "issue:43756", "right": "issue:46399", "accept": false, - "reason": "Smollm3 RoPE-layer mismatch and Qwen3.5 apply_rotary_pos_emb registration error do not share the same failure mode." + "reason": "SmolLM3 RoPE-layer issue is unrelated to Qwen3.5 rotary-pos-embedding registration failure." }, { "left": "issue:43742", "right": "issue:44589", "accept": false, - "reason": "Key error loading MobileLLM-125M is unrelated to missing Float8 storage type handling." - }, - { - "left": "issue:44568", - "right": "issue:46491", - "accept": false, - "reason": "These are both tokenizer regressions, but one is special-token addition failure and the other is a leading-space encode/decode regression in CodeLlama." + "reason": "Different loader/runtime failures with no clear shared root cause." } ] }, - "evaluator_result": null, - "evaluator_used": false, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays mostly conservative. The two accepted pairs are reasonably supported by the titles and context, and the rejected pairs are generally distinct enough to avoid over-merging. The chosen anchor issue (45488) is defensible from the activity/reference signals. " + }, + "evaluator_used": true, "retried": false, - "accepted_nontrivial_soft_edge": false, + "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null }, @@ -6882,14 +6862,14 @@ "issue:43720", "issue:43749", "issue:43756", - "issue:43819", "issue:43906", "issue:44466", "issue:44488", + "issue:44568", "issue:44625", "issue:44737", - "issue:44811", "issue:44821", + "issue:44843", "issue:44877", "issue:44898", "issue:44913", @@ -6912,219 +6892,215 @@ "issue:45987", "issue:46121", "issue:46463", - "issue:46468" + "issue:46468", + "issue:46491" ], "soft_pairs": [ - "issue:43756|issue:45325", + "issue:44568|issue:46491", "issue:45042|issue:45081", + "issue:43756|issue:45325", "issue:43906|issue:46468", "issue:45823|issue:46463", "issue:44737|issue:45588", "issue:45405|issue:45600", - "issue:43526|issue:45479", - "issue:43404|issue:45987", "issue:43531|issue:44821", + "issue:43404|issue:45987", "issue:43452|issue:44488", + "issue:43526|issue:45479", "issue:45357|issue:45920", "issue:43906|issue:45381", "issue:44625|issue:44877", "issue:44913|issue:46121", "issue:43720|issue:44898", "issue:43697|issue:45216", - "issue:43643|issue:45375", "issue:24643|issue:29127", "issue:28282|issue:32090", + "issue:43643|issue:45375", "issue:45200|issue:46463", "issue:28282|issue:35141", "issue:41762|issue:44936", "issue:44466|issue:45005", - "issue:43819|issue:44811", - "issue:43688|issue:43749" + "issue:43688|issue:43749", + "issue:43452|issue:44843" ], - "prepared_review_unit_hash": "ff166f2499e11529f23d92be4cf9bf78ebd3a083a4cbf22dff9b27b42419c9d4", + "prepared_review_unit_hash": "195cf426f9986be599b98f0653aaacf0c5a651125b7e611a0a5d9828af07df3d", "component_budget": { "node_count": 473, "item_count": 473, "soft_pair_count": 1738, - "serialized_chars": 901383, - "estimated_input_tokens": 225346, - "estimated_eval_tokens": 450948 + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 }, "budget": { - "node_count": 47, - "item_count": 47, - "soft_pair_count": 25, - "serialized_chars": 35123, - "estimated_input_tokens": 8781, - "estimated_eval_tokens": 17818 + "node_count": 48, + "item_count": 48, + "soft_pair_count": 26, + "serialized_chars": 36107, + "estimated_input_tokens": 9027, + "estimated_eval_tokens": 18310 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Mostly heterogeneous issue set with no broad duplicate cluster. The only credible overlap is the tied-weights serialization/regression pair around `lm_head.weight` (44466/45005); the rest are distinct bugs across unrelated code paths.", - "confidence": 0.72, - "canonical_issue_reason": "issue:44466 is the clearest, most self-contained regression report in the set: it names the concrete symptom, the affected code path (`lm_head.weight` / tied weights serialization), and lines up with the only near-duplicate issue (45005).", + "summary": "Most pairs are clearly unrelated despite superficial keyword overlap. The only strong duplicate-like pair is the tied-weights v5 serialization reports (#44466 and #45005).", + "confidence": 0.81, + "canonical_issue_reason": "#45005 is the better canonical issue because it is broader and more representative of the tied-weights regression, while #44466 reads like a specific manifestation of the same problem (device-dependent lm_head.weight serialization).", "canonical_pr_reason": null, - "best_issue_reason": "issue:44466 is the best representative issue because it is specific, actionable, and tied to a concrete serialization bug rather than a broader umbrella report.", + "best_issue_reason": "#45005 is the best issue to keep as the primary tracker for the tied-weights bug: it describes the underlying regression at a higher level and can absorb narrower repros like #44466.", "best_pr_reason": null, "soft_edge_verdicts": [ { - "left": "issue:43756", - "right": "issue:45325", + "left": "issue:44568", + "right": "issue:46491", "accept": false, - "reason": "Both are positional/rope-related, but they describe different mechanisms and model families: Smollm3 layer dropping vs Qwen2.5-VL rope indexing. Not the same bug." + "reason": "Different models and different failure modes: tokenizer special-token insertion vs CodeLlama encode/decode spacing regression." }, { "left": "issue:45042", "right": "issue:45081", "accept": false, - "reason": "Different subsystems and failure modes: PIL image processors requiring torchvision vs a Mistral tokenizer regex patch crash." + "reason": "Unrelated bugs: PIL/torchvision dependency handling vs Mistral regex/tokenizer crash." + }, + { + "left": "issue:43756", + "right": "issue:45325", + "accept": false, + "reason": "Both mention RoPE, but they affect different models and different logic paths; not the same bug." }, { "left": "issue:43906", "right": "issue:46468", "accept": false, - "reason": "One is an isolated reproduction of issue 38071; the other is a broken Nemotron example. Different underlying problems." + "reason": "An isolated reproduction of one issue is not evidence of the same bug as the Nemotron example failure; the titles indicate different problems." }, { "left": "issue:45823", "right": "issue:46463", "accept": false, - "reason": "Gemma4 device-mismatch in forward pass vs T5 decoder-input validation error messaging. Different bugs." + "reason": "Gemma4 device-map/PLE mismatch is unrelated to T5 decoder input validation." }, { "left": "issue:44737", "right": "issue:45588", "accept": false, - "reason": "XLNet CPU device placement issue vs flash-attention crash on `s_aux=None`. Not the same code path." + "reason": "XLNet CPU arange bug and flash-attention s_aux=None crash are different code paths and symptoms." }, { "left": "issue:45405", "right": "issue:45600", "accept": false, - "reason": "Dependency version bump issue vs stale auto-mappings referencing removed configs. Different causes and fixes." + "reason": "PEFT version bump is packaging metadata; auto_mappings.py CI breakage is a removed-config reference bug." }, { - "left": "issue:43526", - "right": "issue:45479", + "left": "issue:43531", + "right": "issue:44821", "accept": false, - "reason": "BEiT image processor label reduction bug vs sequence-classification zero-loss bug. Unrelated." + "reason": "Qwen3-MoE sliding_window behavior and AutoImageProcessor URL loading are unrelated." }, { "left": "issue:43404", "right": "issue:45987", "accept": false, - "reason": "Mistral3 tied-weight regression vs `StaticCache.get_seq_length()` returning a tensor. Different functionality." - }, - { - "left": "issue:43531", - "right": "issue:44821", - "accept": false, - "reason": "Qwen3-MoE sliding-window behavior vs loading an AutoImageProcessor from URL. No concrete overlap." + "reason": "lm_head tying in Mistral3 and StaticCache.get_seq_length() returning a tensor are separate defects." }, { "left": "issue:43452", "right": "issue:44488", "accept": false, - "reason": "Both concern loading, but one is `gguf_file` handling for tokenizer/CausalLM and the other is a specific model load failure. Too different to treat as the same bug." + "reason": "Both concern loading, but one is gguf_file path handling and the other is a model/tokenizer load failure for a specific repo; no shared concrete bug." }, { - "left": "issue:45357", - "right": "issue:45920", + "left": "issue:43526", + "right": "issue:45479", "accept": false, - "reason": "Qwen3.5 save_pretrained visual-key serialization bug vs AutoTokenizer producing wrong token IDs across many models. Different layers and symptoms." + "reason": "BEiT reduce_labels returning one label is unrelated to zero loss with num_labels=1." }, { - "left": "issue:43906", - "right": "issue:45381", + "left": "issue:45357", + "right": "issue:45920", "accept": false, - "reason": "Isolated reproduction of a previous issue vs Qwen2.5-VL video `vision_position_ids` regression. Not the same defect." + "reason": "Qwen3.5 save_pretrained key serialization and AutoTokenizer wrong token IDs are different regressions." }, { "left": "issue:44625", "right": "issue:44877", "accept": false, - "reason": "Qwen3.5 `num_labels` propagation bug vs strict config blocking `granite_speech` loading. Different config-handling issues." + "reason": "Both involve config loading, but one is num_labels propagation and the other is strict config rejection for granite_speech; not the same bug." }, { "left": "issue:44913", "right": "issue:46121", "accept": false, - "reason": "GPTNeoX `rotary_pct` reload regression vs `convert_rope_params_to_dict` TypeError on JSON-loaded lists. Related theme, but not the same bug." + "reason": "GPTNeoX rotary_pct resetting on reload and rope-validation TypeError for JSON-loaded lists are distinct configuration issues." }, { "left": "issue:43720", "right": "issue:44898", "accept": false, - "reason": "BitNet packed-weight loading failure vs Perceiver interpolation failure. Different model families and code paths." + "reason": "BitNet packed-weight unpacking during accelerate loading is unrelated to Perceiver interpolation with non-default resolution." }, { "left": "issue:43697", "right": "issue:45216", "accept": false, - "reason": "RTDetrV2 output drift regression vs Qwen3.5 checkpoint save_pretrained regression. Separate issues." - }, - { - "left": "issue:43643", - "right": "issue:45375", - "accept": false, - "reason": "Generic `trust_remote_code` missing-fields issue vs a specific Qwen3_5MoeVisionConfig field dropped by strict config handling. Similar theme, but not the same underlying bug." + "reason": "RTDetrV2 output regression and Qwen3.5 save_pretrained checkpoint corruption are different problems." }, { "left": "issue:24643", "right": "issue:29127", "accept": false, - "reason": "DeepSpeed training `weight` dimensionality error vs LayoutLMv3 box-information error message quality. Completely different." + "reason": "DeepSpeed weight shape runtime error and LayoutLMv3 error-message clarity are unrelated." }, { "left": "issue:28282", "right": "issue:32090", "accept": false, - "reason": "Missing PyTorch import dependency vs Trainer broadcast of `NoneType`. Different failure modes." + "reason": "PyTorch-missing ImportError and Trainer _gpu_broadcast_one NoneType crash are unrelated." }, { - "left": "issue:45200", - "right": "issue:46463", + "left": "issue:43643", + "right": "issue:45375", "accept": false, - "reason": "Gemma 4 missing `mm_token_type_ids` default vs T5 decoder-input validation messaging. Not the same bug." + "reason": "Missing fields with trust_remote_code and strict-dropping deepstack_visual_indexes are both config-related, but they are not the same concrete bug." }, { - "left": "issue:28282", - "right": "issue:35141", + "left": "issue:45200", + "right": "issue:46463", "accept": false, - "reason": "ImportError for missing PyTorch vs token-embedding resize causing output embedding reinit. Unrelated." + "reason": "Gemma 4 multimodal token defaults and T5 decoder input validation are unrelated." }, { "left": "issue:41762", "right": "issue:44936", "accept": false, - "reason": "Gemma3 ZeRO-3 loading crash vs Trainer evaluate-after-train failure. Different problem areas." + "reason": "Gemma3 ZeRO-3 loading failure and trainer.evaluate() after trainer.train() are separate bugs." }, { "left": "issue:44466", "right": "issue:45005", "accept": true, - "reason": "Both describe tied-weights regressions in v5 around `lm_head`/serialization behavior; they look like the same underlying save/load code-path problem and could plausibly be fixed together." + "reason": "Both describe the same underlying tied-weights serialization regression in v5: inconsistent lm_head.weight/save-pretrained behavior for translation models, with #44466 being a specific manifestation." }, { - "left": "issue:43819", - "right": "issue:44811", + "left": "issue:43688", + "right": "issue:43749", "accept": false, - "reason": "DAC latent reconstruction mismatch vs Whisper `batch_decode()` ignoring `skip_special_tokens`. Different APIs and bugs." + "reason": "Aux-loss normalization in MoE models and broken FSDP CPU RAM efficient loading are unrelated." }, { - "left": "issue:43688", - "right": "issue:43749", + "left": "issue:43452", + "right": "issue:44843", "accept": false, - "reason": "Aux-loss normalization bug in OLMoE/GPT Oss vs FSDP CPU RAM efficient loading breakage. No direct overlap." + "reason": "gguf_file loading failure and the offline mistral_regex/model_info bug are different tokenizer/loading issues." } ] }, "evaluator_result": { "accept": true, - "feedback": "Grounded overall. The only supported soft-edge overlap is the tied-weights regression pair 44466/45005, and the rejections are phrased conservatively based on distinct titles/code paths. No clear overstatement from the packet." + "feedback": "The primary/canonical issue choice is grounded in the packet: #45005 is broader and #44466 is a narrower tied-weights serialization regression. The soft-edge rejections are conservative and stay within what the titles support, with no obvious overstatement of evidence." }, "evaluator_used": true, "retried": false, @@ -7161,6 +7137,7 @@ "issue:43742", "issue:43749", "issue:43756", + "issue:43819", "issue:43927", "issue:44361", "issue:44442", @@ -7172,8 +7149,8 @@ "issue:44568", "issue:44625", "issue:44704", - "issue:44734", "issue:44743", + "issue:44811", "issue:44843", "issue:44936", "issue:44991", @@ -7191,242 +7168,234 @@ "issue:46170", "issue:46178", "issue:46402", - "issue:46491", - "issue:46612" + "issue:46491" ], "soft_pairs": [ + "issue:43819|issue:44811", "issue:44493|issue:46050", - "issue:43452|issue:44843", "issue:44361|issue:44442", - "issue:38175|issue:43295", "issue:43653|issue:44361", + "issue:38175|issue:43295", "issue:45372|issue:46170", "issue:44936|issue:45823", "issue:42898|issue:46491", "issue:45542|issue:46178", "issue:43452|issue:43577", - "issue:44514|issue:45381", "issue:43295|issue:45356", + "issue:44514|issue:45381", "issue:43531|issue:43742", - "issue:43577|issue:43697", "issue:43653|issue:44568", + "issue:43577|issue:43697", "issue:43646|issue:43749", "issue:44843|issue:44991", - "issue:43688|issue:45704", "issue:29159|issue:33357", + "issue:43688|issue:45704", + "issue:43756|issue:46402", "issue:44464|issue:44743", "issue:43653|issue:43927", - "issue:43756|issue:46402", "issue:44484|issue:44704", "issue:43688|issue:45440", "issue:43688|issue:45701", "issue:44479|issue:44625", "issue:43550|issue:44361", - "issue:45491|issue:46002", - "issue:44734|issue:46612" + "issue:45491|issue:46002" ], - "prepared_review_unit_hash": "3c4814ed86df7b3362e484a1562fe01eef354995d15eda28bd1ee84e329112a0", + "prepared_review_unit_hash": "6fbdc79224d1e33431e1146c7031e1fc79e6f3b0bb6abfb8a3de1b6c1b49c28a", "component_budget": { "node_count": 473, "item_count": 473, "soft_pair_count": 1738, - "serialized_chars": 901383, - "estimated_input_tokens": 225346, - "estimated_eval_tokens": 450948 + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 }, "budget": { "node_count": 48, "item_count": 48, - "soft_pair_count": 29, - "serialized_chars": 36532, - "estimated_input_tokens": 9133, - "estimated_eval_tokens": 18522 + "soft_pair_count": 28, + "serialized_chars": 36190, + "estimated_input_tokens": 9048, + "estimated_eval_tokens": 18352 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is very broad: it mixes tokenizer regressions, model-loading failures, generation/cache bugs, backend/device issues, and model-specific numerical problems. I do not see any soft pair that is clearly the same underlying bug, so all candidate merges should be rejected. The most representative issue is the broad tokenizer-regression report, but even that is not a true duplicate of the rest.", - "confidence": 0.71, - "canonical_issue_reason": "Issue 45701 is the broadest tokenizer-regression report in the set and the most representative anchor, though the cluster is too heterogeneous to treat as one true duplicate family.", + "summary": "This set is mostly a grab bag of distinct tokenizer, model-loading, and generation regressions. I don\u2019t see any soft-edge pair that clearly describes the same underlying bug/change, so I would keep them separate.", + "confidence": 0.89, + "canonical_issue_reason": "issue:29159 is the broadest and most central representative of the tokenizer-behavior theme: it explicitly covers inconsistency between slow and fast tokenizer paths, which is the closest umbrella problem among the set.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 45701 is the best single representative because it captures a generic version-dependent tokenization change, which is the closest shared theme among several of these reports.", + "best_issue_reason": "issue:29159 best represents the cluster because it is generic, cross-model, and directly about tokenizer parity, unlike the more model-specific regressions.", "best_pr_reason": null, "soft_edge_verdicts": [ { - "left": "issue:44493", - "right": "issue:46050", + "left": "issue:43819", + "right": "issue:44811", "accept": false, - "reason": "Different failures: unexpected position-id keys vs quantized save/load corruption." + "reason": "Different subsystems: DAC latent decoding vs Whisper batch_decode skip_special_tokens behavior." }, { - "left": "issue:43452", - "right": "issue:44843", + "left": "issue:44493", + "right": "issue:46050", "accept": false, - "reason": "Both involve loading paths, but one is gguf handling and the other is an offline model_info call in tokenizer regex patching." + "reason": "Unrelated bugs: position-id key warnings in configs vs quantized model save/load corruption." }, { "left": "issue:44361", "right": "issue:44442", "accept": false, - "reason": "Different tokenizers and different error modes; no shared concrete code path." - }, - { - "left": "issue:38175", - "right": "issue:43295", - "accept": false, - "reason": "Unrelated models and symptoms: zero probabilities vs processor.tokenizer/image passthrough regression." + "reason": "Different tokenizer classes and failures: MLuke task AttributeError vs FastSpeech2ConformerTokenizer loading." }, { "left": "issue:43653", "right": "issue:44361", "accept": false, - "reason": "BigBird special-token registration bug is unrelated to MLukeTokenizer attribute errors." + "reason": "Both are tokenizer bugs, but one is BigBird special-token registration/decode and the other is MLuke task AttributeError; not the same concrete bug." }, { - "left": "issue:45372", - "right": "issue:46170", + "left": "issue:38175", + "right": "issue:43295", + "accept": false, + "reason": "Different problems: zero probabilities in a model vs processor/tokenizer regression for custom image inputs." + }, + { + "left": "issue:45372", + "right": "issue:46170", "accept": false, - "reason": "Different root causes: mistral_common import breakage vs Windows/ROCm torch.distributed dependency issues." + "reason": "Both are import/load failures, but the causes are different: mistral_common ReasoningEffort import vs Windows ROCm torch.distributed dependency issues." }, { "left": "issue:44936", "right": "issue:45823", "accept": false, - "reason": "Trainer lifecycle bug and Gemma4 device-map mismatch are different problems." + "reason": "Training loop evaluate-after-train failure is unrelated to Gemma4 device-map mismatches in forward." }, { "left": "issue:42898", "right": "issue:46491", "accept": false, - "reason": "Both are tokenization regressions, but one is a v5 cleanup behavior change and the other is a CodeLlama leading-space round-trip bug." + "reason": "Both involve tokenization behavior, but 42898 is a general v5 cleanup-space policy change while 46491 is a CodeLlama round-trip spacing regression." }, { "left": "issue:45542", "right": "issue:46178", "accept": false, - "reason": "Both concern backend availability, but they name different missing backends and likely different code paths." + "reason": "Different backend-registration issues: missing TensorFlow backend handling vs tensorflow_text backend mapping." }, { "left": "issue:43452", "right": "issue:43577", "accept": false, - "reason": "gguf loading failure is unrelated to BLIP2 dtype propagation." + "reason": "Unrelated: gguf loading breaks from_pretrained paths, while the other is dtype propagation for Blip2 submodules." }, { - "left": "issue:44514", - "right": "issue:45381", + "left": "issue:43295", + "right": "issue:45356", "accept": false, - "reason": "Both are Qwen2.5-VL issues, but one is batched chat-template padding and the other is video position IDs." + "reason": "Processor/image-tokenizer regression is not the same bug as Kimi-K2.5 codec handling and warning behavior." }, { - "left": "issue:43295", - "right": "issue:45356", + "left": "issue:44514", + "right": "issue:45381", "accept": false, - "reason": "Different regression symptoms and likely different fixes: processor/tokenizer passthrough vs Kimi-K2.5 codec/regex warning." + "reason": "Both are Qwen VL issues, but one is batched chat-template padding and the other is video vision_position_ids; different code paths." }, { "left": "issue:43531", "right": "issue:43742", "accept": false, - "reason": "Sliding-window behavior in Qwen3-MoE is unrelated to MobileLLM loading key errors." + "reason": "Different domains: Qwen3-MoE sliding_window logic vs MobileLLM-125M loading key errors." }, { - "left": "issue:43577", - "right": "issue:43697", + "left": "issue:43653", + "right": "issue:44568", "accept": false, - "reason": "Dtype propagation in BLIP2 and RTDetrV2 output divergence are separate bugs." + "reason": "Both concern special tokens, but BigBird mask-token registration and mDeBERTa BOS/EOS insertion are distinct tokenizer bugs." }, { - "left": "issue:43653", - "right": "issue:44568", + "left": "issue:43577", + "right": "issue:43697", "accept": false, - "reason": "Different tokenizer special-token issues: missing mask-token registration vs BOS/EOS not added." + "reason": "Different failures: Blip2 dtype loading vs RTDetrV2 output divergence in v5." }, { "left": "issue:43646", "right": "issue:43749", "accept": false, - "reason": "Custom model initialization breakage is not the same as FSDP CPU RAM efficient loading." + "reason": "Custom model initialization breakage is unrelated to FSDP CPU/RAM-efficient loading." }, { "left": "issue:44843", "right": "issue:44991", "accept": false, - "reason": "Both affect tokenizer loading, but one is offline model_info behavior and the other is an est-roberta loading failure." + "reason": "Both are tokenizer loading issues, but one is offline-mode model_info misuse and the other is a model-specific tokenizer load failure." + }, + { + "left": "issue:29159", + "right": "issue:33357", + "accept": false, + "reason": "Generic tokenizer parity issue vs Mac bus error on a CLIP community model; not the same bug." }, { "left": "issue:43688", "right": "issue:45704", "accept": false, - "reason": "Aux-loss normalization and apex RMSNorm memory leak are unrelated." + "reason": "Loss normalization in MoE models is unrelated to a T5/apex RMSNorm memory leak." }, { - "left": "issue:29159", - "right": "issue:33357", + "left": "issue:43756", + "right": "issue:46402", "accept": false, - "reason": "General tokenizer inconsistency is unrelated to a MacOS bus error in a CLIP model." + "reason": "Different model-specific problems: RoPE layer dropping vs missing Sam3 text encoder weight loading." }, { "left": "issue:44464", "right": "issue:44743", "accept": false, - "reason": "Both touch generation/cache behavior, but one is compiled chunked generation and the other is recurrent-state reset in a specific modular model." + "reason": "Both touch generation state, but compiled-forward chunked generation inconsistency is not the same as Qwen3.5 recurrent-state reset logic." }, { "left": "issue:43653", "right": "issue:43927", "accept": false, - "reason": "Special-token registration bug is not the same as DiaConfig losing custom token IDs on save/load." - }, - { - "left": "issue:43756", - "right": "issue:46402", - "accept": false, - "reason": "Smollm3 RoPE-layer mismatch and Sam3 text-encoder weight loading are different architecture-specific issues." + "reason": "Different special-token issues: BigBird mask-token registration vs DiaConfig custom token IDs lost on save/load." }, { "left": "issue:44484", "right": "issue:44704", "accept": false, - "reason": "save_pretrained shard-size behavior is unrelated to AutoProcessor cached_file kwargs forwarding." + "reason": "save_pretrained max_shard_size question is unrelated to AutoProcessor kwargs not reaching cached_file." }, { "left": "issue:43688", "right": "issue:45440", "accept": false, - "reason": "MoE auxiliary-loss normalization does not match DeepseekV3 implementation divergence." + "reason": "Aux-loss normalization in MoE models is unrelated to DeepseekV3MoE implementation divergence." }, { "left": "issue:43688", "right": "issue:45701", "accept": false, - "reason": "One is MoE loss normalization; the other is a broad tokenizer behavior change." + "reason": "Generic MoE loss bug and broad tokenizer-version regression are not the same underlying issue." }, { "left": "issue:44479", "right": "issue:44625", "accept": false, - "reason": "Video-input regression and num_labels propagation bug are different Qwen3.5 issues." + "reason": "Video-input regression in Qwen variants is different from Qwen3.5 num_labels propagation." }, { "left": "issue:43550", "right": "issue:44361", "accept": false, - "reason": "torch.compile/SDPA model failure is unrelated to tokenizer AttributeError failures." + "reason": "torch.compile/SDPA failure in Bamba is unrelated to MLukeTokenizer task AttributeError." }, { "left": "issue:45491", "right": "issue:46002", "accept": false, - "reason": "Both mention NaNs, but in different components and scenarios; no shared bug path is evident." - }, - { - "left": "issue:44734", - "right": "issue:46612", - "accept": false, - "reason": "Both involve cache handling, but one is server KV-cache continuation indexing and the other is beam-search cache reorder omission across models." + "reason": "Both mention NaNs, but the triggers differ: sliding-window all-padding embeddings vs classification-head init/checkout." } ] }, @@ -7474,7 +7443,6 @@ "issue:43720", "issue:43742", "issue:43825", - "issue:43844", "issue:43927", "issue:43994", "issue:44361", @@ -7486,8 +7454,8 @@ "issue:44561", "issue:44610", "issue:44625", + "issue:44734", "issue:44821", - "issue:44849", "issue:44964", "issue:44977", "issue:45290", @@ -7497,9 +7465,11 @@ "issue:46050", "issue:46192", "issue:46209", - "issue:46365" + "issue:46365", + "issue:46612" ], "soft_pairs": [ + "issue:44734|issue:46612", "issue:33666|issue:36010", "issue:32090|issue:33666", "issue:44464|issue:44977", @@ -7518,8 +7488,8 @@ "issue:43262|issue:43295", "issue:44554|issue:46365", "issue:43278|issue:43688", - "issue:44977|issue:46209", "issue:43720|issue:44361", + "issue:44977|issue:46209", "issue:39692|issue:43493", "issue:44964|issue:45405", "issue:43493|issue:43825", @@ -7528,245 +7498,216 @@ "issue:44625|issue:45584", "issue:45823|issue:46050", "issue:43994|issue:46192", - "issue:28218|issue:33357", - "issue:39692|issue:43295", - "issue:43278|issue:43844", - "issue:43493|issue:43844", - "issue:33357|issue:33666", - "issue:44849|issue:46209" + "issue:28218|issue:33357" ], - "prepared_review_unit_hash": "39d5d7cb3ac2905f6ce6886d13b2a3b2eedc513dc3508cd2e543edb36c648bb8", + "prepared_review_unit_hash": "3fdcd97b1e243ebc1a6e2f89b0624b8799670155c59d57c6d2e24beb5f5c1072", "component_budget": { "node_count": 473, "item_count": 473, "soft_pair_count": 1738, - "serialized_chars": 901383, - "estimated_input_tokens": 225346, - "estimated_eval_tokens": 450948 + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 }, "budget": { "node_count": 48, "item_count": 48, - "soft_pair_count": 34, - "serialized_chars": 37957, - "estimated_input_tokens": 9490, - "estimated_eval_tokens": 19236 + "soft_pair_count": 30, + "serialized_chars": 36779, + "estimated_input_tokens": 9195, + "estimated_eval_tokens": 18646 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This set is not a true duplicate cluster. The soft-similarity edges mostly connect unrelated issues that only share broad themes like loading, tokenization, training, or generation. No pair looks like the same underlying bug or change, so all soft edges should be rejected.", - "confidence": 0.94, - "canonical_issue_reason": "None of the issues describe the same defect closely enough to make a safe canonical target; the cluster is too heterogeneous.", + "summary": "The cluster is highly heterogeneous: the soft pairs mostly share broad Transformer/model-loading/generation themes but not the same underlying bug. I would reject all soft edges. If forced to pick an issue anchor, issue 43504 is the most central by linkage density, but it is still not a strong true-duplicate hub.", + "confidence": 0.92, + "canonical_issue_reason": "issue:43504 has the densest soft-similarity neighborhood in this set (linked to 43720, 43550, and 44610) and is a concrete load-time regression, making it the best anchor if one must choose.", "canonical_pr_reason": null, - "best_issue_reason": "No single issue is a good cluster representative because the overlap is only broad subsystem similarity, not duplicate content.", + "best_issue_reason": "issue:43504 is the most representative issue in the cluster, but only weakly so; the cluster overall is too mixed to have a strong canonical issue.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:44734", + "right": "issue:46612", + "accept": false, + "reason": "Both are generation bugs, but they hit different code paths: KV-cache continuation indexing vs beam-search cache reorder for several model families." + }, { "left": "issue:33666", "right": "issue:36010", "accept": false, - "reason": "Multi-GPU training vs an import error for GenerationMixin are unrelated failures." + "reason": "Multi-GPU training trouble is unrelated to a GenerationMixin import failure; different symptoms and subsystems." }, { "left": "issue:32090", "right": "issue:33666", "accept": false, - "reason": "Trainer `_gpu_broadcast_one` TypeError and Qwen2-VL multi-GPU training are different bugs." + "reason": "Trainer/_gpu_broadcast_one TypeError is a distributed training error, not the same as Qwen2-VL multi-GPU training support/failure." }, { "left": "issue:44464", "right": "issue:44977", "accept": false, - "reason": "Both concern generation, but chunked compiled forward output drift is not the same as flash-attention generation failure." + "reason": "Both concern generation quality, but one is chunked generation with compiled forward and the other is Qwen3.5 flash-attention behavior." }, { "left": "issue:41720", "right": "issue:44464", "accept": false, - "reason": "CUDA assert on Qwen3 auto device mapping is a different problem from chunked generation inconsistency." + "reason": "A device-mapping cudaErrorAssert on A800 is a different failure mode from inconsistent outputs in compiled chunked generation." }, { "left": "issue:43742", "right": "issue:44821", "accept": false, - "reason": "MobileLLM key error when loading is unrelated to AutoImageProcessor URL loading." + "reason": "Loading a causal LM for MobileLLM is unrelated to loading AutoImageProcessor from URL; different object types and failure modes." }, { "left": "issue:44466", "right": "issue:46050", "accept": false, - "reason": "Serialization of tied lm_head weights is a different issue from sinq quantized save/load breakage." + "reason": "Both touch save/load semantics, but one is tied-weight serialization by device and the other is sinq quantization persistence." }, { "left": "issue:43504", "right": "issue:43720", "accept": false, - "reason": "BEiT legacy-field loading and BitNet packed-weight unpacking are different model-loading bugs." + "reason": "Beit legacy-field loading and BitNet packed-weight unpacking are distinct model-specific loading bugs." }, { "left": "issue:43504", "right": "issue:43550", "accept": false, - "reason": "These affect different models and code paths: pretrained loading vs torch.compile with SDPA." + "reason": "A Beit pretrained preset load issue is unrelated to Bamba-9B-v2 torch.compile/SDPA failures." }, { "left": "issue:43643", "right": "issue:44561", "accept": false, - "reason": "Both involve trust_remote_code, but one is missing fields in AutoConfig and the other is removal of is_torch_fx_available." + "reason": "AutoConfig returning missing fields under trust_remote_code and removal of is_torch_fx_available are different compatibility regressions." }, { "left": "issue:43577", "right": "issue:44554", "accept": false, - "reason": "dtype mismatch in BLIP2 evaluation is unrelated to the MPS attention correctness bug." + "reason": "Blip2 dtype casting inconsistency and an MPS attention correctness bug are unrelated." }, { "left": "issue:43504", "right": "issue:44610", "accept": false, - "reason": "A legacy config field load failure is unrelated to OmDet-Turbo processor/model image-size mismatch." + "reason": "Legacy-field loading for Beit is not the same as OmDet-Turbo's processor/model input-size mismatch." }, { "left": "issue:44514", "right": "issue:45290", "accept": false, - "reason": "Both touch apply_chat_template, but batched padding crashes and tool-call/no-content crashes are distinct cases." + "reason": "Both involve apply_chat_template, but one crashes on batched padding=False and the other on assistant tool-call messages with no content." }, { "left": "issue:43720", "right": "issue:43927", "accept": false, - "reason": "BitNet accelerate loading and DiaConfig custom token-id persistence are different bugs." + "reason": "Packed-weight accelerate loading and DiaConfig custom token IDs after save/load are different persistence bugs." }, { "left": "issue:43606", "right": "issue:43927", "accept": false, - "reason": "CPU offload device mismatch is unrelated to DiaConfig save/load token ID loss." + "reason": "CPU offload device mismatch for bark-small is unrelated to DiaConfig losing custom token IDs." }, { "left": "issue:33666", "right": "issue:36331", "accept": false, - "reason": "Qwen2-VL multi-GPU training and a CustomTrainer signature mismatch are different trainer issues." + "reason": "Qwen2-VL multi-GPU training and CustomTrainer.compute_loss signature mismatch are different trainer/distributed issues." }, { "left": "issue:43262", "right": "issue:43295", "accept": false, - "reason": "Audio chat-template sampling-rate defaulting and processor.tokenizer regression are different processor bugs." + "reason": "Audio chat-template sample-rate defaults and processor.tokenizer regression are distinct processor API bugs." }, { "left": "issue:44554", "right": "issue:46365", "accept": false, - "reason": "MPS attention correctness is unrelated to a Trainer import conflict with datasets." + "reason": "An MPS attention correctness issue is unrelated to importing Trainer failing when datasets is installed." }, { "left": "issue:43278", "right": "issue:43688", "accept": false, - "reason": "Embedding dtype drift during eval is unrelated to auxiliary-loss normalization." + "reason": "Model dtype drift in eval and auxiliary-loss normalization are different training-time bugs." }, { - "left": "issue:44977", - "right": "issue:46209", + "left": "issue:43720", + "right": "issue:44361", "accept": false, - "reason": "Qwen3.5 flash-attention generation and FP8 inference on B200 are separate backend problems." + "reason": "BitNet accelerate loading and MLukeTokenizer task AttributeError are unrelated model/tokenizer failures." }, { - "left": "issue:43720", - "right": "issue:44361", + "left": "issue:44977", + "right": "issue:46209", "accept": false, - "reason": "BitNet packed-weight loading and MLukeTokenizer task AttributeError are unrelated." + "reason": "Qwen3.5 flash-attention generation problems and FP8 inference failure on B200 are different hardware/kernel issues." }, { "left": "issue:39692", "right": "issue:43493", "accept": false, - "reason": "SigLIP2 doc-example errors and HF-vs-JAX implementation discrepancy are different issues." + "reason": "SigLIP2 docs example errors and HF-vs-JAX implementation discrepancy are related to the same model family but not the same underlying bug." }, { "left": "issue:44964", "right": "issue:45405", "accept": false, - "reason": "Phi-4 multimodal loading failure is unrelated to the MIN_PEFT_VERSION packaging issue." + "reason": "A model-loading regression for Phi-4 multimodal is not the same as a too-new MIN_PEFT_VERSION requirement." }, { "left": "issue:43493", "right": "issue:43825", "accept": false, - "reason": "SigLIP2 implementation mismatch and pipeline translation warning are unrelated." + "reason": "SigLIP2 implementation mismatch and an incorrect pipeline error message about translation support are unrelated." }, { "left": "issue:43452", "right": "issue:44451", "accept": false, - "reason": "gguf_file breaks tokenizer/model loading APIs, while ScandiBERT load failure is a different model-specific issue." + "reason": "gguf_file breaking from_pretrained paths and ScandiBERT load failure are different loading issues." }, { "left": "issue:43525", "right": "issue:43572", "accept": false, - "reason": "Missing pad_token_id in Llama4Config and missing pad_token_idx in StableLmConfig are similar in shape but not the same defect." + "reason": "Both mention missing pad-token-related fields, but they affect different configs, fields, and models." }, { "left": "issue:44625", "right": "issue:45584", "accept": false, - "reason": "Qwen3.5 num_labels propagation and Whisper empty-transcription handling are unrelated." + "reason": "Qwen3.5 num_labels propagation and Whisper empty-transcription failure are unrelated model behaviors." }, { "left": "issue:45823", "right": "issue:46050", "accept": false, - "reason": "Gemma4 device-map mismatch and sinq save/load quantization breakage are different bugs." + "reason": "Gemma4 PLE device mismatch with auto device map is not the same as sinq quantization save/load breakage." }, { "left": "issue:43994", "right": "issue:46192", "accept": false, - "reason": "SigLIP2 nonsensical outputs and return_tensors being ignored are unrelated API issues." + "reason": "SigLIP2 nonsense outputs and return_tensors being ignored are different API bugs." }, { "left": "issue:28218", "right": "issue:33357", "accept": false, - "reason": "Tokenizer added-space behavior and MacOS bus error in CLIP are clearly unrelated." - }, - { - "left": "issue:39692", - "right": "issue:43295", - "accept": false, - "reason": "SigLIP2 doc-example mistakes and processor.tokenizer / image-passing regression are different issues." - }, - { - "left": "issue:43278", - "right": "issue:43844", - "accept": false, - "reason": "Evaluation dtype drift and DeepSpeed gradient explosion are unrelated training behaviors." - }, - { - "left": "issue:43493", - "right": "issue:43844", - "accept": false, - "reason": "SigLIP2 implementation mismatch and ZeRO-3 gradient growth are unrelated." - }, - { - "left": "issue:33357", - "right": "issue:33666", - "accept": false, - "reason": "MacOS CLIP bus error and Qwen2-VL multi-GPU training do not share the same bug." - }, - { - "left": "issue:44849", - "right": "issue:46209", - "accept": false, - "reason": "Qwen3.5 hidden-state bug and FP8 B200 inference failure are separate model/backend issues." + "reason": "Tokenizer extra-spacing after added token and MacOS bus error with a CLIP model are unrelated." } ] }, @@ -7790,12 +7731,15 @@ "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-10000-473", "nodes": [ - "issue:29942", - "issue:30333", + "issue:33357", + "issue:33666", + "issue:39692", "issue:43257", "issue:43278", + "issue:43295", "issue:43299", "issue:43334", + "issue:43493", "issue:43504", "issue:43540", "issue:43572", @@ -7806,7 +7750,6 @@ "issue:43701", "issue:43824", "issue:43844", - "issue:43873", "issue:43927", "issue:43994", "issue:44112", @@ -7829,28 +7772,32 @@ "issue:45405", "issue:45520", "issue:45584", - "issue:45588", "issue:45759", "issue:45812", "issue:45823", - "issue:46050", + "issue:46209", "issue:46285", "issue:46424", "issue:46489" ], "soft_pairs": [ + "issue:43278|issue:43844", + "issue:43493|issue:43844", + "issue:39692|issue:43295", + "issue:44849|issue:46209", + "issue:33357|issue:33666", "issue:44488|issue:44843", "issue:43646|issue:44466", "issue:44568|issue:46489", "issue:43278|issue:43701", "issue:43927|issue:44361", - "issue:43540|issue:44977", "issue:44568|issue:45812", - "issue:43257|issue:43299", + "issue:43540|issue:44977", "issue:44792|issue:45405", + "issue:43257|issue:43299", "issue:43575|issue:45405", - "issue:43504|issue:43606", "issue:44589|issue:45520", + "issue:43504|issue:43606", "issue:44849|issue:45357", "issue:43334|issue:43572", "issue:43606|issue:44112", @@ -7864,213 +7811,222 @@ "issue:44464|issue:45584", "issue:45229|issue:45823", "issue:43994|issue:44493", - "issue:43844|issue:44964", - "issue:43873|issue:46050", - "issue:29942|issue:30333", - "issue:45588|issue:46050" + "issue:43844|issue:44964" ], - "prepared_review_unit_hash": "1d8d8b7ff80b31af974621121679db7d101ed31860bc3d24ed0e3c6958ef023b", + "prepared_review_unit_hash": "bcfb3589755502632e33adcf63cc6a7d4a70eff678f6636b6f0d08c7f1a80a3f", "component_budget": { "node_count": 473, "item_count": 473, "soft_pair_count": 1738, - "serialized_chars": 901383, - "estimated_input_tokens": 225346, - "estimated_eval_tokens": 450948 + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 }, "budget": { - "node_count": 47, - "item_count": 47, - "soft_pair_count": 29, - "serialized_chars": 35928, - "estimated_input_tokens": 8982, - "estimated_eval_tokens": 18220 + "node_count": 48, + "item_count": 48, + "soft_pair_count": 31, + "serialized_chars": 37130, + "estimated_input_tokens": 9283, + "estimated_eval_tokens": 18822 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Loose cluster of mostly unrelated Transformers bug reports. The soft edges are driven by shared keywords like tokenizer, Qwen, offload, or v5 regression, but they do not look like the same concrete underlying defect.", - "confidence": 0.91, - "canonical_issue_reason": "No single issue clearly subsumes the cluster; the items span different models, error modes, and code paths.", + "summary": "This cluster is mostly a loose collection of unrelated Transformers regressions, with many tokenizer/config/model-loading issues across different model families. None of the soft pairs look like the same underlying bug or change, so I rejected them all. If one issue has to represent the cluster, the broadest and most user-impactful is the Granite tokenizer regression.", + "confidence": 0.43, + "canonical_issue_reason": "issue:45812 is the best umbrella representative: it describes a broad, silent v4\u2192v5 tokenizer regression affecting all Granite models, which is more general and higher-impact than the more model-specific reports here.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 45812 is the broadest representative of the cluster because it describes a general v4\u2192v5 tokenizer regression, but it still does not duplicate the rest.", + "best_issue_reason": "issue:45812 has the clearest cluster-representative value because it is broad, regression-framed, and likely to matter to multiple downstream users; the rest are mostly narrower, unrelated edge cases.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:43278", + "right": "issue:43844", + "accept": false, + "reason": "Different bugs: dtype change in evaluation vs. exploding gradients with HfDeepSpeedConfig/ZeRO-3." + }, + { + "left": "issue:43493", + "right": "issue:43844", + "accept": false, + "reason": "SigLIP2 implementation discrepancy is unrelated to the DeepSpeed gradient problem." + }, + { + "left": "issue:39692", + "right": "issue:43295", + "accept": false, + "reason": "Documentation/example errors for SigLIP2 are not the same as the v4.57.5 processor/tokenizer regression." + }, + { + "left": "issue:44849", + "right": "issue:46209", + "accept": false, + "reason": "Different model families and failure modes: Qwen3.5 hidden-states behavior vs FP8 inference on B200." + }, + { + "left": "issue:33357", + "right": "issue:33666", + "accept": false, + "reason": "CLIP bus error on macOS is unrelated to Qwen2-VL multi-GPU training." + }, { "left": "issue:44488", "right": "issue:44843", "accept": false, - "reason": "Both involve loading/tokenizer behavior, but one is a specific model load failure and the other is an HF_HUB_OFFLINE regression in _patch_mistral_regex; different bugs." + "reason": "Both involve loading/tokenizer code, but one is a specific model load failure and the other is an offline-mode regression from unconditional hub calls." }, { "left": "issue:43646", "right": "issue:44466", "accept": false, - "reason": "One is custom model initialization breaking in v5, the other is inconsistent lm_head serialization tied to device state; not the same defect." + "reason": "Custom model initialization breakage is not the same as device-dependent serialization of lm_head.weight." }, { "left": "issue:44568", "right": "issue:46489", "accept": false, - "reason": "Both are tokenizer regressions, but they affect different models and symptoms (BOS/EOS handling vs DeepSeek-Coder output after a prior fix)." + "reason": "Both are tokenizer regressions, but they concern different models and different concrete failures." }, { "left": "issue:43278", "right": "issue:43701", "accept": false, - "reason": "Embedding dtype drift in eval and resume_from_checkpoint key mismatch are unrelated failures." + "reason": "Embedding dtype mismatch is unrelated to resume_from_checkpoint key mismatch." }, { "left": "issue:43927", "right": "issue:44361", "accept": false, - "reason": "Different model families and different failure modes: lost custom token IDs vs MLuke tokenizer AttributeError." + "reason": "DiaConfig save/load token ID loss is a different issue from MLukeTokenizer task-time AttributeError." + }, + { + "left": "issue:44568", + "right": "issue:45812", + "accept": false, + "reason": "Granite token ID regression and mdeberta add_special_tokens regression are separate tokenizer bugs." }, { "left": "issue:43540", "right": "issue:44977", "accept": false, - "reason": "Video-input processing error in Qwen3OmniMoe is unrelated to Qwen3.5 flash-attention generation issues." + "reason": "Video input processing error in Qwen3OmniMoe is not the same as Qwen3.5 flash-attention generation issues." }, { - "left": "issue:44568", - "right": "issue:45812", + "left": "issue:44792", + "right": "issue:45405", "accept": false, - "reason": "Both are tokenizer bugs, but they hit different pretrained models and different code paths; not mergeable as one fix." + "reason": "A janus model test failure is unrelated to the PEFT minimum-version bump." }, { "left": "issue:43257", "right": "issue:43299", "accept": false, - "reason": "Qwen3 MoE weight conversion with accelerate+deepspeed is a different problem from Qwen3VL MoE loading in v5.0.0.dev0." + "reason": "Both are Qwen3 MOE-related, but one is accelerate/deepspeed weight conversion and the other is Qwen3VL loading; not the same bug." }, { - "left": "issue:44792", + "left": "issue:43575", "right": "issue:45405", "accept": false, - "reason": "Janus image-generation test failure and a PEFT version bump are unrelated." + "reason": "Tensor-parallel OOM is unrelated to a dependency version bump." }, { - "left": "issue:43575", - "right": "issue:45405", + "left": "issue:44589", + "right": "issue:45520", "accept": false, - "reason": "Tensor-parallel OOM for Qwen2-57B is unrelated to the PEFT release/version issue." + "reason": "Float8 storage lookup failure and flash_attn import KeyError are different platform/runtime issues." }, { "left": "issue:43504", "right": "issue:43606", "accept": false, - "reason": "BEiT legacy-field load failure and Bark CPU-offload device mismatch are different model-specific bugs." - }, - { - "left": "issue:44589", - "right": "issue:45520", - "accept": false, - "reason": "Float8 storage lookup failure and flash_attn import KeyError are distinct runtime/import problems." + "reason": "Legacy-field load failure for Beit is not the same as Bark CPU-offload device mismatch." }, { "left": "issue:44849", "right": "issue:45357", "accept": false, - "reason": "Qwen3.5 hidden-state output bug and incorrect save_pretrained visual encoder keys are different regressions." + "reason": "Qwen3.5 hidden-states bug and incorrect visual encoder save_pretrained keys are distinct regression symptoms." }, { "left": "issue:43334", "right": "issue:43572", "accept": false, - "reason": "Missing pad_token_id in Qwen3-VL and missing pad_token_idx in StableLmConfig are separate config issues." + "reason": "Missing pad_token_id in Qwen3-VL text config is unrelated to StableLm pad_token_idx." }, { "left": "issue:43606", "right": "issue:44112", "accept": false, - "reason": "Both mention device/offload tests, but they involve different models and different failures; not the same code-path bug." + "reason": "A real CPU-offload device mismatch is not the same as a stale GraniteSpeech CI test." }, { "left": "issue:43824", "right": "issue:44589", "accept": false, - "reason": "ImportError for a Qwen2.5 VL class is unrelated to the Float8 storage-object TypeError." + "reason": "Qwen2.5-VL import error and Float8 storage lookup failure are unrelated." }, { "left": "issue:43334", "right": "issue:45759", "accept": false, - "reason": "Missing pad_token_id in Qwen3-VL is unrelated to AutoModelForCausalLM.from_config not unwrapping text_config for Qwen multimodal configs." + "reason": "Both touch Qwen multimodal config loading, but the concrete failure modes differ: missing pad_token_id vs text_config unwrapping." }, { "left": "issue:44464", "right": "issue:46424", "accept": false, - "reason": "Chunked generation/compiled forward inconsistency is unrelated to cache max_cache_length only applying to quantized cache." + "reason": "Chunked generation inconsistency with compiled forward is not the same as max_cache_length handling in cache config." }, { "left": "issue:44493", "right": "issue:46285", "accept": false, - "reason": "Unexpected position-id keys across many models is not the same as CLIP text-model structure divergence." + "reason": "Unexpected position_id keys across many models is a different structural regression than CLIP text-model wrapper inconsistency." }, { "left": "issue:43504", "right": "issue:43653", "accept": false, - "reason": "Different tokenizer/model issues: BEiT legacy-field loading vs BigBirdTokenizer mask token registration." + "reason": "Beit legacy-field loading and BigBirdTokenizer special-token registration are unrelated." }, { "left": "issue:45200", "right": "issue:45229", "accept": false, - "reason": "Defaulting mm_token_type_ids to zeros for fine-tuning is unrelated to a Gemma4 multi-GPU CUDA OOM." + "reason": "Text-only mm_token_type_ids defaulting is unrelated to Gemma4 multi-GPU CUDA OOM." }, { "left": "issue:43572", "right": "issue:44291", "accept": false, - "reason": "StableLmConfig pad_token_idx and init_empty_weights unexpected _is_hf_initialized argument are separate compatibility bugs." + "reason": "StableLm pad_token_idx regression and init_empty_weights unexpected argument are different initialization problems." }, { "left": "issue:44464", "right": "issue:45584", "accept": false, - "reason": "Generation inconsistency with compiled forward and Whisper empty-transcription behavior after token alignment are different issues." + "reason": "Compiled-forward chunked generation inconsistency is unrelated to Whisper empty-transcription failure after align_special_tokens." }, { "left": "issue:45229", "right": "issue:45823", "accept": false, - "reason": "Gemma4 multi-GPU OOM and PLE device mismatch during forward are distinct runtime problems." + "reason": "Both are Gemma4 issues, but one is inference OOM and the other is a device mismatch during forward." }, { "left": "issue:43994", "right": "issue:44493", "accept": false, - "reason": "Siglip2 nonsensical outputs and unexpected position-id keys are unrelated regression reports." + "reason": "SigLIP2 nonsensical outputs and widespread position_id key warnings are not the same bug." }, { "left": "issue:43844", "right": "issue:44964", "accept": false, - "reason": "ZeRO-3 gradient growth and Phi-4 multimodal loading failure are different code paths." - }, - { - "left": "issue:43873", - "right": "issue:46050", - "accept": false, - "reason": "Quantization/offloading mismatch is broader than the specific sinq save/load breakage; not the same defect." - }, - { - "left": "issue:29942", - "right": "issue:30333", - "accept": false, - "reason": "Flash Attention 2 test failures and MLFlow reporting on training failure are unrelated." - }, - { - "left": "issue:45588", - "right": "issue:46050", - "accept": false, - "reason": "Flash attention s_aux=None crash and sinq save/load corruption are different subsystems and symptoms." + "reason": "DeepSpeed gradient anomaly is unrelated to Phi-4-multimodal model loading failure." } ] }, @@ -8097,6 +8053,8 @@ "issue:15354", "issue:28218", "issue:29159", + "issue:29942", + "issue:30333", "issue:32090", "issue:34567", "issue:37428", @@ -8114,13 +8072,11 @@ "issue:43575", "issue:43696", "issue:43701", - "issue:43824", "issue:43844", "issue:43873", "issue:43957", "issue:44186", "issue:44423", - "issue:44521", "issue:44554", "issue:44610", "issue:44734", @@ -8130,7 +8086,6 @@ "issue:45084", "issue:45198", "issue:45200", - "issue:45290", "issue:45356", "issue:45406", "issue:45446", @@ -8138,258 +8093,246 @@ "issue:45588", "issue:45759", "issue:45823", + "issue:46050", "issue:46121", "issue:46178", "issue:46192", "issue:46369" ], "soft_pairs": [ - "issue:29159|issue:32090", + "issue:43873|issue:46050", + "issue:45588|issue:46050", + "issue:29942|issue:30333", "issue:42490|issue:43696", + "issue:29159|issue:32090", "issue:43065|issue:46369", "issue:45084|issue:45520", - "issue:44423|issue:44734", "issue:28218|issue:34567", + "issue:44423|issue:44734", "issue:44554|issue:45446", "issue:15354|issue:38175", "issue:45406|issue:46121", "issue:43493|issue:43701", "issue:45072|issue:45198", "issue:45588|issue:45759", - "issue:43295|issue:43450", "issue:43540|issue:45200", + "issue:43295|issue:43450", "issue:37428|issue:43957", "issue:44186|issue:46192", "issue:45200|issue:45823", "issue:45823|issue:46192", "issue:43295|issue:44610", "issue:43295|issue:43873", - "issue:43450|issue:43575", "issue:45520|issue:46178", + "issue:43450|issue:43575", "issue:43550|issue:43844", "issue:43232|issue:44849", "issue:43572|issue:44987", "issue:44554|issue:45356", - "issue:39692|issue:46192", - "issue:44521|issue:45290", - "issue:43572|issue:43957", - "issue:37428|issue:43824", - "issue:42490|issue:43493" + "issue:39692|issue:46192" ], - "prepared_review_unit_hash": "8700b057ae9f839d0664a5a5743c9e6c859630c74fc1dac4ac27b384378901e1", + "prepared_review_unit_hash": "9146da04106aa5e94d1713322633bf0b6211a9a9965cab368a50d9014e45bcd0", "component_budget": { "node_count": 473, "item_count": 473, "soft_pair_count": 1738, - "serialized_chars": 901383, - "estimated_input_tokens": 225346, - "estimated_eval_tokens": 450948 + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 }, "budget": { "node_count": 48, "item_count": 48, - "soft_pair_count": 31, - "serialized_chars": 37161, - "estimated_input_tokens": 9291, - "estimated_eval_tokens": 18838 + "soft_pair_count": 30, + "serialized_chars": 36779, + "estimated_input_tokens": 9195, + "estimated_eval_tokens": 18646 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is heterogeneous: it mixes unrelated issues across tokenizers, multimodal processors, serving, training, loading, and attention. None of the soft pairs looks like the same concrete bug/change, so all soft edges are rejected.", - "confidence": 0.95, - "canonical_issue_reason": "No single issue cleanly represents the cluster; the items span unrelated failures rather than one underlying bug.", + "summary": "This cluster is a loose grab-bag of transformer bug reports spanning tokenizer regressions, multimodal processor issues, generation/serve crashes, flash-attention, quantization, and loading/import edge cases. The soft pairs mostly share only broad subsystem overlap, not the same underlying defect.", + "confidence": 0.31, + "canonical_issue_reason": "issue:43295 is the broadest user-facing regression here around processor/tokenizer behavior in a recent release, and it sits closest to the multimodal/tokenization theme that recurs across several nearby issues.", "canonical_pr_reason": null, - "best_issue_reason": "issue:43295 is the closest representative of the small multimodal processor/tokenizer regression subgroup, but it still does not explain the rest of the cluster.", + "best_issue_reason": "issue:43295 is the most representative anchor because it describes a concrete regression with clear user impact and a general processor/tokenizer path, making it the best single issue to summarize this otherwise heterogeneous set.", "best_pr_reason": null, "soft_edge_verdicts": [ { - "left": "issue:29159", - "right": "issue:32090", + "left": "issue:43873", + "right": "issue:46050", + "accept": false, + "reason": "Quantization offloading behavior and save/load corruption are different failure modes; same broad area, not the same bug." + }, + { + "left": "issue:45588", + "right": "issue:46050", + "accept": false, + "reason": "Flash-attention sink handling crash is unrelated to sinq save_pretrained/from_pretrained breakage." + }, + { + "left": "issue:29942", + "right": "issue:30333", "accept": false, - "reason": "Tokenizer inconsistency and Trainer broadcast TypeError are unrelated failures." + "reason": "FlashAttention2 CI failures and MLflow reporting on job failure are unrelated subsystems." }, { "left": "issue:42490", "right": "issue:43696", "accept": false, - "reason": "Local-path import warning and CUDA OOM are different problems." + "reason": "Trailing-slash import warning is a path-parsing issue; the other is a model OOM on load." + }, + { + "left": "issue:29159", + "right": "issue:32090", + "accept": false, + "reason": "Tokenizer slow/fast inconsistency is unrelated to the trainer NoneType GPU broadcast error." }, { "left": "issue:43065", "right": "issue:46369", "accept": false, - "reason": "Both mention Sam3, but one is a dummy Conv2d issue and the other is a memory overflow." + "reason": "A dummy Conv2d placeholder in a decoder is not the same as memory overflow when using the SAM3 lite text model." }, { "left": "issue:45084", "right": "issue:45520", "accept": false, - "reason": "torch.compile/non-template-node error and flash_attn import KeyError are unrelated." + "reason": "Compilation of non-template nodes and flash_attn backend import mapping are different code paths." }, { - "left": "issue:44423", - "right": "issue:44734", + "left": "issue:28218", + "right": "issue:34567", "accept": false, - "reason": "Both are serve crashes, but they hit different code paths and symptoms." + "reason": "Added-token spacing and num_input_tokens_seen bookkeeping are unrelated tokenizer vs trainer-state issues." }, { - "left": "issue:28218", - "right": "issue:34567", + "left": "issue:44423", + "right": "issue:44734", "accept": false, - "reason": "Added-token spacing and TrainerState token counter update are unrelated." + "reason": "Both involve serve, but one is a multimodal continuous-batching crash and the other is KV-cache continuation indexing." }, { "left": "issue:44554", "right": "issue:45446", "accept": false, - "reason": "MPS attention correctness bug and PyTorch version-check import bug are unrelated." + "reason": "MPS attention correctness and a PyTorch version check for AuxRequest import are unrelated defects." }, { "left": "issue:15354", "right": "issue:38175", "accept": false, - "reason": "torch.jit.script export failure and SigLIP2 zero-probabilities are unrelated." + "reason": "TorchScript generator-expression export failure is unrelated to SigLIP2 zero probabilities." }, { "left": "issue:45406", "right": "issue:46121", "accept": false, - "reason": "Gemma4Processor _tokenizer crash and rope-validation TypeError are different bugs." + "reason": "Gemma4Processor missing _tokenizer and rope param list conversion are different bugs in different layers." }, { "left": "issue:43493", "right": "issue:43701", "accept": false, - "reason": "SigLIP2 implementation discrepancy and resume checkpoint key mismatch are unrelated." + "reason": "SigLIP2 implementation parity and checkpoint resume key mismatch do not describe the same underlying problem." }, { "left": "issue:45072", "right": "issue:45198", "accept": false, - "reason": "dtype mismatch inference and Wav2Vec2 save/tokenization failure are different issues." + "reason": "Dtype mismatches in bfloat16 inference are unrelated to Wav2Vec2 save_pretrained/tokenization failure." }, { "left": "issue:45588", "right": "issue:45759", "accept": false, - "reason": "flash_attention sink-less crash and text_config unwrap bug are not the same code-path." + "reason": "Flash-attention s_aux=None crash and composite Qwen config unwrapping are different model-integration bugs." }, { - "left": "issue:43295", - "right": "issue:43450", + "left": "issue:43540", + "right": "issue:45200", "accept": false, - "reason": "Both touch processors, but one is a processor.tokenizer regression and the other is batched video shape handling." + "reason": "Video input processing error in Qwen3OmniMoe is not the same as Gemma4 mm_token_type_ids defaults." }, { - "left": "issue:43540", - "right": "issue:45200", + "left": "issue:43295", + "right": "issue:43450", "accept": false, - "reason": "Video-input processing error and mm_token_type_ids defaulting are unrelated." + "reason": "Processor/tokenizer regression in one issue and batched video output shape in the other are distinct multimodal bugs." }, { "left": "issue:37428", "right": "issue:43957", "accept": false, - "reason": "Flash-attention import error and meta-device model loading breakage are different failures." + "reason": "Missing flash-support import and meta-device loading breakage are different import/loading problems." }, { "left": "issue:44186", "right": "issue:46192", "accept": false, - "reason": "Tokenizer crash on NER/batched padding and ignored return_tensors are separate bugs." + "reason": "LayoutLMv2 tokenizer crashes on NER/batching differ from return_tensors being ignored with text_kwargs." }, { "left": "issue:45200", "right": "issue:45823", "accept": false, - "reason": "mm_token_type_ids defaulting and Gemma4 PLE device mismatch are unrelated." + "reason": "Defaulting mm_token_type_ids and device-map-induced device mismatch are not the same defect." }, { "left": "issue:45823", "right": "issue:46192", "accept": false, - "reason": "Device-placement mismatch and return_tensors handling are different issues." + "reason": "Device mismatch during forward is unrelated to return_tensors being ignored." }, { "left": "issue:43295", "right": "issue:44610", "accept": false, - "reason": "Both are multimodal processor issues, but the concrete bug is different: processor.tokenizer regression vs image-size mismatch." + "reason": "A processor/tokenizer regression is not the same as OmDet-Turbo producing the wrong image size." }, { - "left": "issue:43295", - "right": "issue:43873", + "left": "issue:45520", + "right": "issue:46178", "accept": false, - "reason": "Processor/tokenizer regression and quantization offloading behavior are unrelated." + "reason": "flash_attn KeyError on Python 3.13 and tensorflow_text backend mapping validation are separate import-registry issues." }, { "left": "issue:43450", "right": "issue:43575", "accept": false, - "reason": "Video processor shape bug and tensor-parallel OOM on load are unrelated." - }, - { - "left": "issue:45520", - "right": "issue:46178", - "accept": false, - "reason": "Missing flash_attn backend and missing tensorflow_text backend mapping are different import/config issues." + "reason": "Batched video shape bug and tensor-parallel OOM for Qwen2-57B are unrelated." }, { "left": "issue:43550", "right": "issue:43844", "accept": false, - "reason": "torch.compile SDPA failure and DeepSpeed ZeRO-3 gradient growth are unrelated." + "reason": "torch.compile SDPA failure and ZeRO-3 gradient explosion are different training/runtime bugs." }, { "left": "issue:43232", "right": "issue:44849", "accept": false, - "reason": "Generation kwargs update after sync_gpus and output_hidden_states bug are different generation-path problems." + "reason": "Generation kwargs update after sync_gpus and output_hidden_states behavior in Qwen3.5 are not the same issue." }, { "left": "issue:43572", "right": "issue:44987", "accept": false, - "reason": "Config pad_token_idx regression and model-loading failure are unrelated." + "reason": "Missing pad_token_idx in StableLmConfig and a transformers>=5.1 loading failure are separate regression reports." }, { "left": "issue:44554", "right": "issue:45356", "accept": false, - "reason": "MPS attention correctness bug and Kimi tokenizer codec regression are unrelated." + "reason": "MPS attention correctness and Kimi tokenizer codec regression are unrelated." }, { "left": "issue:39692", "right": "issue:46192", "accept": false, - "reason": "Docs/example model-processor mismatch and ignored return_tensors are not the same bug." - }, - { - "left": "issue:44521", - "right": "issue:45290", - "accept": false, - "reason": "Both involve apply_chat_template, but one is zero assistant masks and the other is a crash on tool-call assistant messages." - }, - { - "left": "issue:43572", - "right": "issue:43957", - "accept": false, - "reason": "Missing pad_token_idx in config and meta-device loading failures are unrelated." - }, - { - "left": "issue:37428", - "right": "issue:43824", - "accept": false, - "reason": "Different import errors in different modules/models; not the same underlying issue." - }, - { - "left": "issue:42490", - "right": "issue:43493", - "accept": false, - "reason": "Local-path slash warning and SigLIP2 implementation discrepancy are unrelated." + "reason": "Documentation example errors for SigLIP2 are not the same as return_tensors being ignored when text_kwargs is passed." } ] }, @@ -8415,20 +8358,21 @@ "nodes": [ "issue:28218", "issue:35141", + "issue:37428", "issue:38175", "issue:41720", + "issue:42490", "issue:42915", "issue:43262", "issue:43278", "issue:43334", - "issue:43335", "issue:43388", "issue:43450", "issue:43452", "issue:43493", "issue:43525", "issue:43526", - "issue:43550", + "issue:43572", "issue:43653", "issue:43819", "issue:43824", @@ -8436,34 +8380,36 @@ "issue:43856", "issue:43901", "issue:43950", + "issue:43957", "issue:43975", "issue:44112", - "issue:44206", "issue:44442", "issue:44466", - "issue:44561", + "issue:44479", + "issue:44521", "issue:44610", "issue:44792", "issue:44857", "issue:44898", - "issue:45083", "issue:45127", "issue:45198", "issue:45200", + "issue:45290", "issue:45325", "issue:45381", "issue:45412", "issue:45563", - "issue:45584", - "issue:45759", "issue:46192", "issue:46209", - "issue:46234", "issue:46285", "issue:46512" ], "soft_pairs": [ + "issue:44521|issue:45290", + "issue:43572|issue:43957", + "issue:37428|issue:43824", "issue:45381|issue:46209", + "issue:42490|issue:43493", "issue:44466|issue:45127", "issue:38175|issue:43493", "issue:43653|issue:44112", @@ -8476,218 +8422,200 @@ "issue:43334|issue:43525", "issue:43262|issue:43450", "issue:43824|issue:43975", - "issue:44442|issue:45198", "issue:45200|issue:45325", - "issue:43526|issue:46192", - "issue:43388|issue:43526", + "issue:44442|issue:45198", + "issue:44479|issue:46192", "issue:44792|issue:45563", + "issue:43388|issue:43526", "issue:43819|issue:44112", "issue:43827|issue:43901", - "issue:44610|issue:44857", "issue:41720|issue:46209", - "issue:44561|issue:46234", - "issue:38175|issue:45412", - "issue:43335|issue:43550", - "issue:44206|issue:45584", - "issue:43819|issue:45083", - "issue:45759|issue:46285" + "issue:44610|issue:44857" ], - "prepared_review_unit_hash": "e5aaac4886adeb71c2bc5092aa6bbe1eee16b0afad0002b9a5e6110197a25ac0", + "prepared_review_unit_hash": "e7e9decde8a025d02ca0d1e9d9f344f1dff01a249304f5e114ee3b0b0f1f1e84", "component_budget": { "node_count": 473, "item_count": 473, "soft_pair_count": 1738, - "serialized_chars": 901383, - "estimated_input_tokens": 225346, - "estimated_eval_tokens": 450948 + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 }, "budget": { - "node_count": 48, - "item_count": 48, - "soft_pair_count": 28, - "serialized_chars": 36507, - "estimated_input_tokens": 9127, - "estimated_eval_tokens": 18510 + "node_count": 47, + "item_count": 47, + "soft_pair_count": 26, + "serialized_chars": 35354, + "estimated_input_tokens": 8839, + "estimated_eval_tokens": 17934 }, "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 subsystems or model families, not the same concrete bug. I rejected every soft edge and found no clear canonical duplicate issue or PR.", - "confidence": 0.93, - "canonical_issue_reason": "No canonical issue stands out because the items are not a duplicate set; they span unrelated bugs across tokenization, model loading, inference, docs, and config handling.", + "summary": "All soft pairs look like distinct issues rather than duplicates; they mostly share broad subsystems or API surfaces but not the same concrete bug. I therefore rejected every soft edge and selected one strong open issue as the best representative issue overall.", + "confidence": 0.94, + "canonical_issue_reason": "issue:44521 is the strongest open, user-facing report here: it has multiple inbound references, clear reproducible behavior, and a focused multimodal `apply_chat_template` regression.", "canonical_pr_reason": null, - "best_issue_reason": "No single issue is a good global representative for deduplication, since no soft pair looks like the same underlying bug/change.", + "best_issue_reason": "issue:44521 is the most actionable overall because it is open, well-scoped, and appears to impact a common multimodal code path with active discussion.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:44521", + "right": "issue:45290", + "accept": false, + "reason": "Both involve `apply_chat_template`, but one is about zeroed assistant masks for multimodal inputs and the other is a crash with tool-call messages; different failure modes and likely different fixes." + }, + { + "left": "issue:43572", + "right": "issue:43957", + "accept": false, + "reason": "These are separate loading/config bugs: missing `pad_token_idx` in one model config versus failures when loading models on `torch.device(\"meta\")`." + }, + { + "left": "issue:37428", + "right": "issue:43824", + "accept": false, + "reason": "Both are import errors, but for different symbols and likely unrelated code paths (`_flash_supports_window_size` vs `Qwen2_5_VLForConditionalGeneration`)." + }, { "left": "issue:45381", "right": "issue:46209", "accept": false, - "reason": "Both are Qwen-related inference failures, but one is a vision-position-id bug and the other is an FP8/B200 hardware crash." + "reason": "One is a Qwen2.5-VL video position-id regression; the other is an FP8 inference failure on B200. Different hardware/model issues." + }, + { + "left": "issue:42490", + "right": "issue:43493", + "accept": false, + "reason": "Local-path slash import warning and SigLIP2 implementation mismatch are unrelated bugs." }, { "left": "issue:44466", "right": "issue:45127", "accept": false, - "reason": "Both involve tied-weight serialization concerns, but they describe different failures: lm_head serialization vs LoRA merge collapse." + "reason": "One concerns inconsistent `lm_head.weight` serialization by device; the other concerns LoRA merge behavior with extended vocab and tied embeddings. Not the same bug." }, { "left": "issue:38175", "right": "issue:43493", "accept": false, - "reason": "Same model family, but one reports zero probabilities and the other a broader HF-vs-JAX discrepancy; not the same concrete bug." + "reason": "Both mention SigLIP2, but one is a specific zero-probability symptom and the other is a broader HF-vs-JAX discrepancy; insufficient evidence they are the same concrete issue." }, { "left": "issue:43653", "right": "issue:44112", "accept": false, - "reason": "Completely different areas: BigBird tokenizer special-token decoding vs a GraniteSpeech CI device-override test." + "reason": "A tokenizer special-token registration bug and a GraniteSpeech CI device-override test failure are different problems." }, { "left": "issue:43278", "right": "issue:43856", "accept": false, - "reason": "Different symptoms and code paths: dtype mismatch during eval vs memory usage during Qwen3 MoE training." + "reason": "Different topics: dtype drift in evaluation versus memory inefficiency during Qwen3 MoE training." }, { "left": "issue:28218", "right": "issue:35141", "accept": false, - "reason": "Tokenizer spacing after added tokens is unrelated to embedding reinitialization during resize/post_init." + "reason": "Tokenizer spacing after added tokens is unrelated to output embedding reinitialization after resizing embeddings." }, { "left": "issue:43452", "right": "issue:46512", "accept": false, - "reason": "One is gguf loading in AutoTokenizer/AutoModel; the other is stale config.dtype after from_config. Different bugs." + "reason": "GGUF loading breakage and stale `config.dtype` after `from_config(dtype=...)` are separate model-loading regressions." }, { "left": "issue:43950", "right": "issue:46285", "accept": false, - "reason": "Non-persistent buffer corruption in from_pretrained is unrelated to CLIP text-model structural inconsistency." + "reason": "Non-persistent buffer corruption on load is not the same as CLIP text model structural inconsistency." }, { "left": "issue:44898", "right": "issue:45412", "accept": false, - "reason": "Perceiver resolution failure and RT-DETR memory leak are unrelated model bugs." + "reason": "Perceiver interpolation failure and RT-DETR memory not being released are unrelated." }, { "left": "issue:42915", "right": "issue:43824", "accept": false, - "reason": "Qwen3Moe FP8 failure is unrelated to Qwen2.5-VL import/export errors." + "reason": "Qwen3Moe FP8 failure and a missing Qwen2.5-VL import are different issues." }, { "left": "issue:43334", "right": "issue:43525", "accept": false, - "reason": "Both mention missing pad_token_id, but they affect different model families and are not clearly the same code-path fix." + "reason": "Both are missing `pad_token_id` on configs, but they affect different model classes and likely require separate fixes." }, { "left": "issue:43262", "right": "issue:43450", "accept": false, - "reason": "Audio chat-template sampling-rate handling is unrelated to batched video processor output shapes." + "reason": "Audio chat-template sampling-rate defaulting and video batch-shape handling are different processor bugs." }, { "left": "issue:43824", "right": "issue:43975", "accept": false, - "reason": "ImportError for a Qwen2.5-VL class is unrelated to DeepSeek detokenization regressions." - }, - { - "left": "issue:44442", - "right": "issue:45198", - "accept": false, - "reason": "Tokenizer loading failure for FastSpeech2Conformer is not the same bug as Wav2Vec2 save_pretrained/tokenization failure." + "reason": "Importing a missing class and incorrect detokenization for DeepSeek are unrelated." }, { "left": "issue:45200", "right": "issue:45325", "accept": false, - "reason": "Defaulting mm_token_type_ids to zeros is unrelated to Qwen2.5-VL rope-index scaling." + "reason": "Gemma 4 multimodal token-type defaults and Qwen2.5-VL temporal position-id scaling are different multimodal regressions." }, { - "left": "issue:43526", - "right": "issue:46192", + "left": "issue:44442", + "right": "issue:45198", "accept": false, - "reason": "Label reduction in BeitImageProcessorFast and return_tensors being ignored are different input-processing bugs." + "reason": "FastSpeech2Conformer tokenizer loading and Wav2Vec2 save/tokenization failures are distinct audio-model problems." }, { - "left": "issue:43388", - "right": "issue:43526", + "left": "issue:44479", + "right": "issue:46192", "accept": false, - "reason": "Both involve labels, but one is gather_for_metrics batching and the other is image-processor label reduction; different paths." + "reason": "A video-input regression in Qwen models is unrelated to `return_tensors` being ignored when `text_kwargs` is passed." }, { "left": "issue:44792", "right": "issue:45563", "accept": false, - "reason": "Janus image-generation test failure is unrelated to the stale paged-generate warning." + "reason": "A failed Janus image-generation test and a stale warning for `num_return_sequences` are not the same bug." + }, + { + "left": "issue:43388", + "right": "issue:43526", + "accept": false, + "reason": "Metric gathering drops tuple labels in the last batch, whereas BeitImageProcessorFast reduces labels to a single item; different code paths." }, { "left": "issue:43819", "right": "issue:44112", "accept": false, - "reason": "DAC from_latents/STE mismatch is unrelated to a GraniteSpeech CI device override test." + "reason": "DAC latent reconstruction mismatch and a GraniteSpeech CI test failure are unrelated." }, { "left": "issue:43827", "right": "issue:43901", "accept": false, - "reason": "Both are docs issues, but they concern different pipelines and different stale references." - }, - { - "left": "issue:44610", - "right": "issue:44857", - "accept": false, - "reason": "Different models and failures: processor image size mismatch vs AMP/CUDA crash in LwDetrImageLoss." + "reason": "Both are pipeline docs issues, but they concern different pipelines and different deprecated/changed behaviors." }, { "left": "issue:41720", "right": "issue:46209", "accept": false, - "reason": "Both are hardware-specific inference failures, but they involve different models, devices, and failure modes." - }, - { - "left": "issue:44561", - "right": "issue:46234", - "accept": false, - "reason": "Both mention missing utility helpers, but they are different functions with different breakages." - }, - { - "left": "issue:38175", - "right": "issue:45412", - "accept": false, - "reason": "SigLIP2 probability issues are unrelated to RT-DETR memory release behavior." - }, - { - "left": "issue:43335", - "right": "issue:43550", - "accept": false, - "reason": "SwitchTransformers config construction and Bamba torch.compile/SDPA failure are unrelated." - }, - { - "left": "issue:44206", - "right": "issue:45584", - "accept": false, - "reason": "Feature-extractor argument misuse and Whisper empty-transcription generation failure are unrelated." - }, - { - "left": "issue:43819", - "right": "issue:45083", - "accept": false, - "reason": "DAC latent conversion and qwen3_omni_moe feature-length helper behavior do not describe the same bug." + "reason": "Different hardware/model failures: auto device mapping on A800 versus FP8 inference on B200." }, { - "left": "issue:45759", - "right": "issue:46285", + "left": "issue:44610", + "right": "issue:44857", "accept": false, - "reason": "Composite Qwen config unwrapping and CLIP text-model structure differences are different issues." + "reason": "OmDet-Turbo preprocessing shape mismatch and LwDetrImageLoss AMP/CUDA crash are separate detection-model issues." } ] }, @@ -8711,22 +8639,17 @@ "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-10000-473", "nodes": [ - "issue:26413", "issue:28282", "issue:29127", "issue:29942", - "issue:31501", "issue:32090", "issue:33357", - "issue:34689", - "issue:35141", "issue:37428", - "issue:39692", + "issue:38175", "issue:43278", "issue:43335", "issue:43425", "issue:43504", - "issue:43526", "issue:43550", "issue:43606", "issue:43611", @@ -8736,11 +8659,12 @@ "issue:43825", "issue:43931", "issue:44112", + "issue:44206", "issue:44351", "issue:44448", - "issue:44493", "issue:44514", "issue:44521", + "issue:44561", "issue:44743", "issue:44855", "issue:44857", @@ -8749,25 +8673,35 @@ "issue:45200", "issue:45245", "issue:45290", + "issue:45412", "issue:45520", + "issue:45584", "issue:45588", "issue:45593", "issue:45759", "issue:45859", "issue:46170", + "issue:46234", + "issue:46285", "issue:46326", "issue:46431", "issue:46512", "issue:46566" ], "soft_pairs": [ + "issue:44561|issue:46234", + "issue:38175|issue:45412", + "issue:43335|issue:43550", + "issue:44206|issue:45584", + "issue:45759|issue:46285", + "issue:43819|issue:45083", "issue:44857|issue:45593", "issue:45520|issue:46170", "issue:43606|issue:45083", - "issue:29942|issue:32090", "issue:43335|issue:43653", - "issue:43550|issue:44112", + "issue:29942|issue:32090", "issue:37428|issue:43425", + "issue:43550|issue:44112", "issue:28282|issue:33357", "issue:44448|issue:44743", "issue:43611|issue:43931", @@ -8778,216 +8712,212 @@ "issue:43278|issue:43825", "issue:45290|issue:46326", "issue:46431|issue:46566", - "issue:29127|issue:32090", "issue:43720|issue:44112", - "issue:43335|issue:45083", + "issue:29127|issue:32090", "issue:37428|issue:44351", + "issue:43335|issue:45083", "issue:43335|issue:44112", "issue:45127|issue:45245", "issue:43335|issue:45593", "issue:45588|issue:45859", - "issue:45200|issue:45588", - "issue:39692|issue:43526", - "issue:34689|issue:35141", - "issue:26413|issue:31501", - "issue:43819|issue:44493" + "issue:45200|issue:45588" ], - "prepared_review_unit_hash": "878cb571ebf3527db3c939ad4e0e66522a63fd55ad0d150cd06c0f0e25288187", + "prepared_review_unit_hash": "6c79e578bb7ba8ef8afa85f564e4cb96a1cb26cd8ef4ba6c7f7413a0dd440cd1", "component_budget": { "node_count": 473, "item_count": 473, "soft_pair_count": 1738, - "serialized_chars": 901383, - "estimated_input_tokens": 225346, - "estimated_eval_tokens": 450948 + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 }, "budget": { "node_count": 48, "item_count": 48, - "soft_pair_count": 30, - "serialized_chars": 37039, - "estimated_input_tokens": 9260, - "estimated_eval_tokens": 18776 + "soft_pair_count": 32, + "serialized_chars": 37603, + "estimated_input_tokens": 9401, + "estimated_eval_tokens": 19058 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This is a heterogeneous set of unrelated Transformers bugs spanning model loading, chat templating, device/offload, training, and CI failures. None of the soft pairs look like the same underlying bug or change; they mostly share a broad subsystem or error class, not a concrete code path.", - "confidence": 0.96, - "canonical_issue_reason": null, + "summary": "The cluster is heterogeneous: most items are separate bugs across unrelated subsystems, so none of the soft pairs should be merged as duplicates. The only loosely coherent area is chat-template/multimodal handling, but even there the issues describe different failure modes.", + "confidence": 0.6, + "canonical_issue_reason": "issue:44521 is the best representative issue: it is open, has the most discussion/inbound references, and sits in the most coherent subgroup in this set (chat_template / multimodal input handling).", "canonical_pr_reason": null, - "best_issue_reason": "issue:44521 is the most active and best-referenced item in the set, but it is still not a duplicate representative for the others; the cluster is too diverse for a true canonical issue.", + "best_issue_reason": "issue:44521 is the strongest global issue candidate because it is the most active and broadly scoped within the only clearly related theme in the cluster.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:44561", + "right": "issue:46234", + "accept": false, + "reason": "Both are import-utils-related, but they are different missing symbols and different breakages: `is_torch_fx_available` removal vs missing `is_falcon_mamba_ssm_available`." + }, + { + "left": "issue:38175", + "right": "issue:45412", + "accept": false, + "reason": "Different symptoms and code paths: zero probabilities in SigLIP2 vs RT-DETR memory not released on delete." + }, + { + "left": "issue:43335", + "right": "issue:43550", + "accept": false, + "reason": "One is a config bug in SwitchTransformers layer creation; the other is a Bamba SDPA/torch.compile failure. Not the same bug." + }, + { + "left": "issue:44206", + "right": "issue:45584", + "accept": false, + "reason": "Different feature areas: `LasrFeatureExtractor` center-arg regression vs Whisper empty transcription after `align_special_tokens`." + }, + { + "left": "issue:45759", + "right": "issue:46285", + "accept": false, + "reason": "Both involve model construction, but one is Qwen multimodal `text_config` unwrapping and the other is CLIP text-model structure inconsistency." + }, + { + "left": "issue:43819", + "right": "issue:45083", + "accept": false, + "reason": "Different model-specific helper bugs: DAC `from_latents`/STE mismatch vs qwen3_omni_moe feature-length helper behavior." + }, { "left": "issue:44857", "right": "issue:45593", "accept": false, - "reason": "Both are vision/detection losses, but one is an AMP/CUDA crash in LwDetrImageLoss and the other is D-FINE missing auxiliary losses when denoising is off; different models and different failure modes." + "reason": "Unrelated training-time failures: float16 AMP crash in LwDetrImageLoss vs D-FINE auxiliary-loss handling when denoising is off." }, { "left": "issue:45520", "right": "issue:46170", "accept": false, - "reason": "Both involve import/runtime environment problems, but one is a Python 3.13 flash_attn import key error and the other is a Windows ROCm/torch.distributed dependency issue; not the same bug." + "reason": "Both are import/dependency issues, but they are on different platforms and dependencies: flash_attn on Python 3.13 vs torch.distributed on Windows ROCm." }, { "left": "issue:43606", "right": "issue:45083", "accept": false, - "reason": "CPU offload device mismatch in bark-small vs a helper length-calculation bug in qwen3_omni_moe; no shared code path or concrete fix." - }, - { - "left": "issue:29942", - "right": "issue:32090", - "accept": false, - "reason": "Flash Attention 2 test failures and Trainer _gpu_broadcast_one NoneType errors are unrelated failures in different parts of the stack." + "reason": "Different problems: CPU offload device mismatch in bark-small vs a qwen3_omni_moe helper-function issue." }, { "left": "issue:43335", "right": "issue:43653", "accept": false, - "reason": "SwitchTransformers sparse-layer construction and BigBirdTokenizer special-token registration are distinct bugs in different components." + "reason": "SwitchTransformers sparse-layer creation and BigBirdTokenizer special-token registration are unrelated bugs." }, { - "left": "issue:43550", - "right": "issue:44112", + "left": "issue:29942", + "right": "issue:32090", "accept": false, - "reason": "torch.compile/SDPA failure in Bamba is unrelated to a GraniteSpeech CI test stale device override; same general area of model execution, but not the same issue." + "reason": "Flash Attention 2 test failures and Trainer `_gpu_broadcast_one` NoneType broadcast errors are different code paths." }, { "left": "issue:37428", "right": "issue:43425", "accept": false, - "reason": "Missing flash-support import vs general Torch 2.10 incompatibility; one is a specific symbol/import regression, the other is a version-compatibility report." + "reason": "One is a missing import from flash-attention utils; the other is broad Torch 2.10 incompatibility. Not the same issue." + }, + { + "left": "issue:43550", + "right": "issue:44112", + "accept": false, + "reason": "Bamba SDPA/compile failure and GraniteSpeech CI stale device override are distinct failures." }, { "left": "issue:28282", "right": "issue:33357", "accept": false, - "reason": "AutoModel missing PyTorch and a MacOS bus error on a CLIP model are entirely different failures." + "reason": "Generic torch-missing ImportError vs MacOS bus error with a community CLIP model are not the same underlying bug." }, { "left": "issue:44448", "right": "issue:44743", "accept": false, - "reason": "Pegasus output regression across v4/v5 is not the same as recurrent-state reset in modular_qwen3_5.py; different models and code paths." + "reason": "Pegasus output regression across v4/v5 and Qwen3 recurrent-state reset are separate model behavior bugs." }, { "left": "issue:43611", "right": "issue:43931", "accept": false, - "reason": "Base-model-prefix loading breakage and Qwen3-VL weight-shape mismatch are separate model-loading issues with different root causes." - }, - { - "left": "issue:45759", - "right": "issue:46512", - "accept": false, - "reason": "Both are from_config-related, but one is about unwrapping text_config for composite multimodal configs and the other is about stale config.dtype after dtype initialization." - }, - { - "left": "issue:44514", - "right": "issue:44521", - "accept": false, - "reason": "Both hit apply_chat_template, but one is a batched padding=False crash and the other is all-zero assistant masks for multimodal inputs; different bugs in the same area." - }, - { - "left": "issue:43335", - "right": "issue:43504", - "accept": false, - "reason": "Sparse-layer zero-count behavior in SwitchTransformers is unrelated to Beit pretrained model loading with a legacy field." - }, - { - "left": "issue:44855", - "right": "issue:45083", - "accept": false, - "reason": "Python 3.13 @torch.jit.script parsing/import error is unrelated to qwen3_omni_moe feature-length helper behavior." + "reason": "Base-model-prefix loading breakage and Qwen3-VL weight-shape mismatch are different loading failures." }, { "left": "issue:43278", "right": "issue:43825", "accept": false, - "reason": "Embedding dtype drift during evaluate is unrelated to a pipeline error-message bug about translation support in v5." + "reason": "A dtype-staleness bug in `from_config` is unrelated to a pipeline error-message regression." }, { "left": "issue:45290", "right": "issue:46326", "accept": false, - "reason": "Both involve chat templating, but one is assistant messages with tool calls and no content, while the other is continue_final_message on mistral_common-based tokenizers; not the same concrete defect." + "reason": "Both touch chat template logic, but one crashes on assistant tool-call messages and the other concerns `continue_final_message` for mistral_common tokenizers." }, { "left": "issue:46431", "right": "issue:46566", "accept": false, - "reason": "Missing chat template on a processor and DiffusionGemma meta-device weights under device_map=auto are unrelated." - }, - { - "left": "issue:29127", - "right": "issue:32090", - "accept": false, - "reason": "LayoutLMv3 box-validation messaging and Trainer GPU broadcast NoneType are different errors with no shared fix." + "reason": "Processor chat-template absence and DiffusionGemma meta-device loading are unrelated." }, { "left": "issue:43720", "right": "issue:44112", "accept": false, - "reason": "BitNet packed-weight unpacking during accelerate loading is unrelated to a GraniteSpeech CI stale-device test." + "reason": "BitNet packed-weight unpacking during accelerate loading is not the same as a GraniteSpeech CI device override test failure." }, { - "left": "issue:43335", - "right": "issue:45083", + "left": "issue:29127", + "right": "issue:32090", "accept": false, - "reason": "SwitchTransformers sparse-layer creation and qwen3_omni_moe output-length helper behavior are unrelated." + "reason": "LayoutLMV3 error-message clarity issue vs Trainer broadcast TypeError; different functionality and severity." }, { "left": "issue:37428", "right": "issue:44351", "accept": false, - "reason": "Missing flash-support helper import and missing HybridCache import are both import errors but for different symbols and regressions." - }, - { - "left": "issue:45127", - "right": "issue:45245", - "accept": false, - "reason": "LoRA merge collapse with extended vocabulary is not the same as hitting the 2^24 category limit." + "reason": "Different missing imports: `_flash_supports_window_size` vs `HybridCache`." }, { - "left": "issue:45588", - "right": "issue:45859", + "left": "issue:43335", + "right": "issue:45083", "accept": false, - "reason": "One is a flash_attention.py crash on s_aux=None for sink-less models; the other is CPU-offload incompatibility in Qwen3_5MoeTextRotaryEmbedding. Different layers, different root causes." + "reason": "SwitchTransformers sparse-layer construction and qwen3_omni_moe feature-length helper behavior are unrelated." }, { - "left": "issue:45200", - "right": "issue:45588", + "left": "issue:43335", + "right": "issue:44112", "accept": false, - "reason": "Gemma 4 mm_token_type_ids defaults and flash_attention s_aux=None are unrelated multimodal/attention bugs." + "reason": "Different subsystems and failure modes: model config bug vs CI stale device override." }, { - "left": "issue:39692", - "right": "issue:43526", + "left": "issue:45127", + "right": "issue:45245", "accept": false, - "reason": "Documentation example errors for SigLIP2 are not the same as BeitImageProcessorFast reduce_labels returning only one label." + "reason": "LoRA merge/model collapse with tied embeddings is unrelated to category cardinality limits." }, { - "left": "issue:34689", - "right": "issue:35141", + "left": "issue:43335", + "right": "issue:45593", "accept": false, - "reason": "Llama 3.2 vision model loading regression and output-embedding reinitialization after resizing token embeddings are different model-loading/training behaviors." + "reason": "SwitchTransformers sparse-layer config bug and D-FINE auxiliary-loss omission are different issues." }, { - "left": "issue:26413", - "right": "issue:31501", + "left": "issue:45588", + "right": "issue:45859", "accept": false, - "reason": "Checkpoint-resume failing on an empty epoch iterator and FineWeb training not starting are different training startup failures." + "reason": "Both mention attention-related paths, but one is an `s_aux=None` crash in flash_attention integration and the other is a CPU-offload incompatibility in Qwen3_5MoeTextRotaryEmbedding." }, { - "left": "issue:43819", - "right": "issue:44493", + "left": "issue:45200", + "right": "issue:45588", "accept": false, - "reason": "DAC.from_latents mismatch with missing STE and unexpected position-id keys are unrelated model-behavior regressions." + "reason": "Gemma 4 `mm_token_type_ids` defaulting and a flash_attention `s_aux=None` crash are separate bugs." } ] }, @@ -9013,9 +8943,11 @@ "nodes": [ "issue:10000", "issue:15354", + "issue:26413", "issue:27957", "issue:28218", "issue:29942", + "issue:31501", "issue:32090", "issue:33357", "issue:34689", @@ -9032,8 +8964,8 @@ "issue:43575", "issue:43606", "issue:43701", + "issue:43819", "issue:43854", - "issue:43927", "issue:44206", "issue:44360", "issue:44448", @@ -9051,8 +8983,6 @@ "issue:45381", "issue:45459", "issue:45478", - "issue:45588", - "issue:45698", "issue:45823", "issue:45920", "issue:46192", @@ -9061,24 +8991,28 @@ "issue:46655" ], "soft_pairs": [ + "issue:39692|issue:43526", + "issue:34689|issue:35141", + "issue:26413|issue:31501", "issue:42490|issue:43854", + "issue:43819|issue:44493", "issue:45459|issue:46234", "issue:45310|issue:45478", "issue:43606|issue:44857", - "issue:39692|issue:44493", "issue:44448|issue:44779", + "issue:39692|issue:44493", "issue:44206|issue:45823", "issue:28218|issue:32090", - "issue:44857|issue:45072", "issue:44360|issue:44561", + "issue:44857|issue:45072", "issue:43701|issue:45083", "issue:38175|issue:44493", "issue:32090|issue:33357", "issue:42490|issue:43575", - "issue:40444|issue:43526", "issue:45290|issue:45381", - "issue:29942|issue:35141", "issue:43550|issue:45083", + "issue:29942|issue:35141", + "issue:40444|issue:43526", "issue:45290|issue:46655", "issue:45920|issue:46192", "issue:44448|issue:44977", @@ -9087,221 +9021,231 @@ "issue:44977|issue:45229", "issue:44855|issue:46399", "issue:10000|issue:15354", - "issue:43278|issue:43381", - "issue:44857|issue:45588", - "issue:43927|issue:45698" + "issue:43278|issue:43381" ], - "prepared_review_unit_hash": "49615afe91a44e432bd9685a955286a41cb35831acb77a3e2c0d2a7f4f60627a", + "prepared_review_unit_hash": "efe77f0dc5ce3529c0e120cd5027a3acd33548e460256645020ebad854c99f93", "component_budget": { "node_count": 473, "item_count": 473, "soft_pair_count": 1738, - "serialized_chars": 901383, - "estimated_input_tokens": 225346, - "estimated_eval_tokens": 450948 + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 }, "budget": { "node_count": 48, "item_count": 48, - "soft_pair_count": 29, - "serialized_chars": 36423, - "estimated_input_tokens": 9106, - "estimated_eval_tokens": 18468 + "soft_pair_count": 31, + "serialized_chars": 36984, + "estimated_input_tokens": 9246, + "estimated_eval_tokens": 18748 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly unrelated issue reports. I agree there is one clear duplicate pair: issue:45310 and issue:45478 report the same Qwen3.5 MoE from_pretrained regression. Among the pair, issue:45310 is the earlier report, while issue:45478 has slightly more discussion activity; no pull requests are present.", - "confidence": 0.83, - "canonical_issue_reason": "issue:45310 is a reasonable canonical for the duplicate pair because it is the earlier report and the titles/body describe the same Qwen3.5 MoE from_pretrained failure as issue:45478.", + "summary": "This cluster is mostly heterogeneous: nearly all pairs are unrelated bug reports that only share broad Transformer/regression language. The only clear duplicate is the repeated Qwen3.5 MoE `from_pretrained` issue (45310/45478).", + "confidence": 0.9, + "canonical_issue_reason": "issue:45310 is the earlier, fuller report of the same Qwen3.5 MoE `from_pretrained` regression as issue:45478, so it is the better canonical issue for that duplicate pair.", "canonical_pr_reason": null, - "best_issue_reason": "issue:45478 is the strongest representative issue in this set because it has slightly more discussion activity than issue:45310 and describes the same underlying Qwen3.5 MoE from_pretrained regression.", + "best_issue_reason": "issue:45920 is the broadest and most representative standalone issue in the set: it describes a multi-model tokenizer regression, is open, and already has inbound references.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:39692", + "right": "issue:43526", + "accept": false, + "reason": "SigLIP2 doc-example errors and BEiT label reduction are different components and failure modes." + }, + { + "left": "issue:34689", + "right": "issue:35141", + "accept": false, + "reason": "Llama 3.2 model-loading failure is unrelated to token embedding resizing/post_init reinitialization." + }, + { + "left": "issue:26413", + "right": "issue:31501", + "accept": false, + "reason": "A checkpoint-resume epoch_iterator bug is not the same as FineWeb SLM startup failure." + }, { "left": "issue:42490", "right": "issue:43854", "accept": false, - "reason": "Both involve loading, but one is a trailing-slash import warning and the other is a model-loading test failure; different bugs." + "reason": "Local-path trailing-slash warning and GLM-4.7-Flash loading correctness are different loading issues." + }, + { + "left": "issue:43819", + "right": "issue:44493", + "accept": false, + "reason": "DAC `from_latents` mismatch and position-id key warnings concern different model paths." }, { "left": "issue:45459", "right": "issue:46234", "accept": false, - "reason": "Tokenizer error masking with missing protobuf is unrelated to a missing import_utils helper." + "reason": "Tokenizer decode-error masking with missing protobuf is unrelated to a missing import-utils availability symbol." }, { "left": "issue:45310", "right": "issue:45478", "accept": true, - "reason": "Near-identical titles and descriptions indicate the same Qwen3.5 MoE from_pretrained regression." + "reason": "Same Qwen3.5 MoE `from_pretrained` regression, with matching title and symptom; 45478 is a later duplicate." }, { "left": "issue:43606", "right": "issue:44857", "accept": false, - "reason": "CPU offload device mismatch and AMP/CUDA loss crash are different failure modes and code paths." + "reason": "CPU offload device mismatch for `suno/bark-small` is not the same bug as AMP/CUDA float16 loss failure." }, { - "left": "issue:39692", - "right": "issue:44493", + "left": "issue:44448", + "right": "issue:44779", "accept": false, - "reason": "SigLIP2 doc-example failures are unrelated to the position-id key regression." + "reason": "Pegasus v4/v5 output drift and DeepSeek tokenizer regression are both regressions, but not the same code-path bug." }, { - "left": "issue:44448", - "right": "issue:44779", + "left": "issue:39692", + "right": "issue:44493", "accept": false, - "reason": "Both are v5 regressions, but they affect different models and different code paths." + "reason": "SigLIP2 documentation errors do not match the generic position-id key issue affecting many models." }, { "left": "issue:44206", "right": "issue:45823", "accept": false, - "reason": "Unsupported feature-extractor arg and Gemma4 device mismatch are unrelated issues." + "reason": "Unsupported `center` arg in feature extraction is unrelated to Gemma4 PLE device placement mismatch." }, { "left": "issue:28218", "right": "issue:32090", "accept": false, - "reason": "Tokenizer spacing behavior is unrelated to the Trainer broadcast error." + "reason": "Added-token spacing behavior and `_gpu_broadcast_one` NoneType errors are unrelated tokenizer vs trainer bugs." }, { - "left": "issue:44857", - "right": "issue:45072", + "left": "issue:44360", + "right": "issue:44561", "accept": false, - "reason": "Both are device/precision related, but they are separate model-specific crashes with different causes." + "reason": "A missing ReLU in DSA indexer is unrelated to removal of `is_torch_fx_available` breaking `trust_remote_code` models." }, { - "left": "issue:44360", - "right": "issue:44561", + "left": "issue:44857", + "right": "issue:45072", "accept": false, - "reason": "A missing ReLU in DSA indexer is unrelated to removal of is_torch_fx_available." + "reason": "Both mention dtype/precision, but they affect different models and different execution paths." }, { "left": "issue:43701", "right": "issue:45083", "accept": false, - "reason": "Checkpoint key mismatch and multimodal feature-length helper behavior are different bugs." + "reason": "Checkpoint key mismatch and qwen3_omni_moe feature-length behavior are distinct issues." }, { "left": "issue:38175", "right": "issue:44493", "accept": false, - "reason": "Unexpected zero probabilities for SigLIP2 is not the same as the key-with-position-id warning." + "reason": "Zero probabilities in SigLIP2 are not the same as unexpected position-id key handling." }, { "left": "issue:32090", "right": "issue:33357", "accept": false, - "reason": "Trainer NoneType broadcast error and MacOS bus error are distinct failures." + "reason": "Trainer broadcast NoneType failure and MacOS bus error are different runtime failures." }, { "left": "issue:42490", "right": "issue:43575", "accept": false, - "reason": "Local-path warning/import behavior is unrelated to Qwen2 tensor-parallel OOM." + "reason": "Import warning for a trailing slash is unrelated to TP load OOM for Qwen2-57B." }, { - "left": "issue:40444", - "right": "issue:43526", + "left": "issue:45290", + "right": "issue:45381", "accept": false, - "reason": "Multi-image VLM finetuning failure and BeitImageProcessorFast label reduction bug are unrelated." + "reason": "Chat-template tool-call crash and qwen2.5-vl video position-id bug are different multimodal code paths." }, { - "left": "issue:45290", - "right": "issue:45381", + "left": "issue:43550", + "right": "issue:45083", "accept": false, - "reason": "Chat-template tool-call handling and Qwen2.5-VL video position-id bug are different multimodal paths." + "reason": "Bamba SDPA compile failure and qwen3_omni_moe feature-length behavior are unrelated." }, { "left": "issue:29942", "right": "issue:35141", "accept": false, - "reason": "Flash Attention 2 test failures are unrelated to embedding reinitialization after resize." + "reason": "Flash Attention 2 test failures are not the same bug as embedding reinitialization after resize." }, { - "left": "issue:43550", - "right": "issue:45083", + "left": "issue:40444", + "right": "issue:43526", "accept": false, - "reason": "Bamba torch.compile/SDPA failure is unrelated to qwen3_omni_moe output-length helper behavior." + "reason": "Qwen2.5-VL iterable-dataset finetuning failure and BEiT label reduction are unrelated." }, { "left": "issue:45290", "right": "issue:46655", "accept": false, - "reason": "Chat-template tool-call crash and SmolVLM video padding bug are different issues." + "reason": "Chat template tool-call handling and SmolVLM video padding are different bugs." }, { "left": "issue:45920", "right": "issue:46192", "accept": false, - "reason": "Tokenizer ID regressions across models are unrelated to return_tensors being ignored with text_kwargs." + "reason": "A v5 tokenizer wrong-token-ID regression is broader and different from `return_tensors` being ignored when `text_kwargs` is passed." }, { "left": "issue:44448", "right": "issue:44977", "accept": false, - "reason": "Pegasus v4/v5 output divergence and Qwen3.5 flash-attention generation failure are separate model-specific issues." + "reason": "Pegasus generation drift and Qwen3.5 flash-attention generation failure are distinct model-specific problems." }, { "left": "issue:27957", "right": "issue:34689", "accept": false, - "reason": "XLMRoberta Flash Attention 2 and Llama 3.2 Vision model loading are not the same bug." + "reason": "XLMRoberta flash-attention support and Llama 3.2 model loading are unrelated." }, { "left": "issue:43278", "right": "issue:43329", "accept": false, - "reason": "Embedding dtype drift in eval and undefined video-token helper references are unrelated." + "reason": "BF16-to-FP32 dtype change in evaluation is unrelated to the missing video helper functions in multimodal token counting." }, { "left": "issue:44977", "right": "issue:45229", "accept": false, - "reason": "Qwen3.5 flash-attention generation bug is unrelated to Gemma4 multi-GPU OOM." + "reason": "Qwen3.5 flash-attention generation failure is unrelated to Gemma4 multi-GPU CUDA OOM." }, { "left": "issue:44855", "right": "issue:46399", "accept": false, - "reason": "Python 3.13 import parsing error and Qwen3.5 kernel registration error are different decorator-related failures." + "reason": "Python 3.13 `IndentationError` in a scripted model import is not the same as a missing decorator for rotary-pos-emb registration." }, { "left": "issue:10000", "right": "issue:15354", "accept": false, - "reason": "German DistilBertModel issue and GeneratorExp torch.jit.script export failure are unrelated." + "reason": "German DistilBertModel issue and GeneratorExp/JIT export failure are unrelated." }, { "left": "issue:43278", "right": "issue:43381", "accept": false, - "reason": "Training/eval dtype mismatch is unrelated to gradient checkpointing in eval mode." - }, - { - "left": "issue:44857", - "right": "issue:45588", - "accept": false, - "reason": "AMP/CUDA loss crash and flash_attention.py s_aux=None AttributeError are different crash sites." - }, - { - "left": "issue:43927", - "right": "issue:45698", - "accept": false, - "reason": "DiaConfig token-id persistence and wrong custom-module loading are both serialization-related, but not the same underlying bug." + "reason": "Embedding dtype drift in eval and gradient checkpointing in eval mode are different concerns." } ] }, "evaluator_result": { - "accept": false, - "feedback": "The duplicate-pair judgment on issue:45310 vs issue:45478 looks fine, but the writeup is not fully grounded: `best_issue_reason` says 45310 has the most context and discussion, which the packet does not support (45478 has more discussion_activity, and the body sizes are similar). I\u2019d accept the duplicate detection itself, but reject the summary because it overstates the evidence." + "accept": true, + "feedback": "The summary is well grounded in the packet: the cluster is indeed mostly heterogeneous, and the duplicate Qwen3.5 MoE from_pretrained pair (45310/45478) is the only clearly supported duplicate. The soft-edge calls are conservative overall and do not overstate equivalence between different bugs." }, "evaluator_used": true, - "retried": true, + "retried": false, "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null @@ -9341,17 +9285,17 @@ "issue:43701", "issue:43746", "issue:43873", + "issue:43927", "issue:43976", "issue:44186", - "issue:44206", "issue:44361", "issue:44448", "issue:44466", - "issue:44479", "issue:44561", "issue:44568", "issue:44743", "issue:44779", + "issue:44857", "issue:44871", "issue:44977", "issue:45198", @@ -9362,237 +9306,245 @@ "issue:45491", "issue:45563", "issue:45588", + "issue:45698", "issue:45803", "issue:45823", "issue:46121", "issue:46209" ], "soft_pairs": [ + "issue:44857|issue:45588", + "issue:43927|issue:45698", "issue:43746|issue:46209", "issue:44561|issue:44779", "issue:37428|issue:42175", "issue:43873|issue:45229", "issue:32090|issue:33290", "issue:25251|issue:32090", - "issue:41720|issue:45229", "issue:43976|issue:45440", "issue:44448|issue:44466", + "issue:41720|issue:45229", "issue:42490|issue:43873", "issue:44568|issue:45491", "issue:45588|issue:45823", "issue:44977|issue:45563", + "issue:45372|issue:45588", "issue:43278|issue:43575", "issue:33357|issue:39692", - "issue:29127|issue:34567", - "issue:45372|issue:45588", "issue:45198|issue:45803", + "issue:29127|issue:34567", "issue:43412|issue:45588", - "issue:43381|issue:43701", "issue:44871|issue:44977", + "issue:43381|issue:43701", "issue:43653|issue:44743", "issue:44361|issue:45198", "issue:44186|issue:45198", "issue:43611|issue:46121", "issue:45372|issue:45406", - "issue:42915|issue:42994", - "issue:44206|issue:44479" + "issue:42915|issue:42994" ], - "prepared_review_unit_hash": "c66c1d739706e0c7d82f102b982a947ae9fd582fbf36bd8bfd47ccc415b75b2d", + "prepared_review_unit_hash": "f28ff716a012f25e1ee77cc95c817c29d9e0f29a322327198b801aed4782b135", "component_budget": { "node_count": 473, "item_count": 473, "soft_pair_count": 1738, - "serialized_chars": 901383, - "estimated_input_tokens": 225346, - "estimated_eval_tokens": 450948 + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 }, "budget": { - "node_count": 47, - "item_count": 47, - "soft_pair_count": 28, - "serialized_chars": 35691, - "estimated_input_tokens": 8923, - "estimated_eval_tokens": 18102 + "node_count": 48, + "item_count": 48, + "soft_pair_count": 29, + "serialized_chars": 36517, + "estimated_input_tokens": 9130, + "estimated_eval_tokens": 18516 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is not a true duplicate group: the issues span unrelated bugs across tokenizers, quantization/offloading, Gemma/Qwen regressions, training/runtime errors, and packaging/import problems. All proposed soft edges look superficially similar at the subsystem level but do not share the same concrete bug.", - "confidence": 0.96, - "canonical_issue_reason": "No single issue is a clear canonical representative because the set is heterogeneous and the similarity links are only broad topical overlap.", + "summary": "The soft pairs are thematic false positives rather than true duplicates; each pair targets a different bug surface, so there is no safe canonical issue or PR for this cluster.", + "confidence": 0.95, + "canonical_issue_reason": "No single issue is a defensible canonical: the cluster spans unrelated regressions across tokenizers, training, inference, packaging, and docs, and the soft pairs do not share the same concrete bug.", "canonical_pr_reason": null, - "best_issue_reason": "If one issue must represent the cluster, issue 43873 is the broadest and most actively discussed problem, but it is still not a duplicate of the others.", + "best_issue_reason": "No issue is a strong global representative because the items are not one underlying problem; any pick would be arbitrary and misleading for dedup triage.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:44857", + "right": "issue:45588", + "accept": false, + "reason": "Different failures: one is an AMP/CUDA crash in LwDetrImageLoss, the other is a flash-attention AttributeError on `s_aux=None`." + }, + { + "left": "issue:43927", + "right": "issue:45698", + "accept": false, + "reason": "Both involve save/load behavior, but one is DiaConfig custom token IDs and the other is wrong custom module resolution after `save_pretrained`; different root causes." + }, { "left": "issue:43746", "right": "issue:46209", "accept": false, - "reason": "Different failures: PEFT local checkpoint loading vs FP8 inference on B200." + "reason": "Unrelated problems: PEFT adapter loading from local checkpoints vs FP8 inference failure on B200." }, { "left": "issue:44561", "right": "issue:44779", "accept": false, - "reason": "Unrelated regressions: removed helper breaking trust_remote_code vs DeepSeek tokenizer output correctness." + "reason": "Both are v5 regressions, but one is a removed utility breaking remote-code models and the other is a tokenizer output bug for DeepSeek; not the same code path." }, { "left": "issue:37428", "right": "issue:42175", "accept": false, - "reason": "Import error for flash attention utility vs TensorFlow missing from torch-only install extras." + "reason": "Different areas entirely: flash-attention import failure vs missing TensorFlow in a torch-only install." }, { "left": "issue:43873", "right": "issue:45229", "accept": false, - "reason": "Both involve memory/quantization themes, but one is offloading behavior and the other is multi-GPU OOM on a specific model." + "reason": "Both mention memory/offloading, but one is quantized offloading behavior and the other is multi-GPU CUDA OOM for Gemma4; not the same fix." }, { "left": "issue:32090", "right": "issue:33290", "accept": false, - "reason": "Trainer broadcast NoneType error is unrelated to Adafactor+DeepSpeed OOM." + "reason": "Trainer broadcast TypeError and DeepSpeed Adafactor OOM are distinct training failures with different mechanisms." }, { "left": "issue:25251", "right": "issue:32090", "accept": false, - "reason": "Pipeline top_k nesting bug is unrelated to Trainer GPU broadcast failure." - }, - { - "left": "issue:41720", - "right": "issue:45229", - "accept": false, - "reason": "Different models and root causes: auto device mapping assert vs Gemma4 OOM." + "reason": "Pipeline top_k nesting behavior is unrelated to the Trainer `_gpu_broadcast_one` TypeError." }, { "left": "issue:43976", "right": "issue:45440", "accept": false, - "reason": "Python version compatibility issue is unrelated to DeepSeekV3MoE divergence from remote implementation." + "reason": "Python version compatibility issue vs DeepSeekV3MoE divergence from remote implementation; no shared underlying bug." }, { "left": "issue:44448", "right": "issue:44466", "accept": false, - "reason": "Generation output mismatch and lm_head serialization inconsistency are different code paths." + "reason": "Both are v5 behavior changes, but one is generation/output mismatch and the other is serialization of tied weights; different concrete bugs." + }, + { + "left": "issue:41720", + "right": "issue:45229", + "accept": false, + "reason": "Both are large-model inference failures, but one is a cudaErrorAssert on Qwen3 auto device mapping and the other is CUDA OOM on Gemma4 multi-GPU." }, { "left": "issue:42490", "right": "issue:43873", "accept": false, - "reason": "Path-trailing-slash warning is unrelated to quantization offloading behavior." + "reason": "Import warning for trailing slash paths is unrelated to quantized offloading behavior." }, { "left": "issue:44568", "right": "issue:45491", "accept": false, - "reason": "Tokenizer special-token bug is unrelated to Gemma3 NaN embeddings with sliding-window attention." + "reason": "Tokenizer special-token handling and Gemma3 NaN embeddings with mixed-length batching are different issues." }, { "left": "issue:45588", "right": "issue:45823", "accept": false, - "reason": "Flash-attention s_aux=None crash is different from Gemma4 device-placement mismatch." + "reason": "Both touch Gemma4/attention, but one is a flash-attention crash and the other is a device-mapping mismatch in PLE forward; not mergeable as one fix." }, { "left": "issue:44977", "right": "issue:45563", "accept": false, - "reason": "Flash-attention generation regression is unrelated to a stale warning in paged generate()." + "reason": "Flash-attention generation bug and a stale warning in paged generate() are unrelated." + }, + { + "left": "issue:45372", + "right": "issue:45406", + "accept": false, + "reason": "Gemma4 processor loading fails for different reasons: missing `ReasoningEffort` import vs missing `_tokenizer` on serve; same family, different bugs." }, { "left": "issue:43278", "right": "issue:43575", "accept": false, - "reason": "Training/eval dtype change is unrelated to tensor-parallel OOM when loading Qwen2-57B." + "reason": "Embedding dtype drift during eval is unrelated to tensor-parallel OOM on Qwen2-57B." }, { "left": "issue:33357", "right": "issue:39692", "accept": false, - "reason": "MacOS bus error on CLIP is not the same as SigLIP2 docs/model-processor mismatch and quantization failure." - }, - { - "left": "issue:29127", - "right": "issue:34567", - "accept": false, - "reason": "LayoutLMv3 error message clarity and TrainerState token counting are unrelated issues." + "reason": "CLIP bus error on macOS and SigLIP2 doc example errors are different issues, with no shared code-path evidence." }, { - "left": "issue:45372", - "right": "issue:45588", + "left": "issue:45198", + "right": "issue:45803", "accept": false, - "reason": "Gemma4 processor import dependency failure is unrelated to flash_attention s_aux handling." + "reason": "Wav2Vec2 save/tokenization failure and Fuyu batch tensor conversion swallowing errors are distinct tokenizer/data-processing bugs." }, { - "left": "issue:45198", - "right": "issue:45803", + "left": "issue:29127", + "right": "issue:34567", "accept": false, - "reason": "Wav2Vec2 save/tokenization failure is unrelated to Fuyu's bare-except exception handling." + "reason": "LayoutLMv3 error-message clarity and TrainerState token-count tracking are unrelated." }, { "left": "issue:43412", "right": "issue:45588", "accept": false, - "reason": "Executorch export/training-guard and unsupported erfinv are unrelated to flash_attention s_aux crashes." + "reason": "Executorch export failures for Gemma3n and a flash-attention crash on sink-less models do not share the same underlying bug." }, { - "left": "issue:43381", - "right": "issue:43701", + "left": "issue:44871", + "right": "issue:44977", "accept": false, - "reason": "Eval-mode gradient checkpointing limitation is unrelated to resume_from_checkpoint key mismatch." + "reason": "Gemma3 EOS token config mismatch and Qwen3.5 flash-attention generation failure are separate model-specific issues." }, { - "left": "issue:44871", - "right": "issue:44977", + "left": "issue:43381", + "right": "issue:43701", "accept": false, - "reason": "Gemma eos_token_id config inconsistency is unrelated to Qwen3.5 flash-attention generation failure." + "reason": "Eval-mode checkpointing restriction and resume checkpoint key mismatch are different training-state problems." }, { "left": "issue:43653", "right": "issue:44743", "accept": false, - "reason": "BigBirdTokenizer special-token registration bug is unrelated to Qwen recurrent-state reset with cache." + "reason": "BigBirdTokenizer special-token registration and Qwen3.5 recurrent-state reset with cache are unrelated." }, { "left": "issue:44361", "right": "issue:45198", "accept": false, - "reason": "MLuke tokenizer AttributeError is unrelated to Wav2Vec2 save_pretrained/tokenization failure." + "reason": "Both affect tokenizers, but MLuke task AttributeError and Wav2Vec2 save/tokenization failure are different bugs." }, { "left": "issue:44186", "right": "issue:45198", "accept": false, - "reason": "LayoutLMv2 NER/padding crash is unrelated to Wav2Vec2 save/tokenization issues." + "reason": "LayoutLMv2 NER/batching crash and Wav2Vec2 save/tokenization failure are not the same code path." }, { "left": "issue:43611", "right": "issue:46121", "accept": false, - "reason": "base_model_prefix loading regression is unrelated to rope-parameter validation TypeError." + "reason": "base_model_prefix loading regression and rope-validation TypeError are unrelated implementation bugs." }, { "left": "issue:45372", - "right": "issue:45406", + "right": "issue:45588", "accept": false, - "reason": "Both mention Gemma4 processors, but one is an import dependency problem and the other is a missing _tokenizer attribute in serve." + "reason": "Gemma4 processor import failure and flash-attention `s_aux=None` crash are separate failures even though both mention Gemma4/attention." }, { "left": "issue:42915", "right": "issue:42994", "accept": false, - "reason": "Quantized inference with FineGrainedFP8Config is not the same as quantized model saving failure." - }, - { - "left": "issue:44206", - "right": "issue:44479", - "accept": false, - "reason": "Feature-extractor center-arg regression and Qwen video-input regression are different model/code-path failures." + "reason": "Qwen3Moe FP8 config failure and quantized model saving failure are different quantization-related problems, not the same bug." } ] }, @@ -9620,7 +9572,6 @@ "issue:29127", "issue:29942", "issue:30064", - "issue:33290", "issue:33357", "issue:35141", "issue:42490", @@ -9640,11 +9591,13 @@ "issue:43931", "issue:43994", "issue:44186", + "issue:44206", "issue:44265", "issue:44279", "issue:44387", "issue:44442", "issue:44462", + "issue:44479", "issue:45072", "issue:45198", "issue:45245", @@ -9665,9 +9618,10 @@ "issue:46655" ], "soft_pairs": [ + "issue:44206|issue:44479", + "issue:45459|issue:45712", "issue:43065|issue:45412", "issue:29942|issue:30064", - "issue:45459|issue:45712", "issue:43065|issue:43994", "issue:43756|issue:45440", "issue:43653|issue:44442", @@ -9676,248 +9630,247 @@ "issue:45245|issue:45563", "issue:29127|issue:33357", "issue:44265|issue:45198", - "issue:43994|issue:45412", "issue:43066|issue:44462", + "issue:43994|issue:45412", "issue:27957|issue:29942", "issue:45440|issue:45593", "issue:43575|issue:43701", - "issue:29127|issue:35141", "issue:43421|issue:45491", + "issue:29127|issue:35141", "issue:42490|issue:43381", "issue:43526|issue:45834", "issue:43575|issue:44387", "issue:44442|issue:45072", - "issue:43526|issue:44186", "issue:42915|issue:43931", "issue:43994|issue:46369", "issue:43526|issue:46655", + "issue:43526|issue:44186", "issue:43824|issue:44279", "issue:46121|issue:46399", - "issue:27957|issue:29127", "issue:45250|issue:46640", - "issue:43334|issue:43531", - "issue:27957|issue:33290" + "issue:27957|issue:29127", + "issue:43334|issue:43531" ], - "prepared_review_unit_hash": "76e0b209e70719b1ee584e99c29249b0feda7f3542bb3d5da1de2d6850c5d1c6", + "prepared_review_unit_hash": "68c19df0be18907903005f8681cc4b99d7c3be2e012a7b2aef0b6d01fd8dec88", "component_budget": { "node_count": 473, "item_count": 473, "soft_pair_count": 1738, - "serialized_chars": 901383, - "estimated_input_tokens": 225346, - "estimated_eval_tokens": 450948 + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 }, "budget": { - "node_count": 47, - "item_count": 47, + "node_count": 48, + "item_count": 48, "soft_pair_count": 32, - "serialized_chars": 36504, - "estimated_input_tokens": 9126, - "estimated_eval_tokens": 18508 + "serialized_chars": 37148, + "estimated_input_tokens": 9287, + "estimated_eval_tokens": 18830 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a set of unrelated issues that only share broad keywords like tokenizer, memory, or Flash Attention 2. None of the soft edges look like the same underlying bug/change, so I would not merge any of them.", - "confidence": 0.93, - "canonical_issue_reason": "No safe canonical issue: the items span multiple unrelated subsystems and the soft pairs are only surface-level topic matches, not the same concrete defect.", + "summary": "Most soft matches are broad-topic similarities rather than true duplicates. The only noticeable mini-theme is Flash Attention 2, but even there the issues describe different concrete failures, so I would not merge any soft pair.", + "confidence": 0.88, + "canonical_issue_reason": "issue:29942 is the clearest actionable Flash Attention 2 bug report in this set and the best representative of the only loosely related subgroup.", "canonical_pr_reason": null, - "best_issue_reason": "No single issue is a good representative for the cluster because there is no clear duplicate group to anchor on.", + "best_issue_reason": "issue:29942 is the most representative issue here because it is concrete and broad enough to serve as a reference for the FA2-related subset, even though the overall cluster is heterogeneous.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:44206", + "right": "issue:44479", + "accept": false, + "reason": "Different failures: unsupported feature-processor arg vs Qwen video-input regression; same area, not the same bug." + }, + { + "left": "issue:45459", + "right": "issue:45712", + "accept": false, + "reason": "Tokenizer error masking from missing protobuf is unrelated to leftover dummy classes leaking into exports." + }, { "left": "issue:43065", "right": "issue:45412", "accept": false, - "reason": "Different problems: Sam3PixelDecoder dummy conv vs RT-DETR memory leak." + "reason": "Dummy conv in Sam3PixelDecoder and RT-DETR memory leak are unrelated code paths and symptoms." }, { "left": "issue:29942", "right": "issue:30064", "accept": false, - "reason": "Both mention vision-model regressions, but Flash Attention 2 test failures and void segmentation-map processing are unrelated bugs." - }, - { - "left": "issue:45459", - "right": "issue:45712", - "accept": false, - "reason": "Tokenizer error masking vs leftover dummy classes; different code paths and fixes." + "reason": "Flash Attention 2 test failures and void segmentation-map processing are unrelated issues." }, { "left": "issue:43065", "right": "issue:43994", "accept": false, - "reason": "Sam3PixelDecoder dummy module and Siglip2 bad outputs are separate model issues." + "reason": "Sam3PixelDecoder dummy layer vs SigLIP2 bad outputs are different model bugs." }, { "left": "issue:43756", "right": "issue:45440", "accept": false, - "reason": "Both are model-implementation mismatches, but for different models and mechanisms (RoPE layers vs DeepSeekV3 divergence)." + "reason": "Different model implementations and failure modes; no shared concrete bug." }, { "left": "issue:43653", "right": "issue:44442", "accept": false, - "reason": "BigBirdTokenizer special-token registration and AutoTokenizer loading FastSpeech2ConformerTokenizer are distinct tokenizer bugs." + "reason": "BigBirdTokenizer special-token registration and FastSpeech2ConformerTokenizer loading are separate tokenizer issues." }, { "left": "issue:45372", "right": "issue:45459", "accept": false, - "reason": "Gemma processor import failure and protobuf-masked tokenizer errors do not describe the same defect." + "reason": "Dependency import failure in Gemma 4 processor loading is unrelated to protobuf error masking in tokenizers." }, { "left": "issue:45412", "right": "issue:46369", "accept": false, - "reason": "Both involve memory, but one is a deletion leak and the other is an overflow when loading a specific model." + "reason": "Both mention memory, but one is a deletion/GC leak and the other is an overflow when loading a model; not the same bug." }, { "left": "issue:45245", "right": "issue:45563", "accept": false, - "reason": "Category-limit runtime error and stale generate warning are unrelated." + "reason": "Category-limit runtime error and stale paged-generate warning are unrelated." }, { "left": "issue:29127", "right": "issue:33357", "accept": false, - "reason": "LayoutLMv3 box-validation messaging and CLIP bus error on macOS are unrelated." + "reason": "LayoutLMv3 box validation error and MacOS bus error on CLIP are unrelated." }, { "left": "issue:44265", "right": "issue:45198", "accept": false, - "reason": "torch.export exportability failure and Wav2Vec2 save/tokenization failure are different code paths." + "reason": "torch.export failure with torch_compilable_check is unrelated to Wav2Vec2 save/tokenization failure." }, { - "left": "issue:43994", - "right": "issue:45412", + "left": "issue:43066", + "right": "issue:44462", "accept": false, - "reason": "Siglip2 incorrect outputs and RT-DETR memory leak are not the same bug." + "reason": "Wrong tokenizer decoder type in v5 and AutoTokenizer ignoring tokenizer.json are different loading/configuration bugs." }, { - "left": "issue:43066", - "right": "issue:44462", + "left": "issue:43994", + "right": "issue:45412", "accept": false, - "reason": "Wrong tokenizer decoder type and ignoring tokenizer.json are different AutoTokenizer behaviors." + "reason": "SigLIP2 nonsensical outputs and RT-DETR memory release issues are unrelated." }, { "left": "issue:27957", "right": "issue:29942", "accept": false, - "reason": "XLMRoberta Flash Attention 2 support and Flash Attention 2 test failures are only broadly related by topic, not the same concrete issue." + "reason": "Both touch Flash Attention 2, but one is XLMRoberta-specific support and the other is FA2 test failures; not the same concrete bug." }, { "left": "issue:45440", "right": "issue:45593", "accept": false, - "reason": "DeepSeekV3 implementation divergence and D-FINE auxiliary-loss behavior are unrelated model bugs." + "reason": "DeepseekV3 divergence and D-FINE auxiliary-loss behavior are different issues." }, { - "left": "issue:43575", - "right": "issue:43701", - "accept": false, - "reason": "Tensor-parallel OOM and resume_from_checkpoint key mismatch are unrelated." - }, - { - "left": "issue:29127", - "right": "issue:35141", + "left": "issue:43575", + "right": "issue:43701", "accept": false, - "reason": "LayoutLMv3 error clarity and embedding reinitialization after resize are different defects." + "reason": "Tensor-parallel OOM when loading a model and resume-from-checkpoint key mismatch are unrelated." }, { "left": "issue:43421", "right": "issue:45491", "accept": false, - "reason": "Runtime post-processor updates and Gemma3 NaN embeddings are different tokenizer/model issues." + "reason": "Tokenizer post-processor update behavior and Gemma3 NaN embeddings come from different layers of the stack." + }, + { + "left": "issue:29127", + "right": "issue:35141", + "accept": false, + "reason": "Error-message clarity for LayoutLMv3 boxes and embedding reinitialization after resize are unrelated." }, { "left": "issue:42490", "right": "issue:43381", "accept": false, - "reason": "Trailing-slash import warning and gradient-checkpointing-in-eval limitation are unrelated." + "reason": "Import warning for trailing-slash paths and gradient-checkpointing eval-mode restriction are unrelated." }, { "left": "issue:43526", "right": "issue:45834", "accept": false, - "reason": "BeitImageProcessorFast label reduction and Kosmos2.5 long-OCR index error are unrelated." + "reason": "BEiT image label reduction bug and Kosmos2.5 OCR index error are different processor problems." }, { "left": "issue:43575", "right": "issue:44387", "accept": false, - "reason": "Tensor-parallel OOM for one Qwen model and int4 CUDA reserved-memory regression are different memory issues." + "reason": "Both involve OOM, but the causes differ: tensor-parallel loading vs increased reserved memory under int4 quantization." }, { "left": "issue:44442", "right": "issue:45072", "accept": false, - "reason": "FastSpeech2Conformer tokenizer loading and bf16 dtype mismatches are unrelated." - }, - { - "left": "issue:43526", - "right": "issue:44186", - "accept": false, - "reason": "Both are tokenizer/processor bugs, but different classes of failure and different model families." + "reason": "Tokenizer loading failure and bfloat16 dtype mismatch are unrelated." }, { "left": "issue:42915", "right": "issue:43931", "accept": false, - "reason": "Qwen3Moe FP8 config failure and Qwen3-VL weight-shape mismatch are separate Qwen3 issues." + "reason": "FineGrainedFP8Config failure for Qwen3Moe and Qwen3-VL weight-shape mismatch are different issues." }, { "left": "issue:43994", "right": "issue:46369", "accept": false, - "reason": "Siglip2 bad outputs and Sam3-LiteText memory overflow are unrelated." + "reason": "SigLIP2 output quality bug and Sam3-LiteText memory overflow are unrelated." }, { "left": "issue:43526", "right": "issue:46655", "accept": false, - "reason": "BeitImageProcessorFast label handling and SmolVLM video padding are different processor bugs." + "reason": "BEiTImageProcessorFast label reduction and SmolVLM video padding are unrelated processor bugs." + }, + { + "left": "issue:43526", + "right": "issue:44186", + "accept": false, + "reason": "Different components and failure modes: BEiT image label reduction vs LayoutLMv2 tokenizer crashes." }, { "left": "issue:43824", "right": "issue:44279", "accept": false, - "reason": "Missing Qwen2_5_VL import and a generic dependency issue are not the same underlying problem." + "reason": "Missing Qwen2.5-VL import and a generic dependency issue are not the same concrete problem." }, { "left": "issue:46121", "right": "issue:46399", "accept": false, - "reason": "RoPE validation TypeError and Qwen3.5 kernel registration crash are unrelated." - }, - { - "left": "issue:27957", - "right": "issue:29127", - "accept": false, - "reason": "Flash Attention 2 support and LayoutLMv3 box error messaging are unrelated." + "reason": "RoPE-validation TypeError and Qwen3.5 kernel registration crash are unrelated." }, { "left": "issue:45250", "right": "issue:46640", "accept": false, - "reason": "Both are vague Flash Attention 2 reports, but there is no evidence they refer to the same concrete bug." + "reason": "Both are vague Flash Attention 2 reports, but the titles do not identify the same underlying bug or code path." }, { - "left": "issue:43334", - "right": "issue:43531", + "left": "issue:27957", + "right": "issue:29127", "accept": false, - "reason": "Qwen3-VL pad_token_id loading error and Qwen3-MoE sliding_window issue are different failures." + "reason": "XLMRoberta Flash Attention 2 support and LayoutLMv3 box error are unrelated." }, { - "left": "issue:27957", - "right": "issue:33290", + "left": "issue:43334", + "right": "issue:43531", "accept": false, - "reason": "Flash Attention 2 support and Adafactor/DeepSpeed OOM are unrelated." + "reason": "Qwen3-VL missing pad_token_id and Qwen3-MoE sliding_window issue are different model/config bugs." } ] }, @@ -9941,6 +9894,7 @@ "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-10000-473", "nodes": [ + "issue:27957", "issue:33290", "issue:34567", "issue:34689", @@ -9976,7 +9930,6 @@ "issue:44871", "issue:44898", "issue:44977", - "issue:44987", "issue:45003", "issue:45216", "issue:45245", @@ -9987,10 +9940,10 @@ "issue:45600", "issue:45859", "issue:45907", - "issue:46178", "issue:46234" ], "soft_pairs": [ + "issue:27957|issue:33290", "issue:43761|issue:45859", "issue:33290|issue:34689", "issue:33290|issue:35141", @@ -10018,217 +9971,210 @@ "issue:43761|issue:44188", "issue:43232|issue:43388", "issue:43749|issue:43931", - "issue:43931|issue:45216", - "issue:44987|issue:46178" + "issue:43931|issue:45216" ], - "prepared_review_unit_hash": "980b5b53a2f852eedb3dcfe6def56c474a67953f51d265410d5f7138c3f8bd3d", + "prepared_review_unit_hash": "df76f2e05279feb25d0e73e35b876f57a88c4fd019cf04a90d5b24076d24e387", "component_budget": { "node_count": 473, "item_count": 473, "soft_pair_count": 1738, - "serialized_chars": 901383, - "estimated_input_tokens": 225346, - "estimated_eval_tokens": 450948 + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 }, "budget": { - "node_count": 48, - "item_count": 48, + "node_count": 47, + "item_count": 47, "soft_pair_count": 29, - "serialized_chars": 36561, - "estimated_input_tokens": 9141, - "estimated_eval_tokens": 18538 + "serialized_chars": 35916, + "estimated_input_tokens": 8979, + "estimated_eval_tokens": 18214 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is heterogeneous: the soft pairs span unrelated bugs, regressions, packaging, docs, and model-specific loading issues. I would not merge any of the candidate pairs.", - "confidence": 0.96, - "canonical_issue_reason": "No single canonical issue stands out because the items do not share one underlying bug; they are mostly unrelated regressions across different subsystems.", - "canonical_pr_reason": "No pull requests are present in the cluster.", - "best_issue_reason": "No global best issue for deduping this cluster: the reports are too diverse to serve as one representative duplicate target.", - "best_pr_reason": "No pull requests are present in the cluster.", + "summary": "This cluster is mostly a grab-bag of unrelated transformer bugs/regressions; none of the soft-linked pairs look like true duplicates. I would anchor the cluster on the most concrete regression report, but not merge any of the candidate pairs.", + "confidence": 0.23, + "canonical_issue_reason": "No single issue is a clear duplicate canonical for the whole set, but issue #43761 is the strongest anchor because it describes a specific, reproducible v5 regression in one code path with a clear symptom and model target.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue #43761 is the most triage-friendly representative: specific regression, explicit failure mode, and narrow scope. Many others are either broader, more speculative, or belong to different subsystems.", + "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:27957", + "right": "issue:33290", + "accept": false, + "reason": "Different problems: Flash Attention 2 support for XLM-Roberta vs Deepspeed+Adafactor OOM." + }, { "left": "issue:43761", "right": "issue:45859", "accept": false, - "reason": "Different failures: CLIP hidden states vs Qwen3_5Moe CPU-offload rotary embedding." + "reason": "Both are regressions, but they hit different models and different paths: CLIPVisionModel output states vs Qwen3.5 CPU offload." }, { "left": "issue:33290", "right": "issue:34689", "accept": false, - "reason": "OOM in Adafactor/DeepSpeed is unrelated to a model-loading regression for Llama 3.2 Vision." + "reason": "OOM during optimizer/deepspeed use is unrelated to model loading breakage for Llama 3.2 Vision." }, { "left": "issue:33290", "right": "issue:35141", "accept": false, - "reason": "Optimizer memory blowup is not the same bug as token embedding reinitialization after resize." + "reason": "Optimizer memory issue vs embedding resize/post_init reinitialization bug." }, { "left": "issue:45588", "right": "issue:45600", "accept": false, - "reason": "Flash-attention s_aux crash and stale auto_mappings references are unrelated." + "reason": "One is a flash_attention sinkless-model crash; the other is stale auto_mappings references breaking CI." }, { "left": "issue:33290", "right": "issue:34567", "accept": false, - "reason": "Training OOM is unrelated to TrainerState token-count bookkeeping." + "reason": "Training memory OOM is unrelated to TrainerState token-count tracking." }, { "left": "issue:43262", "right": "issue:44206", "accept": false, - "reason": "Audio chat-template sample-rate defaulting and feature-extractor center-arg validation are different issues." + "reason": "Audio processor sampling-rate default mismatch is unrelated to a feature extractor center-arg crash." }, { "left": "issue:43381", "right": "issue:43873", "accept": false, - "reason": "Eval-mode gradient checkpointing restriction is not the same as quantization/offloading breakage." + "reason": "Gradient-checkpointing eval-mode incompatibility is not the same as quantization offloading failure." }, { "left": "issue:44977", "right": "issue:45245", "accept": false, - "reason": "Qwen3.5 flash-attention generation bug is unrelated to the category-cardinality runtime error." + "reason": "Qwen3.5 flash-attention generation bug is unrelated to the 2^24 category-count limit error." }, { "left": "issue:44464", "right": "issue:45491", "accept": false, - "reason": "Chunked generation with compiled forward and Gemma3 NaN embeddings have different code paths and symptoms." + "reason": "Compiled-forward chunked generation inconsistency is unrelated to Gemma3 mixed-length NaN embeddings." }, { "left": "issue:43646", "right": "issue:44077", "accept": false, - "reason": "Generic v5 custom-init breakage is not the same concrete bug as patchtsmixer's optional post_init constraint." + "reason": "Custom model initialization breakage is not the same as patchtsmixer optional post_init validation." }, { "left": "issue:44442", "right": "issue:44898", "accept": false, - "reason": "Tokenizer loading failure and Perceiver interpolation failure are unrelated." + "reason": "Tokenizer loading failure is unrelated to Perceiver non-default-resolution image classification." }, { "left": "issue:44743", "right": "issue:44811", "accept": false, - "reason": "Recurrent-state reset with cache does not match Whisper batch_decode skip_special_tokens handling." + "reason": "Qwen3.5 recurrent-state reset bug is unrelated to Whisper batch_decode skip_special_tokens handling." }, { "left": "issue:42490", "right": "issue:44743", "accept": false, - "reason": "Local-path import warning and Qwen recurrent-state reset are unrelated." + "reason": "Local-path import warning does not match the Qwen3.5 cache/recurrent-state bug." }, { "left": "issue:45003", "right": "issue:45440", "accept": false, - "reason": "Unsafe sys.modules access is unrelated to DeepseekV3MoE divergence from the remote implementation." + "reason": "Unsafe sys.modules access is unrelated to DeepseekV3MoE divergence from remote implementation." }, { "left": "issue:34689", "right": "issue:43611", "accept": false, - "reason": "Both are loading regressions, but they affect different model/version paths and have different root causes." + "reason": "Llama 3.2 Vision model-loading regression is not the same as base_model_prefix loading support in v5." }, { "left": "issue:43720", "right": "issue:44442", "accept": false, - "reason": "BitNet accelerate-loading unpacking bug is unrelated to FastSpeech2ConformerTokenizer loading." + "reason": "BitNet packed-weight loading bug is unrelated to FastSpeech2ConformerTokenizer auto-loading." }, { "left": "issue:42175", "right": "issue:46234", "accept": false, - "reason": "Packaging backend omission and missing import-utils symbol are different problems." + "reason": "Backend packaging/install issue is unrelated to missing import_utils symbol for falcon_mamba_ssm." }, { "left": "issue:43723", "right": "issue:45479", "accept": false, - "reason": "Tokenizer loading in v5 is unrelated to the zero-loss single-label classification bug." + "reason": "Tokenizer loading in v5 is unrelated to single-label classification zero-loss behavior." + }, + { + "left": "issue:43316", + "right": "issue:43525", + "accept": false, + "reason": "Gemma3TextConfig API discrepancy is unrelated to Llama4Config missing pad_token_id." }, { "left": "issue:43493", "right": "issue:43827", "accept": false, - "reason": "SigLIP2 implementation mismatch and outdated docs about pipeline() are not the same issue." + "reason": "SigLIP2 implementation mismatch is unrelated to docs still referencing removed pipeline()." }, { "left": "issue:44077", "right": "issue:44479", "accept": false, - "reason": "Patchtsmixer post_init policy and Qwen video-input regression affect different models and code paths." + "reason": "Patchtsmixer post_init restriction is unrelated to video-input regression in Qwen VL models." }, { "left": "issue:43761", "right": "issue:44079", "accept": false, - "reason": "Missing hidden states and ModelOutput key assignment are unrelated behaviors." + "reason": "CLIPVisionModel hidden_states regression is unrelated to ModelOutput None-key assignment." }, { "left": "issue:44062", "right": "issue:45907", "accept": false, - "reason": "AddedToken special-kwarg duplication is unrelated to list_repo_files revision forwarding." + "reason": "AddedToken keyword conflict is unrelated to list_repo_files not forwarding revision." }, { "left": "issue:43873", "right": "issue:44871", "accept": false, - "reason": "Quantization/offloading behavior is unrelated to eos_token_id config mismatch." + "reason": "Quantization offloading issue is unrelated to Gemma-3 eos_token_id config mismatch." }, { "left": "issue:43761", "right": "issue:44188", "accept": false, - "reason": "CLIP hidden_states regression and attention-kernel divergence under torch.compile are different bugs." + "reason": "These are both runtime regressions, but one is hidden_states propagation and the other is attention-kernel divergence under torch.compile." }, { "left": "issue:43232", "right": "issue:43388", "accept": false, - "reason": "Generation kwargs update after sync_gpus is unrelated to gather_for_metrics label truncation." + "reason": "Generation kwargs update after sync_gpus is unrelated to gather_for_metrics dropping tuple labels." }, { "left": "issue:43749", "right": "issue:43931", "accept": false, - "reason": "FSDP CPU RAM efficient loading breakage is not the same as Qwen3-VL weight-shape mismatch." + "reason": "FSDP CPU RAM efficient loading breakage is unrelated to Qwen3-VL weight-shape mismatch." }, { "left": "issue:43931", "right": "issue:45216", "accept": false, - "reason": "Model load shape mismatch and incorrect save_pretrained checkpoint output are different stages and causes." - }, - { - "left": "issue:44987", - "right": "issue:46178", - "accept": false, - "reason": "Loading failure for a specific model is unrelated to missing tensorflow_text backend registration." - }, - { - "left": "issue:43262", - "right": "issue:44206", - "accept": false, - "reason": "Audio processor template defaults and unsupported feature-extractor argument are distinct issues." - }, - { - "left": "issue:44079", - "right": "issue:45479", - "accept": false, - "reason": "ModelOutput key assignment and zero-loss classification are unrelated." + "reason": "Model load shape mismatch is unrelated to Qwen3.5 save_pretrained checkpoint corruption." } ] }, @@ -10285,32 +10231,32 @@ "issue:44855", "issue:44871", "issue:44913", + "issue:44987", "issue:45137", "issue:45584", - "issue:45632", "issue:45698", "issue:45759", "issue:45803", "issue:46050", - "issue:46170", + "issue:46178", "issue:46234", "issue:46291", "issue:46365", "issue:46369", "issue:46421", - "issue:46424", "issue:46506" ], "soft_pairs": [ + "issue:44987|issue:46178", "issue:43316|issue:45759", - "issue:43643|issue:44913", "issue:44855|issue:45137", + "issue:43643|issue:44913", "issue:44442|issue:45803", "issue:43701|issue:46506", "issue:44190|issue:44496", "issue:42907|issue:45698", - "issue:42617|issue:43575", "issue:44464|issue:46421", + "issue:42617|issue:43575", "issue:25251|issue:28282", "issue:27957|issue:32090", "issue:43550|issue:43701", @@ -10324,223 +10270,215 @@ "issue:43316|issue:43716", "issue:42175|issue:43316", "issue:44077|issue:44534", - "issue:44871|issue:46050", "issue:45803|issue:46365", + "issue:44871|issue:46050", "issue:43646|issue:44913", "issue:44610|issue:46369", "issue:43452|issue:44496", - "issue:44162|issue:44792", - "issue:45632|issue:46424", - "issue:45803|issue:46170" + "issue:44162|issue:44792" ], - "prepared_review_unit_hash": "7f469ef2ff7edf1b3891b82c4940dd2bcf107894834c82c605beed674273e491", + "prepared_review_unit_hash": "83eeef5be26049c0ebfa198a669e92fc98ec893b1819a7c6810078ce14e3a340", "component_budget": { "node_count": 473, "item_count": 473, "soft_pair_count": 1738, - "serialized_chars": 901383, - "estimated_input_tokens": 225346, - "estimated_eval_tokens": 450948 + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 }, "budget": { - "node_count": 48, - "item_count": 48, - "soft_pair_count": 30, - "serialized_chars": 36509, - "estimated_input_tokens": 9128, - "estimated_eval_tokens": 18512 + "node_count": 47, + "item_count": 47, + "soft_pair_count": 29, + "serialized_chars": 35635, + "estimated_input_tokens": 8909, + "estimated_eval_tokens": 18074 }, "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. The soft pairs mostly share broad vocabulary like loading, config, cache, or generation, but they point to different models, different symptoms, and different code paths, so none look like true duplicates.", - "confidence": 0.96, - "canonical_issue_reason": "No single issue is a credible canonical duplicate for the cluster: it spans unrelated bugs in config loading, model initialization, generation, distributed training, cache behavior, and packaging/imports.", + "summary": "These issues are mostly unrelated Transformers bugs spanning config loading, generation, backend availability, and model-specific regressions. I do not see any soft pair that is the same underlying bug/change, so the cluster should not be merged as duplicates. If a single anchor is needed, issue 45698 is the broadest and most active open loading-related report, but it is not a true duplicate of the others.", + "confidence": 0.84, + "canonical_issue_reason": "No true duplicate anchor is apparent across this set; issue 45698 is the broadest and most active open report about save_pretrained/from_pretrained custom-module loading, so it is the least-bad cluster anchor if one must be chosen.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 45698 is the most broadly framed and actively discussed open report, but it still represents only one of several distinct save/load-related bugs in the set.", + "best_issue_reason": "Issue 45698 has the widest scope, is still open, and has the strongest discussion/reference signal among the listed issues, making it the best representative issue for this noisy loading-regression cluster.", "best_pr_reason": null, "soft_edge_verdicts": [ { - "left": "issue:43316", - "right": "issue:45759", + "left": "issue:44987", + "right": "issue:46178", "accept": false, - "reason": "Both involve config/model API shape, but one is a Gemma3TextConfig discrepancy and the other is Qwen multimodal text_config unwrapping; different models and failure modes." + "reason": "Different failures: one is model loading under transformers>=5.1.0, the other is an invalid backend name for tensorflow_text." }, { - "left": "issue:43643", - "right": "issue:44913", + "left": "issue:43316", + "right": "issue:45759", "accept": false, - "reason": "Both are config-serialization issues, but one is missing fields with trust_remote_code and the other is rotary_pct resetting on reload; unrelated." + "reason": "Both touch multimodal config APIs, but the concrete problems differ: Gemma3TextConfig API mismatch vs Qwen composite text_config unwrapping." }, { "left": "issue:44855", "right": "issue:45137", "accept": false, - "reason": "Python 3.13 import parsing bug vs DeepSpeed ZeRO3 deque underflow; unrelated." + "reason": "Completely different code paths: Python 3.13 TorchScript parsing vs DeepSpeed ZeRO3 deque handling." + }, + { + "left": "issue:43643", + "right": "issue:44913", + "accept": false, + "reason": "Both are config persistence issues, but one is trust_remote_code field loss and the other is GPTNeoX rotary_pct not reloading." }, { "left": "issue:44442", "right": "issue:45803", "accept": false, - "reason": "Tokenizer load failure vs Fuyu exception handling swallowing errors; different code paths." + "reason": "Unrelated tokenizer loading failure vs a bare-except bug in Fuyu tensor conversion." }, { "left": "issue:43701", "right": "issue:46506", "accept": false, - "reason": "Resume-checkpoint key mismatch vs CPU/CUDA output divergence; unrelated." + "reason": "Checkpoint resume key mismatch is unrelated to PPDocLayoutV3 CPU/CUDA result differences." }, { "left": "issue:44190", "right": "issue:44496", "accept": false, - "reason": "Data-loading script issue vs unrecognized model/config; unrelated." + "reason": "Different loading problems: local dataset handling in a training script vs missing model_type in config.json." }, { "left": "issue:42907", "right": "issue:45698", "accept": false, - "reason": "Both touch save/load, but one is dequantized model saving and the other is wrong custom-module loading after save_pretrained; not the same bug." + "reason": "Both involve save/load, but one is dequantized model saving and the other is wrong custom-module resolution after save_pretrained." }, { - "left": "issue:42617", - "right": "issue:43575", + "left": "issue:44464", + "right": "issue:46421", "accept": false, - "reason": "Parallel script runtime issue vs TP-induced OOM on a specific model; different problems." + "reason": "Both mention chunked generation/compiled execution, but one is output inconsistency and the other is lazy cache initialization causing recompilation." }, { - "left": "issue:44464", - "right": "issue:46421", + "left": "issue:42617", + "right": "issue:43575", "accept": false, - "reason": "Both mention chunked/compiled execution, but one is inconsistent generation outputs and the other is recompilation from lazy cache init; separate bugs." + "reason": "3d_parallel.py failure and Qwen2 tp OOM are different runtime/resource issues." }, { "left": "issue:25251", "right": "issue:28282", "accept": false, - "reason": "Pipeline output nesting issue vs missing PyTorch import; unrelated." + "reason": "Pipeline top_k nesting bug is unrelated to missing PyTorch import support." }, { "left": "issue:27957", "right": "issue:32090", "accept": false, - "reason": "Flash Attention model support vs Trainer broadcast error; unrelated." + "reason": "Flash Attention 2 model support and Trainer _gpu_broadcast_one NoneType error are different failures." }, { "left": "issue:43550", "right": "issue:43701", "accept": false, - "reason": "torch.compile/SDPA failure vs checkpoint key mismatch; unrelated." + "reason": "torch.compile/SDPA model bug is unrelated to checkpoint resume key mismatch." }, { "left": "issue:43575", "right": "issue:43742", "accept": false, - "reason": "OOM on tensor-parallel load vs key error for a model config; unrelated." + "reason": "OOM on a large Qwen model and a key error loading MobileLLM are different loading issues." }, { "left": "issue:46234", "right": "issue:46291", "accept": false, - "reason": "Both are import-related, but one is a missing availability helper and the other is a kernels dependency/version check; different issues." + "reason": "Missing import_utils symbol and kernels version/revision import failure are distinct dependency/API problems." }, { "left": "issue:43606", "right": "issue:43701", "accept": false, - "reason": "CPU offload device mismatch vs checkpoint resume key mismatch; unrelated." + "reason": "CPU offload device mismatch does not match the checkpoint resume key mismatch bug." }, { "left": "issue:43295", "right": "issue:43825", "accept": false, - "reason": "Regression in processor/tokenizer access vs pipeline translation error message; different layers and symptoms." + "reason": "Custom processor/tokenizer regression is unrelated to a v5 pipeline error-message bug." }, { "left": "issue:16998", "right": "issue:22355", "accept": false, - "reason": "Model_max_length question vs missing transformers.onnx module; unrelated." + "reason": "model_max_length on DeBERTa-V3 and missing transformers.onnx are unrelated concerns." }, { "left": "issue:43295", "right": "issue:44871", "accept": false, - "reason": "Processor/tokenizer regression vs Gemma eos_token_id mismatch; not the same bug." + "reason": "One is processor.tokenizer access regression, the other is eos_token_id config inconsistency in Gemma-3." }, { "left": "issue:44871", "right": "issue:45584", "accept": false, - "reason": "Both involve special-token behavior, but one is a config mismatch and the other is Whisper empty-transcription generation failure; different code paths." + "reason": "Both involve special-token behavior, but one is Gemma-3 eos_token_id mismatch and the other is Whisper empty-transcription generation after align_special_tokens." }, { "left": "issue:43316", "right": "issue:43716", "accept": false, - "reason": "Config API discrepancy vs image preprocessor/model dtype mismatch; unrelated." + "reason": "Gemma3TextConfig API mismatch and Mistral-3 image-preprocessor dtype mismatch are different multimodal issues." }, { "left": "issue:42175", "right": "issue:43316", "accept": false, - "reason": "Backend packaging issue vs model config API discrepancy; unrelated." + "reason": "Backend packaging omission with pip install '.[torch]' is unrelated to Gemma3TextConfig API behavior." }, { "left": "issue:44077", "right": "issue:44534", "accept": false, - "reason": "Model post_init allowance vs non-persistent buffer junk; unrelated." + "reason": "Optional post_init restriction for patchtsmixer is unrelated to non-persistent buffers being filled with junk in v5." }, { - "left": "issue:44871", - "right": "issue:46050", + "left": "issue:45803", + "right": "issue:46365", "accept": false, - "reason": "Gemma eos token config inconsistency vs quantized save/load breakage; unrelated." + "reason": "Bare except swallowing errors in Fuyu is unrelated to importing Trainer conflicting with datasets." }, { - "left": "issue:45803", - "right": "issue:46365", + "left": "issue:44871", + "right": "issue:46050", "accept": false, - "reason": "Exception handling bug vs Trainer import affected by datasets; unrelated." + "reason": "Gemma-3 eos_token_id mismatch is not the same as quantized model save/load breakage after sinq." }, { "left": "issue:43646", "right": "issue:44913", "accept": false, - "reason": "General custom model init regression vs GPTNeoX rotary_pct reload defaulting; different scope." + "reason": "Custom model initialization breakage in v5 is different from GPTNeoX rotary_pct not persisting on reload." }, { "left": "issue:44610", "right": "issue:46369", "accept": false, - "reason": "Processor input-size mismatch vs memory overflow on sam3-litetext; unrelated." + "reason": "OmDet-Turbo preprocessing size mismatch and sam3-litetext memory overflow are unrelated." }, { "left": "issue:43452", "right": "issue:44496", "accept": false, - "reason": "gguf_file breaks tokenizer/model loading vs unrecognized model config; both loading-related but distinct failure modes." + "reason": "gguf_file loading breakage and missing model_type in config.json are separate model-loading failures." }, { "left": "issue:44162", "right": "issue:44792", "accept": false, - "reason": "ESM2 model broken vs janus image generation test failure; unrelated." - }, - { - "left": "issue:45632", - "right": "issue:46424", - "accept": false, - "reason": "Both mention cache/path behavior, but one is trust_remote_code cache collisions and the other is max_cache_length handling; different bugs." - }, - { - "left": "issue:45803", - "right": "issue:46170", - "accept": false, - "reason": "Exception handling bug vs Windows ROCm import error; unrelated." + "reason": "ESM2 breakage and a janus image-generation test failure are unrelated model-specific regressions." } ] }, @@ -10582,8 +10520,6 @@ "issue:43673", "issue:43688", "issue:43761", - "issue:43931", - "issue:44060", "issue:44190", "issue:44242", "issue:44479", @@ -10606,21 +10542,26 @@ "issue:45600", "issue:45632", "issue:45698", + "issue:45803", + "issue:46170", "issue:46178", "issue:46202", "issue:46234", + "issue:46424", "issue:46617", "issue:46640" ], "soft_pairs": [ + "issue:45632|issue:46424", + "issue:45803|issue:46170", "issue:43650|issue:46202", "issue:43404|issue:43441", "issue:43065|issue:44617", "issue:43540|issue:44514", "issue:43761|issue:44479", "issue:45127|issue:45200", - "issue:43388|issue:43688", "issue:43572|issue:46178", + "issue:43388|issue:43688", "issue:45542|issue:46234", "issue:43352|issue:46640", "issue:43504|issue:45698", @@ -10634,189 +10575,194 @@ "issue:44534|issue:44913", "issue:35141|issue:42490", "issue:44617|issue:45584", - "issue:43761|issue:44242", "issue:45632|issue:46617", - "issue:44855|issue:45588", - "issue:43931|issue:44060" + "issue:43761|issue:44242", + "issue:44855|issue:45588" ], - "prepared_review_unit_hash": "c7beeaa9be506701b494ce4c8cdad09e8613586a696ff6dfbab2a6f423c1b281", + "prepared_review_unit_hash": "46f4ac241fdffe7c689eb46c54b47449c4174f5a7c9d024a9532ec676bc53625", "component_budget": { "node_count": 473, "item_count": 473, "soft_pair_count": 1738, - "serialized_chars": 901383, - "estimated_input_tokens": 225346, - "estimated_eval_tokens": 450948 + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 }, "budget": { - "node_count": 47, - "item_count": 47, - "soft_pair_count": 25, - "serialized_chars": 34643, - "estimated_input_tokens": 8661, - "estimated_eval_tokens": 17578 + "node_count": 48, + "item_count": 48, + "soft_pair_count": 26, + "serialized_chars": 35524, + "estimated_input_tokens": 8881, + "estimated_eval_tokens": 18018 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is a grab bag of unrelated Transformers bugs spanning model loading, cache paths, processors, Flash Attention, multimodal/video regressions, and configuration issues. None of the soft pairs look like the same underlying bug closely enough to merge.", - "confidence": 0.89, - "canonical_issue_reason": "I\u2019d use issue #45698 as the representative issue: it is open, has the highest discussion activity, and describes a clear reproducible loading bug with broad user impact (`from_pretrained` selecting the wrong custom module after `save_pretrained`).", + "summary": "This cluster is highly heterogeneous: the soft edges mostly share only broad subsystem keywords (loading, cache, flash attention, video, etc.) but do not describe the same underlying bug or a mergeable combined fix. I would keep the issues separate; issue:45698 is the best anchor only because it is the broadest and most referenced loading regression in the set.", + "confidence": 0.94, + "canonical_issue_reason": "issue:45698 is the most central/general item here: it concerns a broad `from_pretrained`/`save_pretrained` loading regression, has the strongest discussion/inbound activity, and is the best representative anchor if one issue must be chosen.", "canonical_pr_reason": null, - "best_issue_reason": "Issue #45698 is the strongest standalone issue in the set because it is well-scoped, actively discussed, and likely to attract the most useful fix/workaround for a concrete loading regression.", + "best_issue_reason": "issue:45698 is the best overall issue candidate because it is active, broadly scoped, and more likely to serve as a canonical reference than the narrower model-specific or unrelated bugs.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:45632", + "right": "issue:46424", + "accept": false, + "reason": "Both are cache-related, but one is a trust_remote_code path collision and the other is max_cache_length behavior for quantized cache; different bugs." + }, + { + "left": "issue:45803", + "right": "issue:46170", + "accept": false, + "reason": "Different failure modes: one is a bare-except in tensor conversion, the other is a Windows/ROCm torch.distributed import dependency issue." + }, { "left": "issue:43650", "right": "issue:46202", "accept": false, - "reason": "Completely different problems: one is a vague placeholder report, the other is a Gemma pipeline roundtrip issue on a specific token. No shared code path." + "reason": "Completely unrelated: placeholder text body vs Gemma pipeline roundtrip/tokenization failure." }, { "left": "issue:43404", "right": "issue:43441", "accept": false, - "reason": "Both are model-specific, but one is about untied lm_head weights and the other is Flash Attention support/failure. Different bugs." + "reason": "One is weight tying in Mistral3; the other is FlashAttention 2 support for Ministral-3. Same family, different bug." }, { "left": "issue:43065", "right": "issue:44617", "accept": false, - "reason": "Sam3PixelDecoder dummy Conv2d is a model implementation detail; CUDA OOM in Sam3Video is a resource problem, not the same defect." + "reason": "Sam3PixelDecoder dummy Conv2d is an architecture/implementation issue, while Sam3Video CUDA OOM is a memory usage problem." }, { "left": "issue:43540", "right": "issue:44514", "accept": false, - "reason": "Different models and different failure modes: Qwen3OmniMoe video processing vs Qwen2_5_VL chat templating with batching/padding." + "reason": "Both touch Qwen3 models, but one is video-input processing and the other is chat template batching/padding." }, { "left": "issue:43761", "right": "issue:44479", "accept": false, - "reason": "CLIPVisionModel hidden_states regression and Qwen VL video-input regression are unrelated code paths." + "reason": "CLIPVisionModel hidden_states regression is unrelated to the Qwen video input regression." }, { "left": "issue:45127", "right": "issue:45200", "accept": false, - "reason": "One is LoRA merge / tied embeddings collapse; the other is missing `mm_token_type_ids` for text-only fine-tuning. Not the same bug." + "reason": "LoRA merge with extended vocabulary and Gemma 4 mm_token_type_ids defaulting to zeros are different model-behavior bugs." }, { - "left": "issue:43388", - "right": "issue:43688", + "left": "issue:43572", + "right": "issue:46178", "accept": false, - "reason": "Gathering metrics on tuple labels and auxiliary-loss normalization in MoE models are different training bugs." + "reason": "StableLmConfig pad_token_idx regression is unrelated to a missing tensorflow_text backend mapping." }, { - "left": "issue:43572", - "right": "issue:46178", + "left": "issue:43388", + "right": "issue:43688", "accept": false, - "reason": "StableLmConfig missing `pad_token_idx` and backend mapping error for `tensorflow_text` are unrelated configuration/import issues." + "reason": "Both involve training utilities/losses, but one drops labels in gather_for_metrics and the other misnormalizes auxiliary loss." }, { "left": "issue:45542", "right": "issue:46234", "accept": false, - "reason": "TensorFlow/tensorboard backend detection and a missing `is_falcon_mamba_ssm_available` helper are different import-system problems." + "reason": "Different integration problems: missing TensorFlow backend handling vs missing import_utils symbol." }, { "left": "issue:43352", "right": "issue:46640", "accept": false, - "reason": "Both mention Flash Attention, but one is a specific Nemotron unsupported-feature error while the other is a vague general flash-attention issue; not enough evidence of the same fix." + "reason": "One is a specific model not supporting Flash Attention 2 yet; the other is a vague generic flash-attention issue without the same concrete code path." }, { "left": "issue:43504", "right": "issue:45698", "accept": false, - "reason": "Legacy field preventing BEiT preset loading is not the same as loading the wrong custom module after save/load roundtrip." + "reason": "Legacy field on Beit preset loading is unrelated to wrong custom-module selection after save_pretrained." }, { "left": "issue:43673", "right": "issue:44743", "accept": false, - "reason": "Both involve cache/state handling, but one is GenerationMixin chunked_prefill cache plumbing and the other is recurrent state reset in Qwen modular code. Different failures." + "reason": "Both mention cache, but one is GenerationMixin cache missing in chunked_prefill while the other resets recurrent states in modular_qwen3_5 with seq_len>1." }, { "left": "issue:44792", "right": "issue:44977", "accept": false, - "reason": "A janus image-generation test failure and Qwen3.5 Flash Attention generation issues are separate regressions." + "reason": "Janus image-generation test failure is unrelated to Qwen3.5 flash-attention generation issues." }, { "left": "issue:44792", "right": "issue:45600", "accept": false, - "reason": "A failed janus generation test and broken auto_mappings entries for removed Sam3LiteText configs are unrelated." + "reason": "A failed Janus test case is unrelated to auto_mappings referencing removed Sam3LiteText configs." }, { "left": "issue:41950", "right": "issue:43475", "accept": false, - "reason": "Video-classification processor selection and missing `fpn_position_embeddings` on Sam3VisionEncoderOutput are different SAM3 issues." + "reason": "Video-classification processor lookup and missing fpn_position_embeddings in Sam3Video are different SAM/video bugs." }, { "left": "issue:43525", "right": "issue:43531", "accept": false, - "reason": "Missing `pad_token_id` in Llama4Config and a Qwen3-MoE sliding-window bug are unrelated model/config problems." + "reason": "Missing pad_token_id in Llama4Config and Qwen3-MoE sliding_window behavior are unrelated config/model issues." }, { "left": "issue:44734", "right": "issue:45406", "accept": false, - "reason": "Serve KV-cache continuation tensor indexing bug and Gemma4Processor missing `_tokenizer` are different service-layer failures." + "reason": "KV-cache continuation indexing in serve is unrelated to Gemma4Processor missing `_tokenizer`." }, { "left": "issue:44190", "right": "issue:44964", "accept": false, - "reason": "Local dataset loading in a training example and loading Phi-4 multimodal with latest transformers are unrelated." + "reason": "Local dataset loading in an example script is unrelated to loading a Phi-4 multimodal model." }, { "left": "issue:44534", "right": "issue:44913", "accept": false, - "reason": "Non-persistent buffer initialization in v5 and GPTNeoX rotary_pct reload persistence are different config/runtime bugs." + "reason": "Non-persistent buffer initialization and GPTNeoXConfig rotary_pct reload persistence are different regression classes." }, { "left": "issue:35141", "right": "issue:42490", "accept": false, - "reason": "Token embedding resize/post-init reinitialization and trailing-slash local model import warnings are unrelated." + "reason": "Token-embedding resizing/tie_word_embedding bug is unrelated to the local-path trailing-slash import warning." }, { "left": "issue:44617", "right": "issue:45584", "accept": false, - "reason": "Sam3Video OOM is a memory issue; Whisper empty-transcription failure after special-token alignment is a generation edge case." - }, - { - "left": "issue:43761", - "right": "issue:44242", - "accept": false, - "reason": "A vision-model hidden_states regression and missing load-balancing loss when `output_router_logits=False` are distinct bugs." + "reason": "Sam3Video CUDA OOM is a resource issue, while Whisper empty-transcription failure after align_special_tokens is a generation/tokenization bug." }, { "left": "issue:45632", "right": "issue:46617", "accept": false, - "reason": "Both concern trust_remote_code/cache behavior, but one is a leaf-directory collision and the other is a FileNotFoundError on cached load; different concrete failure modes." + "reason": "Both involve trust_remote_code and cache, but one is a leaf-directory collision and the other is a FileNotFoundError when loading from cache." }, { - "left": "issue:44855", - "right": "issue:45588", + "left": "issue:43761", + "right": "issue:44242", "accept": false, - "reason": "Python 3.13 import parsing issue in DebertaV2Model and Flash Attention `s_aux=None` crash are unrelated." + "reason": "Hidden_states regression in CLIPVisionModel is unrelated to missing load balancing loss when output_router_logits=False." }, { - "left": "issue:43931", - "right": "issue:44060", + "left": "issue:44855", + "right": "issue:45588", "accept": false, - "reason": "Qwen3-VL weight-shape mismatch and Qwen3-Next tied-weights warning are different model-loading/tie-weight issues." + "reason": "Python 3.13 indentation/parsing failure in DebertaV2Model is unrelated to a flash_attention.py AttributeError on sink-less models." } ] }, @@ -10852,7 +10798,9 @@ "issue:43644", "issue:43673", "issue:43827", + "issue:43931", "issue:43994", + "issue:44060", "issue:44077", "issue:44112", "issue:44190", @@ -10880,27 +10828,26 @@ "issue:45563", "issue:45584", "issue:45600", - "issue:45632", "issue:45803", "issue:45907", - "issue:46234", "issue:46369", "issue:46421", "issue:46489", "issue:46655" ], "soft_pairs": [ + "issue:43931|issue:44060", "issue:43827|issue:43994", "issue:43673|issue:46421", - "issue:45381|issue:46369", "issue:45563|issue:45584", + "issue:45381|issue:46369", "issue:44479|issue:44734", "issue:43644|issue:44913", - "issue:25251|issue:27957", "issue:44933|issue:45362", "issue:43065|issue:44610", - "issue:41720|issue:45245", + "issue:25251|issue:27957", "issue:43335|issue:44855", + "issue:41720|issue:45245", "issue:43550|issue:44190", "issue:44077|issue:44877", "issue:44464|issue:45563", @@ -10917,218 +10864,166 @@ "issue:43295|issue:45276", "issue:44938|issue:45542", "issue:45276|issue:45584", - "issue:45632|issue:46234", "issue:44462|issue:45907" ], - "prepared_review_unit_hash": "c5fa16b8cfdd687869e2d3aefe5fac47669006b4c3135cdbfaadcee91b538db4", + "prepared_review_unit_hash": "28abaff7cd1effc8647d14b771b0eb2193dc7f77730622423ef5e8abe0c00923", "component_budget": { "node_count": 473, "item_count": 473, "soft_pair_count": 1738, - "serialized_chars": 901383, - "estimated_input_tokens": 225346, - "estimated_eval_tokens": 450948 + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 29, - "serialized_chars": 36311, - "estimated_input_tokens": 9078, - "estimated_eval_tokens": 18412 + "serialized_chars": 36331, + "estimated_input_tokens": 9083, + "estimated_eval_tokens": 18422 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is heterogeneous: the soft pairs mostly link unrelated issues across docs, tokenizers, video/multimodal processing, config loading, generation, and CI. I found no pair that convincingly describes the same underlying bug/change, so all soft edges are rejected.", - "confidence": 0.95, - "canonical_issue_reason": "No single canonical issue stands out because the items do not share one underlying defect; they span many distinct subsystems and regression classes.", - "canonical_pr_reason": "No pull requests are present in this cluster.", - "best_issue_reason": "No issue is a strong global representative for the whole set; even the closest-looking pairs still describe different failure modes or components.", - "best_pr_reason": "No pull requests are present in this cluster.", + "summary": "This cluster is mostly a grab bag of unrelated Transformers bug reports. The only plausible duplicate pair is the DeepSeek tokenizer regression reports (44779 and 46489); the rest differ by model, subsystem, or failure mode and should stay separate.", + "confidence": 0.87, + "canonical_issue_reason": "issue:46489 is the most specific and actionable report in the only real duplicate pair: it narrows the problem to DeepSeek-Coder v1 tokenizer output on v5+ and explicitly references a gap in PR #44801's fix. issue:44779 is the broader sibling report.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:46489 is the best representative issue overall because it has the clearest scope, an explicit link to the prior fix, and a concrete regression description.", + "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:43931", + "right": "issue:44060", + "accept": false, + "reason": "Different models and different bugs: Qwen3-VL weight shape mismatch vs Qwen3-Next tied-weights warning." + }, { "left": "issue:43827", "right": "issue:43994", "accept": false, - "reason": "Both mention pipeline-related behavior, but one is a docs reference problem and the other is a SigLIP2 model/pipeline output issue; different root causes." + "reason": "Both mention pipeline/vision workflows, but one is docs about removed pipelines and the other is an AutoModel/pipeline output correctness problem for SigLIP2." }, { "left": "issue:43673", "right": "issue:46421", "accept": false, - "reason": "Both concern chunked prefill/cache behavior, but one is a missing GenerationMixin cache in v5.0.0 and the other is lazy cache init causing recompilation; not the same bug." + "reason": "Both involve chunked prefill/cache, but one is a missing GenerationMixin cache bug and the other is lazy cache init causing recompilation; not the same code path." }, { - "left": "issue:45381", - "right": "issue:46369", + "left": "issue:45563", + "right": "issue:45584", "accept": false, - "reason": "Both involve multimodal/video-related failures, but one is a specific Qwen2.5-VL video position-id regression and the other is Sam3-lite memory overflow; different problems." + "reason": "Unrelated failures: a stale warning in paged generate vs Whisper empty-transcription handling after align_special_tokens." }, { - "left": "issue:45563", - "right": "issue:45584", + "left": "issue:45381", + "right": "issue:46369", "accept": false, - "reason": "Both are generation edge cases, but one is a stale warning about num_return_sequences and the other is Whisper failing on empty transcription after token alignment." + "reason": "Different models and symptoms: Qwen2.5-VL video position ids vs Sam3-LiteText memory overflow." }, { "left": "issue:44479", "right": "issue:44734", "accept": false, - "reason": "Both mention v5/server-related breakage, but one is a video-input regression across Qwen models and the other is a KV-cache continuation indexing crash in transformers serve." + "reason": "Both are generation-related, but one is a Qwen video-input regression and the other is a server-side KV-cache indexing crash." }, { "left": "issue:43644", "right": "issue:44913", "accept": false, - "reason": "Both describe config/state persistence issues, but one is about non-persistent buffers being filled with junk and the other is rotary_pct not round-tripping in GPTNeoXConfig." - }, - { - "left": "issue:25251", - "right": "issue:27957", - "accept": false, - "reason": "These are unrelated: one is a pipeline top_k output-shape issue, the other is XLMRoberta with Flash Attention 2." + "reason": "One is a framework-wide non-persistent buffer initialization bug; the other is a GPTNeoX config reload persistence issue." }, { "left": "issue:44933", "right": "issue:45362", "accept": false, - "reason": "One is a missing import from image_utils, the other is Qwen3.5 chat crashes; no shared code path." + "reason": "Completely different problems: missing image_utils import vs Qwen3.5 chat crash." }, { "left": "issue:43065", "right": "issue:44610", "accept": false, - "reason": "One is about a dummy Conv2d in Sam3PixelDecoder, the other is an OmDet-Turbo processor/model input-size mismatch; different components." + "reason": "Both touch vision models, but one is a dummy Conv2d in Sam3PixelDecoder and the other is a processor/model image-size mismatch in OmDet-Turbo." }, { - "left": "issue:41720", - "right": "issue:45245", + "left": "issue:25251", + "right": "issue:27957", "accept": false, - "reason": "These are unrelated runtime failures: Qwen3 auto device mapping cudaErrorAssert versus a category-count limit error." + "reason": "No underlying overlap: pipeline top_k nesting bug vs XLM-RoBERTa Flash Attention 2 support." }, { "left": "issue:43335", "right": "issue:44855", "accept": false, - "reason": "One is a SwitchTransformers sparse-layer construction bug, the other is a Python 3.13 import failure from @torch.jit.script parsing; not the same issue." - }, - { - "left": "issue:43550", - "right": "issue:44190", - "accept": false, - "reason": "Bamba torch.compile/SDPA failure is unrelated to loading a local dataset in run_image_classification_no_trainer.py." - }, - { - "left": "issue:44077", - "right": "issue:44877", - "accept": false, - "reason": "Both involve config restrictions, but one is patchtsmixer post_init allowance and the other is strict config loading for granite_speech; different fixes." - }, - { - "left": "issue:44464", - "right": "issue:45563", - "accept": false, - "reason": "Chunked generation inconsistent outputs with compiled forward is a behavior bug, while num_return_sequences warning is a stale warning; not the same change." + "reason": "Different failure modes and code paths: SwitchTransformers config creates sparse layers incorrectly vs a Python 3.13 IndentationError in DebertaV2Model." }, { - "left": "issue:45381", - "right": "issue:46655", - "accept": false, - "reason": "Both are video-input issues, but they target different models and different bugs: incorrect vision_position_ids versus incorrect padding calculation." - }, - { - "left": "issue:43335", - "right": "issue:44190", - "accept": false, - "reason": "A config construction bug in SwitchTransformers is unrelated to a dataset-loading failure in an example script." - }, - { - "left": "issue:44617", - "right": "issue:46369", - "accept": false, - "reason": "Both mention SAM3 models and memory problems, but one is OOM in Sam3Video and the other is memory overflow in sam3-litetext; different model paths." - }, - { - "left": "issue:45127", - "right": "issue:45357", - "accept": false, - "reason": "Both involve LoRA/save_pretrained/vocabulary-like concerns, but one is model collapse after merging LoRA with extended vocab and the other is incorrect visual encoder keys saved for Qwen3.5." - }, - { - "left": "issue:44779", - "right": "issue:46489", - "accept": false, - "reason": "Both are DeepSeek tokenizer regressions, but the titles indicate different affected tokenizers and failure modes; insufficient evidence of one underlying bug." - }, - { - "left": "issue:39692", - "right": "issue:44190", - "accept": false, - "reason": "SigLIP2 docs/model-processor mismatch has nothing to do with loading a local dataset in an image-classification training script." - }, - { - "left": "issue:44112", - "right": "issue:45600", + "left": "issue:41720", + "right": "issue:45245", "accept": false, - "reason": "Both are CI-related, but one is a stale device override test and the other is a removed Sam3LiteText config reference in auto_mappings.py; different breakages." + "reason": "Unrelated runtime failures: Qwen3 auto device mapping cudaErrorAssert vs category-count limit in data processing." }, { - "left": "issue:44521", - "right": "issue:45381", + "left": "issue:43550", + "right": "issue:44190", "accept": false, - "reason": "Both touch multimodal chat/video behavior, but one is all-zero assistant masks in apply_chat_template and the other is wrong vision_position_ids for Qwen2.5-VL video input." + "reason": "Different subsystems and bugs: torch.compile/SDPA issue in Bamba vs dataset loading in an image classification example." }, { - "left": "issue:43540", - "right": "issue:44734", + "left": "issue:44077", + "right": "issue:44877", "accept": false, - "reason": "A video-input ValueError in Qwen3OmniMoe is unrelated to a KV-cache continuation indexing crash in the serve response path." + "reason": "Both concern config handling, but they are different models and distinct validation issues." }, { - "left": "issue:44737", - "right": "issue:45803", + "left": "issue:44464", + "right": "issue:45563", "accept": false, - "reason": "XLNet CPU device placement in torch.arange and a bare except swallowing errors in FuyuBatchFeature are unrelated bug classes." + "reason": "Chunked generation inconsistency under compilation is unrelated to a stale warning about num_return_sequences." }, { - "left": "issue:43295", - "right": "issue:45276", + "left": "issue:45381", + "right": "issue:46655", "accept": false, - "reason": "One is a regression around processor.tokenizer/images access, the other is resize_token_embeddings not updating gemma4 embedding/output fields; different code paths." + "reason": "Both are video-input bugs, but the models and defect types differ: wrong vision_position_ids vs incorrect padding calculation." }, { - "left": "issue:44938", - "right": "issue:45542", + "left": "issue:43335", + "right": "issue:44190", "accept": false, - "reason": "Python 3.14 import failure and tensorboard-without-TensorFlow backend error are unrelated environment/package issues." + "reason": "No meaningful overlap: SwitchTransformers sparse-layer config bug vs dataset loading script failure." }, { - "left": "issue:45276", - "right": "issue:45584", + "left": "issue:44617", + "right": "issue:46369", "accept": false, - "reason": "Gemma4 embedding resize behavior and Whisper empty-transcription generation are unrelated." + "reason": "Both mention memory pressure, but they affect different Sam3 variants and likely different causes." }, { - "left": "issue:45632", - "right": "issue:46234", + "left": "issue:45127", + "right": "issue:45357", "accept": false, - "reason": "A cache-path collision for local trust_remote_code models is unrelated to a missing import_utils availability helper." + "reason": "Different issues in different workflows: LoRA merge with extended vocab vs save_pretrained visual encoder key handling." }, { - "left": "issue:44462", - "right": "issue:45907", - "accept": false, - "reason": "AutoTokenizer ignoring tokenizer.json and list_repo_files not forwarding revision are different hub/tokenizer issues." + "left": "issue:44779", + "right": "issue:46489", + "accept": true, + "reason": "Both describe the same DeepSeek tokenizer regression on v5+: incorrect outputs compared with v4, with 46489 explicitly calling out the remaining gap in the prior fix." } ] }, - "evaluator_result": null, - "evaluator_used": false, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet, and the soft-edge verdicts stay conservative. The DeepSeek tokenizer pair (44779/46489) is the only clearly duplicate-looking match from the provided titles, while the other pairs are reasonably rejected as different models, subsystems, or failure modes." + }, + "evaluator_used": true, "retried": false, - "accepted_nontrivial_soft_edge": false, + "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null }, @@ -11179,7 +11074,6 @@ "issue:44964", "issue:44987", "issue:45003", - "issue:45042", "issue:45083", "issue:45216", "issue:45245", @@ -11188,12 +11082,15 @@ "issue:45446", "issue:45584", "issue:45600", + "issue:45632", "issue:46082", "issue:46121", + "issue:46234", "issue:46424", "issue:46506" ], "soft_pairs": [ + "issue:45632|issue:46234", "issue:44246|issue:45003", "issue:42796|issue:44521", "issue:46082|issue:46121", @@ -11204,15 +11101,15 @@ "issue:44112|issue:44964", "issue:43643|issue:44496", "issue:27957|issue:35141", - "issue:36246|issue:41628", "issue:44186|issue:44734", + "issue:36246|issue:41628", "issue:43404|issue:45216", "issue:44295|issue:46506", "issue:36246|issue:38617", "issue:43650|issue:44220", "issue:41950|issue:44016", - "issue:32090|issue:35141", "issue:43479|issue:45375", + "issue:32090|issue:35141", "issue:43673|issue:46424", "issue:43531|issue:45600", "issue:44016|issue:44220", @@ -11220,205 +11117,204 @@ "issue:43295|issue:45335", "issue:43526|issue:43901", "issue:44683|issue:45446", - "issue:43976|issue:44987", - "issue:44496|issue:45042" + "issue:43976|issue:44987" ], - "prepared_review_unit_hash": "0bcf2ace67433309a1506db7304d48ffe8d7792968c2c7cd349d3de23b0e0cc8", + "prepared_review_unit_hash": "c9cb3e6141a9065d9230acae04aa49374ff680c3558096a31fb275e1c50320eb", "component_budget": { "node_count": 473, "item_count": 473, "soft_pair_count": 1738, - "serialized_chars": 901383, - "estimated_input_tokens": 225346, - "estimated_eval_tokens": 450948 + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 }, "budget": { - "node_count": 47, - "item_count": 47, + "node_count": 48, + "item_count": 48, "soft_pair_count": 28, - "serialized_chars": 35531, - "estimated_input_tokens": 8883, - "estimated_eval_tokens": 18022 + "serialized_chars": 36129, + "estimated_input_tokens": 9033, + "estimated_eval_tokens": 18322 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "These items are mostly unrelated, spanning distinct import/config/model-specific regressions; none of the soft pairs look like true duplicates.", - "confidence": 0.74, - "canonical_issue_reason": "issue:45003 is the clearest concrete code-path bug with broad impact; it is more actionable than the many model-specific or environment-specific reports here.", + "summary": "This set is mostly a grab bag of unrelated closed issues across imports, multimodal configs, tokenizers, cache behavior, CI, and docs. The soft links are only superficial keyword/subsystem overlaps, so none look like true duplicates or mergeable PR-equivalents.", + "confidence": 0.94, + "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "issue:45003 is the best representative overall because it describes a shared internal bug rather than a downstream symptom or narrow model/regression report.", + "best_issue_reason": "Issue 44964 is the strongest standalone bug report here: it describes a concrete, reproducible model-loading failure with a clear target and user impact. It is not a duplicate hub for the rest of the set, but it is the best representative issue if one must be chosen.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:45632", + "right": "issue:46234", + "accept": false, + "reason": "Both touch trust_remote_code/import utilities, but one is a cache-path collision and the other is a missing availability helper; different bugs." + }, { "left": "issue:44246", "right": "issue:45003", "accept": false, - "reason": "Both concern imports/performance, but one is an import latency complaint and the other is unsafe sys.modules access; different bugs and fixes." + "reason": "Both mention import behavior, but one is slow imports and the other is unsafe sys.modules access; not the same failure mode." }, { "left": "issue:42796", "right": "issue:44521", "accept": false, - "reason": "Tokenizer BPE marker handling and multimodal assistant mask generation are different code paths with different failure modes." + "reason": "Tokenizer BPE-marker handling and multimodal assistant_masks are unrelated bugs in different code paths." }, { "left": "issue:46082", "right": "issue:46121", "accept": false, - "reason": "Both are config-validation related, but one is about explicit head_dim handling and the other is a type conversion bug in rope validation." + "reason": "One is LlamaConfig head_dim validation, the other is rope-parameter JSON list handling; distinct configuration bugs." }, { "left": "issue:42175", "right": "issue:44279", "accept": false, - "reason": "A packaging/backend dependency issue is not the same as a vague dependency problem; too broad to be the same underlying bug." + "reason": "Backend packaging dependency mismatch vs a vague dependency issue; no shared concrete code-path problem." }, { "left": "issue:45245", "right": "issue:45584", "accept": false, - "reason": "Category-count limits and Whisper empty-transcription handling are unrelated defects." + "reason": "Category-count limit and Whisper empty transcription are unrelated runtime issues." }, { "left": "issue:44964", "right": "issue:45600", "accept": false, - "reason": "Phi-4 multimodal loading and stale auto_mappings entries for Sam3LiteText are different model/config breakages." + "reason": "Model loading failure for Phi-4 multimodal vs stale auto_mappings references; different underlying causes." }, { "left": "issue:43854", "right": "issue:44964", "accept": false, - "reason": "Both are model loading failures, but they affect different models and likely different loading paths." + "reason": "Both are model-loading failures, but they involve different models and likely different loader problems." }, { "left": "issue:44112", "right": "issue:44964", "accept": false, - "reason": "A CI test flake in GraniteSpeech is unrelated to a Phi-4 multimodal load regression." + "reason": "A CI test failure in GraniteSpeech is not the same as Phi-4 multimodal load regression." }, { "left": "issue:43643", "right": "issue:44496", "accept": false, - "reason": "Missing fields after trust_remote_code and an unrecognized model_type error are distinct config-loading problems." + "reason": "Both are config/load errors, but one drops fields with trust_remote_code while the other is an unrecognized model_type; not duplicates." }, { "left": "issue:27957", "right": "issue:35141", "accept": false, - "reason": "Flash Attention support for XLM-Roberta and embedding reinitialization after resize_token_embeddings are separate bugs." + "reason": "Flash Attention compatibility and embedding reinitialization after resize_token_embeddings are different bugs." }, { - "left": "issue:36246", - "right": "issue:41628", + "left": "issue:44186", + "right": "issue:44734", "accept": false, - "reason": "These are different import errors for different symbols; the underlying fixes would not be the same." + "reason": "Tokenizer padding/truncation crash vs serve KV-cache continuation indexing bug; different paths and symptoms." }, { - "left": "issue:44186", - "right": "issue:44734", + "left": "issue:36246", + "right": "issue:41628", "accept": false, - "reason": "Tokenizer crashes on NER/padding and server KV-cache continuation indexing are unrelated." + "reason": "Missing Qwen2_5_VLImageProcessor export vs missing AutoImageProcessor import are related only at a high level, not the same bug." }, { "left": "issue:43404", "right": "issue:45216", "accept": false, - "reason": "Weight tying in a multimodal generation model and a broken checkpoint save_pretrained regression are different issues." + "reason": "Weight tying in Mistral3ForConditionalGeneration and save_pretrained checkpoint correctness for Qwen3.5 are separate model-specific issues." }, { "left": "issue:44295", "right": "issue:46506", "accept": false, - "reason": "A buffer/position_ids access error is not the same as CPU/CUDA output divergence in PPDocLayoutV3." + "reason": "Position_ids buffer access error and CPU/CUDA output mismatch in PPDocLayoutV3 are unrelated correctness problems." }, { "left": "issue:36246", "right": "issue:38617", "accept": false, - "reason": "These import errors involve different missing symbols in different modules, not one shared defect." + "reason": "Missing model export and missing configuration_utils symbol are different import/export regressions." }, { "left": "issue:43650", "right": "issue:44220", "accept": false, - "reason": "The placeholder title \"ADD THE DATA\" does not describe the same audio feature-extraction bug." + "reason": "Placeholder text issue and fbank feature extraction bug are not the same underlying problem." }, { "left": "issue:41950", "right": "issue:44016", "accept": false, - "reason": "A video-classification pipeline bug and a notebook syntax error are unrelated." + "reason": "Video pipeline resolving image processors and a notebook syntax error are entirely unrelated." }, { - "left": "issue:32090", - "right": "issue:35141", + "left": "issue:43479", + "right": "issue:45375", "accept": false, - "reason": "Trainer GPU broadcast NoneType handling and embedding resize/post_init reinitialization are different subsystems." + "reason": "Different multimodal config classes and different missing/default fields; no shared bug." }, { - "left": "issue:43479", - "right": "issue:45375", + "left": "issue:32090", + "right": "issue:35141", "accept": false, - "reason": "Different multimodal config bugs in different models; one default-initializes None configs, the other drops a field under strict parsing." + "reason": "Trainer broadcast TypeError and embedding reinitialization after resize are unrelated." }, { "left": "issue:43673", "right": "issue:46424", "accept": false, - "reason": "Generic cache missing in chunked prefill and max_cache_length only applying to quantized cache are not the same bug." + "reason": "Both involve cache, but one is a missing GenerationMixin cache in v5 and the other is max_cache_length handling for Quantized Cache only." }, { "left": "issue:43531", "right": "issue:45600", "accept": false, - "reason": "A sliding_window issue in Qwen3-MoE is unrelated to auto_mappings referencing removed configs." + "reason": "Qwen3-MoE sliding_window behavior and stale auto_mappings references are different issues." }, { "left": "issue:44016", "right": "issue:44220", "accept": false, - "reason": "Notebook syntax and audio feature extraction are completely different problems." + "reason": "Notebook syntax error and audio feature-extraction bug do not share a code-path." }, { "left": "issue:45083", "right": "issue:46506", "accept": false, - "reason": "A helper length-calculation bug in qwen3_omni_moe is not the same as CPU/CUDA numerical divergence." + "reason": "Helper length computation in qwen3_omni_moe vs CPU/CUDA divergence in PPDocLayoutV3 are unrelated." }, { "left": "issue:43295", "right": "issue:45335", "accept": false, - "reason": "Processor/tokenizer regression and decoder.embed_tokens not updating on resize are different failure points." + "reason": "Processor/tokenizer regression and untied decoder.embed_tokens after resize are different problems." }, { "left": "issue:43526", "right": "issue:43901", "accept": false, - "reason": "An image-processor label reduction bug and stale docs for return_all_scores are unrelated." + "reason": "ImageProcessorFast reduce_labels bug and stale docs for return_all_scores are unrelated." }, { "left": "issue:44683", "right": "issue:45446", "accept": false, - "reason": "Both touch flex_attention, but one is a compilation failure on new torch versions while the other is a version-check/import gating bug." + "reason": "Both are flex_attention-related, but one is a torch>=2.9 compilation failure and the other is a version-check/import guard bug; not the same fix." }, { "left": "issue:43976", "right": "issue:44987", "accept": false, - "reason": "A Python version compatibility issue and a specific model-loading failure are different root causes." - }, - { - "left": "issue:44496", - "right": "issue:45042", - "accept": false, - "reason": "Unrecognized model_type and PIL image processors requiring torchvision are unrelated config vs backend dependency bugs." + "reason": "A Python-version compatibility issue and a specific model-loading failure on 5.1.0 are too different to merge." } ] }, @@ -11455,7 +11351,6 @@ "issue:43650", "issue:43673", "issue:43824", - "issue:43854", "issue:43901", "issue:43976", "issue:44016", @@ -11472,6 +11367,7 @@ "issue:44792", "issue:44933", "issue:45020", + "issue:45042", "issue:45200", "issue:45216", "issue:45278", @@ -11492,6 +11388,7 @@ "issue:46515" ], "soft_pairs": [ + "issue:44496|issue:45042", "issue:42907|issue:43441", "issue:43901|issue:44496", "issue:43381|issue:45593", @@ -11500,8 +11397,8 @@ "issue:43441|issue:45588", "issue:46123|issue:46369", "issue:44016|issue:44062", - "issue:44792|issue:46424", "issue:43824|issue:44933", + "issue:44792|issue:46424", "issue:44496|issue:45759", "issue:43299|issue:45216", "issue:44279|issue:46515", @@ -11509,28 +11406,27 @@ "issue:46082|issue:46399", "issue:44279|issue:45405", "issue:44075|issue:44368", - "issue:43577|issue:45216", "issue:42796|issue:43650", + "issue:43577|issue:45216", "issue:39692|issue:43901", "issue:43295|issue:44779", "issue:43976|issue:45405", "issue:44351|issue:45278", + "issue:46082|issue:46512", "issue:43299|issue:45357", "issue:44242|issue:45200", - "issue:46082|issue:46512", "issue:44779|issue:45812", "issue:41950|issue:45020", - "issue:38175|issue:44617", - "issue:43299|issue:43854" + "issue:38175|issue:44617" ], - "prepared_review_unit_hash": "ad6aa6229697ce8e73ebc3f285705c1cc1cf52c5b1721d151fe1db37b15b09e2", + "prepared_review_unit_hash": "15f37f944dd2babb6f9e11ffacce1098a64f683e64760865951fd68e3489baba", "component_budget": { "node_count": 473, "item_count": 473, "soft_pair_count": 1738, - "serialized_chars": 901383, - "estimated_input_tokens": 225346, - "estimated_eval_tokens": 450948 + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 }, "budget": { "node_count": 48, @@ -11545,192 +11441,192 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "These items are mostly unrelated issue reports across different models, pipelines, and regressions. The only overlaps are broad topical similarities (tokenizers, import errors, model loading, save/load regressions), but they do not look like the same underlying bug or change.", - "confidence": 0.92, - "canonical_issue_reason": "issue:45812 is the strongest standalone bug report: it is a clear, concrete v4\u2192v5 regression with a specific symptom and broad impact across the Granite family.", + "summary": "This set is a loose cluster of Transformers regression/model-specific bugs, not a duplicate set. The closest thematic umbrella is recent-version breakage and import/loading regressions, but the individual reports mostly target different code paths and models.", + "confidence": 0.82, + "canonical_issue_reason": "issue:45278 is the broadest representative of the cluster: it captures multiple import failures after upgrading Transformers, which matches several other version-regression reports better than the more model-specific items.", "canonical_pr_reason": null, - "best_issue_reason": "issue:45812 is the best representative issue here because it is specific, reproducible, and describes a single well-bounded failure mode.", + "best_issue_reason": "issue:45020 is the best single issue for a cluster label because it is the most general statement of the shared theme: recent Transformers versions breaking model loading, especially with remote code.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:44496", + "right": "issue:45042", + "accept": false, + "reason": "Both are model-loading failures, but one is an unrecognized-model/config issue and the other is PIL image processor requiring torchvision; different root causes and code paths." + }, { "left": "issue:42907", "right": "issue:43441", "accept": false, - "reason": "Same model family, but one is save/dequantization failure and the other is a FlashAttention runtime crash." + "reason": "Both mention Ministral, but one is about saving dequantized models and the other is FlashAttention initialization; not the same underlying bug." }, { "left": "issue:43901", "right": "issue:44496", "accept": false, - "reason": "Both involve docs/model names, but one is a pipeline docs mismatch and the other is an unrecognized-model config loading error." + "reason": "One is a docs mismatch for TextClassificationPipeline, the other is an unrecognized-model config error; unrelated problems." }, { "left": "issue:43381", "right": "issue:45593", "accept": false, - "reason": "Different bugs in different code paths: gradient checkpointing eval behavior vs missing auxiliary losses in D-FINE." + "reason": "Gradient checkpointing in eval mode and missing auxiliary losses when denoising is off are different training-path bugs." }, { "left": "issue:43976", "right": "issue:44279", "accept": false, - "reason": "One is a Python-version compatibility regression; the other is a vague dependency issue with no concrete shared code path." + "reason": "Python version compatibility and a generic dependency issue are too vague to be the same concrete bug." }, { "left": "issue:43673", "right": "issue:44336", "accept": false, - "reason": "Chunked-prefill cache handling and ANSI codes in loading_report are unrelated subsystems." + "reason": "Chunked prefill cache handling and ANSI codes in loading_report are unrelated." }, { "left": "issue:43441", "right": "issue:45588", "accept": false, - "reason": "Both mention flash attention, but they target different failures: Ministral model init vs s_aux=None handling in a separate integration file." + "reason": "FlashAttention failure for Ministral and a sink-less-model crash in a different flash_attention integration path are not the same bug." }, { "left": "issue:46123", "right": "issue:46369", "accept": false, - "reason": "Pipeline batch-finalization dropping results is unrelated to SAM3 LiteText memory overflow." + "reason": "Both concern memory/performance symptoms in SAM3-related work, but one is a dropped-results batching bug and the other is a memory overflow report." }, { "left": "issue:44016", "right": "issue:44062", "accept": false, - "reason": "Notebook syntax error and AddedToken keyword-argument conflict are different bugs." + "reason": "A notebook syntax error and an AddedToken keyword collision are distinct issues." }, { - "left": "issue:44792", - "right": "issue:46424", + "left": "issue:43824", + "right": "issue:44933", "accept": false, - "reason": "A janus test failure and a cache-length configuration limitation are not the same issue." + "reason": "Both are import-related, but one is a missing Qwen2_5_VL symbol and the other is a nonexistent image_utils import; different symbols and fixes." }, { - "left": "issue:43824", - "right": "issue:44933", + "left": "issue:44792", + "right": "issue:46424", "accept": false, - "reason": "Both are import errors, but for different symbols/modules and likely different fixes." + "reason": "A janus image-generation test failure and a cache config max length bug are unrelated." }, { "left": "issue:44496", "right": "issue:45759", "accept": false, - "reason": "Different model-loading problems: unrecognized model type vs from_config failing to unwrap text_config." + "reason": "Both involve model/config loading, but one is an unrecognized OLMo model and the other is unwrapping text_config for composite Qwen multimodal configs; different underlying problems." }, { "left": "issue:43299", "right": "issue:45216", "accept": false, - "reason": "One is a load-time failure for Qwen3.5 VL MoE models; the other is an incorrect save_pretrained checkpoint regression." + "reason": "Qwen3VL MoE loading failure and Qwen3.5 save_pretrained regression are different code paths." }, { "left": "issue:44279", "right": "issue:46515", "accept": false, - "reason": "These are generic incompatibility reports, but one is about transformers dependency issues and the other is torch 2.4.1 incompatibility." + "reason": "A generic dependency issue and a torch 2.4.1 incompatibility report are too nonspecific to merge as one bug." }, { "left": "issue:45335", "right": "issue:45584", "accept": false, - "reason": "resize_token_embeddings not affecting decoder embeddings is unrelated to Whisper empty-transcription generation after token alignment." + "reason": "resize_token_embeddings not affecting decoder embeddings and Whisper empty-transcription failure after token alignment are unrelated." }, { "left": "issue:46082", "right": "issue:46399", "accept": false, - "reason": "LlamaConfig head_dim validation and Qwen3.5 kernel registration errors are distinct config/runtime issues." + "reason": "A LlamaConfig head_dim validation issue and a missing decorator for apply_rotary_pos_emb registration are different failures." }, { "left": "issue:44279", "right": "issue:45405", "accept": false, - "reason": "A dependency issue and a bumped unreleased PEFT minimum version are different package/version problems." + "reason": "A generic dependency issue and a MIN_PEFT_VERSION bump are not the same bug; one is about requirements, the other about release availability." }, { "left": "issue:44075", "right": "issue:44368", "accept": false, - "reason": "SGD args being ignored is unrelated to a tie_word_embeddings warning in Qwen fine-tuning." + "reason": "SGD args not being used and a tie_word_embeddings warning in Qwen3.5 fine-tuning are unrelated." }, { - "left": "issue:43577", - "right": "issue:45216", + "left": "issue:42796", + "right": "issue:43650", "accept": false, - "reason": "Blip2 dtype propagation and Qwen3.5 save_pretrained correctness are unrelated regression classes." + "reason": "Ministral tokenizer BPE-marker handling and the placeholder 'ADD THE DATA' issue do not describe the same technical problem." }, { - "left": "issue:42796", - "right": "issue:43650", + "left": "issue:43577", + "right": "issue:45216", "accept": false, - "reason": "Tokenizer BPE-marker handling and a low-signal data request are clearly unrelated." + "reason": "Blip2 dtype propagation and Qwen3.5 incorrect save_pretrained checkpointing are different model-specific bugs." }, { "left": "issue:39692", "right": "issue:43901", "accept": false, - "reason": "Both are documentation-related only in the broadest sense; the underlying bugs and affected components differ." + "reason": "SigLIP2 doc example errors and TextClassificationPipeline docs mentioning return_all_scores are both documentation-related, but they are separate examples and behaviors." }, { "left": "issue:43295", "right": "issue:44779", "accept": false, - "reason": "Both mention tokenizer regressions, but one is processor/tokenizer API handling and the other is incorrect tokenization output for Deepseek." + "reason": "A regression around processor.tokenizer/images handling is not the same as Deepseek tokenizer output corruption." }, { "left": "issue:43976", "right": "issue:45405", "accept": false, - "reason": "Python support regression and unreleased dependency bump are different versioning problems." + "reason": "Python version support and unreleased PEFT minimum version are different compatibility problems." }, { "left": "issue:44351", "right": "issue:45278", "accept": false, - "reason": "A missing HybridCache import may be one symptom of broader import errors, but this is not enough to treat them as the same concrete bug." + "reason": "Missing HybridCache import and broad import errors after upgrade are both import-related, but not the same missing symbol or fix." + }, + { + "left": "issue:46082", + "right": "issue:46512", + "accept": false, + "reason": "Explicit head_dim validation and stale model.config.dtype after from_config(dtype=...) are unrelated configuration bugs." }, { "left": "issue:43299", "right": "issue:45357", "accept": false, - "reason": "Both are Qwen3.5 regressions, but one is model loading and the other is incorrect visual encoder keys on save." + "reason": "Qwen3VL loading failure and incorrect visual encoder keys on save_pretrained are separate multimodal model regressions." }, { "left": "issue:44242", "right": "issue:45200", "accept": false, - "reason": "Load-balancing loss logic and Gemma 4 mm_token_type_ids defaults are unrelated training issues." - }, - { - "left": "issue:46082", - "right": "issue:46512", - "accept": false, - "reason": "Both touch config/shape semantics, but one rejects head_dim inputs while the other leaves config.dtype stale after from_config." + "reason": "Load balancing loss gating and Gemma 4 mm_token_type_ids defaults are different training semantics bugs." }, { "left": "issue:44779", "right": "issue:45812", "accept": false, - "reason": "Both are tokenizer regressions, but they affect different model families and different failure modes, so they are not the same bug." + "reason": "Deepseek tokenizer regression and Granite AutoTokenizer wrong token IDs are both tokenizer issues, but they affect different models and likely different code paths." }, { "left": "issue:41950", "right": "issue:45020", "accept": false, - "reason": "A video-classification processor bug and a broad remote_code compatibility regression are different issues." + "reason": "A video-classification pipeline processor lookup bug and broad remote_code model breakage are not the same bug." }, { "left": "issue:38175", "right": "issue:44617", "accept": false, - "reason": "Unexpected zero probabilities in SigLIP2 and SAM3Video CUDA OOM are unrelated model-specific failures." - }, - { - "left": "issue:43299", - "right": "issue:43854", - "accept": false, - "reason": "Different Qwen-related import/loading failures with different missing symbols and scopes." + "reason": "Unexpected zero probabilities in SigLIP2 and SAM3 video CUDA OOM are completely different model/runtime issues." } ] }, @@ -11758,6 +11654,7 @@ "issue:41950", "issue:42175", "issue:42994", + "issue:43299", "issue:43317", "issue:43441", "issue:43479", @@ -11767,12 +11664,12 @@ "issue:43673", "issue:43723", "issue:43761", + "issue:43854", "issue:43901", "issue:43976", "issue:43994", "issue:44016", "issue:44162", - "issue:44246", "issue:44261", "issue:44263", "issue:44291", @@ -11803,11 +11700,12 @@ "issue:46209" ], "soft_pairs": [ + "issue:43299|issue:43854", "issue:43479|issue:43525", "issue:43650|issue:46209", - "issue:44683|issue:45542", "issue:41950|issue:45823", "issue:44466|issue:45276", + "issue:44683|issue:45542", "issue:44521|issue:45325", "issue:45103|issue:45372", "issue:43653|issue:45593", @@ -11827,200 +11725,199 @@ "issue:43901|issue:43994", "issue:43441|issue:44683", "issue:43673|issue:43761", - "issue:43723|issue:44162", "issue:45542|issue:45800", - "issue:43901|issue:44246" + "issue:43723|issue:44162" ], - "prepared_review_unit_hash": "b6d17db80c0b65b5c9106f59050c72b44a530109b9e04f596938877085713a76", + "prepared_review_unit_hash": "298c8000ec09cdfdabaa3272bfdca135dad91c0dab865c83594cf781e98df003", "component_budget": { "node_count": 473, "item_count": 473, "soft_pair_count": 1738, - "serialized_chars": 901383, - "estimated_input_tokens": 225346, - "estimated_eval_tokens": 450948 + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 }, "budget": { - "node_count": 47, - "item_count": 47, + "node_count": 48, + "item_count": 48, "soft_pair_count": 27, - "serialized_chars": 35215, - "estimated_input_tokens": 8804, - "estimated_eval_tokens": 17864 + "serialized_chars": 35825, + "estimated_input_tokens": 8957, + "estimated_eval_tokens": 18170 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is actually a loose collection of unrelated Transformers issues; none of the soft pairs look like true duplicates, so I would not merge any edges. The best representative issue is the tokenizer-loading regression family, with issue 44462 as the most actionable anchor.", - "confidence": 0.93, - "canonical_issue_reason": "Issue 44462 is the most concrete, broadly relevant regression in the set and has follow-up references, making it the best representative anchor even though the overall cluster is heterogeneous.", + "summary": "These issues are largely unrelated single-bug reports across different models, configs, tokenizer paths, and environment problems; I don\u2019t see any true duplicate pairs worth merging.", + "confidence": 0.69, + "canonical_issue_reason": "issue:44521 is the strongest standalone report here: it has a clear regression description, reproducer focus, and active discussion/inbound references, making it the best representative issue in this set.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 44462 has a clear user-facing repro and a well-scoped code path, so it is the strongest issue to keep as the cluster\u2019s representative.", + "best_issue_reason": "issue:44521 is the most actionable and well-scoped issue among a mostly unrelated set, with enough detail to serve as a reference point if needed.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:43299", + "right": "issue:43854", + "accept": false, + "reason": "Both are model-loading failures, but they concern different models and different failure modes (Qwen3VL MoE vs GLM-4.7-Flash unit-test loading)." + }, { "left": "issue:43479", "right": "issue:43525", "accept": false, - "reason": "Different bugs: one is bad default multimodal config initialization, the other is a missing pad_token_id on Llama4Config." + "reason": "Different config bugs: one misinitializes Phi4MultimodalConfig subconfigs, the other lacks pad_token_id on Llama4Config." }, { "left": "issue:43650", "right": "issue:46209", "accept": false, - "reason": "Completely unrelated: a vague data request versus an FP8 inference failure on B200." - }, - { - "left": "issue:44683", - "right": "issue:45542", - "accept": false, - "reason": "Different subsystems and failures: torch flex_attention compilation vs a TensorFlow/backend installation error." + "reason": "Unrelated topics: a vague data request versus an FP8 inference failure on B200." }, { "left": "issue:41950", "right": "issue:45823", "accept": false, - "reason": "Pipeline processor lookup bug and Gemma4 device-mismatch bug are distinct code paths." + "reason": "Different subsystems and symptoms: pipeline processor selection vs Gemma4 device-map forward mismatch." }, { "left": "issue:44466", "right": "issue:45276", "accept": false, - "reason": "Both touch model weights, but one is serialization inconsistency and the other is resize_token_embeddings not updating Gemma4 internals." + "reason": "Both involve Gemma/v5 model state handling, but one is serialization of tied lm_head weights and the other is resize_token_embeddings behavior." + }, + { + "left": "issue:44683", + "right": "issue:45542", + "accept": false, + "reason": "Completely different bugs: torch>=2.9 flex_attention compilation vs a TensorFlow backend dependency/install error." }, { "left": "issue:44521", "right": "issue:45325", "accept": false, - "reason": "Both are multimodal, but one is assistant mask generation and the other is RoPE/position-id scaling." + "reason": "Both are multimodal/VL related, but one is chat-template assistant masks and the other is Qwen2.5-VL rope/position-id scaling." }, { "left": "issue:45103", "right": "issue:45372", "accept": false, - "reason": "Auto-docstring annotation handling and Gemma4 processor import failure are unrelated bugs." + "reason": "Different failure classes: auto_docstring annotation handling vs Gemma processor loading due to a missing mistral_common import." }, { "left": "issue:43653", "right": "issue:45593", "accept": false, - "reason": "Tokenizer special-token registration and D-FINE auxiliary-loss logic are unrelated." + "reason": "Tokenizer special-token registration bug versus a training-loss issue in D-FINE with denoising disabled." }, { "left": "issue:44462", "right": "issue:45081", "accept": false, - "reason": "Both involve tokenizer loading, but the former is generic tokenizer.json resolution while the latter is a Mistral-specific regex/backend tokenizer crash." + "reason": "Both are tokenizer-related, but one is repository tokenizer discovery and the other is a Mistral regex patch crash on a different code path." }, { "left": "issue:44016", "right": "issue:44291", "accept": false, - "reason": "Notebook syntax error and init_empty_weights TypeError are unrelated failure modes." + "reason": "Notebook syntax error and init_empty_weights argument mismatch are unrelated." }, { "left": "issue:43441", "right": "issue:45834", "accept": false, - "reason": "Different model-specific issues: FlashAttention failure for Ministral versus OCR index overflow for Kosmos2.5." + "reason": "Different models and bugs: FlashAttention failure for Ministral-3 versus OCR index error in Kosmos2.5." }, { "left": "issue:33290", "right": "issue:45656", "accept": false, - "reason": "Both mention DeepSpeed, but one is Adafactor OOM and the other is optimizer.step being invoked twice." + "reason": "Both mention deepspeed/optimizer behavior, but one is Adafactor OOM and the other is optimizer.step being called twice." }, { "left": "issue:44361", "right": "issue:45593", "accept": false, - "reason": "Tokenizer AttributeError and D-FINE training-loss behavior are unrelated." + "reason": "Tokenizer AttributeError on tasks is unrelated to D-FINE auxiliary-loss behavior." }, { "left": "issue:45081", "right": "issue:45907", "accept": false, - "reason": "Tokenizer regex patch crash and list_repo_files revision forwarding are unrelated." + "reason": "Different issues: a tokenizer regex patch crash versus repo file listing not forwarding revision." }, { "left": "issue:42994", "right": "issue:43317", "accept": false, - "reason": "Both are quantization-related, but one is save failure and the other is dequantized load/offload failure." + "reason": "Both touch quantization/dequantization, but one is saving quantized models and the other is loading a dequantized model with device_map=auto." }, { "left": "issue:44261", "right": "issue:44263", "accept": false, - "reason": "Different model internals: config rms_norm_eps precision issue vs torch.split return values in GlmMoeDsaIndexer." + "reason": "Distinct bugs in different components: MLA q_a_layernorm config precision vs GlmMoeDsaIndexer split return values." }, { "left": "issue:44623", "right": "issue:44749", "accept": false, - "reason": "Processor save-pretrained file omission and a data-filtering performance regression are not the same bug." + "reason": "Processor save_pretrained file persistence is unrelated to the Chinese report about slower filtering after upgrade." }, { "left": "issue:45600", "right": "issue:45712", "accept": false, - "reason": "Both are repo hygiene/CI issues, but they affect different generated files and different invariants." + "reason": "Both are CI/repo-integrity related, but one is stale auto_mappings references and the other is leftover dummy_pt_objects classes." }, { "left": "issue:44623", "right": "issue:45397", "accept": false, - "reason": "Processor saving missing files and Gemma4 zero3 from_pretrained are distinct load/save paths." + "reason": "Different code paths: processor.save_pretrained missing files versus Gemma-4 zero3 from_pretrained." }, { "left": "issue:42175", "right": "issue:43976", "accept": false, - "reason": "Packaging/install dependency issue and Python version compatibility issue are different release problems." + "reason": "Different environment problems: missing TensorFlow in a torch install versus Python version incompatibility in Transformers 5.1.0." }, { "left": "issue:44671", "right": "issue:44779", "accept": false, - "reason": "Both are v5 regressions, but they concern different models/tokenizers and different incorrect outputs." + "reason": "Both are v5 regressions in model outputs, but they affect different models and likely different tokenization/modeling paths." }, { "left": "issue:43901", "right": "issue:43994", "accept": false, - "reason": "Docs mismatch for return_all_scores and a SigLIP2 inference correctness bug are not duplicates." + "reason": "Docs mismatch for return_all_scores is not the same bug as nonsensical SigLIP2 model/pipeline outputs." }, { "left": "issue:43441", "right": "issue:44683", "accept": false, - "reason": "Both mention attention, but they are different implementations and failures: FlashAttention compatibility vs compiled flex_attention on newer torch." + "reason": "Different failure modes: Ministral FlashAttention compatibility versus compiled flex_attention failing on newer torch." }, { "left": "issue:43673", "right": "issue:43761", "accept": false, - "reason": "Generation cache handling and CLIPVisionModel hidden_states output are separate API regressions." - }, - { - "left": "issue:43723", - "right": "issue:44162", - "accept": false, - "reason": "Tokenizer loading regression and ESM2 breakage are unrelated model-level issues." + "reason": "Unrelated regressions: chunked_prefill cache handling versus CLIPVisionModel hidden_states not returned." }, { "left": "issue:45542", "right": "issue:45800", "accept": false, - "reason": "TensorFlow backend installation bug and torch/transformers version incompatibility are different setup problems." + "reason": "TensorBoard/TensorFlow backend install issue is unrelated to torch/Transformers version incompatibility." }, { - "left": "issue:43901", - "right": "issue:44246", + "left": "issue:43723", + "right": "issue:44162", "accept": false, - "reason": "A docs inconsistency and intermittent import slowness are unrelated." + "reason": "AutoTokenizer loading regression and ESM2 model breakage are different problems with no shared concrete code path." } ] }, @@ -12060,6 +11957,7 @@ "issue:43784", "issue:43824", "issue:43881", + "issue:43901", "issue:43976", "issue:44016", "issue:44188", @@ -12069,7 +11967,6 @@ "issue:44295", "issue:44297", "issue:44336", - "issue:44351", "issue:44373", "issue:44485", "issue:44496", @@ -12094,20 +11991,21 @@ "issue:46665" ], "soft_pairs": [ + "issue:43901|issue:44246", "issue:46291|issue:46399", - "issue:44016|issue:45823", "issue:43525|issue:43881", + "issue:44016|issue:45823", "issue:43824|issue:43976", - "issue:43525|issue:43643", "issue:44188|issue:44336", + "issue:43525|issue:43643", "issue:44589|issue:45362", - "issue:44297|issue:45397", "issue:44336|issue:44373", + "issue:44297|issue:45397", "issue:43784|issue:45542", "issue:41628|issue:43976", "issue:44246|issue:44496", - "issue:46614|issue:46665", "issue:45561|issue:45563", + "issue:46614|issue:46665", "issue:44242|issue:45593", "issue:45103|issue:46399", "issue:42491|issue:43299", @@ -12121,211 +12019,210 @@ "issue:44336|issue:46424", "issue:41950|issue:44220", "issue:43540|issue:46655", - "issue:36246|issue:42757", - "issue:36246|issue:44351" + "issue:36246|issue:42757" ], - "prepared_review_unit_hash": "96716d691d3e4d714620abc84eaa7026b9357a8bb397c79d5bfae413a016a485", + "prepared_review_unit_hash": "b09e42712fb5202ed33e9ebdf4975d9d17d03fa9e7fa1ddad7339ed113de4e26", "component_budget": { "node_count": 473, "item_count": 473, "soft_pair_count": 1738, - "serialized_chars": 901383, - "estimated_input_tokens": 225346, - "estimated_eval_tokens": 450948 + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 29, - "serialized_chars": 36284, - "estimated_input_tokens": 9071, - "estimated_eval_tokens": 18398 + "serialized_chars": 36300, + "estimated_input_tokens": 9075, + "estimated_eval_tokens": 18406 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a mixed bag of Transformers import/export regressions and model-specific loading bugs, but the individual issues do not appear to be duplicates of one another. The soft links are generally only surface-similar, not the same underlying defect.", - "confidence": 0.44, - "canonical_issue_reason": "issue:43824 is the most central top-level import regression in the set: it targets a public Transformers export, has strong discussion activity, and is representative of the broader import/load failures without being too niche.", + "summary": "This cluster is thematically related (mostly Transformers/HF bug reports), but the soft pairs are not true duplicates: they usually differ by model family, API surface, or failure mode. No PRs are present, so no PR canonical applies.", + "confidence": 0.93, + "canonical_issue_reason": "issue:41628 is the broadest, most generic user-facing failure here: a top-level Transformers import error. It best represents the cluster\u2019s import/API breakage theme without overfitting to a single model family.", "canonical_pr_reason": null, - "best_issue_reason": "issue:43824 is the best anchor for this cluster because it is a high-visibility public API import failure and is closest to the recurring theme of broken model/component exports.", + "best_issue_reason": "issue:41628 is the strongest exemplar for triage because it is concise, high-signal, and broadly applicable across the import-related subset of reports.", "best_pr_reason": null, "soft_edge_verdicts": [ { - "left": "issue:46291", - "right": "issue:46399", + "left": "issue:43901", + "right": "issue:44246", "accept": false, - "reason": "Both are import/registration failures, but one is about `transformers.pipelines` and kernels version metadata while the other is a Qwen3.5 kernel decorator crash; different code paths and fixes." + "reason": "Docs mismatch for text classification versus intermittent slow import; unrelated symptoms and code paths." }, { - "left": "issue:44016", - "right": "issue:45823", + "left": "issue:46291", + "right": "issue:46399", "accept": false, - "reason": "Notebook syntax error in docs vs Gemma4 device mismatch during forward; unrelated failures." + "reason": "Different failures: pipeline import conflict with kernels versioning vs Qwen3.5 kernel registration error." }, { "left": "issue:43525", "right": "issue:43881", "accept": false, - "reason": "Different model-loading bugs: `Llama4Config.pad_token_id` missing versus a `glm-4v-9b` load failure." + "reason": "Llama4 missing config field and GLM-4V loading failure are different model-specific bugs." }, { - "left": "issue:43824", - "right": "issue:43976", + "left": "issue:44016", + "right": "issue:45823", "accept": false, - "reason": "Qwen2.5-VL import/export regression versus a Python-version compatibility issue for Transformers 5.1.0; not the same bug." + "reason": "Notebook syntax error and Gemma4 device-mapping bug are unrelated." }, { - "left": "issue:43525", - "right": "issue:43643", + "left": "issue:43824", + "right": "issue:43976", "accept": false, - "reason": "A missing config field in Llama4 is not the same as `trust_remote_code=True` dropping fields from `AutoConfig`." + "reason": "Qwen2.5 VL import error versus Python-version compatibility issue; not the same bug." }, { "left": "issue:44188", "right": "issue:44336", "accept": false, - "reason": "Attention kernel divergence on `torch.compile` versus ANSI codes in `loading_report`; unrelated subsystems." + "reason": "torch.compile attention-kernel divergence is unrelated to ANSI output in loading_report." }, { - "left": "issue:44589", - "right": "issue:45362", + "left": "issue:43525", + "right": "issue:43643", "accept": false, - "reason": "Float8 storage object lookup failure versus Qwen3.5 chat crash; different symptoms and code paths." + "reason": "Missing pad_token_id in Llama4Config is unrelated to trust_remote_code returning incomplete AutoConfig fields." }, { - "left": "issue:44297", - "right": "issue:45397", + "left": "issue:44589", + "right": "issue:45362", "accept": false, - "reason": "Tokenizer config mismatch on save versus Gemma-4 Zero3 `from_pretrained` failure; not the same defect." + "reason": "Storage type lookup failure and Qwen3.5 chat crash are different runtime errors." }, { "left": "issue:44336", "right": "issue:44373", "accept": false, - "reason": "Terminal ANSI escape output issue versus a docstring mismatch; no shared underlying bug." + "reason": "Terminal ANSI formatting bug and wrong docstring are unrelated." + }, + { + "left": "issue:44297", + "right": "issue:45397", + "accept": false, + "reason": "Tokenizer config mismatch and Gemma-4 ZeRO3 loading failure are distinct issues." }, { "left": "issue:43784", "right": "issue:45542", "accept": false, - "reason": "`nn` import NameError in sentence-transformers integration versus backend detection error when only tensorboard is installed." + "reason": "sentence-transformers import NameError and undefined tf backend from tensorboard-only install are different dependency bugs." }, { "left": "issue:41628", "right": "issue:43976", "accept": false, - "reason": "Missing `AutoImageProcessor` export and a Python-version support issue are both import-related, but they are separate problems." + "reason": "Top-level import error versus Python support regression; not the same underlying defect." }, { "left": "issue:44246", - "right": "issue:44496", + "right": "issue:44871", "accept": false, - "reason": "Load-balancing loss handling in MoE versus an unrecognized model/config error; different failure modes." + "reason": "Slow import and Gemma eos_token_id inconsistency are unrelated." }, { - "left": "issue:46614", - "right": "issue:46665", + "left": "issue:45561", + "right": "issue:45563", "accept": false, - "reason": "Stale phonemizer language state versus inconsistent `from_pretrained` behavior inside Python classes; unrelated logic." + "reason": "pytest-xdist race in testing utils versus stale warning in paged generate are different code paths." }, { - "left": "issue:45561", - "right": "issue:45563", + "left": "issue:46614", + "right": "issue:46665", "accept": false, - "reason": "Pytest-xdist file write race versus a stale paging warning; different issues." + "reason": "Stale phonemizer language state and from_pretrained behavior inside Python classes are unrelated." }, { "left": "issue:44242", "right": "issue:45593", "accept": false, - "reason": "Router-logits loss omission and D-FINE auxiliary-loss omission are both loss-related, but they affect different models and code paths." + "reason": "Load-balancing loss omission in MoE routing is not the same as D-FINE auxiliary-loss behavior." }, { "left": "issue:45103", "right": "issue:46399", "accept": false, - "reason": "Auto-docstring annotation parsing crash versus Qwen3.5 kernel registration crash; different components." + "reason": "auto_docstring annotation crash and Qwen3.5 kernel registration bug are unrelated." }, { "left": "issue:42491", "right": "issue:43299", "accept": false, - "reason": "Both mention Qwen3-MoE family models, but one is a LoRA compatibility mismatch across hf4.x/hf5.x while the other is a main-branch loading breakage; not clearly the same bug." + "reason": "Both involve Qwen/MoE, but one is LoRA compatibility across branches and the other is a model loading regression; different bugs." }, { "left": "issue:43526", "right": "issue:45003", "accept": false, - "reason": "`reduce_labels` returning one label and unsafe `sys.modules` access are unrelated implementation bugs." + "reason": "BeitImageProcessorFast reduce_labels bug and sys.modules access in modeling_utils are unrelated." }, { "left": "issue:44295", "right": "issue:45397", "accept": false, - "reason": "Position IDs buffer read error versus Gemma-4 Zero3 loading failure; separate issues." + "reason": "position_ids buffer read error and Gemma-4 ZeRO3 loading failure are different failures." }, { "left": "issue:43531", "right": "issue:45362", "accept": false, - "reason": "Qwen3-MoE sliding-window issue versus Qwen3.5 chat crash; same family, but not the same concrete bug." + "reason": "Qwen3-MoE sliding_window issue and Qwen3.5 chat crash are distinct model behaviors." }, { "left": "issue:44246", "right": "issue:44871", "accept": false, - "reason": "MoE load-balancing loss omission versus Gemma-3 eos token configuration inconsistency; unrelated." + "reason": "Performance complaint about import time and eos_token_id config mismatch do not share the same underlying bug." }, { "left": "issue:44336", "right": "issue:44485", "accept": false, - "reason": "ANSI code emission in loading reports versus GLM-5 RoPE implementation discussion; unrelated." + "reason": "ANSI code emission in loading_report and GLM-5 RoPE implementation are unrelated." }, { "left": "issue:42886", "right": "issue:46132", "accept": false, - "reason": "Offline tokenizer cache loading failure versus `AttentionInterface.register` behavior change; different areas." + "reason": "Offline cache loading bug and AttentionInterface.register behavior change are different subsystems." }, { "left": "issue:43550", "right": "issue:46506", "accept": false, - "reason": "Bamba torch.compile SDPA failure versus CPU/CUDA result divergence in PPDocLayoutV3; not the same bug." + "reason": "torch.compile SDPA failure in Bamba and CPU/CUDA output divergence in PPDocLayoutV3 are not the same bug." }, { "left": "issue:44336", "right": "issue:46424", "accept": false, - "reason": "ANSI escape output problem versus cache-length config behavior; unrelated." + "reason": "Terminal ANSI formatting and cache max length semantics are unrelated." }, { "left": "issue:41950", "right": "issue:44220", "accept": false, - "reason": "Video-classification pipeline using image processors versus `_torch_extract_fbank_features()`; different pipelines and failures." + "reason": "video-classification pipeline image-processor lookup and fbank feature extraction are different media-processing paths." }, { "left": "issue:43540", "right": "issue:46655", "accept": false, - "reason": "Both involve video processing, but they are different models and different bugs: Qwen3OmniMoe video input handling versus SmolVLM padding calculation." + "reason": "Qwen3OmniMoe video-input processing error and SmolVLM padding bug are both video-related but clearly different defects." }, { "left": "issue:36246", "right": "issue:42757", "accept": false, - "reason": "Qwen2.5-VL image processor import error versus missing `is_offline_mode` from huggingface_hub; same symptom class, different dependency and fix." - }, - { - "left": "issue:36246", - "right": "issue:44351", - "accept": false, - "reason": "Both are import errors, but they concern different missing symbols (`Qwen2_5_VLImageProcessor` vs `HybridCache`) and likely separate export regressions." + "reason": "Qwen2.5 VL import failure and huggingface_hub is_offline_mode import failure are different package/API breaks." } ] }, @@ -12364,6 +12261,7 @@ "issue:44242", "issue:44246", "issue:44295", + "issue:44351", "issue:44466", "issue:44479", "issue:44496", @@ -12374,7 +12272,6 @@ "issue:44857", "issue:44877", "issue:44987", - "issue:44995", "issue:45003", "issue:45137", "issue:45200", @@ -12399,238 +12296,238 @@ "issue:46655" ], "soft_pairs": [ + "issue:36246|issue:44351", "issue:44779|issue:45292", - "issue:45859|issue:46399", "issue:45003|issue:45561", - "issue:43901|issue:45003", - "issue:44479|issue:46655", + "issue:45859|issue:46399", "issue:44734|issue:46424", + "issue:44479|issue:46655", + "issue:43901|issue:45003", "issue:43257|issue:45137", - "issue:44242|issue:44857", "issue:43475|issue:46369", "issue:44496|issue:45003", + "issue:44242|issue:44857", "issue:43673|issue:44561", "issue:44295|issue:44749", "issue:44162|issue:45200", "issue:43262|issue:44246", "issue:45375|issue:45759", - "issue:44877|issue:45375", "issue:36246|issue:43596", "issue:45292|issue:45584", - "issue:45137|issue:45600", "issue:44561|issue:46399", "issue:45292|issue:45701", + "issue:45137|issue:45600", + "issue:44877|issue:45375", "issue:44038|issue:46506", "issue:44466|issue:45335", - "issue:45412|issue:46123", "issue:44987|issue:45459", + "issue:45412|issue:46123", "issue:43352|issue:45250", "issue:44479|issue:46170", - "issue:44060|issue:44077", - "issue:44995|issue:45200" + "issue:44060|issue:44077" ], - "prepared_review_unit_hash": "56610eadaeb08f59984e9605e63f92e33ba9687407797ad6e54dbe61f77cc69a", + "prepared_review_unit_hash": "6daba197ba25058752b1833286e70c71657b3663b925274f2dfa2819e596d0e8", "component_budget": { "node_count": 473, "item_count": 473, "soft_pair_count": 1738, - "serialized_chars": 901383, - "estimated_input_tokens": 225346, - "estimated_eval_tokens": 450948 + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 29, - "serialized_chars": 36475, - "estimated_input_tokens": 9119, - "estimated_eval_tokens": 18494 + "serialized_chars": 36472, + "estimated_input_tokens": 9118, + "estimated_eval_tokens": 18492 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "These items are mostly unrelated one-off issues; the soft pairs share broad subsystems or model families but not the same concrete bug. I would not collapse any of the proposed pairs into duplicates, and there is no clear canonical issue/PR for this set.", - "confidence": 0.96, - "canonical_issue_reason": "No single issue stands out as the canonical duplicate target because the set spans many unrelated regressions and model-specific bugs.", - "canonical_pr_reason": "No pull requests are present in the candidate set.", - "best_issue_reason": "No issue is a strong global representative; the topics are too heterogeneous to justify a single best issue beyond being the most general/visible in its own thread.", - "best_pr_reason": "No pull requests are present in the candidate set.", + "summary": "This set is heterogeneous: import/export regressions, model-specific bugs, cache issues, docs mismatches, and unrelated runtime failures. None of the soft pairs look like true duplicates of the same underlying bug or change.", + "confidence": 0.88, + "canonical_issue_reason": "No single issue is a defensible canonical for the whole set: the items span unrelated models and subsystems, and the apparent similarities are mostly broad keyword overlap (imports, Qwen, video, cache, Deepspeed, etc.).", + "canonical_pr_reason": null, + "best_issue_reason": "If a representative issue is needed, issue:44479 is the broadest user-facing regression report here, covering multiple Qwen VL variants and a concrete v5.3.0 video-input breakage.", + "best_pr_reason": null, "soft_edge_verdicts": [ { - "left": "issue:44779", - "right": "issue:45292", + "left": "issue:36246", + "right": "issue:44351", "accept": false, - "reason": "Both mention token embeddings/version regressions, but one is DeepSeek tokenizer output changes and the other is resize_token_embeddings not updating output embeddings; different code paths and failures." + "reason": "Both are import errors, but they involve different missing symbols (`Qwen2_5_VLImageProcessor` vs `HybridCache`) and likely different export paths." }, { - "left": "issue:45859", - "right": "issue:46399", + "left": "issue:44779", + "right": "issue:45292", "accept": false, - "reason": "Both involve Qwen3.5/moe and rotary-related behavior, but one is CPU offload incompatibility in a text rotary embedding forward path while the other is a missing decorator causing a kernel registration crash." + "reason": "Tokenizer output regression and `resize_token_embeddings` not updating output embeddings are different bugs in different code paths." }, { "left": "issue:45003", "right": "issue:45561", "accept": false, - "reason": "Both touch import/internal utilities, but one reports unsafe sys.modules access in modeling_utils and the other is a pytest-xdist race in testing utilities; different bugs and fixes." + "reason": "`sys.modules` access in model loading and a pytest-xdist file race are unrelated failures." }, { - "left": "issue:43901", - "right": "issue:45003", + "left": "issue:45859", + "right": "issue:46399", + "accept": false, + "reason": "Both touch Qwen3.5, but one is CPU-offload incompatibility in rotary embedding and the other is a missing decorator causing kernel registration failure." + }, + { + "left": "issue:44734", + "right": "issue:46424", "accept": false, - "reason": "The first is a pipeline docs/behavior mismatch, while the second is a modeling_utils import-safety issue; no shared underlying defect." + "reason": "KV-cache continuation crash and cache max-length handling are separate cache features with different symptoms." }, { "left": "issue:44479", "right": "issue:46655", "accept": false, - "reason": "Both are video/multimodal regressions, but one is a broad v5.3.0 input regression across Qwen video models and the other is a SmolVLM padding-calculation bug in the video processor." + "reason": "Both concern video processing, but 44479 is a broad Qwen VL input regression while 46655 is a SmolVLM padding-calculation bug." }, { - "left": "issue:44734", - "right": "issue:46424", + "left": "issue:43901", + "right": "issue:45003", "accept": false, - "reason": "KV-cache continuation tensor indexing in serve is unrelated to cache max length handling for quantized cache; same general caching theme but different code paths." + "reason": "Docs inconsistency for `return_all_scores` is not the same bug as unsafe `sys.modules` access." }, { "left": "issue:43257", "right": "issue:45137", "accept": false, - "reason": "One is a Qwen3 MOE weight-conversion issue with accelerate+deepspeed, the other is a DeepSpeed ZeRO3 deque underflow; both involve DeepSpeed but not the same failure." - }, - { - "left": "issue:44242", - "right": "issue:44857", - "accept": false, - "reason": "Load-balancing loss omission in MoE routing is unrelated to an AMP/CUDA crash in LwDetrImageLoss." + "reason": "Both involve DeepSpeed/zero3-related loading, but one is Qwen3 MOE weight conversion and the other is a deque underflow during init." }, { "left": "issue:43475", "right": "issue:46369", "accept": false, - "reason": "SAM 3 video attribute access failure and SAM3 LiteText memory overflow are different bugs in different execution paths." + "reason": "Sam3VisionEncoderOutput missing an attribute is a structural output bug; memory overflow in sam3-litetext is a separate resource issue." }, { "left": "issue:44496", "right": "issue:45003", "accept": false, - "reason": "Unrecognized model/config loading failure is not the same as unsafe sys.modules access in modeling_utils." + "reason": "Unrecognized model loading and unsafe `sys.modules` access are different failure modes." + }, + { + "left": "issue:44242", + "right": "issue:44857", + "accept": false, + "reason": "Missing load-balancing loss and an AMP/CUDA loss crash are not the same defect." }, { "left": "issue:43673", "right": "issue:44561", "accept": false, - "reason": "Chunked prefill cache handling and removal of is_torch_fx_available causing trust_remote_code breakage are unrelated regressions." + "reason": "Chunked prefill cache handling and removal of `is_torch_fx_available` affect different APIs and code paths." }, { "left": "issue:44295", "right": "issue:44749", "accept": false, - "reason": "One is a buffer registration/position_ids access error; the other is a performance regression during filtering after a version upgrade." + "reason": "A buffer read error for `position_ids` and a v4-to-v5 slowdown report are unrelated." }, { "left": "issue:44162", "right": "issue:45200", "accept": false, - "reason": "ESM2 breakage and Gemma 4 mm_token_type_ids defaulting are separate model-specific bugs." + "reason": "ESM2 breakage and Gemma 4 token-type defaults are unrelated model-specific issues." }, { "left": "issue:43262", "right": "issue:44246", "accept": false, - "reason": "Audio apply_chat_template sampling-rate defaulting is unrelated to slow import / captured-info style behavior." + "reason": "Audio chat-template sampling-rate defaulting and slow import times are different subsystems." }, { "left": "issue:45375", "right": "issue:45759", "accept": false, - "reason": "One is strict config dropping a Qwen3_5MoeVisionConfig field; the other is AutoModelForCausalLM.from_config not unwrapping text_config for composite Qwen multimodal configs." - }, - { - "left": "issue:44877", - "right": "issue:45375", - "accept": false, - "reason": "Both are config-loading issues, but one is strict config rejecting granite_speech while the other is a missing field being silently dropped for Qwen3_5MoeVisionConfig." + "reason": "Missing config field in Qwen3_5MoeVisionConfig and `from_config` not unwrapping `text_config` are distinct composite-config bugs." }, { "left": "issue:36246", "right": "issue:43596", "accept": false, - "reason": "An ImportError for Qwen2_5_VLImageProcessor is unrelated to a DeepSpeed ZeRO3 BertModel IndexError." + "reason": "ImportError for a VL image processor and a zero3/BertModel index error are unrelated." }, { "left": "issue:45292", "right": "issue:45584", "accept": false, - "reason": "resize_token_embeddings output_embeddings inconsistency is unrelated to Whisper empty-transcription generation after align_special_tokens." - }, - { - "left": "issue:45137", - "right": "issue:45600", - "accept": false, - "reason": "Both are CI/runtime failures, but one is a ZeRO3 deque issue and the other is stale auto_mappings references to removed Sam3LiteText configs." + "reason": "Token embedding resize behavior and Whisper empty-transcription generation failure are different bugs." }, { "left": "issue:44561", "right": "issue:46399", "accept": false, - "reason": "A missing utility removed in v5 breaking trust_remote_code models is not the same as a Qwen3.5 kernel registration/decorator crash." + "reason": "A v5 trust_remote_code regression from removing `is_torch_fx_available` is not the same as the Qwen3.5 kernel-registration error." }, { "left": "issue:45292", "right": "issue:45701", "accept": false, - "reason": "resize_token_embeddings affecting output embeddings is a model weight-update bug, while tokenization changes across versions are a tokenizer behavior regression." + "reason": "Both mention tokenization/embedding effects, but one is embedding resizing and the other is changed tokenization behavior across versions." + }, + { + "left": "issue:45137", + "right": "issue:45600", + "accept": false, + "reason": "Zero3 deque underflow and stale auto_mappings references are unrelated." + }, + { + "left": "issue:44877", + "right": "issue:45375", + "accept": false, + "reason": "Strict config loading for granite_speech and a missing Qwen3_5MoeVisionConfig field are different config-validation problems." }, { "left": "issue:44038", "right": "issue:46506", "accept": false, - "reason": "Qwen5/Qwen3-VL-Moe-related bug and CPU-vs-CUDA output divergence in PPDocLayoutV3 are different multimodal model issues." + "reason": "Qwen3-VL-Moe regression in v5.0 and CPU/CUDA output differences for PPDocLayoutV3 are not the same issue." }, { "left": "issue:44466", "right": "issue:45335", "accept": false, - "reason": "Both mention resize_token_embeddings, but one is a generic tied-weight serialization inconsistency and the other is specifically t5gemma decoder.embed_tokens not updating." + "reason": "Both concern embedding/weight updates, but one is tied-weight serialization and the other is `resize_token_embeddings` not reaching decoder embeddings." }, { - "left": "issue:45412", - "right": "issue:46123", + "left": "issue:44987", + "right": "issue:45459", "accept": false, - "reason": "RT-DETR memory retention after deletion is unrelated to MaskGenerationPipeline dropping partial-batch results." + "reason": "Model loading failure on v5.1+ and tokenizer error masking when protobuf is absent are separate problems." }, { - "left": "issue:44987", - "right": "issue:45459", + "left": "issue:45412", + "right": "issue:46123", "accept": false, - "reason": "Loading physical-intelligence/fast on transformers>=5.1.0 and hiding tokenizer errors when protobuf is absent are separate failure modes." + "reason": "Memory not released for RT-DETR and final partial batch handling in MaskGenerationPipeline are unrelated." }, { "left": "issue:43352", "right": "issue:45250", "accept": false, - "reason": "NemotronHForCausalLM not supporting Flash Attention 2.0 is specific to a model/backend support gap, not the generic 'Flash Attention 2.0' issue title." + "reason": "One is a model-specific Flash Attention 2 unsupported error; the other is just a generic Flash Attention 2 topic, not the same concrete bug." }, { "left": "issue:44479", "right": "issue:46170", "accept": false, - "reason": "The first is a multimodal video regression; the second is a Windows/ROCm import error due to torch.distributed dependency." + "reason": "Video-input regression in Qwen VL models and Windows/ROCm torch.distributed import/type errors are unrelated." }, { "left": "issue:44060", "right": "issue:44077", "accept": false, - "reason": "Incorrect tied-weights warning in Qwen3-Next and patchtsmixer post_init policy are unrelated configuration/validation issues." - }, - { - "left": "issue:44995", - "right": "issue:45200", - "accept": false, - "reason": "Stale indexer cache on second forward pass in GlmMoeDsa is unrelated to Gemma 4 requiring mm_token_type_ids for text-only fine-tuning." + "reason": "Incorrect tied-weights warning and `patchtsmixer` post_init policy are different issues." } ] }, @@ -12682,6 +12579,7 @@ "issue:44877", "issue:44977", "issue:44995", + "issue:45200", "issue:45229", "issue:45230", "issue:45356", @@ -12703,6 +12601,7 @@ "issue:46431" ], "soft_pairs": [ + "issue:44995|issue:45200", "issue:44368|issue:44977", "issue:43653|issue:44162", "issue:43475|issue:43540", @@ -12714,12 +12613,12 @@ "issue:43065|issue:43408", "issue:43906|issue:46202", "issue:43931|issue:44368", - "issue:44188|issue:46202", "issue:45685|issue:46431", "issue:45987|issue:46082", "issue:44483|issue:46202", - "issue:45584|issue:46170", + "issue:44188|issue:46202", "issue:44188|issue:46421", + "issue:45584|issue:46170", "issue:43065|issue:46123", "issue:44297|issue:44623", "issue:44361|issue:44749", @@ -12735,220 +12634,226 @@ "issue:43475|issue:43643", "issue:45375|issue:45478" ], - "prepared_review_unit_hash": "9fd0382ab0cd65761e0b3bce7bad290b3ae3832851234034ce7487995966b13c", + "prepared_review_unit_hash": "92684c07073cc0f7f8a79df623b195e412e5a9881b34883d1c3829b2c6da6586", "component_budget": { "node_count": 473, "item_count": 473, "soft_pair_count": 1738, - "serialized_chars": 901383, - "estimated_input_tokens": 225346, - "estimated_eval_tokens": 450948 + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 }, "budget": { - "node_count": 47, - "item_count": 47, - "soft_pair_count": 31, - "serialized_chars": 36313, - "estimated_input_tokens": 9079, - "estimated_eval_tokens": 18414 + "node_count": 48, + "item_count": 48, + "soft_pair_count": 32, + "serialized_chars": 37186, + "estimated_input_tokens": 9297, + "estimated_eval_tokens": 18850 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster is highly heterogeneous: most soft edges connect issues that share a broad subsystem or version context but not the same concrete failure mode. I would reject all proposed duplicate links and keep the cluster split.", - "confidence": 0.41, - "canonical_issue_reason": "issue:45375 is the most specific root-cause-style report in the set: it names the exact missing field, the strict-deserialization behavior, and a concrete affected config family.", + "summary": "The set is heterogeneous: tokenizer/save/load regressions, generation/cache issues, model-specific config bugs, and unrelated framework/runtime problems. None of the proposed soft edges looks like the same underlying bug or a mergeable duplicate PR.", + "confidence": 0.93, + "canonical_issue_reason": "No clear canonical issue: the items span unrelated regressions across tokenizers, generation, configs, save/load, and model-specific bugs.", "canonical_pr_reason": null, - "best_issue_reason": "issue:45375 is the best anchor for this cluster because it is precise, actionable, and points to a likely underlying config-loading regression rather than a downstream symptom.", + "best_issue_reason": "No single issue is a safe duplicate anchor; the cluster is too mixed to consolidate around one report.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:44995", + "right": "issue:45200", + "accept": false, + "reason": "Different failures: stale indexer cache on second forward pass vs Gemma mm_token_type_ids defaulting for fine-tuning." + }, { "left": "issue:44368", "right": "issue:44977", "accept": false, - "reason": "Both mention Qwen3.5, but one is a warning/tie-weights config issue and the other is a flash-attention generation failure; different bugs." + "reason": "Unrelated: a warning about tie_word_embeddings vs a flash-attention generation bug in Qwen3.5." }, { "left": "issue:43653", "right": "issue:44162", "accept": false, - "reason": "BigBirdTokenizer special-token registration and ESM2 breakage are unrelated model/tokenizer failures." + "reason": "Different models and symptoms: BigBirdTokenizer special-token decode bug vs ESM2 regression." }, { "left": "issue:43475", "right": "issue:43540", "accept": false, - "reason": "Sam3 vision-encoder missing attribute and Qwen3Omni video-processing ValueError are different code paths and model families." + "reason": "Both are SAM3/Qwen video-related, but one is a missing attribute and the other is a video-input ValueError; not the same bug." }, { "left": "issue:45397", "right": "issue:45656", "accept": false, - "reason": "Gemma-4 ZeRO3 loading and double optimizer-step under deepspeed are separate training/runtime bugs." + "reason": "Different code paths: Gemma zero3 from_pretrained failure vs deepspeed optimizer stepping twice." }, { "left": "issue:42886", "right": "issue:45229", "accept": false, - "reason": "Offline cache loading for a tokenizer and multi-GPU CUDA OOM are unrelated failures." + "reason": "Tokenizer cache/offline loading bug is unrelated to Gemma4 multi-GPU OOM." }, { "left": "issue:44877", "right": "issue:46170", "accept": false, - "reason": "Strict config loading for granite_speech is unrelated to the Windows/ROCm torch.distributed import problem." + "reason": "Strict config loading for granite_speech is unrelated to Windows/ROCm torch.distributed import failures." }, { "left": "issue:44995", "right": "issue:45588", "accept": false, - "reason": "A stale indexer cache crash on second forward pass is not the same as the flash_attention s_aux=None AttributeError." + "reason": "Different bugs: stale cache on forward pass vs flash_attention.py crashing on s_aux=None." }, { "left": "issue:44368", "right": "issue:45563", "accept": false, - "reason": "These are both warnings in generation-related flows, but one is about tie_word_embeddings and the other is a stale num_return_sequences warning." + "reason": "A tie_word_embeddings warning is not the same as a stale num_return_sequences warning in paged generate()." }, { "left": "issue:43065", "right": "issue:43408", "accept": false, - "reason": "Both are SAM3-related, but one is a dummy Conv2d in the pixel decoder and the other is a model-type mismatch warning." + "reason": "Related SAM3 area, but dummy Conv2d in the decoder is not the same as a model-type warning on instantiation." }, { "left": "issue:43906", "right": "issue:46202", "accept": false, - "reason": "An isolated reproduction of an old issue and a Gemma pipeline roundtrip failure on \u2581 are not the same bug." + "reason": "Unrelated: isolated reproduction of an old issue vs Gemma pipeline roundtrip failure on \u2581." }, { "left": "issue:43931", "right": "issue:44368", "accept": false, - "reason": "Qwen3-VL weight-shape mismatch and a Qwen3.5 tie-embeddings warning are unrelated loading/config problems." - }, - { - "left": "issue:44188", - "right": "issue:46202", - "accept": false, - "reason": "torch.compile attention-kernel divergence is unrelated to the Gemma pipeline roundtrip token issue." + "reason": "Qwen3-VL weight-shape mismatch is unrelated to a tie_word_embeddings warning." }, { "left": "issue:45685", "right": "issue:46431", "accept": false, - "reason": "MPS histogram on Int and missing chat template are different errors in different subsystems." + "reason": "MPS histogram/int error is unrelated to missing chat template in processor.apply_chat_template()." }, { "left": "issue:45987", "right": "issue:46082", "accept": false, - "reason": "StaticCache returning a tensor instead of int and LlamaConfig head_dim validation are distinct API/validation issues." + "reason": "StaticCache return-type contract bug is unrelated to LlamaConfig head_dim validation." }, { "left": "issue:44483", "right": "issue:46202", "accept": false, - "reason": "A chat-completions API rejection is unrelated to the Gemma \u2581 roundtrip tokenizer bug." + "reason": "Chat/completions request rejection is unrelated to Gemma pipeline roundtrip tokenization behavior." }, { - "left": "issue:45584", - "right": "issue:46170", + "left": "issue:44188", + "right": "issue:46202", "accept": false, - "reason": "Whisper empty-transcription handling and Windows/ROCm distributed import failure are different problems." + "reason": "torch.compile attention-kernel divergence is unrelated to Gemma roundtrip decoding on \u2581." }, { "left": "issue:44188", "right": "issue:46421", "accept": false, - "reason": "Attention-kernel divergence under torch.compile is unrelated to cache lazy initialization recompilation." + "reason": "Both mention cache/compile, but one is branching-induced kernel divergence and the other is lazy cache init causing recompilation; different bugs." + }, + { + "left": "issue:45584", + "right": "issue:46170", + "accept": false, + "reason": "Whisper empty-transcription failure is unrelated to Windows/ROCm import issues." }, { "left": "issue:43065", "right": "issue:46123", "accept": false, - "reason": "SAM3 decoder details and MaskGenerationPipeline dropping final partial batches do not share the same failure mode." + "reason": "Different SAM3 and pipeline bugs: dummy Conv2d vs is_last never true on the final batch." }, { "left": "issue:44297", "right": "issue:44623", "accept": false, - "reason": "Tokenizer.save_pretrained class metadata mismatch and processor.save_pretrained missing files are related only at a very broad serialization level." + "reason": "Both involve save_pretrained, but tokenizer_class mismatch and missing processor files are different defects." }, { "left": "issue:44361", "right": "issue:44749", "accept": false, - "reason": "MLukeTokenizer AttributeError and a 10x slowdown after upgrade are unrelated tokenization/performance issues." + "reason": "MLukeTokenizer task AttributeError is unrelated to a performance regression in filtering after upgrade." }, { "left": "issue:44749", "right": "issue:45656", "accept": false, - "reason": "A data-filtering slowdown and a deepspeed double-step bug are not the same underlying change." + "reason": "Performance slowdown after upgrade is unrelated to double optimizer.step() under deepspeed." }, { "left": "issue:43701", "right": "issue:44749", "accept": false, - "reason": "Checkpoint resume key mismatch and a preprocessing slowdown do not match." + "reason": "resume_from_checkpoint key mismatch is unrelated to the filtering performance regression." }, { "left": "issue:44623", "right": "issue:45656", "accept": false, - "reason": "Processor save_pretrained file emission and double optimizer stepping are unrelated." + "reason": "Missing processor.save_pretrained files is unrelated to deepspeed optimizer double-step behavior." }, { "left": "issue:43646", "right": "issue:44162", "accept": false, - "reason": "Custom model initialization breakage in Transformers 5 and ESM2 breakage are different model-loading regressions." + "reason": "Custom model initialization breakage in v5 is unrelated to ESM2 being broken." }, { "left": "issue:44871", "right": "issue:46132", "accept": false, - "reason": "Gemma eos_token_id inconsistency and AttentionInterface.register behavior changes are unrelated." + "reason": "Gemma EOS config mismatch is unrelated to AttentionInterface.register changing function behavior." }, { "left": "issue:43232", "right": "issue:43673", "accept": false, - "reason": "Generation kwargs update after sync_gpus and missing cache during chunked prefill are different generation internals." + "reason": "Both generation-related, but one is sync_gpus/update_model_kwargs and the other is missing cache during chunked prefill; not the same bug." }, { "left": "issue:42994", "right": "issue:43482", "accept": false, - "reason": "Quantized model saving failure and Qwen2.5-GGUF loading failure are separate serialization/loading bugs." + "reason": "Quantized model saving failure is unrelated to Qwen2.5-GGUF loading failure." }, { "left": "issue:44869", "right": "issue:45356", "accept": false, - "reason": "Whisper trailing replacement-character crash and Kimi-K2.5 codec regression are different tokenizer failures." + "reason": "Whisper timestamp decode crash on trailing replacement char is unrelated to Kimi-K2.5 tokenizer codec handling." }, { "left": "issue:45230", "right": "issue:45397", "accept": false, - "reason": "Generic bug report and Gemma-4 ZeRO3 from_pretrained failure are not the same artifact." + "reason": "\"Bug report\" gives no concrete overlap with Gemma4 zero3 from_pretrained." }, { "left": "issue:43475", "right": "issue:43643", "accept": false, - "reason": "A Sam3 vision encoder output attribute error and missing fields from AutoConfig with trust_remote_code are unrelated." + "reason": "Different problems: missing Sam3VisionEncoderOutput attribute vs AutoConfig missing fields under trust_remote_code." }, { "left": "issue:45375", "right": "issue:45478", "accept": false, - "reason": "Both are Qwen3.5-Moe related, but one is a specific missing config field and the other is a broad from_pretrained error; not enough evidence they are the same bug." + "reason": "Related Qwen3.5 family, but missing deepstack_visual_indexes being dropped by strict is not the same as a from_pretrained error in transformers>=5.4.0." } ] }, @@ -12976,7 +12881,6 @@ "issue:30333", "issue:30990", "issue:33290", - "issue:35141", "issue:36010", "issue:41628", "issue:42175", @@ -13023,25 +12927,25 @@ ], "soft_pairs": [ "issue:28282|issue:30990", + "issue:42371|issue:45092", "issue:44038|issue:44623", "issue:41628|issue:44933", - "issue:42371|issue:45092", "issue:45310|issue:45375", "issue:43475|issue:45381", - "issue:44964|issue:45137", "issue:42994|issue:43525", + "issue:44964|issue:45137", "issue:43723|issue:46326", "issue:43784|issue:46399", "issue:45406|issue:46202", - "issue:43873|issue:44945", "issue:43232|issue:44188", + "issue:43873|issue:44945", "issue:44945|issue:45753", "issue:41628|issue:44351", "issue:45081|issue:45103", "issue:42673|issue:43704", "issue:44038|issue:45310", - "issue:44222|issue:45310", "issue:42994|issue:43824", + "issue:44222|issue:45310", "issue:36010|issue:41628", "issue:42831|issue:44222", "issue:43278|issue:45137", @@ -13052,235 +12956,228 @@ "issue:45561|issue:45600", "issue:43976|issue:44933", "issue:30333|issue:33290", - "issue:42175|issue:45712", - "issue:30990|issue:35141" + "issue:42175|issue:45712" ], - "prepared_review_unit_hash": "e590191e974862c4c396ae0e42bbf2a9705e4c136701843e70f4061c241da6bc", + "prepared_review_unit_hash": "09aab3d309f2c9970b28d66888e6136301ed32391577735218848974ce0b5258", "component_budget": { "node_count": 473, "item_count": 473, "soft_pair_count": 1738, - "serialized_chars": 901383, - "estimated_input_tokens": 225346, - "estimated_eval_tokens": 450948 + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 }, "budget": { - "node_count": 48, - "item_count": 48, - "soft_pair_count": 32, - "serialized_chars": 37152, - "estimated_input_tokens": 9288, - "estimated_eval_tokens": 18832 + "node_count": 47, + "item_count": 47, + "soft_pair_count": 31, + "serialized_chars": 36243, + "estimated_input_tokens": 9061, + "estimated_eval_tokens": 18378 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a loose mix of Transformers regression reports, especially import/API breakages and model-specific bugs, but only one soft pair looks like a true duplicate. Most other pairs share a broad subsystem or wording pattern without the same underlying code-path failure.", - "confidence": 0.39, - "canonical_issue_reason": "issue:44933 is the broadest representative of the import-regression subset and sits near several other import-related soft matches (41628, 44351, 36010, 43976), so it is the best canonical anchor for this mixed cluster.", + "summary": "This cluster is mostly heterogeneous, but it contains one clear duplicate pair about a Qwen3ForCausalLM VRAM leak. The rest of the soft edges look like broad thematic similarity rather than the same underlying bug.", + "confidence": 0.62, + "canonical_issue_reason": "issue:42673 is the best canonical issue because it has an exact duplicate twin (issue:43704) with the same title and underlying VRAM leak bug, making it the strongest representative of the only clear duplicate in the set.", "canonical_pr_reason": null, - "best_issue_reason": "issue:42673 is the clearest actionable issue because it has an essentially identical duplicate (43704) with the same VRAM-leak behavior and title, making it the best single issue to keep.", + "best_issue_reason": "issue:42673 is the most suitable issue to anchor triage: it is concrete, has a direct duplicate, and is cleaner than the other mostly unrelated or loosely similar import/model bugs.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:28282", "right": "issue:30990", "accept": false, - "reason": "Missing PyTorch dependency error vs a sentence-transformers loading hang; different failures and code paths." + "reason": "PyTorch import failure vs Sentence Transformers loading hang; unrelated symptoms and code paths." + }, + { + "left": "issue:42371", + "right": "issue:45092", + "accept": false, + "reason": "TF32 API warning/settings issue vs old InternVL2 meta-init incompatibility; different bugs." }, { "left": "issue:44038", "right": "issue:44623", "accept": false, - "reason": "Qwen3-VL-Moe model bug vs processor.save_pretrained missing files; unrelated symptoms." + "reason": "Qwen3-VL-Moe load bug vs processor.save_pretrained missing files; not the same failure." }, { "left": "issue:41628", "right": "issue:44933", "accept": false, - "reason": "Both are import regressions, but they involve different symbols/modules and likely different fixes." - }, - { - "left": "issue:42371", - "right": "issue:45092", - "accept": false, - "reason": "TF32 API guidance vs old InternVL2 meta-init incompatibility; not the same bug." + "reason": "Different import errors in different modules; similar surface area only, not same bug." }, { "left": "issue:45310", "right": "issue:45375", "accept": false, - "reason": "Different Qwen3.5 issues: one is a from_pretrained error, the other is a missing config field silently dropped." + "reason": "Both are Qwen3.5/Moe-related, but one is from_pretrained failure and the other is a missing config field; not clearly the same bug." }, { "left": "issue:43475", "right": "issue:45381", "accept": false, - "reason": "Different vision/video bugs in different models and code paths." + "reason": "Different model families and different runtime failures; no evidence of one underlying defect." }, { - "left": "issue:44964", - "right": "issue:45137", + "left": "issue:42994", + "right": "issue:43525", "accept": false, - "reason": "Model loading failure vs DeepSpeed ZeRO3 deque IndexError; unrelated." + "reason": "Quantized model save failure vs missing pad_token_id on Llama4Config; unrelated." }, { - "left": "issue:42994", - "right": "issue:43525", + "left": "issue:44964", + "right": "issue:45137", "accept": false, - "reason": "Quantized save failure vs missing pad_token_id on Llama4Config; different underlying problems." + "reason": "Multimodal model loading error vs DeepSpeed ZeRO3 deque IndexError; unrelated." }, { "left": "issue:43723", "right": "issue:46326", "accept": false, - "reason": "Tokenizer loading failure vs continue_final_message behavior for mistral_common tokenizers; not the same issue." + "reason": "Tokenizer loading in v5 vs continue_final_message on mistral_common tokenizers; different code paths and bugs." }, { "left": "issue:43784", "right": "issue:46399", "accept": false, - "reason": "Importing sentence-transformers breaks on nn vs a Qwen3.5 kernel registration crash; unrelated." + "reason": "Sentence-transformers import NameError vs Qwen3.5 kernel registration error; unrelated." }, { "left": "issue:45406", "right": "issue:46202", "accept": false, - "reason": "Gemma4Processor missing _tokenizer vs Gemma pipeline roundtrip failure on \u2581; different bugs." + "reason": "Gemma processor attribute error vs Gemma pipeline roundtrip token issue; not the same bug." }, { - "left": "issue:43873", - "right": "issue:44945", + "left": "issue:43232", + "right": "issue:44188", "accept": false, - "reason": "Quantization/offloading behavior vs pipeline parallelism output corruption; different code paths." + "reason": "Generation kwargs update after sync_gpus vs torch.compile attention-kernel branching; different issues." }, { - "left": "issue:43232", - "right": "issue:44188", + "left": "issue:43873", + "right": "issue:44945", "accept": false, - "reason": "Generation kwargs update bug vs attention-kernel divergence under torch.compile; not the same defect." + "reason": "Quantization/offloading behavior vs pipeline-parallel output correctness; unrelated." }, { "left": "issue:44945", "right": "issue:45753", "accept": false, - "reason": "Pipeline parallelism output bug vs Qwen3_5 image infinite loop; different symptoms and models." + "reason": "Pipeline parallelism output bug vs Qwen3.5 image-specific infinite loop; different failures." }, { "left": "issue:41628", "right": "issue:44351", "accept": false, - "reason": "Both are import-name errors, but they are for different exported classes and likely separate namespace regressions." + "reason": "Both are import-name problems, but for different symbols and likely different missing-export causes." }, { "left": "issue:45081", "right": "issue:45103", "accept": false, - "reason": "Tokenizer regex patch crash vs auto_docstring annotation parsing crash; unrelated." + "reason": "Mistral tokenizer patch crash vs auto_docstring annotations crash; unrelated." }, { "left": "issue:42673", "right": "issue:43704", "accept": true, - "reason": "Same title, same VRAM leak in Qwen3ForCausalLM with multiple dataloader threads; these are clear duplicates." + "reason": "Exact duplicate titles and the same Qwen3ForCausalLM VRAM leak in multiple dataloader threads." }, { "left": "issue:44038", "right": "issue:45310", "accept": false, - "reason": "Both mention Qwen MoE, but they are different model/version failures and not the same concrete bug." + "reason": "Both mention Qwen/Moe, but they target different versions and different from_pretrained failures." }, { - "left": "issue:44222", - "right": "issue:45310", + "left": "issue:42994", + "right": "issue:43824", "accept": false, - "reason": "FP8 save_pretrained failure vs Qwen3.5 from_pretrained error; different issues." + "reason": "Quantized model saving vs missing Qwen2_5_VLForConditionalGeneration import; unrelated." }, { - "left": "issue:42994", - "right": "issue:43824", + "left": "issue:44222", + "right": "issue:45310", "accept": false, - "reason": "Quantized model saving vs missing Qwen2_5_VL import; unrelated." + "reason": "FP8 save_pretrained issue vs Qwen3.5 from_pretrained error; different code paths." }, { "left": "issue:36010", "right": "issue:41628", "accept": false, - "reason": "Different missing imports from different APIs; same broad v5 breakage, but not one bug." + "reason": "Both are import errors, but for different APIs and different breakages; too broad to merge." }, { "left": "issue:42831", "right": "issue:44222", "accept": false, - "reason": "FP8 accuracy issue vs FP8 save_pretrained failure; related theme but different defect." + "reason": "Accuracy regression in FineGrainedFP8 vs FP8 save_pretrained bug; not the same defect." }, { "left": "issue:43278", "right": "issue:45137", "accept": false, - "reason": "dtype drift in embedding layers vs DeepSpeed deque IndexError; unrelated." + "reason": "Embedding dtype mismatch during eval vs DeepSpeed deque IndexError; unrelated." }, { "left": "issue:43232", "right": "issue:44792", "accept": false, - "reason": "Generation sync_gpus bug vs janus image-generation test failure; different failures." + "reason": "Generation state bug vs janus image-generation test failure; no clear shared root cause." }, { "left": "issue:42175", "right": "issue:44933", "accept": false, - "reason": "Packaging/backend extra issue vs missing import from image_utils; not the same bug." + "reason": "Packaging/backend dependency omission vs missing image_utils import; different problems." }, { "left": "issue:44496", "right": "issue:45561", "accept": false, - "reason": "Unrecognized model/config issue vs xdist race on captured_info.txt; unrelated." + "reason": "Unrecognized model/config issue vs xdist file race; unrelated." }, { "left": "issue:43504", "right": "issue:45561", "accept": false, - "reason": "Legacy config field issue vs xdist test-file race; different problems." + "reason": "Legacy config field loading bug vs test-worker file race; unrelated." }, { "left": "issue:45561", "right": "issue:45600", "accept": false, - "reason": "Pytest-xdist file race vs stale auto_mappings references; unrelated." + "reason": "Captured-info race vs broken auto_mappings references; different CI failures." }, { "left": "issue:43976", "right": "issue:44933", "accept": false, - "reason": "Python-version compatibility regression vs missing import from image_utils; different concrete failures." + "reason": "Python-version compatibility problem vs missing import from image_utils; unrelated." }, { "left": "issue:30333", "right": "issue:33290", "accept": false, - "reason": "MLflow reporting after failure vs OOM in deepspeed adafactor; unrelated." + "reason": "MLflow reporting on failed jobs vs DeepSpeed OOM; unrelated." }, { "left": "issue:42175", "right": "issue:45712", "accept": false, - "reason": "Backend extra packaging problem vs leftover dummy classes leaking into dir(transformers); different issues." - }, - { - "left": "issue:30990", - "right": "issue:35141", - "accept": false, - "reason": "Loading hang for sentence-transformers vs embedding reinitialization during post_init; unrelated." + "reason": "Both touch torch packaging/dummy objects, but one is missing TF backend in torch install and the other is leftover dummy classes; different fixes." } ] }, "evaluator_result": { "accept": true, - "feedback": "The summary is grounded in the packet and stays conservative. The only accepted soft pair, 42673/43704, is well supported by the identical titles and matching VRAM-leak description. The remaining pair rejections are cautious and do not overclaim duplicate status from broad thematic similarity alone. The canonical anchor choice is somewhat judgment-based, but it is still defensible from the import-related soft matches shown." + "feedback": "The summary is well grounded in the packet. The only clear canonical pair is issue:42673 and issue:43704, which share the exact same title and appear to describe the same Qwen3ForCausalLM VRAM leak. The other pairwise rejections are cautious and consistent with the issue titles and metadata, without overstating evidence." }, "evaluator_used": true, "retried": false, @@ -13305,6 +13202,7 @@ "issue:27957", "issue:28282", "issue:30333", + "issue:30990", "issue:34689", "issue:35141", "issue:36331", @@ -13328,7 +13226,6 @@ "issue:44242", "issue:44295", "issue:44297", - "issue:44448", "issue:44568", "issue:44623", "issue:44734", @@ -13347,11 +13244,11 @@ "issue:45812", "issue:46077", "issue:46123", - "issue:46132", "issue:46231" ], "soft_pairs": [ "issue:46077|issue:46123", + "issue:30990|issue:35141", "issue:42898|issue:44568", "issue:43377|issue:44242", "issue:44623|issue:45310", @@ -13380,230 +13277,222 @@ "issue:30333|issue:35141", "issue:22355|issue:28282", "issue:44297|issue:45656", - "issue:42831|issue:44623", - "issue:44448|issue:46132", - "issue:44222|issue:44623" + "issue:42831|issue:44623" ], - "prepared_review_unit_hash": "ffd4f362aa6c4a4a6cd6b969eb6819d161e94886ab9f15793d9ecfe7c2a7f24a", + "prepared_review_unit_hash": "abfae1369dfe5e3527608f917cdf1d762b5845db3f858c0ad730189427dfa183", "component_budget": { "node_count": 473, "item_count": 473, "soft_pair_count": 1738, - "serialized_chars": 901383, - "estimated_input_tokens": 225346, - "estimated_eval_tokens": 450948 + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 }, "budget": { - "node_count": 48, - "item_count": 48, - "soft_pair_count": 32, - "serialized_chars": 37536, - "estimated_input_tokens": 9384, - "estimated_eval_tokens": 19024 + "node_count": 47, + "item_count": 47, + "soft_pair_count": 31, + "serialized_chars": 36621, + "estimated_input_tokens": 9156, + "estimated_eval_tokens": 18568 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This is a loose cluster of mostly unrelated issues. A few are in the same broad areas (tokenizers, multimodal models, save_pretrained, docs/examples), but none of the soft pairs look like the same underlying bug or change, so all soft edges are rejected. Issue 42898 is the best representative because it is the broadest tokenizer/v5 regression and sits closest to several other tokenizer-related reports.", - "confidence": 0.78, - "canonical_issue_reason": "issue:42898 is the broadest, most central tokenizer regression in the set and is the closest thematic match to several other tokenizer-related issues.", + "summary": "The candidate pairs are mostly loose topical similarities, not true duplicates; they span unrelated tokenizer, training, multimodal, serving, and docs bugs. No pair looks safe to merge as the same underlying issue/change.", + "confidence": 0.95, + "canonical_issue_reason": "No single true canonical issue stands out because the set is heterogeneous; issue:42898 is the closest umbrella candidate only because several pairs are tokenizer/regression themed, but it does not subsume the rest.", "canonical_pr_reason": null, - "best_issue_reason": "issue:42898 is the best cluster representative because it describes a generic v5 tokenizer behavior regression rather than a model-specific or one-off failure.", + "best_issue_reason": "issue:42898 is the broadest and most representative regression-style issue in this set, but only as a weak umbrella candidate, not a duplicate target for the others.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:46077", "right": "issue:46123", "accept": false, - "reason": "Both are generation-related, but one is a Gemma4 attention-mask bug and the other is a pipeline batching bug; different code paths." + "reason": "Different bugs: Gemma4 attention-mask construction vs pipeline batch-finalization logic." + }, + { + "left": "issue:30990", + "right": "issue:35141", + "accept": false, + "reason": "Unrelated symptoms and code paths: loading hangs vs embedding resize/post-init reinitialization." }, { "left": "issue:42898", "right": "issue:44568", "accept": false, - "reason": "Both involve tokenizers in v5, but one is clean-up spacing behavior and the other is add_special_tokens/BOS-EOS handling; not the same bug." + "reason": "Both tokenizer regressions, but different models and failures; not the same concrete bug." }, { "left": "issue:43377", "right": "issue:44242", "accept": false, - "reason": "MIMI padding-mask mismatch and MoE load-balancing loss omission are unrelated training/inference behaviors." + "reason": "Different functionality: MIMI batching/padding-mask mismatch vs MoE load-balancing loss calculation." }, { "left": "issue:44623", "right": "issue:45310", "accept": false, - "reason": "Processor save_pretrained file emission and Qwen3.5 MoE from_pretrained failure are different issues." + "reason": "Processor save-pretrained file emission vs Qwen3.5 MoE from_pretrained loading error are separate issues." }, { "left": "issue:43479", "right": "issue:44869", "accept": false, - "reason": "Default config initialization and Whisper timestamp decode crash affect different subsystems." + "reason": "Multimodal config default initialization is unrelated to Whisper timestamp decoding crash." }, { "left": "issue:42222", "right": "issue:45200", "accept": false, - "reason": "VitPose breakage and Gemma4 mm_token_type_ids defaults are unrelated model-specific bugs." + "reason": "Different model families and bugs: vitpose breakage vs Gemma4 text-only mm_token_type_ids defaulting." }, { "left": "issue:44016", "right": "issue:46231", "accept": false, - "reason": "Notebook syntax error and docs parameter-name mismatch are both docs/examples, but not the same defect." + "reason": "Notebook syntax error is unrelated to a docs parameter-name mismatch." }, { "left": "issue:42898", "right": "issue:45812", "accept": false, - "reason": "Both are tokenizer regressions, but one is spacing cleanup and the other is Granite token ID generation; distinct code paths." + "reason": "Both are tokenizer regressions, but they affect different model families and likely different code paths." }, { "left": "issue:43723", "right": "issue:44568", "accept": false, - "reason": "AutoTokenizer loading failure and add_special_tokens behavior regression are different tokenizer problems." + "reason": "General tokenizer-loading failure vs missing BOS/EOS insertion for one tokenizer." }, { "left": "issue:43874", "right": "issue:45092", "accept": false, - "reason": "Missing image patch count method and remote-code meta-init incompatibility are separate multimodal issues." + "reason": "Different multimodal bugs: missing image-patch helper vs remote-code/meta-init incompatibility." }, { "left": "issue:43479", "right": "issue:44079", "accept": false, - "reason": "Config defaulting and ModelOutput key assignment are unrelated internals." + "reason": "Config default construction bug is unrelated to ModelOutput key assignment behavior." }, { "left": "issue:42898", "right": "issue:44779", "accept": false, - "reason": "Both are v5 tokenizer regressions, but they concern different models and different failure modes." + "reason": "Same broad tokenizer-regression theme, but the affected models and failures differ; not one duplicate." }, { "left": "issue:43784", "right": "issue:43824", "accept": false, - "reason": "One is a sentence-transformers import NameError, the other a missing Qwen2.5-VL class import; unrelated." + "reason": "Separate import failures in different integration points and different missing symbols." }, { "left": "issue:44734", "right": "issue:45103", "accept": false, - "reason": "KV-cache indexing crash and auto_docstring annotation handling are unrelated bugs." + "reason": "Serving KV-cache indexing crash is unrelated to auto-docstring annotation handling." }, { "left": "issue:42898", "right": "issue:45356", "accept": false, - "reason": "Both are tokenizer regressions, but they affect different models and different tokenizer logic." + "reason": "Both are tokenizer regressions, but they concern different model-specific codec/regex behavior." }, { "left": "issue:44295", "right": "issue:45656", "accept": false, - "reason": "Position_ids buffer access and deepspeed double optimizer step are unrelated runtime issues." + "reason": "Position-ids buffer access bug is unrelated to deepspeed double optimizer stepping." }, { "left": "issue:42898", "right": "issue:43295", "accept": false, - "reason": "Tokenizer spacing behavior and processor.tokenizer/image passing regression are different API breakages." + "reason": "Tokenizer cleanup regression and processor.tokenizer access regression are different code paths." }, { "left": "issue:27957", "right": "issue:30333", "accept": false, - "reason": "Flash Attention 2 on XLMRoberta and MLFlow job-reporting failure are unrelated." + "reason": "Flash Attention 2 support for XLMRoberta is unrelated to MLFlow reporting after training failure." }, { "left": "issue:45405", "right": "issue:45712", "accept": false, - "reason": "A version-pin problem and leftover dummy classes in dummy_pt_objects are unrelated maintenance issues." + "reason": "Version bump for peft dependency is unrelated to leftover dummy-class cleanup." }, { "left": "issue:30333", "right": "issue:34689", "accept": false, - "reason": "MLFlow reporting and Llama 3.2 vision model-loading regression are unrelated." + "reason": "No shared underlying bug: MLFlow failure reporting vs model-loading regression for Llama 3.2 Vision." }, { "left": "issue:44060", "right": "issue:45125", "accept": false, - "reason": "Both are Qwen-related, but one is a warning/tied-weights bug and the other is missing tensor-parallel plan support." + "reason": "Different model-specific tensor-parallel / tied-weights issues." }, { "left": "issue:44075", "right": "issue:46231", "accept": false, - "reason": "Optimizer arguments not used is unrelated to a docs parameter-name mismatch." + "reason": "Optimizer argument handling bug is unrelated to documentation parameter naming." }, { "left": "issue:43116", "right": "issue:44190", "accept": false, - "reason": "Two example-script failures, but one is multi-label classification output and the other is local dataset loading." + "reason": "Different example scripts and failures: classification output bug vs local dataset loading." }, { "left": "issue:44222", "right": "issue:44297", "accept": false, - "reason": "MoE FP8 save_pretrained and tokenizer.save_pretrained class metadata are different serialization bugs." + "reason": "FP8 save_pretrained issue and tokenizer_config class mismatch are unrelated." }, { "left": "issue:44933", "right": "issue:45712", "accept": false, - "reason": "Image-utils import exposure and dummy_pt_objects cleanup are both packaging-related, but not the same issue." + "reason": "Missing import from image_utils is unrelated to dummy-class cleanup." }, { "left": "issue:30333", "right": "issue:36331", "accept": false, - "reason": "MLFlow reporting and custom trainer loss-signature incompatibility are unrelated training workflow bugs." + "reason": "MLFlow reporting after failure and CustomTrainer signature breakage are different issues." }, { "left": "issue:30333", "right": "issue:35141", "accept": false, - "reason": "MLFlow reporting and embedding reinitialization after resize are different problems." + "reason": "Training-job reporting bug is unrelated to embedding reinitialization on resize." }, { "left": "issue:22355", "right": "issue:28282", "accept": false, - "reason": "Missing transformers.onnx module and missing PyTorch dependency are both import errors, but they arise from different missing components." + "reason": "Both are import errors, but one is missing transformers.onnx and the other is missing PyTorch entirely." }, { "left": "issue:44297", "right": "issue:45656", "accept": false, - "reason": "Tokenizer save metadata mismatch and double optimizer step under deepspeed are unrelated." + "reason": "Tokenizer save-pretrained metadata mismatch is unrelated to deepspeed optimizer stepping." }, { "left": "issue:42831", "right": "issue:44623", "accept": false, - "reason": "FineGrainedFP8 accuracy and processor.save_pretrained missing files are different concerns." - }, - { - "left": "issue:44448", - "right": "issue:46132", - "accept": false, - "reason": "Pegasus v4/v5 output regression and AttentionInterface.register behavior change are both behavioral, but not the same code-path problem." - }, - { - "left": "issue:44222", - "right": "issue:44623", - "accept": false, - "reason": "Both mention save_pretrained, but one is FP8 MoE serialization and the other is processor file generation; different artifacts and fixes." + "reason": "FineGrainedFP8 accuracy issue is unrelated to processor save-pretrained missing files." } ] }, @@ -13633,7 +13522,6 @@ "issue:42757", "issue:42796", "issue:42831", - "issue:42907", "issue:43122", "issue:43232", "issue:43475", @@ -13650,6 +13538,7 @@ "issue:44297", "issue:44360", "issue:44373", + "issue:44448", "issue:44484", "issue:44485", "issue:44623", @@ -13664,7 +13553,6 @@ "issue:45357", "issue:45381", "issue:45834", - "issue:46050", "issue:46082", "issue:46132", "issue:46209", @@ -13677,6 +13565,8 @@ "issue:46665" ], "soft_pairs": [ + "issue:44448|issue:46132", + "issue:44222|issue:44623", "issue:42796|issue:43723", "issue:43479|issue:43644", "issue:45245|issue:46132", @@ -13685,10 +13575,10 @@ "issue:44297|issue:45310", "issue:45071|issue:46082", "issue:44373|issue:44485", - "issue:43723|issue:44977", "issue:44263|issue:45092", - "issue:43122|issue:46209", + "issue:43723|issue:44977", "issue:44222|issue:44484", + "issue:43122|issue:46209", "issue:43232|issue:46424", "issue:44206|issue:45103", "issue:44190|issue:46082", @@ -13706,230 +13596,235 @@ "issue:45092|issue:45357", "issue:44279|issue:46328", "issue:43644|issue:46305", - "issue:42831|issue:45310", "issue:44263|issue:44373", - "issue:42907|issue:46050" + "issue:42831|issue:45310" ], - "prepared_review_unit_hash": "00ecfbe6bbbed168d177856433259adadf2b602048b630c8de173807a825b1ae", + "prepared_review_unit_hash": "1dcb1e0ea53d216cea08b67edcd5a17fc25a033edbf9dadbde4f9d819650cff5", "component_budget": { "node_count": 473, "item_count": 473, "soft_pair_count": 1738, - "serialized_chars": 901383, - "estimated_input_tokens": 225346, - "estimated_eval_tokens": 450948 + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 }, "budget": { - "node_count": 48, - "item_count": 48, - "soft_pair_count": 32, - "serialized_chars": 36945, - "estimated_input_tokens": 9237, - "estimated_eval_tokens": 18730 + "node_count": 47, + "item_count": 47, + "soft_pair_count": 33, + "serialized_chars": 36710, + "estimated_input_tokens": 9178, + "estimated_eval_tokens": 18612 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is a heterogeneous mix of mostly unrelated Transformers regressions, import errors, tokenizer/model quirks, and model-specific bugs. None of the soft pairs look like the same underlying bug/change closely enough to merge as duplicates.", - "confidence": 0.87, - "canonical_issue_reason": "issue:43644 is the broadest, most systemic report in the set: a Transformers 5.0.0 buffer-initialization regression. It is more central than the highly model-specific or one-off API issues elsewhere in the cluster.", + "summary": "This cluster is heterogeneous: it mixes unrelated tokenizer, import, config, generation, save_pretrained, cache, and model-specific bugs, so there is no single duplicate center.", + "confidence": 0.97, + "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "issue:43644 is the best representative issue because it describes a core framework-level regression rather than a narrow model- or checkpoint-specific symptom, so it is the most plausible anchor for this noisy cluster.", + "best_issue_reason": null, "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:44448", + "right": "issue:46132", + "accept": false, + "reason": "Different problems: Pegasus v4/v5 output regression vs AttentionInterface.register behavior change." + }, + { + "left": "issue:44222", + "right": "issue:44623", + "accept": false, + "reason": "Different save_pretrained bugs: FP8 MoE saving vs processor file omission." + }, { "left": "issue:42796", "right": "issue:43723", "accept": false, - "reason": "Both are tokenizer/loading failures, but they describe different concrete failures: one is BPE-marker handling for Ministral-3-8B-Instruct, the other is AutoTokenizer loading in v5. Not the same bug." + "reason": "Tokenizer BPE marker handling vs AutoTokenizer loading failure; different failure modes." }, { "left": "issue:43479", "right": "issue:43644", "accept": false, - "reason": "One is a multimodal config default-initialization bug; the other is a framework-wide non-persistent buffer initialization regression. Related only at a very high level." + "reason": "Config initialization from None vs junk non-persistent buffers; unrelated code paths." }, { "left": "issue:45245", "right": "issue:46132", "accept": false, - "reason": "Both concern attention/quantization internals, but one is a category-count runtime limit and the other is AttentionInterface.register behavior. Different code paths and symptoms." + "reason": "Category-count limit error vs registered attention function behavior; unrelated." }, { "left": "issue:43479", "right": "issue:46665", "accept": false, - "reason": "These are unrelated multimodal initialization issues: config defaults vs inconsistent from_pretrained behavior inside Python classes." + "reason": "Default multimodal config init vs from_pretrained behavior inside Python classes." }, { "left": "issue:41628", "right": "issue:42757", "accept": false, - "reason": "Both are import errors from moved/changed symbols, but they are different symbols and different compatibility breakages, not one underlying bug." + "reason": "Missing AutoImageProcessor import vs missing is_offline_mode import; different symbols." }, { "left": "issue:44297", "right": "issue:45310", "accept": false, - "reason": "Tokenizer save_pretrained metadata mismatch is unrelated to the Qwen3.5 Moe from_pretrained error. Same family, different failure." + "reason": "Tokenizer config mismatch on save vs Qwen3.5 MoE load error; different stages and causes." }, { "left": "issue:45071", "right": "issue:46082", "accept": false, - "reason": "These are both v5/config-related, but one is type-checking breakage in PretrainedConfig and the other is a LlamaConfig validation issue about head_dim. Not the same bug." + "reason": "PretrainedConfig type-checking regression vs head_dim validation in LlamaConfig." }, { "left": "issue:44373", "right": "issue:44485", "accept": false, - "reason": "Wrong docstring for position_ids is documentation-only; GLM-5 RoPE implementation is a code-level modeling issue." - }, - { - "left": "issue:43723", - "right": "issue:44977", - "accept": false, - "reason": "Both involve model loading/generation in v5, but one is tokenizer loading and the other is flash-attention generation behavior for Qwen3.5. Different path." + "reason": "Docstring bug vs RoPE implementation issue; not the same underlying defect." }, { "left": "issue:44263", "right": "issue:45092", "accept": false, - "reason": "DSA indexer split-return behavior and InternVL2 meta-initialization incompatibility are different concrete bugs despite both touching model internals." + "reason": "DSA indexer torch.split handling vs remote-code meta-init incompatibility; unrelated." }, { - "left": "issue:43122", - "right": "issue:46209", + "left": "issue:43723", + "right": "issue:44977", "accept": false, - "reason": "Tokenization-version drift and FP8 inference on B200 are completely different issues." + "reason": "Tokenizer loading in v5 vs flash-attention generation failure; different code paths." }, { "left": "issue:44222", "right": "issue:44484", "accept": false, - "reason": "FP8 save_pretrained for MoE is unrelated to the default 50GB sharding threshold question." + "reason": "FP8 save_pretrained bug vs max_shard_size default question; not duplicates." + }, + { + "left": "issue:43122", + "right": "issue:46209", + "accept": false, + "reason": "Tokenization version regression vs FP8 inference failure on B200; unrelated." }, { "left": "issue:43232", "right": "issue:46424", "accept": false, - "reason": "Both mention cache/generation settings, but one is a generation update bug after sync_gpus and the other is max_cache_length being honored only for Quantized Cache." + "reason": "Generation kwargs after sync_gpus vs cache max_length configuration; different bugs." }, { "left": "issue:44206", "right": "issue:45103", "accept": false, - "reason": "LaserFeatureExtractor center-arg crash and auto_docstring AttributeError are unrelated implementation bugs." + "reason": "Unsupported center arg in feature extractor vs auto_docstring AttributeError; unrelated." }, { "left": "issue:44190", "right": "issue:46082", "accept": false, - "reason": "Local dataset loading in a training script and LlamaConfig head_dim validation are not the same bug or code path." + "reason": "Local dataset loading script issue vs LlamaConfig head_dim validation; different areas." }, { "left": "issue:46209", "right": "issue:46431", "accept": false, - "reason": "FP8 inference on B200 and missing chat template on a processor are unrelated." + "reason": "FP8 on B200 inference failure vs missing chat template error; unrelated." }, { "left": "issue:43232", "right": "issue:45834", "accept": false, - "reason": "Generation kwargs update/sync_gpus and Kosmos2.5 OCR long-input index errors are different failures." + "reason": "Generation bookkeeping bug vs long OCR index error in Kosmos2.5; different models/issues." }, { "left": "issue:44734", "right": "issue:46421", "accept": false, - "reason": "Both involve serving/caching behavior, but one is a KV-cache continuation tensor indexing crash and the other is lazy cache initialization causing recompilation. Different symptoms and fixes." + "reason": "Serve KV-cache continuation indexing bug vs lazy cache init recompilation; different cache problems." }, { "left": "issue:43475", "right": "issue:46305", "accept": false, - "reason": "SAM 3 Video missing attribute and MoonshineStreaming sliding-window off-by-one are different model-specific bugs." + "reason": "SAM 3 video attribute error vs MoonshineStreaming off-by-one masking; unrelated." }, { "left": "issue:44295", "right": "issue:44623", "accept": false, - "reason": "Reading position_ids after registering it as a buffer and processor.save_pretrained missing files are unrelated save/load issues." + "reason": "position_ids buffer read error vs processor.save_pretrained missing files; different APIs." }, { "left": "issue:46305", "right": "issue:46655", "accept": false, - "reason": "Both are video/streaming related, but one is a sliding-window mask off-by-one and the other is incorrect padding calculation in a different processor." + "reason": "Streaming mask off-by-one vs SmolVLM video padding calculation; different video bugs." }, { "left": "issue:43723", "right": "issue:44871", "accept": false, - "reason": "Tokenizer loading failure and Gemma-3 eos_token_id inconsistency are different issues." + "reason": "Tokenizer loading failure vs eos_token_id inconsistency between tokenizer and model config." }, { "left": "issue:44263", "right": "issue:44360", "accept": false, - "reason": "Both mention the DSA indexer, but one is about torch.split return values and the other about a missing ReLU. Same subsystem, not the same bug." + "reason": "DSA indexer torch.split issue vs missing ReLU in DSA indexer; different logic errors." }, { "left": "issue:45381", "right": "issue:46305", "accept": false, - "reason": "Qwen2.5-VL video vision_position_ids and MoonshineStreaming sliding_window_mask_function are separate video-processing bugs." + "reason": "Qwen2.5-VL vision_position_ids bug vs MoonshineStreaming window bug; unrelated models." }, { "left": "issue:44261", "right": "issue:45092", "accept": false, - "reason": "MLA q_a_layernorm eps precision issue and InternVL2 meta-init incompatibility are unrelated model-specific regressions." + "reason": "MLA RMS norm epsilon precision issue vs meta-init incompatibility; unrelated." }, { "left": "issue:30990", "right": "issue:38175", "accept": false, - "reason": "Sentence Transformers loading hang and SIGLIP2 zero-probabilities are unrelated model issues." + "reason": "Sentence Transformers loading hang vs zero probabilities in SigLIP2; different symptoms and models." }, { "left": "issue:45092", "right": "issue:45357", "accept": false, - "reason": "Both mention Qwen/visual encoders, but one is old InternVL2 meta-init incompatibility and the other is incorrect visual encoder keys saved in Qwen3.5. Different bugs." + "reason": "Old InternVL2 meta-init incompatibility vs incorrect visual encoder keys saved in Qwen3.5." }, { "left": "issue:44279", "right": "issue:46328", "accept": false, - "reason": "Generic dependency issue with transformers and a SequenceFeatureExtractor.pad performance issue are not the same underlying defect." + "reason": "General dependency issue vs SequenceFeatureExtractor.pad performance regression; unrelated." }, { "left": "issue:43644", "right": "issue:46305", "accept": false, - "reason": "Framework-wide non-persistent buffer junk initialization and MoonshineStreaming mask off-by-one are different bugs." - }, - { - "left": "issue:42831", - "right": "issue:45310", - "accept": false, - "reason": "FineGrainedFP8 accuracy and Qwen3.5 Moe from_pretrained failure are distinct FP8/model-loading issues." + "reason": "Non-persistent buffer corruption vs streaming mask off-by-one; different bug classes." }, { "left": "issue:44263", "right": "issue:44373", "accept": false, - "reason": "DSA indexer return values and a wrong docstring for position_ids do not describe the same bug." + "reason": "Runtime indexer bug vs docstring correction; not the same issue." }, { - "left": "issue:42907", - "right": "issue:46050", + "left": "issue:42831", + "right": "issue:45310", "accept": false, - "reason": "Failing to save dequantized Ministrals/Devstrals and sinq quantization broken after save/from_pretrained are different save/load quantization problems." + "reason": "FineGrainedFP8 accuracy issue vs Qwen3.5 MoE load error; distinct FP8-related problems." } ] }, @@ -13988,7 +13883,6 @@ "issue:44297", "issue:44373", "issue:44483", - "issue:44488", "issue:44568", "issue:44704", "issue:44829", @@ -13998,11 +13892,13 @@ "issue:45593", "issue:45753", "issue:46002", + "issue:46050", "issue:46132", "issue:46369", "issue:46402" ], "soft_pairs": [ + "issue:42907|issue:46050", "issue:44829|issue:46002", "issue:42831|issue:44038", "issue:44483|issue:45406", @@ -14013,235 +13909,227 @@ "issue:42831|issue:45753", "issue:39401|issue:44568", "issue:42757|issue:45278", - "issue:42886|issue:43066", "issue:27957|issue:30990", + "issue:42886|issue:43066", "issue:42886|issue:42994", "issue:43232|issue:45593", "issue:43408|issue:46369", - "issue:42796|issue:43122", "issue:44188|issue:44373", + "issue:42796|issue:43122", "issue:43550|issue:46132", - "issue:43646|issue:46002", "issue:44261|issue:44373", + "issue:43646|issue:46002", "issue:42371|issue:44263", "issue:43576|issue:43784", - "issue:42907|issue:44945", "issue:44295|issue:44297", + "issue:42907|issue:44945", "issue:30990|issue:33357", "issue:43756|issue:44060", - "issue:43976|issue:44945", - "issue:39401|issue:44488", "issue:44117|issue:44704", - "issue:41628|issue:45278" + "issue:43976|issue:44945" ], - "prepared_review_unit_hash": "4a1d6b8855f0c1905f36105b802479360a1241a0de9959fbda2a698d605f76a2", + "prepared_review_unit_hash": "e059ffbb337cb2caa2e197ab74185c2087154fd29aeba39c9e8b64505293bb82", "component_budget": { "node_count": 473, "item_count": 473, "soft_pair_count": 1738, - "serialized_chars": 901383, - "estimated_input_tokens": 225346, - "estimated_eval_tokens": 450948 + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 }, "budget": { "node_count": 48, "item_count": 48, - "soft_pair_count": 30, - "serialized_chars": 36172, - "estimated_input_tokens": 9043, - "estimated_eval_tokens": 18342 + "soft_pair_count": 29, + "serialized_chars": 35884, + "estimated_input_tokens": 8971, + "estimated_eval_tokens": 18198 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This is a heterogeneous issue cluster dominated by unrelated regression reports around tokenizers, imports, model loading, quantization, and serving. None of the soft pairs look like the same underlying bug/change, so I would not merge any of them.", - "confidence": 0.9, - "canonical_issue_reason": null, + "summary": "This cluster is mostly a mixed bag of unrelated Transformers issues across tokenization, imports, quantization, attention, and model-specific regressions. None of the soft pairs look like true duplicates of the same concrete bug; they are only loosely similar by subsystem or symptom.", + "confidence": 0.81, + "canonical_issue_reason": "Issue 45278 is the broadest representative of the import-regression/version-upgrade subset, which is the closest recurring theme in this set.", "canonical_pr_reason": null, - "best_issue_reason": "issue:42886 is the cleanest narrowly scoped regression report in the set (offline tokenizer cache loading) and the best anchor if one must pick an issue, but it is not a cluster-wide duplicate representative.", + "best_issue_reason": "Issue 45278 is the best overall representative because it captures the widest, most general failure mode here: post-upgrade import breakage.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:42907", + "right": "issue:46050", + "accept": false, + "reason": "Both concern quantized save/load, but one is dequantized Ministral/Devstral save_pretrained failure and the other is sinq round-trip breakage; different bug paths." + }, { "left": "issue:44829", "right": "issue:46002", "accept": false, - "reason": "Different failure modes: FA3 degenerate training vs NaNs at checkout/init; no shared code-path evidence." + "reason": "Both mention training instability, but one is flash_attention_3 degenerate training and the other is NaNs after checkout/init; not the same defect." }, { "left": "issue:42831", "right": "issue:44038", "accept": false, - "reason": "Different bugs: FineGrainedFP8 accuracy regression vs Qwen3-VL-Moe model-specific failure." + "reason": "FineGrainedFP8 accuracy loss and Qwen3-VL-Moe model issues are unrelated concrete problems." }, { "left": "issue:44483", "right": "issue:45406", "accept": false, - "reason": "One is API request rejection, the other is a processor attribute crash; unrelated symptoms and paths." + "reason": "One is an API request rejection issue, the other is a Gemma4Processor `_tokenizer` crash; unrelated." }, { "left": "issue:22355", "right": "issue:27957", "accept": false, - "reason": "Importing transformers.onnx and Flash Attention 2 on XLMRoberta are unrelated issues." + "reason": "ONNX import failure and XLMRoberta Flash Attention 2 support are different issues." }, { "left": "issue:41628", "right": "issue:43824", "accept": false, - "reason": "Both are import errors, but for different symbols/models and likely different broken exports." + "reason": "Different missing imports (`AutoImageProcessor` vs `Qwen2_5_VLForConditionalGeneration`), so these are separate bugs." }, { "left": "issue:43408", "right": "issue:46402", "accept": false, - "reason": "Both mention SAM3, but one is a class-mismatch warning and the other is missing text encoder weights." + "reason": "Both involve sam3, but one is a model-type warning and the other is missing text encoder weights; not the same code-path problem." }, { "left": "issue:22355", "right": "issue:28218", "accept": false, - "reason": "Import failure vs tokenizer spacing bug; no common underlying defect." + "reason": "Transformers ONNX import error and tokenizer added-token spacing bug are unrelated." }, { "left": "issue:42831", "right": "issue:45753", "accept": false, - "reason": "Accuracy regression in FP8 versus image-specific infinite loop; distinct problems." + "reason": "FP8 accuracy regression and a Qwen3_5 image-induced infinite loop are different failures." }, { "left": "issue:39401", "right": "issue:44568", "accept": false, - "reason": "Different tokenizer bugs: wrong offset mapping vs missing BOS/EOS insertion for a model tokenizer." + "reason": "Both are tokenizer-related, but one is wrong offset mappings and the other is missing BOS/EOS insertion; different behavior." }, { "left": "issue:42757", "right": "issue:45278", "accept": false, - "reason": "The latter is a broad upgrade breakage, not the same specific missing import as is_offline_mode." + "reason": "A single `is_offline_mode` import error is not the same concrete bug as the broader post-upgrade import failures." }, { - "left": "issue:42886", - "right": "issue:43066", + "left": "issue:27957", + "right": "issue:30990", "accept": false, - "reason": "Offline cache loading and wrong decoder type are different tokenizer issues." + "reason": "Flash Attention 2 support for XLMRoberta and Sentence Transformers loading hang are unrelated." }, { - "left": "issue:27957", - "right": "issue:30990", + "left": "issue:42886", + "right": "issue:43066", "accept": false, - "reason": "Flash Attention 2 model support and a loading hang in Sentence Transformers are unrelated." + "reason": "Offline cache loading failure and wrong tokenizer decoder type in v5 are separate tokenizer regressions." }, { "left": "issue:42886", "right": "issue:42994", "accept": false, - "reason": "Tokenizer offline-cache bug versus quantized model saving failure; different subsystems." + "reason": "Offline tokenizer cache loading and quantized model saving are different code paths." }, { "left": "issue:43232", "right": "issue:45593", "accept": false, - "reason": "Generation kwargs update after sync_gpus is unrelated to denoising auxiliary loss behavior." + "reason": "Generation kwargs/sync bug and D-FINE auxiliary-loss behavior are unrelated." }, { "left": "issue:43408", "right": "issue:46369", "accept": false, - "reason": "Both are SAM3-related, but a model-type warning is not the same as a memory overflow." + "reason": "SAM3 model-type warning versus SAM3 litetext memory overflow are different bugs." }, { - "left": "issue:42796", - "right": "issue:43122", + "left": "issue:44188", + "right": "issue:44373", "accept": false, - "reason": "Model-specific BPE marker handling versus general version-to-version tokenization drift; not enough to treat as one bug." + "reason": "A torch.compile attention-kernel divergence and a docstring typo are clearly not duplicates." }, { - "left": "issue:44188", - "right": "issue:44373", + "left": "issue:42796", + "right": "issue:43122", "accept": false, - "reason": "torch.compile attention divergence and a docstring error are completely unrelated." + "reason": "Both are tokenizer regressions, but one is BPE-marker handling for Ministral and the other is broader tokenization drift across versions." }, { "left": "issue:43550", "right": "issue:46132", "accept": false, - "reason": "torch.compile+SDPA failure and AttentionInterface.register behavior change are different code paths." + "reason": "Bamba SDPA/torch.compile failure and AttentionInterface.register behavior changes are different attention-related issues." }, { - "left": "issue:43646", - "right": "issue:46002", + "left": "issue:44261", + "right": "issue:44373", "accept": false, - "reason": "Custom model initialization breakage is not the same as NaNs in classification heads after checkout/init." + "reason": "MLA layernorm precision/config bug and a wrong docstring are unrelated." }, { - "left": "issue:44261", - "right": "issue:44373", + "left": "issue:43646", + "right": "issue:46002", "accept": false, - "reason": "Precision/config bug in MLA layernorm versus a wrong docstring; no overlap." + "reason": "Custom model initialization breakage and NaNs in classification heads are different failures with different fixes." }, { "left": "issue:42371", "right": "issue:44263", "accept": false, - "reason": "TF32 API settings and torch.split return values in a MoE indexer are unrelated." + "reason": "TF32 configuration warning and a torch.split return-value issue are unrelated." }, { "left": "issue:43576", "right": "issue:43784", "accept": false, - "reason": "CLI env command breakage and a sentence-transformers import NameError are different issues." + "reason": "Broken `transformers env` CLI and sentence-transformers import NameError are separate issues." }, { - "left": "issue:42907", - "right": "issue:44945", + "left": "issue:44295", + "right": "issue:44297", "accept": false, - "reason": "Quantized model saving failure versus pipeline-parallelism output corruption; different bug classes." + "reason": "Reading a registered position_ids buffer and tokenizer_class mismatch in saved config are different problems." }, { - "left": "issue:44295", - "right": "issue:44297", + "left": "issue:42907", + "right": "issue:44945", "accept": false, - "reason": "Reading position_ids after buffer registration is unrelated to tokenizer_class mismatch on save_pretrained." + "reason": "Quantized save/pretrained failure and pipeline-parallelism output corruption are unrelated." }, { "left": "issue:30990", "right": "issue:33357", "accept": false, - "reason": "Loading hang in Sentence Transformers versus a macOS bus error on a CLIP model; no common path." + "reason": "A loading hang and a MacOS bus error on CLIP are different failure modes." }, { "left": "issue:43756", "right": "issue:44060", "accept": false, - "reason": "RoPE layer drop in Smollm3 and a tied-weights warning in Qwen3-Next are unrelated model bugs." - }, - { - "left": "issue:43976", - "right": "issue:44945", - "accept": false, - "reason": "Python version compatibility regression and pipeline-parallelism output issues are not the same bug." - }, - { - "left": "issue:39401", - "right": "issue:44488", - "accept": false, - "reason": "Wrong offset_mapping and a model loading failure are different tokenizer/model issues." + "reason": "RoPE-layer dropping and tied-weights warning behavior are not the same bug." }, { "left": "issue:44117", "right": "issue:44704", "accept": false, - "reason": "Tokenizer mapping assumption bug versus AutoProcessor kwargs forwarding; related plumbing but not the same defect." + "reason": "Tokenizer mapping None handling and AutoProcessor kwargs passthrough are separate plumbing bugs." }, { - "left": "issue:41628", - "right": "issue:45278", + "left": "issue:43976", + "right": "issue:44945", "accept": false, - "reason": "A specific missing export and a broad import-breakage report after upgrade are too different to merge." + "reason": "Python version compatibility and pipeline-parallelism output issues are unrelated." } ] }, @@ -14271,11 +14159,11 @@ "issue:36296", "issue:38617", "issue:39401", + "issue:41628", "issue:42175", "issue:42757", "issue:42898", "issue:42907", - "issue:43066", "issue:43116", "issue:43122", "issue:43278", @@ -14294,6 +14182,7 @@ "issue:44351", "issue:44360", "issue:44485", + "issue:44488", "issue:44521", "issue:44625", "issue:44704", @@ -14314,26 +14203,28 @@ "issue:46614" ], "soft_pairs": [ + "issue:39401|issue:44488", + "issue:41628|issue:45278", "issue:42907|issue:44964", "issue:43784|issue:45600", "issue:42907|issue:43937", "issue:44360|issue:44485", - "issue:44625|issue:44829", "issue:44704|issue:45290", + "issue:44625|issue:44829", "issue:39401|issue:44521", "issue:43122|issue:46326", - "issue:43122|issue:44521", "issue:44038|issue:44297", + "issue:43122|issue:44521", "issue:43116|issue:45200", "issue:43408|issue:43475", - "issue:44038|issue:44295", "issue:39401|issue:46614", + "issue:44038|issue:44295", "issue:42907|issue:43452", "issue:44038|issue:44222", "issue:22355|issue:29159", "issue:45216|issue:45478", - "issue:43122|issue:44779", "issue:43824|issue:44351", + "issue:43122|issue:44779", "issue:36296|issue:43116", "issue:42757|issue:44351", "issue:43278|issue:44945", @@ -14343,223 +14234,228 @@ "issue:44188|issue:46328", "issue:38617|issue:45278", "issue:42898|issue:46614", - "issue:43937|issue:44485", - "issue:43066|issue:44779" + "issue:43937|issue:44485" ], - "prepared_review_unit_hash": "71fa8584e3333465859550caacb6f9d90e1bb46b343fdb210068d28e6cc4080d", + "prepared_review_unit_hash": "4ec113c6bd7e03c28d22c5eeef637e40d4447fa4c83d0276c6207cd276a0e622", "component_budget": { "node_count": 473, "item_count": 473, "soft_pair_count": 1738, - "serialized_chars": 901383, - "estimated_input_tokens": 225346, - "estimated_eval_tokens": 450948 + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 }, "budget": { - "node_count": 47, - "item_count": 47, - "soft_pair_count": 31, - "serialized_chars": 36126, - "estimated_input_tokens": 9032, - "estimated_eval_tokens": 18320 + "node_count": 48, + "item_count": 48, + "soft_pair_count": 32, + "serialized_chars": 36999, + "estimated_input_tokens": 9250, + "estimated_eval_tokens": 18756 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This set is a loose collection of unrelated issue reports (imports, tokenizer regressions, multimodal/model-specific bugs, and performance/behavior changes). None of the soft pairs looks safe to merge as the same underlying defect.", - "confidence": 0.93, - "canonical_issue_reason": "No single canonical issue: the items do not form one duplicate family, they span different code paths, models, and release regressions.", + "summary": "This cluster is heterogeneous: it mixes unrelated import errors, tokenizer regressions, model-specific bugs, and training/runtime issues. The pairwise similarities look lexical rather than describing the same underlying defect, so I would not merge any of the soft pairs as duplicates.", + "confidence": 0.74, + "canonical_issue_reason": "issue:45278 is the broadest representative of the import-error/version-upgrade subgroup and the most generic issue in the set, but the cluster overall does not look like one true duplicate family.", "canonical_pr_reason": null, - "best_issue_reason": "If one issue must stand in for the group, issue 44521 is the strongest standalone representative because it is concrete, open, and has the most discussion/inbound attention\u2014but it does not represent the whole set.", + "best_issue_reason": "issue:45278 is the best single representative because it is broad and clearly describes a common user-facing failure mode, even though it is not a safe duplicate of most others.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:39401", + "right": "issue:44488", + "accept": false, + "reason": "Different failures: tokenizer offset mapping bug vs a model-loading/import problem for sleng-bert." + }, + { + "left": "issue:41628", + "right": "issue:45278", + "accept": false, + "reason": "Both are import-related, but one is a specific missing symbol and the other is a broad upgrade-induced import-error report." + }, { "left": "issue:42907", "right": "issue:44964", "accept": false, - "reason": "Different problems: one is about saving dequantized Ministral/Devstral models, the other about loading Phi-4 multimodal." + "reason": "Unrelated code paths: dequantized save_pretrained failure vs loading a multimodal model." }, { "left": "issue:43784", "right": "issue:45600", "accept": false, - "reason": "Different failure modes and code paths: sentence-transformers import error vs broken auto_mappings references for Sam3LiteText configs." + "reason": "Both involve import-time breakage, but the missing-name issue in sentence-transformers is distinct from auto_mappings referencing removed configs." }, { "left": "issue:42907", "right": "issue:43937", "accept": false, - "reason": "Both are model-specific, but one is dequantized save behavior and the other is GLM-5 generation config validation; not the same bug." + "reason": "Different model behaviors: saving dequantized weights vs GenerationConfig validation." }, { "left": "issue:44360", "right": "issue:44485", "accept": false, - "reason": "Unrelated topics: DSA indexer discussion vs GLM-5 RoPE implementation." + "reason": "Completely different subsystems and symptoms; no shared bug mechanism is evident." }, { - "left": "issue:44625", - "right": "issue:44829", + "left": "issue:44704", + "right": "issue:45290", "accept": false, - "reason": "Different subsystems: num_labels propagation in Qwen3.5 config vs flash_attention_3 training degeneration." + "reason": "Different APIs and failures: cached_file kwargs propagation vs chat template crashes on tool-call messages." }, { - "left": "issue:44704", - "right": "issue:45290", + "left": "issue:44625", + "right": "issue:44829", "accept": false, - "reason": "Both involve high-level API behavior, but one is cached_file kwargs plumbing and the other is chat template crashes with tool-call messages." + "reason": "One is config propagation for Qwen3.5 labels; the other is a flash_attention_3 training regression." }, { "left": "issue:39401", "right": "issue:44521", "accept": false, - "reason": "Different tokenizer-related bugs: wrong offset_mapping for Qwen3 vs assistant_masks being all-zero for multimodal chat templates." + "reason": "Tokenizer offset mapping bug is unrelated to multimodal assistant_mask generation." }, { "left": "issue:43122", "right": "issue:46326", "accept": false, - "reason": "Both are tokenizer-related regressions, but one is general version-to-version tokenization differences and the other is continue_final_message handling for mistral_common tokenizers." + "reason": "Both are tokenizer-related, but they concern different tokenizer stacks and different failure modes." }, { - "left": "issue:43122", - "right": "issue:44521", + "left": "issue:44038", + "right": "issue:44297", "accept": false, - "reason": "Different functionality: tokenizer output drift vs multimodal apply_chat_template mask generation." + "reason": "Different problems: Qwen3-VL-Moe loading vs tokenizer_class metadata written by save_pretrained." }, { - "left": "issue:44038", - "right": "issue:44297", + "left": "issue:43122", + "right": "issue:44521", "accept": false, - "reason": "Unrelated: Qwen3-VL-Moe bug vs tokenizer.save_pretrained class metadata mismatch." + "reason": "Different functionality: text tokenization drift vs multimodal chat-template masking." }, { "left": "issue:43116", "right": "issue:45200", "accept": false, - "reason": "Different areas: example classification script returning empty results vs Gemma 4 multimodal token type defaults." + "reason": "Different features: classification example script output vs Gemma 4 mm_token_type_ids defaults." }, { "left": "issue:43408", "right": "issue:43475", "accept": false, - "reason": "Same SAM 3 Video family, but one is a model-type warning and the other is a missing attribute in vision encoder output; not clearly the same defect." + "reason": "These are separate SAM 3 video issues with different object/model attribute problems." }, { - "left": "issue:44038", - "right": "issue:44295", + "left": "issue:39401", + "right": "issue:46614", "accept": false, - "reason": "Different bugs: Qwen3-VL-Moe issue vs position_ids buffer read error." + "reason": "Tokenizer offset_mapping vs phonemizer backend state are unrelated." }, { - "left": "issue:39401", - "right": "issue:46614", + "left": "issue:44038", + "right": "issue:44295", "accept": false, - "reason": "Different tokenizer/phonemizer concerns: offset mapping bug vs stale espeak backend language state." + "reason": "Different bugs: Qwen3-VL-Moe loading vs reading position_ids after buffer registration." }, { "left": "issue:42907", "right": "issue:43452", "accept": false, - "reason": "Both touch loading/saving paths, but dequantized save failure and gguf_file loading breakage are separate code paths." + "reason": "Saving dequantized models and gguf_file loading failures are different code paths." }, { "left": "issue:44038", "right": "issue:44222", "accept": false, - "reason": "Different model/runtime issues: Qwen3-VL-Moe regression vs FP8 save_pretrained for moe models." + "reason": "Both mention model support, but one is a load-time bug and the other is FP8 save_pretrained handling." }, { "left": "issue:22355", "right": "issue:29159", "accept": false, - "reason": "No overlap in underlying bug: missing transformers.onnx module vs slow/fast tokenizer inconsistency." + "reason": "Missing transformers.onnx module is unrelated to slow/fast tokenizer inconsistency." }, { "left": "issue:45216", "right": "issue:45478", "accept": false, - "reason": "Related model family and version, but one is corrupted save_pretrained checkpoints and the other is a from_pretrained load error; not enough evidence for one duplicate." + "reason": "Related version and model family, but save_pretrained checkpoint corruption and from_pretrained errors are not the same concrete bug." }, { - "left": "issue:43122", - "right": "issue:44779", + "left": "issue:43824", + "right": "issue:44351", "accept": false, - "reason": "Both are v5 tokenizer regressions, but one reports generic tokenization drift while the other is Deepseek-specific incorrect results." + "reason": "Both are import errors, but they concern different missing classes from different modules." }, { - "left": "issue:43824", - "right": "issue:44351", + "left": "issue:43122", + "right": "issue:44779", "accept": false, - "reason": "Both are import errors, but they are for different symbols and likely different root causes." + "reason": "Both are tokenizer regressions, but there is no evidence they share the same tokenizer implementation or root cause." }, { "left": "issue:36296", "right": "issue:43116", "accept": false, - "reason": "Completely different areas: tensor parallel training bug vs multi-label classification example behavior." + "reason": "Tensor-parallel training bug is unrelated to the multi-label classification example script issue." }, { "left": "issue:42757", "right": "issue:44351", "accept": false, - "reason": "Different missing imports from different packages/modules; no shared underlying change." + "reason": "Different import failures in different packages/modules." }, { "left": "issue:43278", "right": "issue:44945", "accept": false, - "reason": "Different runtime problems: dtype drift between train/eval vs incorrect output under pipeline parallelism." + "reason": "Embedding dtype mismatch during eval is unrelated to pipeline-parallel output corruption." }, { "left": "issue:42175", "right": "issue:44938", "accept": false, - "reason": "No common bug: backend dependency selection in install extras vs Python 3.14 loading failure." + "reason": "One is missing TensorFlow in an install extra; the other is a Python 3.14 load failure." }, { "left": "issue:22355", "right": "issue:29127", "accept": false, - "reason": "Unrelated issues: missing transformers.onnx module vs layoutlmv3 error-message clarity." + "reason": "No module named transformers.onnx is unrelated to the LayoutLMv3 error-message issue." }, { "left": "issue:43493", "right": "issue:45753", "accept": false, - "reason": "Different model-specific issues: SigLIP2 implementation discrepancy vs Qwen3_5 infinite loop on a specific image." + "reason": "Different multimodal/model-specific bugs: SigLIP2 implementation drift vs Qwen3_5 image loop." }, { "left": "issue:44188", "right": "issue:46328", "accept": false, - "reason": "Different concerns: torch.compile attention-kernel branching vs SequenceFeatureExtractor.pad performance overhead." + "reason": "Diverging attention kernels under torch.compile is unrelated to SequenceFeatureExtractor.pad inefficiency." }, { "left": "issue:38617", "right": "issue:45278", "accept": false, - "reason": "Both mention imports, but one is a single missing symbol and the other is a broader upgrade-wide import failure; not enough to treat as the same bug." + "reason": "Both are import failures, but they are different missing symbols and do not look like the same bug." }, { "left": "issue:42898", "right": "issue:46614", "accept": false, - "reason": "Different behavior changes: v5 tokenization cleanup semantics vs phonemizer backend state leakage." + "reason": "Tokenization-space cleanup behavior and phonemizer language state are different regressions." }, { "left": "issue:43937", "right": "issue:44485", "accept": false, - "reason": "Same model family (GLM-5) but different problems: invalid GenerationConfig vs RoPE implementation review." - }, - { - "left": "issue:43066", - "right": "issue:44779", - "accept": false, - "reason": "Both are tokenizer regressions in v5, but one is wrong decoder type metadata and the other is Deepseek-specific incorrect tokenization." + "reason": "Both are GLM-5 related, but they target different implementation areas and failure modes." } ] }, @@ -14591,6 +14487,7 @@ "issue:42673", "issue:42796", "issue:43064", + "issue:43066", "issue:43116", "issue:43441", "issue:43493", @@ -14609,7 +14506,6 @@ "issue:44291", "issue:44355", "issue:44380", - "issue:44466", "issue:44625", "issue:44704", "issue:44779", @@ -14633,9 +14529,10 @@ "issue:46655" ], "soft_pairs": [ + "issue:43066|issue:44779", "issue:42673|issue:45072", - "issue:43116|issue:43493", "issue:43976|issue:45446", + "issue:43116|issue:43493", "issue:43611|issue:46002", "issue:44355|issue:46231", "issue:42371|issue:43874", @@ -14662,223 +14559,216 @@ "issue:44291|issue:45071", "issue:45405|issue:45446", "issue:42673|issue:43906", - "issue:39401|issue:44779", - "issue:44466|issue:46002" + "issue:39401|issue:44779" ], - "prepared_review_unit_hash": "0926859b89437363742b8dbe62907a8b5b94053a579bd6042bb8a63bbfc0fd9a", + "prepared_review_unit_hash": "7d6b90dd6fe33f850d98488c1307984d15e6455092207d242fd9c607f8fa609a", "component_budget": { "node_count": 473, "item_count": 473, "soft_pair_count": 1738, - "serialized_chars": 901383, - "estimated_input_tokens": 225346, - "estimated_eval_tokens": 450948 + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 31, - "serialized_chars": 36882, - "estimated_input_tokens": 9221, - "estimated_eval_tokens": 18698 + "serialized_chars": 36802, + "estimated_input_tokens": 9201, + "estimated_eval_tokens": 18658 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is a loose grab bag of mostly unrelated issue reports, with only weak lexical overlap around v5, tokenizers, flash attention, and training/runtime regressions. I would not merge any of the soft pairs as duplicates.", - "confidence": 0.93, - "canonical_issue_reason": "If a single issue must stand in for the cluster, 43611 is the broadest v5 regression about model loading, but there is no strong true canonical because the items cover many unrelated bugs.", + "summary": "This cluster is mostly a loose thematic grouping of Transformers v5 regressions, but the soft pairs are not true duplicates: they span different models, code paths, and failure modes. No PRs are present.", + "confidence": 0.37, + "canonical_issue_reason": "issue:33453 is the broadest umbrella issue in the set, covering a tokenizer-loading regression theme that overlaps several of the tokenizer-focused reports.", "canonical_pr_reason": null, - "best_issue_reason": "43611 is the most general and representative of the v5 breakage theme; the others are narrower model-, tokenizer-, or subsystem-specific reports.", + "best_issue_reason": "issue:44779 is the strongest concrete anchor for the cluster: a clear, user-visible tokenizer regression in v5 with a specific before/after behavior change.", "best_pr_reason": null, "soft_edge_verdicts": [ { - "left": "issue:42673", - "right": "issue:45072", + "left": "issue:43066", + "right": "issue:44779", "accept": false, - "reason": "VRAM leak in multi-dataloader-thread inference vs dtype mismatch in bfloat16 inference; different failure modes." + "reason": "Both are tokenizer regressions, but one is about decoder type handling while the other is DeepSeek output correctness; different bugs." }, { - "left": "issue:43116", - "right": "issue:43493", + "left": "issue:42673", + "right": "issue:45072", "accept": false, - "reason": "Example script returning empty multi-label results is unrelated to SigLIP2 HF-vs-JAX discrepancy." + "reason": "VRAM leak in multi-thread dataloader use is unrelated to the bfloat16 dtype mismatch in inference." }, { "left": "issue:43976", "right": "issue:45446", "accept": false, - "reason": "Python version compatibility issue vs a flex_attention import/version-check bug; not the same code path." + "reason": "Different version-compatibility failures: Python version gating vs PyTorch version check for a flex_attention import." + }, + { + "left": "issue:43116", + "right": "issue:43493", + "accept": false, + "reason": "Multi-label classification returning empty results is unrelated to a SigLIP2 HF-vs-JAX model discrepancy." }, { "left": "issue:43611", "right": "issue:46002", "accept": false, - "reason": "Model loading with base_model_prefix in v5 vs NaNs in classification heads; unrelated regressions." + "reason": "Model loading with base_model_prefix is a loading-path regression, while the other is NaNs in classification heads after checkout/init." }, { "left": "issue:44355", "right": "issue:46231", "accept": false, - "reason": "Compiled Python file errors and a docs parameter-name mismatch are unrelated." + "reason": "Compiled Python file errors are unrelated to a documentation parameter-name mismatch in set_training()." }, { "left": "issue:42371", "right": "issue:43874", "accept": false, - "reason": "TF32 settings behavior vs missing image-processor method for multimodal token counting; no overlap." + "reason": "TF32 behavior control is unrelated to the GLM46V image processor missing method / token-counting AttributeError." }, { "left": "issue:44038", "right": "issue:45478", "accept": false, - "reason": "Different Qwen MoE bugs with different symptoms and likely different loading paths; not one concrete fix." + "reason": "Both involve Qwen3-family models, but they are different from_pretrained failures on different model variants and versions." }, { "left": "issue:44625", "right": "issue:46002", "accept": false, - "reason": "num_labels propagation bug is unrelated to NaNs in classification heads." + "reason": "num_labels propagation in Qwen3.5 is a config-plumbing bug, not the same as NaNs in classification heads." }, { "left": "issue:45357", "right": "issue:45478", "accept": false, - "reason": "save_pretrained key serialization vs from_pretrained loading error; related family but different concrete problems." + "reason": "Saving wrong visual encoder keys and failing from_pretrained are different Qwen3.5 bugs in different code paths." }, { "left": "issue:45083", "right": "issue:45446", "accept": false, - "reason": "Helper length-calculation bug in qwen3_omni_moe vs a PyTorch version check issue; unrelated." + "reason": "Unexpected helper output lengths in qwen3_omni_moe is unrelated to a flex_attention import version check." }, { "left": "issue:44075", "right": "issue:44936", "accept": false, - "reason": "Optimizer argument handling and trainer.evaluate failing after train are different bugs." + "reason": "Ignoring SGD args and failing evaluate() after train() are distinct training-loop problems." }, { "left": "issue:42222", "right": "issue:43441", "accept": false, - "reason": "Vitpose model breakage vs Ministral FlashAttention failure; different models and failures." + "reason": "VitPose model breakage and Ministral FlashAttention failure affect different model families and mechanisms." }, { "left": "issue:43116", "right": "issue:43825", "accept": false, - "reason": "Empty multi-label outputs in an example script are not the same as a pipeline error-message regression." + "reason": "An example script returning empty multilabel results is unrelated to pipeline()'s translation-task error message." }, { "left": "issue:43992", "right": "issue:44704", "accept": false, - "reason": "Missing embed token weights in UMT5 loading vs AutoProcessor kwargs forwarding; different subsystems." + "reason": "Missing embed_tokens.weight in UMT5 loading and AutoProcessor not forwarding kwargs are separate loading/processor issues." }, { "left": "issue:28348", "right": "issue:46640", "accept": false, - "reason": "Feature request for GPT2 FlashAttention 2 support vs a generic open flash-attention issue; not the same change." + "reason": "GPT2 FlashAttention 2 support request is not the same bug as a generic flash-attention issue report." }, { "left": "issue:43064", "right": "issue:43673", "accept": false, - "reason": "FSDP2/PEFT optimizer-state mismatch vs missing GenerationMixin cache during chunked prefill; unrelated." + "reason": "Wrong optimizer states under FSDP2/PEFT/cpu_ram_efficient_loading is unrelated to missing GenerationMixin cache in chunked prefill." }, { "left": "issue:42673", "right": "issue:42796", "accept": false, - "reason": "VRAM leak in threaded inference vs tokenizer BPE marker handling; unrelated." + "reason": "A VRAM leak in multithreaded loading is unrelated to tokenizer BPE marker handling." }, { "left": "issue:42673", "right": "issue:46655", "accept": false, - "reason": "VRAM leak in Qwen3ForCausalLM vs incorrect video padding in SmolVLM; unrelated." + "reason": "The VRAM leak report and the video padding calculation bug involve different subsystems and symptoms." }, { "left": "issue:43937", "right": "issue:45563", "accept": false, - "reason": "GenerationConfig validation error vs stale warning for num_return_sequences; different generation behaviors." + "reason": "An invalid GenerationConfig error is different from a stale warning about num_return_sequences." }, { "left": "issue:43827", "right": "issue:45446", "accept": false, - "reason": "Docs referencing removed pipeline APIs vs a flex_attention import/version check; unrelated." + "reason": "Docs still referencing pipeline() after removal is unrelated to the flex_attention PyTorch version check." }, { "left": "issue:33453", "right": "issue:39401", "accept": false, - "reason": "Tokenizer loading regression and Qwen3 offset_mapping bug are both tokenizer-related but not the same underlying bug." + "reason": "Both are tokenizer regressions, but one is a broad loading regression and the other is a specific offset_mapping bug in Qwen3." }, { "left": "issue:45356", "right": "issue:46614", "accept": false, - "reason": "Different tokenizer regressions in different model families and code paths." + "reason": "Kimi tokenizer codec/warning regression is unrelated to stale espeak backend language state in phonemization." }, { "left": "issue:44380", "right": "issue:46077", "accept": false, - "reason": "GPT2 attention scaling ignored under SDPA/FlashAttention vs Gemma4 bidirectional mask construction; different attention logic." + "reason": "GPT2 attention scaling being ignored under SDPA/FlashAttention is a different attention-path bug from Gemma4 building causal masks." }, { "left": "issue:43673", "right": "issue:46328", "accept": false, - "reason": "Generation cache missing vs sequence feature padding performance issue; unrelated." + "reason": "Missing generation cache during chunked prefill is unrelated to SequenceFeatureExtractor.pad conversion overhead." }, { "left": "issue:44117", "right": "issue:45588", "accept": false, - "reason": "Tokenizer mapping assumption bug vs FlashAttention s_aux=None crash; unrelated." + "reason": "Tokenizer mapping None handling and flash_attention.py crashing on s_aux=None are unrelated code paths." }, { "left": "issue:45071", "right": "issue:46614", "accept": false, - "reason": "PretrainedConfig type-checking regression vs tokenizer phonemizer stale-language bug; unrelated." + "reason": "PretrainedConfig type checking regression is unrelated to stale phonemizer language backend state." }, { "left": "issue:44291", "right": "issue:45071", "accept": false, - "reason": "Both are v5 regressions, but one is init_empty_weights argument handling and the other is PretrainedConfig type checking; not the same fix." + "reason": "Unexpected _is_hf_initialized handling during init_empty_weights is a different loading bug than PretrainedConfig type checking." }, { "left": "issue:45405", "right": "issue:45446", "accept": false, - "reason": "PEFT minimum-version bump issue vs flex_attention PyTorch version check; unrelated." - }, - { - "left": "issue:42673", - "right": "issue:43906", - "accept": false, - "reason": "VRAM leak in threaded inference vs isolated reproduction of another issue; not enough evidence of same underlying bug." + "reason": "An unreleased PEFT version bump is unrelated to the PyTorch version check for flex_attention imports." }, { "left": "issue:39401", "right": "issue:44779", "accept": false, - "reason": "Different tokenizer regressions affecting different model families and outputs." - }, - { - "left": "issue:44466", - "right": "issue:46002", - "accept": false, - "reason": "Device-dependent lm_head serialization issue vs classification-head NaNs; unrelated." + "reason": "Both are tokenizer issues, but one is Qwen3 offset mapping and the other is DeepSeek tokenization correctness; different underlying bugs." } ] }, @@ -14905,7 +14795,6 @@ "issue:38617", "issue:39401", "issue:41628", - "issue:42175", "issue:42222", "issue:42371", "issue:42617", @@ -14914,7 +14803,6 @@ "issue:43122", "issue:43408", "issue:43450", - "issue:43673", "issue:43742", "issue:43937", "issue:43986", @@ -14924,6 +14812,7 @@ "issue:44261", "issue:44263", "issue:44410", + "issue:44466", "issue:44484", "issue:44485", "issue:44493", @@ -14943,6 +14832,7 @@ "issue:45542", "issue:45561", "issue:45656", + "issue:46002", "issue:46178", "issue:46209", "issue:46231", @@ -14952,20 +14842,21 @@ "issue:46617" ], "soft_pairs": [ + "issue:44466|issue:46002", "issue:43937|issue:44977", - "issue:43450|issue:43986", "issue:44410|issue:45310", + "issue:43450|issue:43986", "issue:45341|issue:45561", "issue:45341|issue:46328", "issue:45071|issue:46512", "issue:44829|issue:45216", "issue:44117|issue:45081", - "issue:43122|issue:46298", "issue:38617|issue:41628", "issue:42617|issue:45341", + "issue:43122|issue:46298", "issue:42222|issue:46178", - "issue:45362|issue:46328", "issue:42371|issue:44261", + "issue:45362|issue:46328", "issue:43408|issue:43742", "issue:44877|issue:46231", "issue:42371|issue:46617", @@ -14974,223 +14865,215 @@ "issue:43937|issue:44496", "issue:44933|issue:45278", "issue:44410|issue:45478", - "issue:44075|issue:45656", "issue:38617|issue:44220", + "issue:44075|issue:45656", "issue:43066|issue:43122", "issue:44263|issue:44485", "issue:42222|issue:45542", - "issue:43097|issue:44493", - "issue:43673|issue:45341", - "issue:42175|issue:42222" + "issue:43097|issue:44493" ], - "prepared_review_unit_hash": "bf2854b06f696634daa8463217b34bef8487569d26455a20c0efb8033999f9b6", + "prepared_review_unit_hash": "5685fd38b8ec01572c212b0f0008c3aa3406d4612749c2bb79d0864cce5ce7da", "component_budget": { "node_count": 473, "item_count": 473, "soft_pair_count": 1738, - "serialized_chars": 901383, - "estimated_input_tokens": 225346, - "estimated_eval_tokens": 450948 + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 }, "budget": { "node_count": 48, "item_count": 48, - "soft_pair_count": 30, - "serialized_chars": 36230, - "estimated_input_tokens": 9058, - "estimated_eval_tokens": 18372 + "soft_pair_count": 29, + "serialized_chars": 35985, + "estimated_input_tokens": 8997, + "estimated_eval_tokens": 18250 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is highly heterogeneous: mostly unrelated Transformers issues spanning tokenization, imports, config loading, model-specific regressions, training, and testing utilities. There are no pull requests here, and none of the soft pairs look like true duplicates of the same concrete bug/change.", - "confidence": 0.74, - "canonical_issue_reason": "issue:45278 is the broadest user-facing regression report in the set and the best umbrella representative for upgrade-related import failures; the other issues are much narrower or belong to different code paths/models.", + "summary": "Mostly unrelated issues were grouped together; the soft pairs are only title-level similarities and do not clear the strict duplicate threshold. I would not merge any of these pairs as duplicates.", + "confidence": 0.88, + "canonical_issue_reason": "No single canonical duplicate stands out: the items span unrelated regressions across imports, tokenization, model loading, testing, optimizer behavior, and config handling.", "canonical_pr_reason": null, - "best_issue_reason": "issue:45278 is the most suitable single representative because it captures a general import-regression class rather than a one-off model-specific symptom.", + "best_issue_reason": "Issue 45278 is the broadest umbrella candidate because it captures the only obvious repeated theme here: upgrade-related import regressions.", "best_pr_reason": null, "soft_edge_verdicts": [ { - "left": "issue:43937", - "right": "issue:44977", + "left": "issue:44466", + "right": "issue:46002", "accept": false, - "reason": "Both are generation-related, but one is a GenerationConfig validation error and the other is a Qwen3.5 flash-attention behavior bug; different failure modes and code paths." + "reason": "Both mention model weight behavior, but one is a serialization inconsistency and the other is NaNs in classification heads; different bugs and code paths." }, { - "left": "issue:43450", - "right": "issue:43986", + "left": "issue:43937", + "right": "issue:44977", "accept": false, - "reason": "One is a batched video-processor shape bug; the other is an AutoProcessor crash when torchvision is missing. Same area, different underlying issue." + "reason": "GenerationConfig validation for GLM-5 is unrelated to Qwen3.5 flash-attention generation failures." }, { "left": "issue:44410", "right": "issue:45310", "accept": false, - "reason": "Different models and symptoms: missing qwen3next projections versus Qwen3.5 MoE from_pretrained failure." + "reason": "One is a missing layer projection in qwen3next, the other is a Qwen3.5 MoE from_pretrained error; not the same defect." + }, + { + "left": "issue:43450", + "right": "issue:43986", + "accept": false, + "reason": "Incorrect batched video processor shapes and missing-torchvision AutoProcessor crashes are different problems." }, { "left": "issue:45341", "right": "issue:45561", "accept": false, - "reason": "Both touch testing utilities, but one is a small bug in testing_utils.py and the other is an xdist file race in patched testing utils; not the same concrete bug." + "reason": "A testing_utils bug and a pytest-xdist file race are not the same underlying issue." }, { "left": "issue:45341", "right": "issue:46328", "accept": false, - "reason": "Testing utils bug versus SequenceFeatureExtractor.pad performance issue; unrelated code paths." + "reason": "A testing helper bug and a SequenceFeatureExtractor.pad performance issue are unrelated." }, { "left": "issue:45071", "right": "issue:46512", "accept": false, - "reason": "PretrainedConfig type-checking regression and from_config(dtype=...) stale config dtype are different config/model-init bugs." + "reason": "Type-checking breakage in PretrainedConfig and stale config.dtype after from_config are distinct issues." }, { "left": "issue:44829", "right": "issue:45216", "accept": false, - "reason": "Flash-attention training degeneration is unrelated to the Qwen3.5 save_pretrained regression." + "reason": "Flash-attention-3 training degeneration and save_pretrained checkpoint regression are different code paths." }, { "left": "issue:44117", "right": "issue:45081", "accept": false, - "reason": "Tokenizer mapping None handling and Mistral regex patch crash are different tokenizer-loading problems." - }, - { - "left": "issue:43122", - "right": "issue:46298", - "accept": false, - "reason": "Tokenization behavior change versus a noisy logging message; no shared bug." + "reason": "TOKENIZER_MAPPING_NAMES returning None and the Mistral regex patch crash are unrelated tokenizer bugs." }, { "left": "issue:38617", "right": "issue:41628", "accept": false, - "reason": "Both are import errors, but they concern different missing symbols and different import surfaces." + "reason": "Both are import errors, but they involve different missing symbols from different modules and do not look like one concrete bug." }, { "left": "issue:42617", "right": "issue:45341", "accept": false, - "reason": "A 3d_parallel.py runtime failure and a testing_utils bug do not appear related." + "reason": "Running 3d_parallel.py and a testing_utils bug are unrelated." }, { - "left": "issue:42222", - "right": "issue:46178", + "left": "issue:43122", + "right": "issue:46298", "accept": false, - "reason": "VitPose model breakage and missing tensorflow_text backend mapping are unrelated." + "reason": "Tokenizer behavior regression and noisy logging are unrelated." }, { - "left": "issue:45362", - "right": "issue:46328", + "left": "issue:42222", + "right": "issue:46178", "accept": false, - "reason": "Qwen3.5 chat crash versus a padding performance issue; different code paths entirely." + "reason": "Broken vitpose models and a missing tensorflow_text backend entry are different issues." }, { "left": "issue:42371", "right": "issue:44261", "accept": false, - "reason": "TF32 API warning/control issue and MLA rms_norm_eps precision issue are unrelated." + "reason": "TF32 API guidance and MLA q_a_layernorm precision error are unrelated." + }, + { + "left": "issue:45362", + "right": "issue:46328", + "accept": false, + "reason": "A chat crash for Qwen3.5-35B and a padding performance issue are not the same bug." }, { "left": "issue:43408", "right": "issue:43742", "accept": false, - "reason": "A model-type warning for sam3_video/sam3_tracker is not the same as the MobileLLM-125M key error." + "reason": "A model-type warning for sam3_video/sam3_tracker and a MobileLLM key error are different failures." }, { "left": "issue:44877", "right": "issue:46231", "accept": false, - "reason": "Strict config loading for granite_speech and a documentation parameter typo are unrelated." + "reason": "Strict config loading for granite_speech and a set_training doc parameter mismatch are unrelated." }, { "left": "issue:42371", "right": "issue:46617", "accept": false, - "reason": "TF32 behavior control and trust_remote_code cache FileNotFoundError are different issues." + "reason": "TF32 settings guidance and trust_remote_code cache FileNotFoundError are unrelated." }, { "left": "issue:44484", "right": "issue:45310", "accept": false, - "reason": "A shard-size question and a Qwen3.5 MoE loading regression are not the same bug." + "reason": "save_pretrained sharding size behavior and Qwen3.5 MoE from_pretrained failure do not share a concrete bug." }, { "left": "issue:39401", "right": "issue:46209", "accept": false, - "reason": "Tokenizer offset mapping and FP8 inference on B200 are unrelated." + "reason": "Qwen3 tokenizer offset mapping and FP8 inference on B200 are unrelated." }, { "left": "issue:43937", "right": "issue:44496", "accept": false, - "reason": "GenerationConfig validation error versus unrecognized model/config.json issue; different failure paths." + "reason": "GenerationConfig validation for GLM-5 and an unrecognized model/config error are different problems." }, { "left": "issue:44933", "right": "issue:45278", "accept": false, - "reason": "Both involve imports, but one is a specific image_utils import problem and the other is a broad upgrade-related import regression." + "reason": "A specific missing import from image_utils and a broad report of many import errors after upgrade are related in theme but not clearly the same bug." }, { "left": "issue:44410", "right": "issue:45478", "accept": false, - "reason": "Missing qwen3next projections and Qwen3.5 MoE from_pretrained error are different model bugs." + "reason": "Missing qwen3next projections and Qwen3.5 MoE from_pretrained errors are not the same defect." }, { - "left": "issue:44075", - "right": "issue:45656", + "left": "issue:38617", + "right": "issue:44220", "accept": false, - "reason": "SGD args ignored and optimizer.step being called twice with deepspeed are different optimizer issues." + "reason": "A missing import from configuration_utils and a feature-extraction helper issue are unrelated." }, { - "left": "issue:38617", - "right": "issue:44220", + "left": "issue:44075", + "right": "issue:45656", "accept": false, - "reason": "ImportError for layer_type_validation and _torch_extract_fbank_features are unrelated." + "reason": "SGD args being ignored and optimizer.step() being called twice with deepspeed are different optimizer bugs." }, { "left": "issue:43066", "right": "issue:43122", "accept": false, - "reason": "Both are tokenizer/v5 regressions, but one is wrong decoder type and the other is changed tokenization output; not clearly the same underlying bug." + "reason": "Both concern tokenizer regressions, but one is about decoder type and the other about differing tokenization output; not enough to treat as the same bug." }, { "left": "issue:44263", "right": "issue:44485", "accept": false, - "reason": "torch.split return behavior in GlmMoeDsaIndexer is unrelated to GLM-5 RoPE implementation." + "reason": "A torch.split return-value issue in GlmMoeDsaIndexer and GLM-5 RoPE implementation are unrelated." }, { "left": "issue:42222", "right": "issue:45542", "accept": false, - "reason": "VitPose model breakage and missing TensorFlow backend with tensorboard-only install are different problems." + "reason": "Broken vitpose models and a TensorBoard-without-TensorFlow backend error are unrelated." }, { "left": "issue:43097", "right": "issue:44493", "accept": false, - "reason": "Removal of tie_embeddings_and_encoder_decoder and unexpected position-id keys are separate regressions." - }, - { - "left": "issue:43673", - "right": "issue:45341", - "accept": false, - "reason": "Generation cache missing during chunked_prefill and a testing_utils bug are unrelated." - }, - { - "left": "issue:42175", - "right": "issue:42222", - "accept": false, - "reason": "A packaging/backend dependency issue and a broken VitPose model issue are unrelated." + "reason": "Removal of tie_embeddings_and_encoder_decoder and unexpected position-id keys are different regressions." } ] }, @@ -15217,6 +15100,8 @@ "issue:27957", "issue:38617", "issue:41628", + "issue:42175", + "issue:42222", "issue:42757", "issue:43012", "issue:43064", @@ -15235,7 +15120,6 @@ "issue:44261", "issue:44263", "issue:44315", - "issue:44351", "issue:44360", "issue:44410", "issue:44483", @@ -15255,7 +15139,6 @@ "issue:45542", "issue:45588", "issue:45753", - "issue:46050", "issue:46123", "issue:46172", "issue:46231", @@ -15264,6 +15147,8 @@ "issue:46664" ], "soft_pairs": [ + "issue:43673|issue:45341", + "issue:42175|issue:42222", "issue:43874|issue:44261", "issue:41628|issue:45071", "issue:45071|issue:45542", @@ -15274,8 +15159,8 @@ "issue:44483|issue:45464", "issue:44484|issue:45478", "issue:27957|issue:43352", - "issue:43931|issue:44038", "issue:43441|issue:44829", + "issue:43931|issue:44038", "issue:43097|issue:43452", "issue:45003|issue:45341", "issue:43986|issue:45341", @@ -15287,243 +15172,227 @@ "issue:43673|issue:46123", "issue:43531|issue:45341", "issue:45440|issue:46172", - "issue:43986|issue:45372", "issue:43064|issue:44315", + "issue:43986|issue:45372", "issue:43097|issue:45588", "issue:38617|issue:42757", "issue:46655|issue:46664", - "issue:44410|issue:45375", - "issue:43097|issue:46050", - "issue:44038|issue:44410", - "issue:43986|issue:44484", - "issue:38617|issue:44351" + "issue:44410|issue:45375" ], - "prepared_review_unit_hash": "dd15d08fac81d279ab218b60cc6596383b2206081fbab3ab8253782ba268916c", + "prepared_review_unit_hash": "233c266f51f099656c69033308378f35196e58444db584bade6299e237f744a2", "component_budget": { "node_count": 473, "item_count": 473, "soft_pair_count": 1738, - "serialized_chars": 901383, - "estimated_input_tokens": 225346, - "estimated_eval_tokens": 450948 + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 }, "budget": { "node_count": 48, "item_count": 48, - "soft_pair_count": 33, - "serialized_chars": 37361, - "estimated_input_tokens": 9341, - "estimated_eval_tokens": 18938 + "soft_pair_count": 31, + "serialized_chars": 36792, + "estimated_input_tokens": 9198, + "estimated_eval_tokens": 18652 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is heterogeneous: it mixes unrelated import regressions, Flash Attention/model-support issues, configuration/loading bugs, and training/runtime problems. The soft pairs look like same subsystem themes but not the same underlying bug/change, so they should stay separate.", - "confidence": 0.81, - "canonical_issue_reason": "No true canonical issue stands out; the cluster spans multiple distinct regressions rather than one duplicate bug.", + "summary": "This cluster is a heterogeneous mix of unrelated Transformers issues (import regressions, model-specific architecture bugs, backend/package dependency problems, and API behavior changes). I do not see any soft pair that clearly describes the same underlying bug or change, so all candidate edges should be rejected.", + "confidence": 0.96, + "canonical_issue_reason": "Issue 45372 is the strongest single anchor: it is an active, concrete regression report with a precise import failure, version scope, and clear downstream impact on Gemma 4 processor loading.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 41628 is the cleanest representative of the largest subtheme (import failures), but it still does not unify the whole cluster.", + "best_issue_reason": "45372 is the most representative issue to keep: it has a specific reproducible failure, explicit affected versions, and active references, making it more actionable than the other mostly model-specific reports.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:43673", + "right": "issue:45341", + "accept": false, + "reason": "Different subsystems and failure modes: generation cache/chunked prefill vs a testing utility bug." + }, + { + "left": "issue:42175", + "right": "issue:42222", + "accept": false, + "reason": "One is a packaging/backend dependency issue; the other is a VitPose model breakage." + }, { "left": "issue:43874", "right": "issue:44261", "accept": false, - "reason": "Different bugs: missing image patch-count method vs a RoPE/layernorm precision mismatch." + "reason": "Different models and bugs: missing image-patch method vs missing rms_norm_eps precision issue." }, { "left": "issue:41628", "right": "issue:45071", "accept": false, - "reason": "Both are import/type-related, but one is an AutoImageProcessor import failure and the other is a PretrainedConfig type-checking regression." + "reason": "Unrelated regressions: missing AutoImageProcessor import vs PreTrainedConfig type-checking breakage." }, { "left": "issue:45071", "right": "issue:45542", "accept": false, - "reason": "Unrelated failures: config type checking vs TensorBoard-only-without-TF backend handling." + "reason": "Both touch backend-related code, but one is type checking and the other is a TensorFlow/tensorboard dependency problem." }, { "left": "issue:43012", "right": "issue:43577", "accept": false, - "reason": "Both involve dtype, but one is a compile-time PyTorch warning and the other is wrong dtype propagation when loading Blip2." + "reason": "Both involve dtype/precision, but one is a compilation warning and the other is incorrect loaded module dtypes." }, { "left": "issue:43931", "right": "issue:44410", "accept": false, - "reason": "Different model-loading bugs: Qwen3-VL shape mismatch vs missing attn projections in qwen3next." + "reason": "Different Qwen model bugs: weight-shape mismatch on load vs missing projections in layer 0." }, { "left": "issue:44263", "right": "issue:46172", "accept": false, - "reason": "Distinct code paths: a torch.split return-value issue vs a DeepSeek comb transpose/dtype bug." + "reason": "Different model families and code paths: a torch.split indexing issue vs DeepSeek-V4 dtype/transpose behavior." }, { "left": "issue:44360", "right": "issue:46172", "accept": false, - "reason": "Different defects in different models/code paths; no shared underlying change." + "reason": "Different underlying defects: missing ReLU in DSA indexer vs DeepSeek-V4 tensor conversion/transpose behavior." }, { "left": "issue:44483", "right": "issue:45464", "accept": false, - "reason": "Both hit chat/completions, but one is request rejection and the other is a streaming inference failure on Qwen3.5-0.8B." + "reason": "Both mention chat/completions, but one is a general API acceptance bug and the other is a Qwen3.5 streaming inference issue." }, { "left": "issue:44484", "right": "issue:45478", "accept": false, - "reason": "A save_pretrained shard-size question is unrelated to a Qwen3.5-MoE from_pretrained loading error." + "reason": "Completely different topics: save_pretrained shard-size question vs Qwen3.5-Moe loading error." }, { "left": "issue:27957", "right": "issue:43352", "accept": false, - "reason": "Both mention Flash Attention 2, but these are separate model-support gaps for different architectures." + "reason": "Same Flash Attention feature area, but different models and different support gaps; not the same bug." }, { - "left": "issue:43931", - "right": "issue:44038", + "left": "issue:43441", + "right": "issue:44829", "accept": false, - "reason": "Same broad Qwen3-VL area, but weight-shape mismatch and a v5/MoE bug are not the same issue." + "reason": "Different Flash Attention versions and symptoms: model loading failure vs degenerate training behavior." }, { - "left": "issue:43441", - "right": "issue:44829", + "left": "issue:43931", + "right": "issue:44038", "accept": false, - "reason": "Both mention Flash Attention, but one is model support failing and the other is degenerate training with flash_attention_3." + "reason": "Related model family, but one is a specific shape mismatch and the other is a broader Transformers 5/Qwen3-VL-Moe issue." }, { "left": "issue:43097", "right": "issue:43452", "accept": false, - "reason": "Different regressions: removed tie_embeddings API vs gguf_file breaking tokenizer/model loading." + "reason": "Unrelated breakages: removed config helper vs gguf_file tokenizer/model loading failure." }, { "left": "issue:45003", "right": "issue:45341", "accept": false, - "reason": "Unsafe sys.modules access in modeling_utils is unrelated to a small testing_utils bug." + "reason": "Different areas: sys.modules access in modeling_utils vs a testing_utils bug." }, { "left": "issue:43986", "right": "issue:45341", "accept": false, - "reason": "Video-model loading without torchvision and a testing_utils bug are unrelated." + "reason": "Different failure domains: AutoProcessor/video-model torchvision dependency vs testing utilities." }, { "left": "issue:45753", "right": "issue:46369", "accept": false, - "reason": "Different symptoms and subsystems: infinite loop on one image vs memory overflow in sam3-litetext." + "reason": "Different models and symptoms: infinite loop on a specific image vs memory overflow in sam3-litetext." }, { "left": "issue:44933", "right": "issue:46231", "accept": false, - "reason": "A missing import from image_utils is unrelated to a documentation parameter-name mismatch." + "reason": "Unrelated: missing image_utils import vs documentation mismatch in set_training()." }, { "left": "issue:43874", "right": "issue:44263", "accept": false, - "reason": "Different model-specific bugs: missing image patch counting vs torch.split handling." + "reason": "Different model components and bugs; no shared code path beyond general multimodal/model internals." }, { "left": "issue:44261", "right": "issue:44485", "accept": false, - "reason": "Layernorm epsilon precision issue is not the same as the GLM-5 RoPE implementation bug." + "reason": "One is a missing config epsilon precision bug; the other is a RoPE implementation discussion." }, { "left": "issue:44117", "right": "issue:44987", "accept": false, - "reason": "Generic TOKENIZER_MAPPING_NAMES handling and a specific physical-intelligence/fast loading failure are separate regressions." + "reason": "Different loading failures: tokenizer mapping None handling vs a specific model load failure." }, { "left": "issue:43673", "right": "issue:46123", "accept": false, - "reason": "Generation cache missing in chunked prefill is unrelated to MaskGenerationPipeline dropping final partial-batch results." + "reason": "Different pipelines and bugs: generation cache missing vs mask pipeline batch flag logic." }, { "left": "issue:43531", "right": "issue:45341", "accept": false, - "reason": "A Qwen3-MoE sliding_window bug is unrelated to a testing_utils bug." + "reason": "Qwen3-MoE sliding window behavior is unrelated to the testing_utils bug." }, { "left": "issue:45440", "right": "issue:46172", "accept": false, - "reason": "Both are DeepSeek-family issues, but they affect different implementations and different failure modes." + "reason": "Same DeepSeek naming, but different model versions and different underlying implementation bugs." }, { - "left": "issue:43986", - "right": "issue:45372", + "left": "issue:43064", + "right": "issue:44315", "accept": false, - "reason": "Video-model/torchvision loading failure and missing ReasoningEffort import for Gemma 4 processor loading are unrelated." + "reason": "Different training/runtime issues: FSDP2 optimizer-state corruption vs Liger Kernel not applied with model_init." }, { - "left": "issue:43064", - "right": "issue:44315", + "left": "issue:43986", + "right": "issue:45372", "accept": false, - "reason": "Wrong optimizer states with FSDP2+PEFT is a different problem from Liger Kernel not being applied with model_init." + "reason": "Both are loading-time import/dependency problems, but they affect different code paths and different missing dependencies." }, { "left": "issue:43097", "right": "issue:45588", "accept": false, - "reason": "Removed tie_embeddings API and a flash_attention.py s_aux=None crash are unrelated." + "reason": "Different bug classes: removed tie_embeddings API vs flash_attention.py crashing on s_aux=None." }, { "left": "issue:38617", "right": "issue:42757", "accept": false, - "reason": "Both are import errors, but they involve different missing symbols from different packages and look like separate regressions." + "reason": "Both are import errors, but for different symbols from different packages." }, { "left": "issue:46655", "right": "issue:46664", "accept": false, - "reason": "Both are SmolVLM video-processor issues, but padding calculation and num_frames semantics are distinct bugs." + "reason": "Same video processor area, but padding calculation and num_frames semantics are distinct bugs." }, { "left": "issue:44410", "right": "issue:45375", "accept": false, - "reason": "Missing projections in qwen3next layer 0 and a missing deepstack_visual_indexes config field are different problems." - }, - { - "left": "issue:43097", - "right": "issue:46050", - "accept": false, - "reason": "API removal in 5.0.0 is unrelated to a quantized-model save/load breakage." - }, - { - "left": "issue:44038", - "right": "issue:44410", - "accept": false, - "reason": "Different Qwen model variants and different failure modes; not the same underlying fix." - }, - { - "left": "issue:43986", - "right": "issue:44484", - "accept": false, - "reason": "Video-model loading dependency failure and max_shard_size behavior are unrelated." - }, - { - "left": "issue:38617", - "right": "issue:44351", - "accept": false, - "reason": "Both are import-name failures, but they concern different missing names and different upgrade breakages." + "reason": "Both are Qwen3 vision/MoE-related, but one is missing projections and the other is a missing config field being dropped." } ] }, @@ -15548,9 +15417,7 @@ "cluster_id": "cluster-10000-473", "nodes": [ "issue:36683", - "issue:42175", - "issue:42490", - "issue:42757", + "issue:38617", "issue:42890", "issue:42907", "issue:43012", @@ -15560,26 +15427,27 @@ "issue:43208", "issue:43352", "issue:43502", - "issue:43526", "issue:43550", "issue:43582", - "issue:43824", "issue:43825", "issue:43873", "issue:43906", "issue:43976", "issue:43986", "issue:43992", + "issue:44016", + "issue:44038", "issue:44062", "issue:44077", "issue:44117", "issue:44162", "issue:44222", + "issue:44351", "issue:44368", "issue:44380", + "issue:44410", "issue:44484", "issue:44655", - "issue:44683", "issue:44704", "issue:44936", "issue:44938", @@ -15589,6 +15457,7 @@ "issue:45584", "issue:45712", "issue:45753", + "issue:46050", "issue:46132", "issue:46172", "issue:46178", @@ -15597,6 +15466,10 @@ "issue:46421" ], "soft_pairs": [ + "issue:43097|issue:46050", + "issue:43986|issue:44484", + "issue:44038|issue:44410", + "issue:38617|issue:44351", "issue:42890|issue:46298", "issue:44062|issue:44117", "issue:43097|issue:43873", @@ -15606,271 +15479,242 @@ "issue:44380|issue:46421", "issue:43986|issue:44117", "issue:44938|issue:46178", - "issue:43824|issue:43986", + "issue:43986|issue:44016", "issue:43208|issue:43550", "issue:43825|issue:44655", "issue:42907|issue:44655", - "issue:43012|issue:43550", "issue:42907|issue:45753", - "issue:45712|issue:46298", + "issue:43012|issue:43550", "issue:43012|issue:44368", + "issue:45712|issue:46298", "issue:43992|issue:44484", "issue:43066|issue:43906", - "issue:43097|issue:46298", "issue:43352|issue:44380", "issue:43352|issue:43976", + "issue:43097|issue:46298", "issue:44162|issue:46172", - "issue:44380|issue:46132", "issue:45584|issue:46365", + "issue:44380|issue:46132", "issue:44077|issue:46365", "issue:44222|issue:45478", "issue:44380|issue:45479", "issue:36683|issue:43582", - "issue:43064|issue:44380", - "issue:42175|issue:46365", - "issue:42490|issue:43012", - "issue:43208|issue:43526", - "issue:42757|issue:43824", - "issue:43012|issue:44683" + "issue:43064|issue:44380" ], - "prepared_review_unit_hash": "f35218af4b742b6a18d9f62bd36cbe18a06dd67673212a0aebb6462afe6f601e", + "prepared_review_unit_hash": "470efb5e9774c15b51c051d42e22aedc04b301e8a7945cce9cad8978460f38d4", "component_budget": { "node_count": 473, "item_count": 473, "soft_pair_count": 1738, - "serialized_chars": 901383, - "estimated_input_tokens": 225346, - "estimated_eval_tokens": 450948 + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 }, "budget": { "node_count": 48, "item_count": 48, - "soft_pair_count": 35, - "serialized_chars": 38055, - "estimated_input_tokens": 9514, - "estimated_eval_tokens": 19284 + "soft_pair_count": 34, + "serialized_chars": 37691, + "estimated_input_tokens": 9423, + "estimated_eval_tokens": 19102 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "No soft pair looks like the same underlying bug or change; they mostly share a broad theme (imports, warnings, saving/loading, backend handling) but split into different code paths. I reject all proposed duplicate links; no PRs are present.", - "confidence": 0.91, - "canonical_issue_reason": "No exact duplicate stands out among the mixed issues; issue:36683 is the clearest standalone report because it has a precise runtime error and the most discussion.", + "summary": "These are mostly unrelated issue reports; I wouldn\u2019t merge any of the soft pairs as duplicates. The strongest representative issue is the v5 breaking-change report in 43097 because it is concrete, broadly relevant, and already referenced elsewhere.", + "confidence": 0.63, + "canonical_issue_reason": "issue:43097 is the clearest high-signal representative: a specific Transformers v5 API removal with inbound references and broad downstream impact.", "canonical_pr_reason": null, - "best_issue_reason": "Issue:43873 is the best overall triage candidate because it is open, well-scoped, and has multiple inbound references.", + "best_issue_reason": "issue:43097 is the best standalone issue to anchor the set because it is specific, impactful, and more central than the other reports here.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:43097", + "right": "issue:46050", + "accept": false, + "reason": "Different problems: a removed v5 embedding-tie API vs broken save/load for quantized sinq models." + }, + { + "left": "issue:43986", + "right": "issue:44484", + "accept": false, + "reason": "One is a torchvision dependency crash in AutoProcessor; the other is a save_pretrained shard-size question." + }, + { + "left": "issue:44038", + "right": "issue:44410", + "accept": false, + "reason": "Both are Qwen-related, but they describe different models and different missing-component bugs." + }, + { + "left": "issue:38617", + "right": "issue:44351", + "accept": false, + "reason": "Both are import errors, but for different missing symbols and likely unrelated code paths." + }, { "left": "issue:42890", "right": "issue:46298", "accept": false, - "reason": "A specific test-seed flakiness report and a noisy logging complaint are unrelated." + "reason": "Test flakiness from missing set_seed is unrelated to a noisy logging message." }, { "left": "issue:44062", "right": "issue:44117", "accept": false, - "reason": "Constructor keyword collision in AddedToken and a tokenizer-mapping None assumption are different bugs." + "reason": "Tokenizer construction error vs mapping lookup returning None; different failures in the tokenizer stack." }, { "left": "issue:43097", "right": "issue:43873", "accept": false, - "reason": "Removed config API and quantization/offloading failure are different code paths and symptoms." + "reason": "Removed v5 API vs quantization/offloading behavior; no shared concrete bug." }, { "left": "issue:44936", "right": "issue:46365", "accept": false, - "reason": "Trainer.evaluate after train and Trainer import failing with datasets are unrelated failures." + "reason": "Trainer.evaluate breakage is not the same as an import failure involving datasets." }, { "left": "issue:43502", "right": "issue:43992", "accept": false, - "reason": "Local-files-only network access and missing embed_tokens weights are distinct issues." + "reason": "Offline file lookup bug vs missing encoder weights on load; different code paths." }, { "left": "issue:44704", "right": "issue:45103", "accept": false, - "reason": "Passing kwargs to cached_file and an auto-docstring annotation crash are unrelated." + "reason": "AutoProcessor kwargs propagation and auto-docstring annotation parsing are unrelated." }, { "left": "issue:44380", "right": "issue:46421", "accept": false, - "reason": "GPT-2 attention scaling and cache lazy-init recompilation are different problems." + "reason": "Attention backend scaling bug vs cache lazy-init recompilation issue; different subsystems." }, { "left": "issue:43986", "right": "issue:44117", "accept": false, - "reason": "Missing torchvision for AutoProcessor video models and tokenizer mapping None are not the same bug." + "reason": "Video-model loading without torchvision is unrelated to tokenizer mapping returning None." }, { "left": "issue:44938", "right": "issue:46178", "accept": false, - "reason": "Python 3.14 load failure and tensorflow_text backend mapping error are unrelated." + "reason": "Python 3.14 import failure vs unsupported backend mapping; different root causes." }, { - "left": "issue:43824", - "right": "issue:43986", + "left": "issue:43986", + "right": "issue:44016", "accept": false, - "reason": "Qwen2.5-VL import failure and a torchvision-dependent video processor crash are different failures." + "reason": "A dependency crash is not the same as a notebook syntax error." }, { "left": "issue:43208", "right": "issue:43550", "accept": false, - "reason": "xLSTM training blockers and Bamba torch.compile/SDPA failure are unrelated." + "reason": "xLSTM training blockers and a Bamba torch.compile SDPA failure are separate bugs." }, { "left": "issue:43825", "right": "issue:44655", "accept": false, - "reason": "Pipeline translation messaging and Pipeline.save_pretrained behavior are different issues." + "reason": "Pipeline task messaging and pipeline serialization are different issues." }, { "left": "issue:42907", "right": "issue:44655", "accept": false, - "reason": "Saving dequantized models and saving Pipeline objects are not the same bug." - }, - { - "left": "issue:43012", - "right": "issue:43550", - "accept": false, - "reason": "A bfloat16 precision warning and a compile-time SDPA failure are different symptoms." + "reason": "Dequantized model save failure is not the same as inability to save Pipeline objects." }, { "left": "issue:42907", "right": "issue:45753", "accept": false, - "reason": "Model saving failure and an image-specific infinite loop are unrelated." + "reason": "Saving dequantized models and an infinite-loop inference bug are unrelated." }, { - "left": "issue:45712", - "right": "issue:46298", + "left": "issue:43012", + "right": "issue:43550", "accept": false, - "reason": "Dummy-class leakage and a noisy logging message are different cleanup/UX issues." + "reason": "A precision warning during compile is not the same as a model failure under SDPA compile." }, { "left": "issue:43012", "right": "issue:44368", "accept": false, - "reason": "A float32 warning during compile and a tie_word_embeddings warning are different config issues." + "reason": "Both mention warnings, but they concern different models and different tie-embedding behavior." + }, + { + "left": "issue:45712", + "right": "issue:46298", + "accept": false, + "reason": "Repository dummy-class cleanup is unrelated to logging noise." }, { "left": "issue:43992", "right": "issue:44484", "accept": false, - "reason": "Missing weights on load and a question about max_shard_size are unrelated." + "reason": "Missing embed_tokens weights on load is not the same as the 50GB shard-size default question." }, { "left": "issue:43066", "right": "issue:43906", "accept": false, - "reason": "Wrong tokenizer decoder type and an isolated reproduction of another issue are not enough to treat as the same bug." - }, - { - "left": "issue:43097", - "right": "issue:46298", - "accept": false, - "reason": "Removed tie_embeddings_and_encoder_decoder and a noisy logging message are unrelated." + "reason": "A tokenizer decoder-type regression is not enough to identify the reproduction issue as the same bug." }, { "left": "issue:43352", "right": "issue:44380", "accept": false, - "reason": "Nemotron Flash Attention 2 support and GPT-2 attention scaling with SDPA/FlashAttention are different code paths." + "reason": "FA2 unsupported for Nemotron and GPT2 attention scaling ignored are different backend issues." }, { "left": "issue:43352", "right": "issue:43976", "accept": false, - "reason": "Flash Attention support for Nemotron and a Python-version compatibility issue are unrelated." + "reason": "Model-specific FA2 support and a Python-version compatibility problem are unrelated." }, { - "left": "issue:44162", - "right": "issue:46172", + "left": "issue:43097", + "right": "issue:46298", "accept": false, - "reason": "ESM2 breakage and a deepseek tensor-transpose bug are different model-specific failures." + "reason": "A v5 API removal is unrelated to noisy logging output." }, { - "left": "issue:44380", - "right": "issue:46132", + "left": "issue:44162", + "right": "issue:46172", "accept": false, - "reason": "GPT-2 attention scaling and AttentionInterface.register changing behavior are different subsystems." + "reason": "ESM2 breakage and a DeepSeek tensor operation issue do not share the same underlying bug." }, { "left": "issue:45584", "right": "issue:46365", "accept": false, - "reason": "Whisper empty-transcription handling and a Trainer import conflict with datasets are unrelated." + "reason": "Whisper empty-transcription failure is unrelated to importing Trainer." }, { "left": "issue:44077", "right": "issue:46365", "accept": false, - "reason": "PatchTSMixer post_init policy and Trainer import failure are different issues." + "reason": "patchtsmixer post_init validation and Trainer import failure are different bugs." }, { "left": "issue:44222", "right": "issue:45478", "accept": false, - "reason": "FP8 MoE save_pretrained and Qwen3.5 MoE from_pretrained error are related only loosely, not the same concrete bug." - }, - { - "left": "issue:44380", - "right": "issue:45479", - "accept": false, - "reason": "Attention scaling config ignored and degenerate zero loss are unrelated." - }, - { - "left": "issue:36683", - "right": "issue:43582", - "accept": false, - "reason": "Gemma3Config missing vocab_size and an Apple Silicon allocator TypeError are unrelated." - }, - { - "left": "issue:43064", - "right": "issue:44380", - "accept": false, - "reason": "Distributed optimizer-state corruption and GPT-2 attention scaling are different failures." - }, - { - "left": "issue:42175", - "right": "issue:46365", - "accept": false, - "reason": "Backend packaging omission and a Trainer import conflict with datasets are unrelated." - }, - { - "left": "issue:42490", - "right": "issue:43012", - "accept": false, - "reason": "Trailing-slash local model warning and a bfloat16 compile warning are different issues." - }, - { - "left": "issue:43208", - "right": "issue:43526", - "accept": false, - "reason": "xLSTM training blockers and BeitImageProcessorFast label reduction are unrelated." - }, - { - "left": "issue:42757", - "right": "issue:43824", - "accept": false, - "reason": "huggingface_hub API import removal and a missing Qwen2.5-VL class import are different failures." + "reason": "Both involve MoE/FP8, but one is save_pretrained and the other is from_pretrained; not enough to treat as the same concrete bug." }, { - "left": "issue:43824", - "right": "issue:43986", + "left": "issue:44351", + "right": "issue:45478", "accept": false, - "reason": "Duplicate of the earlier Qwen2.5-VL vs torchvision mismatch, which is still not the same bug as a plain import error." + "reason": "HybridCache import failure and Qwen3.5 MoE loading error are unrelated." } ] }, @@ -15897,43 +15741,43 @@ "issue:36010", "issue:36683", "issue:38617", + "issue:42175", + "issue:42490", "issue:42757", "issue:42898", "issue:42994", - "issue:43010", - "issue:43064", + "issue:43012", "issue:43066", + "issue:43208", "issue:43299", "issue:43352", "issue:43475", "issue:43502", "issue:43525", + "issue:43526", "issue:43611", "issue:43618", "issue:43756", "issue:43761", + "issue:43824", "issue:43867", "issue:43901", "issue:43937", "issue:43976", - "issue:43992", "issue:44038", - "issue:44075", "issue:44077", "issue:44230", "issue:44380", "issue:44483", "issue:44485", - "issue:44492", "issue:44514", "issue:44625", "issue:44655", + "issue:44683", "issue:44938", "issue:45216", "issue:45276", - "issue:45278", "issue:45290", - "issue:45310", "issue:45600", "issue:45698", "issue:46077", @@ -15943,6 +15787,11 @@ "issue:46365" ], "soft_pairs": [ + "issue:42175|issue:46365", + "issue:42490|issue:43012", + "issue:43208|issue:43526", + "issue:42757|issue:43824", + "issue:43012|issue:44683", "issue:44483|issue:44514", "issue:45600|issue:46298", "issue:44038|issue:44230", @@ -15955,9 +15804,9 @@ "issue:36683|issue:43475", "issue:42994|issue:44655", "issue:44483|issue:45290", + "issue:44625|issue:46365", "issue:42898|issue:43066", "issue:45276|issue:46077", - "issue:44625|issue:46365", "issue:36010|issue:42757", "issue:43352|issue:43937", "issue:43901|issue:44655", @@ -15969,236 +15818,223 @@ "issue:46077|issue:46082", "issue:36683|issue:43525", "issue:43618|issue:45216", - "issue:36683|issue:44625", - "issue:43992|issue:45310", - "issue:43064|issue:44492", - "issue:44075|issue:46365", - "issue:43010|issue:45278" + "issue:36683|issue:44625" ], - "prepared_review_unit_hash": "17a80b4473867376147cf415aa0b3086e9857fe89a89cca2d1259acb403f89bf", + "prepared_review_unit_hash": "92ece040d5a43de40264a436b5abe0d81e4f4552c067f7473da2a8ae9b314212", "component_budget": { "node_count": 473, "item_count": 473, "soft_pair_count": 1738, - "serialized_chars": 901383, - "estimated_input_tokens": 225346, - "estimated_eval_tokens": 450948 + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 }, "budget": { "node_count": 47, "item_count": 47, - "soft_pair_count": 31, - "serialized_chars": 35895, - "estimated_input_tokens": 8974, - "estimated_eval_tokens": 18204 + "soft_pair_count": 32, + "serialized_chars": 36268, + "estimated_input_tokens": 9067, + "estimated_eval_tokens": 18390 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is dominated by Transformers v5 regression and compatibility issues, with one clear CLIP vision output regression pair and the rest mostly separate bugs across loading, config, save, and chat-template paths.", - "confidence": 0.83, - "canonical_issue_reason": "issue:45278 is the strongest umbrella: it is broad, user-facing, and explicitly covers multiple import failures after upgrading to Transformers 5.5.0, which matches the cluster\u2019s main upgrade-breakage theme.", + "summary": "The cluster is highly heterogeneous: most pairs only share superficial keywords like import, load, save, or v5, but they point to different models, configs, and code paths. I would not merge any of the soft edges as duplicates.", + "confidence": 0.92, + "canonical_issue_reason": "No single canonical issue stands out because the items span unrelated bugs across imports, model configs, serialization, attention behavior, and packaging/backend issues.", "canonical_pr_reason": null, - "best_issue_reason": "issue:45278 is the best representative because it captures the widest shared pattern in the cluster\u2014v5 upgrade breakage\u2014without being tied to one narrow model or API path.", + "best_issue_reason": "Issue 45698 is the closest representative only because it has the strongest discussion/inbound activity and is a broad save/load regression, but it still does not unify the rest of the cluster.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:42175", + "right": "issue:46365", + "accept": false, + "reason": "Packaging/backend install behavior vs Trainer import blocked by datasets; different failure modes and root causes." + }, + { + "left": "issue:42490", + "right": "issue:43012", + "accept": false, + "reason": "Local-path import warning vs PyTorch bfloat16 compile warning; unrelated code paths." + }, + { + "left": "issue:43208", + "right": "issue:43526", + "accept": false, + "reason": "xLSTM training bugs and BeitImageProcessorFast reduce_labels are different models and bugs." + }, + { + "left": "issue:42757", + "right": "issue:43824", + "accept": false, + "reason": "Both are import errors, but they concern different missing symbols in different modules." + }, + { + "left": "issue:43012", + "right": "issue:44683", + "accept": false, + "reason": "A warning during bf16 compilation is not the same as compiled flex_attention failing on newer torch." + }, { "left": "issue:44483", "right": "issue:44514", "accept": false, - "reason": "Different surfaces: chat-completions request validation vs batched chat-template crashing." + "reason": "Chat-completions request validation and apply_chat_template batch crash are different layers and symptoms." }, { "left": "issue:45600", "right": "issue:46298", "accept": false, - "reason": "Unrelated: broken auto-mappings CI vs a typo in cache strategies." + "reason": "Stale auto_mappings references vs noisy logging; not the same underlying bug." }, { "left": "issue:44038", "right": "issue:44230", "accept": false, - "reason": "Same model family, but different problems: loading regression vs fp8 inference support." + "reason": "Qwen3-VL-Moe loading regression vs fp8 inference support request; related family, different change." }, { "left": "issue:43618", "right": "issue:43761", - "accept": true, - "reason": "Both are CLIP vision v5 regressions where forward output fields are not being populated correctly." + "accept": false, + "reason": "Both involve CLIP regression, but one drops attentions and the other drops hidden_states; distinct bugs." }, { "left": "issue:43618", "right": "issue:44380", "accept": false, - "reason": "Different bug classes: missing CLIP attentions vs GPT2 attention scaling being ignored under SDPA/FlashAttention." + "reason": "CLIP output regression and GPT2 attention-scaling behavior are unrelated code paths." }, { "left": "issue:45698", "right": "issue:46172", "accept": false, - "reason": "Different functionality: custom-module reload bug vs a DeepSeek-V4 tensor transpose issue." + "reason": "Save/load custom-module resolution bug vs DeepSeek dtype/transpose bug; unrelated." }, { "left": "issue:38617", "right": "issue:43502", "accept": false, - "reason": "Different failures: missing import symbol vs network requests despite local_files_only=True." + "reason": "Missing import from configuration_utils vs local_files_only still triggering network access; different issues." }, { "left": "issue:43756", "right": "issue:44485", "accept": false, - "reason": "Both mention RoPE, but they concern different models and different implementation bugs." + "reason": "Smollm3 RoPE layer count bug vs GLM-5 RoPE implementation discussion; different models and defects." }, { "left": "issue:43299", "right": "issue:43867", "accept": false, - "reason": "Different loading failures with no clear shared code-path or concrete bug." + "reason": "Qwen3VL-Moe loading breakage vs state_dict-sorted load failure; both loading-related but not the same bug." }, { "left": "issue:36683", "right": "issue:43475", "accept": false, - "reason": "Different missing-attribute bugs in different model configs/outputs." + "reason": "Missing vocab_size on Gemma3Config vs missing fpn_position_embeddings on SAM3 output; different objects and models." }, { "left": "issue:42994", "right": "issue:44655", "accept": false, - "reason": "Both are save-related, but one is quantized model saving and the other is Pipeline.save_pretrained." + "reason": "Quantized model saving failure vs pipeline save_pretrained failure; both saving-related but distinct root causes." }, { "left": "issue:44483", "right": "issue:45290", "accept": false, - "reason": "Different chat-stack bugs: HTTP API request rejection vs assistant tool-call templating crash." + "reason": "Request rejection at /v1/chat/completions vs chat_template crash on tool-call messages; different problems." + }, + { + "left": "issue:44625", + "right": "issue:46365", + "accept": false, + "reason": "Qwen3.5 num_labels propagation vs Trainer import conflict with datasets; unrelated." }, { "left": "issue:42898", "right": "issue:43066", "accept": false, - "reason": "Both are tokenizer changes, but one is a cleanup-behavior change and the other is a decoder-type bug." + "reason": "Tokenization-space behavior change vs wrong tokenizer decoder type; both v5/tokenizer-related but not the same bug." }, { "left": "issue:45276", "right": "issue:46077", "accept": false, - "reason": "Both are Gemma4 issues, but one is embedding resizing and the other is attention-mask construction." - }, - { - "left": "issue:44625", - "right": "issue:46365", - "accept": false, - "reason": "Different areas: config propagation bug vs import conflict caused by datasets." + "reason": "Gemma4 embedding resizing bug vs Gemma4 bidirectional-attention masking bug; same model family, different fixes." }, { "left": "issue:36010", "right": "issue:42757", "accept": false, - "reason": "Both are import errors, but they are for different symbols/modules and likely different root causes." + "reason": "GenerationMixin import path and is_offline_mode import path are different missing-symbol regressions." }, { "left": "issue:43352", "right": "issue:43937", "accept": false, - "reason": "Different model/runtime issues: Flash Attention support gate vs invalid GenerationConfig." + "reason": "FlashAttention-2 unsupported for Nemotron vs invalid GenerationConfig for GLM-5; unrelated." }, { "left": "issue:43901", "right": "issue:44655", "accept": false, - "reason": "Docs mismatch for pipeline scoring vs inability to save Pipeline objects." - }, - { - "left": "issue:43611", - "right": "issue:44655", - "accept": false, - "reason": "Different breakages: model loading with base_model_prefix vs pipeline save_pretrained." - }, - { - "left": "issue:36683", - "right": "issue:44077", - "accept": false, - "reason": "Different config problems in different models; no shared underlying bug." + "reason": "Docs mismatch for return_all_scores vs pipeline save_pretrained failure; not a duplicate." }, { "left": "issue:43976", "right": "issue:44938", "accept": false, - "reason": "Both are Python-version compatibility issues, but for different versions and different failures." + "reason": "Python version support regression vs Python 3.14 load failure; different environment problems." }, { "left": "issue:43867", "right": "issue:45216", "accept": false, - "reason": "Different checkpoint/loading bugs with no concrete shared code-path." + "reason": "Sorted state_dict loading error vs Qwen3.5 save_pretrained regression; both serialization-related but different failure modes." }, { "left": "issue:44625", "right": "issue:46077", "accept": false, - "reason": "Different Gemma/Qwen bugs: label propagation vs bidirectional-attention masking." - }, - { - "left": "issue:46077", - "right": "issue:46082", - "accept": false, - "reason": "Related config areas, but the bugs are distinct and not mergeable as one fix." + "reason": "Qwen3.5 label propagation vs Gemma4 attention-mask construction; unrelated." }, { "left": "issue:36683", "right": "issue:43525", "accept": false, - "reason": "Different missing config attributes in different model families." + "reason": "Missing config field on Gemma3 vs missing config field on Llama4; same pattern, but different model-specific bugs." }, { "left": "issue:43618", "right": "issue:45216", "accept": false, - "reason": "Different regressions: CLIP output fields vs Qwen3.5 saved-checkpoint correctness." + "reason": "CLIP attentions regression vs Qwen3.5 save_pretrained regression; unrelated." }, { "left": "issue:36683", "right": "issue:44625", "accept": false, - "reason": "Different config propagation/attribute bugs in different models." - }, - { - "left": "issue:43992", - "right": "issue:45310", - "accept": false, - "reason": "Both are from_pretrained issues, but they affect different models and likely different loading paths." - }, - { - "left": "issue:43064", - "right": "issue:44492", - "accept": false, - "reason": "Unrelated: distributed optimizer-state corruption vs a cache-strategies typo." - }, - { - "left": "issue:44075", - "right": "issue:46365", - "accept": false, - "reason": "Different issues entirely: SGD argument handling vs Trainer import conflict." + "reason": "Different model families and different config propagation bugs." }, { - "left": "issue:43010", - "right": "issue:45278", + "left": "issue:36683", + "right": "issue:44077", "accept": false, - "reason": "Different areas: cache/layer update semantics vs broad import errors after upgrade." + "reason": "Gemma3 missing config attribute vs patchtsmixer post_init policy change; unrelated." } ] }, - "evaluator_result": { - "accept": false, - "feedback": "The overall clustering judgment is mostly conservative, but one soft-edge rationale is not grounded in the packet: the `issue:43064` vs `issue:44492` reason mentions a \"noisy logging message,\" which is actually `issue:46298`, not `issue:44492` (`Typo in Cache strategies`). That evidence mix-up is enough to reject. The rest of the summary is broadly plausible, and the accepted CLIP pair looks reasonable, but the output needs that incorrect pair rationale fixed." - }, - "evaluator_used": true, - "retried": true, - "accepted_nontrivial_soft_edge": true, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, "error_kind": null, "error_message": null }, @@ -16224,25 +16060,23 @@ "issue:42898", "issue:43010", "issue:43064", - "issue:43065", "issue:43502", "issue:43575", "issue:43618", "issue:43723", - "issue:43824", "issue:43867", "issue:43874", "issue:43976", "issue:43992", "issue:44062", + "issue:44075", "issue:44077", "issue:44112", "issue:44222", "issue:44246", "issue:44355", "issue:44360", - "issue:44387", - "issue:44485", + "issue:44492", "issue:44556", "issue:44734", "issue:44855", @@ -16251,6 +16085,8 @@ "issue:44964", "issue:45071", "issue:45125", + "issue:45278", + "issue:45310", "issue:45357", "issue:45375", "issue:45440", @@ -16260,22 +16096,27 @@ "issue:45850", "issue:46202", "issue:46285", + "issue:46365", "issue:46617", "issue:46664" ], "soft_pairs": [ + "issue:43992|issue:45310", + "issue:43064|issue:44492", + "issue:44075|issue:46365", + "issue:43010|issue:45278", "issue:44355|issue:44855", "issue:44938|issue:45071", - "issue:43867|issue:46202", "issue:43867|issue:45375", + "issue:43867|issue:46202", "issue:36683|issue:43064", "issue:42890|issue:45561", - "issue:43010|issue:45850", "issue:41669|issue:44246", + "issue:43010|issue:45850", "issue:36296|issue:43575", - "issue:43867|issue:45357", "issue:43010|issue:43976", "issue:43618|issue:44077", + "issue:43867|issue:45357", "issue:43992|issue:45478", "issue:28282|issue:33453", "issue:43618|issue:46285", @@ -16283,238 +16124,234 @@ "issue:44360|issue:45440", "issue:42890|issue:44964", "issue:42890|issue:44112", - "issue:42898|issue:46664", "issue:42890|issue:45600", + "issue:42898|issue:46664", "issue:43874|issue:46617", "issue:43502|issue:44062", - "issue:44222|issue:45125", "issue:43723|issue:44556", - "issue:44734|issue:46617", - "issue:44387|issue:44556", - "issue:43010|issue:43065", - "issue:43824|issue:46617", - "issue:43502|issue:44485" + "issue:44222|issue:45125", + "issue:44734|issue:46617" ], - "prepared_review_unit_hash": "89c89febcdadb9fc1fae9147f8b87d02f7fa9379fd7848d0566910a74140dc88", + "prepared_review_unit_hash": "9cf530c9aaed79e8dc90656ad39f865144c135458743a50da4dfde3ff19fc7c4", "component_budget": { "node_count": 473, "item_count": 473, "soft_pair_count": 1738, - "serialized_chars": 901383, - "estimated_input_tokens": 225346, - "estimated_eval_tokens": 450948 + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 }, "budget": { - "node_count": 47, - "item_count": 47, + "node_count": 48, + "item_count": 48, "soft_pair_count": 30, - "serialized_chars": 35777, - "estimated_input_tokens": 8945, - "estimated_eval_tokens": 18146 + "serialized_chars": 36255, + "estimated_input_tokens": 9064, + "estimated_eval_tokens": 18384 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly heterogeneous and does not look like one duplicate set. The only plausible duplicate pair is the import-time slowdown report around `import transformers`; the rest are different bugs across loading, config, tensor-parallelism, tests, and model-specific regressions.", - "confidence": 0.67, - "canonical_issue_reason": "issue:41669 is the most concrete root-cause report in the cluster: it identifies a specific import-performance problem and a specific code-path (`import *` in model packages) that plausibly explains the slowdown reports.", + "summary": "This cluster is mostly a grab-bag of unrelated Transformers v5 regression reports spanning model loading, tokenizers, tensor parallelism, tests, and import-time behavior. Only one soft edge looks like a plausible duplicate pair (import slowdown / import * issue).", + "confidence": 0.76, + "canonical_issue_reason": "issue:45278 is the best umbrella issue here: it is the broadest and most representative post-upgrade regression report, with enough detail to route multiple import-related breakages.", "canonical_pr_reason": null, - "best_issue_reason": "issue:41669 is the strongest representative issue here because it is specific, actionable, and tied to the only near-duplicate pair with a clear fix direction.", + "best_issue_reason": "issue:45278 is the strongest single issue to anchor this cluster because it summarizes a wider class of v5 incompatibilities rather than a narrow model-specific bug.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:43992", + "right": "issue:45310", + "accept": false, + "reason": "Different model families and failure modes: missing UMT5 embed weights vs Qwen3.5 MoE from_pretrained error." + }, + { + "left": "issue:43064", + "right": "issue:44492", + "accept": false, + "reason": "Trainer/FSDP optimizer-state bug is unrelated to a cache-strategy typo." + }, + { + "left": "issue:44075", + "right": "issue:46365", + "accept": false, + "reason": "SGD argument handling is unrelated to importing Trainer and dataset dependency conflicts." + }, + { + "left": "issue:43010", + "right": "issue:45278", + "accept": false, + "reason": "Cache/layer no_grad semantics are not the same underlying bug as broad import failures after upgrade." + }, { "left": "issue:44355", "right": "issue:44855", "accept": false, - "reason": "Both are import-time failures, but one is about compiled Python files and the other is a `torch.jit.script` parsing/indentation problem in DebertaV2; different code paths." + "reason": "Both involve Python/import errors, but one is compiled-file execution and the other is a specific JIT parsing/indentation issue." }, { "left": "issue:44938", "right": "issue:45071", "accept": false, - "reason": "One is a Python-version load failure, the other is a `PretrainedConfig` type-checking regression; same broad area, not the same bug." + "reason": "Python 3.14 import failure and PretrainedConfig type-checking regression are different problems." }, { "left": "issue:43867", - "right": "issue:46202", + "right": "issue:45375", "accept": false, - "reason": "`state_dict` loading failure and Gemma pipeline roundtrip/tokenization failure are unrelated subsystems." + "reason": "State-dict sorting/load failure is unrelated to a missing Qwen3.5 vision config field." }, { "left": "issue:43867", - "right": "issue:45375", + "right": "issue:46202", "accept": false, - "reason": "A generic model-load error and a missing `Qwen3_5MoeVisionConfig` field are different model/config bugs." + "reason": "Model load ordering issues are not the same as Gemma pipeline roundtrip tokenization behavior." }, { "left": "issue:36683", "right": "issue:43064", "accept": false, - "reason": "`Gemma3Config.vocab_size` missing and the FSDP2/PEFT optimizer-state bug are different layers of the stack." + "reason": "Gemma3Config missing vocab_size is unrelated to FSDP2/PEFT optimizer state corruption." }, { "left": "issue:42890", "right": "issue:45561", "accept": false, - "reason": "One is a flaky test due to missing `set_seed()`, the other is an xdist race on a captured-info file; both are test issues but not the same bug." - }, - { - "left": "issue:43010", - "right": "issue:45850", - "accept": false, - "reason": "`@torch.no_grad` on cache updates is a specific code fix; the other is a vague broad complaint with no shared concrete failure." + "reason": "Both are test failures, but one is seed nondeterminism and the other is an xdist file race." }, { "left": "issue:41669", "right": "issue:44246", "accept": true, - "reason": "Both point to `import transformers` being unexpectedly slow, and the `import *` removal in models is a plausible direct fix for that exact slowdown." + "reason": "These point to the same underlying import-time slowdown problem; one reports slow imports, the other identifies the import-* source and cost." }, { - "left": "issue:36296", - "right": "issue:43575", + "left": "issue:43010", + "right": "issue:45850", "accept": false, - "reason": "Tensor-parallel training bug versus Qwen2-57B loading OOM; both mention TP, but the failure modes and code paths differ." + "reason": "Generic 'breaks too many models' does not match the specific cache update no_grad bug." }, { - "left": "issue:43867", - "right": "issue:45357", + "left": "issue:36296", + "right": "issue:43575", "accept": false, - "reason": "`state_dict` load ordering and Qwen3.5 save-pretrained visual-key corruption are different regressions." + "reason": "Generic tensor-parallel training bug and a Qwen2 TP OOM are not clearly the same defect." }, { "left": "issue:43010", "right": "issue:43976", "accept": false, - "reason": "`no_grad` on cache/layer updates is unrelated to a Python 3.9/3.10 compatibility issue." + "reason": "Cache update no_grad semantics are unrelated to Python version compatibility." }, { "left": "issue:43618", "right": "issue:44077", "accept": false, - "reason": "Missing CLIP attentions output and an `optional post_init` restriction in PatchTSMixer are unrelated." + "reason": "CLIP attentions regression and patchtsmixer post_init policy are unrelated model issues." + }, + { + "left": "issue:43867", + "right": "issue:45357", + "accept": false, + "reason": "Different Qwen3.5 problems: state-dict/load ordering versus incorrect visual encoder keys on save." }, { "left": "issue:43992", "right": "issue:45478", "accept": false, - "reason": "A missing `embed_tokens.weight` in UMT5 and a Qwen3.5 MoE `from_pretrained` error are different model-loading bugs." + "reason": "Same broad phrase 'from_pretrained error', but the model and missing component differ; not the same bug." }, { "left": "issue:28282", "right": "issue:33453", "accept": false, - "reason": "PyTorch import missing for `AutoModel` and tokenizer loading regression are not the same issue." + "reason": "AutoModel missing PyTorch and tokenizer loading regression are unrelated." }, { "left": "issue:43618", "right": "issue:46285", "accept": false, - "reason": "One is a missing attentions assignment, the other is a structural mismatch between CLIP text model classes." + "reason": "CLIPOutput attentions regression is unrelated to CLIP text model structure divergence." }, { "left": "issue:43618", "right": "issue:44857", "accept": false, - "reason": "CLIP attentions handling and an AMP/CUDA crash in LwDetrImageLoss are unrelated." + "reason": "CLIP attentions handling and LwDetr AMP crash are different code paths." }, { "left": "issue:44360", "right": "issue:45440", "accept": false, - "reason": "A discussion about the DSA indexer and a DeepSeekV3MoE implementation divergence are different topics." + "reason": "DSA indexer/ReLU discussion is unrelated to DeepSeekV3MoE implementation divergence." }, { "left": "issue:42890", "right": "issue:44964", "accept": false, - "reason": "A seeding-related test flake and a latest-transformers model load failure do not share a concrete bug." + "reason": "Seed-related test flakiness is unrelated to Phi-4 multimodal model loading failures." }, { "left": "issue:42890", "right": "issue:44112", "accept": false, - "reason": "A test-seed flake and a GraniteSpeech device-override CI failure are separate test regressions." + "reason": "A flaky integration test and a GraniteSpeech CI stale override issue are different defects." }, { "left": "issue:42890", "right": "issue:45600", "accept": false, - "reason": "Missing `set_seed()` in SAM-HQ tests and stale auto-mapping references for Sam3LiteText are unrelated." + "reason": "Test seeding flakiness is unrelated to auto_mappings references breaking CI." + }, + { + "left": "issue:42898", + "right": "issue:46664", + "accept": false, + "reason": "Tokenizer cleanup behavior in v5 is unrelated to SmolVLM video processor semantics." }, { "left": "issue:43874", "right": "issue:46617", "accept": false, - "reason": "Missing `get_number_of_image_patches` in an image processor and a cache FileNotFoundError for trust_remote_code are different load-path bugs." + "reason": "Missing image-patch counting API and trust_remote_code cache FileNotFoundError are different loading paths." }, { "left": "issue:43502", "right": "issue:44062", "accept": false, - "reason": "`local_files_only=True` still making API calls and duplicated `AddedToken(..., special=...)` kwargs are unrelated tokenizer/loading issues." - }, - { - "left": "issue:44222", - "right": "issue:45125", - "accept": false, - "reason": "FP8 `save_pretrained` for MoE models and a missing tensor-parallel plan are different problems." + "reason": "Offline API-request leakage is unrelated to AddedToken keyword duplication." }, { "left": "issue:43723", "right": "issue:44556", "accept": false, - "reason": "Both involve loading regressions in v5, but one is tokenizer loading and the other is checkpoint reload compatibility after version upgrades; too broad to merge." - }, - { - "left": "issue:44734", - "right": "issue:46617", - "accept": false, - "reason": "KV-cache continuation crash in `transformers serve` and a cache FileNotFoundError are different code paths." - }, - { - "left": "issue:44387", - "right": "issue:44556", - "accept": false, - "reason": "Int4 quantization reserved-memory growth and checkpoint reload compatibility are unrelated regressions." - }, - { - "left": "issue:43010", - "right": "issue:43065", - "accept": false, - "reason": "A cache-update autograd fix and a dummy `nn.Conv2d` in Sam3PixelDecoder are unrelated." - }, - { - "left": "issue:43824", - "right": "issue:46617", - "accept": false, - "reason": "An import-name error for Qwen2.5 VL and a trust_remote_code cache FileNotFoundError are different issues." + "reason": "Tokenizer loading regression and checkpoint reload compatibility regression are not clearly the same bug." }, { "left": "issue:44222", - "right": "issue:44485", + "right": "issue:45125", "accept": false, - "reason": "FP8 MoE save-pretrained behavior and GLM-5 RoPE implementation are unrelated." + "reason": "FP8 save_pretrained behavior and missing tensor-parallel plan are distinct issues." }, { - "left": "issue:43867", - "right": "issue:46202", + "left": "issue:44734", + "right": "issue:46617", "accept": false, - "reason": "`state_dict` loading and Gemma pipeline roundtrip failure are different bugs." + "reason": "KV-cache continuation tensor indexing and cached trust_remote_code file lookup are unrelated." }, { "left": "issue:43867", - "right": "issue:45375", + "right": "issue:46617", "accept": false, - "reason": "Generic load error and missing vision-config field are not the same underlying problem." + "reason": "State-dict ordering/load problems do not match the trust_remote_code cache FileNotFoundError." } ] }, "evaluator_result": { "accept": true, - "feedback": "The summary is well grounded in the packet: the cluster is largely heterogeneous, and the only clearly plausible duplicate pair is issue:41669 with issue:44246 about slow `import transformers` caused by `import *` in model imports. The rejected soft pairs are stated conservatively and the distinctions match the issue titles/descriptions." + "feedback": "The cluster summary is well grounded in the packet: the issues are indeed a mixed set of unrelated v5 regressions, and the one accepted soft edge (41669/44246) is a conservative, mergeable duplicate pair about import-time slowdown from import-* usage. No soft-edge verdict appears overstated." }, "evaluator_used": true, "retried": false, @@ -16547,6 +16384,7 @@ "issue:42548", "issue:42757", "issue:43010", + "issue:43065", "issue:43232", "issue:43299", "issue:43475", @@ -16562,6 +16400,7 @@ "issue:44351", "issue:44355", "issue:44360", + "issue:44387", "issue:44485", "issue:44556", "issue:44655", @@ -16571,11 +16410,9 @@ "issue:44961", "issue:45071", "issue:45125", - "issue:45362", "issue:45375", "issue:45468", "issue:45478", - "issue:45850", "issue:46132", "issue:46143", "issue:46144", @@ -16584,6 +16421,10 @@ "issue:46664" ], "soft_pairs": [ + "issue:44387|issue:44556", + "issue:43010|issue:43065", + "issue:43502|issue:44485", + "issue:43824|issue:46617", "issue:43576|issue:46132", "issue:45125|issue:45478", "issue:30333|issue:33453", @@ -16618,254 +16459,280 @@ "issue:43475|issue:46617", "issue:41628|issue:42548", "issue:44315|issue:46209", - "issue:44485|issue:46664", - "issue:45362|issue:45850" + "issue:44485|issue:46664" ], - "prepared_review_unit_hash": "c3f238976baee1a3c7a85ada26c339aaa7eaa781e92b737be65e86e46c71aa6a", + "prepared_review_unit_hash": "bf82ddf7059586eb0b090f8663715dee38c7e302dd19957277d12ff5fd6e9487", "component_budget": { "node_count": 473, "item_count": 473, "soft_pair_count": 1738, - "serialized_chars": 901383, - "estimated_input_tokens": 225346, - "estimated_eval_tokens": 450948 + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 }, "budget": { "node_count": 47, "item_count": 47, - "soft_pair_count": 36, - "serialized_chars": 37163, - "estimated_input_tokens": 9291, - "estimated_eval_tokens": 18838 + "soft_pair_count": 39, + "serialized_chars": 38108, + "estimated_input_tokens": 9527, + "estimated_eval_tokens": 19310 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This set is mostly a loose collection of transformers regressions, import/export breakages, and docs issues. None of the proposed soft pairs look like true duplicates or the same concrete change, so I would not merge any of them.", + "summary": "These items are mostly unrelated, one-off bugs and documentation issues rather than a single duplicate cluster. The only near-match is the two RoFormer issues, which both point to missing RoFormer/attention-interface plumbing; the rest should stay separate.", "confidence": 0.89, - "canonical_issue_reason": "issue:45850 is the closest umbrella candidate because it explicitly describes broad v5 breakage across many models, which best matches the cluster\u2019s general regression theme, even though it is still very vague.", + "canonical_issue_reason": "If a single issue must serve as the anchor, issue 46617 is the most concrete and generally actionable runtime bug: it describes a reproducible cache-loading failure with a clear code path. That said, the overall set is too heterogeneous for a true canonical duplicate target.", "canonical_pr_reason": null, - "best_issue_reason": "issue:45850 is the best representative issue for this cluster because it is the broadest current report of the shared \u2018v5 breaks models\u2019 theme; the rest are mostly specific, unrelated sub-bugs.", + "best_issue_reason": "Issue 46617 is the strongest standalone candidate because it reports a specific failure mode, affects a common load path, and is easier to validate and triage than the more model-specific or documentation-only reports.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:44387", + "right": "issue:44556", + "accept": false, + "reason": "Both are version-regression reports, but one is int4 memory/OOM and the other is checkpoint reload failure; different symptoms and likely different fixes." + }, + { + "left": "issue:43010", + "right": "issue:43065", + "accept": false, + "reason": "One is about no_grad on cache/layer updates; the other is a Sam3PixelDecoder Conv2d placeholder issue. No shared code-path bug." + }, + { + "left": "issue:43502", + "right": "issue:44485", + "accept": false, + "reason": "local_files_only network leakage is unrelated to GLM-5 RoPE implementation." + }, + { + "left": "issue:43824", + "right": "issue:46617", + "accept": false, + "reason": "Both involve loading/imports, but one is a missing export for a specific model class and the other is a trust_remote_code cache FileNotFoundError; not the same bug." + }, { "left": "issue:43576", "right": "issue:46132", "accept": false, - "reason": "Different code paths: CLI/env command breakage vs AttentionInterface.register behavior. Same project, not the same bug." + "reason": "Broken `transformers env` CLI output and AttentionInterface.register behavior are separate issues in different areas." }, { "left": "issue:45125", "right": "issue:45478", "accept": false, - "reason": "Both mention Qwen3.5 MoE loading, but one is tensor-parallel _tp_plan support and the other is a from_pretrained regression; too different to treat as one bug." + "reason": "Both concern Qwen3.5 MoE, but one is tensor-parallel metadata and the other is a from_pretrained regression; too different to merge." }, { "left": "issue:30333", "right": "issue:33453", "accept": false, - "reason": "Unrelated failures: MLflow reporting on training failure vs tokenizer loading regression." + "reason": "MLflow reporting on failed training jobs is unrelated to tokenizer-loading regression." }, { "left": "issue:44355", "right": "issue:44749", "accept": false, - "reason": "One is about compiled Python file execution errors; the other is a performance regression after upgrade. No shared underlying bug." + "reason": "Compiled-Python-file errors and a data-filtering slowdown after upgrade are unrelated symptoms." }, { "left": "issue:45125", "right": "issue:45375", "accept": false, - "reason": "Different model-specific issues: tensor parallel planning vs a missing config field in Qwen3.5 MoE vision config." + "reason": "Missing `_tp_plan` for tensor parallelism and a missing vision-config field are different model-config bugs." }, { "left": "issue:42548", "right": "issue:44351", "accept": false, - "reason": "Both are import errors, but for different symbols and likely different missing exports; not the same concrete failure." + "reason": "Both are import errors, but for different symbols and likely different packaging/export causes." }, { "left": "issue:43010", "right": "issue:43232", "accept": false, - "reason": "Both touch generation internals, but one is about no_grad decoration on cache updates and the other is about _update_model_kwargs_for_generation after sync_gpus. Not the same bug." + "reason": "Cache/layer update no_grad and generation kwargs update after sync_gpus are separate generation/caching problems." }, { "left": "issue:30990", "right": "issue:31795", "accept": false, - "reason": "Sentence-transformers loading stall vs documentation confusion for forward inputs; unrelated." + "reason": "Sentence Transformers loading hangs and docs confusion about forward arguments are unrelated." }, { "left": "issue:43502", "right": "issue:44222", "accept": false, - "reason": "Local-files-only network leakage vs FP8 moe save_pretrained bug; different subsystems and symptoms." + "reason": "local_files_only network leakage is unrelated to FP8 save_pretrained behavior." }, { "left": "issue:43502", "right": "issue:44961", "accept": false, - "reason": "The second item is not a relevant duplicate of the first; there is no shared bug or code path." + "reason": "A network-policy bug and a placeholder issue are not the same underlying change." }, { "left": "issue:43827", "right": "issue:44961", "accept": false, - "reason": "Docs still referencing removed pipeline API vs an unrelated issue." + "reason": "Doc references to removed pipelines and an unrelated placeholder issue should not be merged." }, { "left": "issue:30333", "right": "issue:36296", "accept": false, - "reason": "MLflow reporting vs tensor-parallel training bug; no overlap in behavior or code path." + "reason": "MLflow reporting and tensor-parallel training are unrelated." }, { "left": "issue:43299", "right": "issue:44556", "accept": false, - "reason": "Both are version-related loading regressions, but they concern different model families and failure modes." + "reason": "Both are upgrade regressions, but they target different model-loading failures and different version ranges." }, { "left": "issue:44355", "right": "issue:45071", "accept": false, - "reason": "Compiled-file execution errors are unrelated to PretrainedConfig type-checking breakage." + "reason": "Compiled-file errors and `PretrainedConfig` type-checking breakage are unrelated." }, { "left": "issue:44355", "right": "issue:44655", "accept": false, - "reason": "Execution of compiled Python files and saving Pipeline objects are distinct problems." + "reason": "Running compiled Python files and saving Pipeline objects are different failures." }, { "left": "issue:33453", "right": "issue:34689", "accept": false, - "reason": "Different loading regressions affecting different artifacts and likely different loader paths." + "reason": "Tokenizer-loading regression and Llama 3.2 model-loading regression are distinct bugs." }, { "left": "issue:27957", "right": "issue:33453", "accept": false, - "reason": "Flash Attention support for XLM-Roberta is unrelated to a tokenizer-loading regression." + "reason": "Flash Attention 2 support for XLM-Roberta and tokenizer-loading regression are unrelated." }, { "left": "issue:42757", "right": "issue:43576", "accept": false, - "reason": "Importing is_offline_mode from huggingface_hub and a broken transformers env subcommand are different issues." + "reason": "Missing `is_offline_mode` import from huggingface_hub and broken env CLI are separate issues." }, { "left": "issue:43576", "right": "issue:45071", "accept": false, - "reason": "CLI/env command breakage vs PretrainedConfig type-checking; no shared underlying change." + "reason": "CLI/env breakage and `PretrainedConfig` type checking are unrelated." }, { "left": "issue:44315", "right": "issue:44360", "accept": false, - "reason": "Liger Kernel model_init behavior and a DSA indexer ReLU discussion are unrelated." + "reason": "Liger Kernel application and a DSA indexer ReLU discussion are unrelated." }, { "left": "issue:43576", "right": "issue:43824", "accept": false, - "reason": "Broken env command and missing Qwen2_5_VLForConditionalGeneration import are not the same bug." + "reason": "Broken env command and missing Qwen2.5-VL export are different code paths." }, { "left": "issue:31795", "right": "issue:38617", "accept": false, - "reason": "Documentation confusion in a model forward signature vs an ImportError for layer_type_validation." + "reason": "Documentation confusion and an import error for `layer_type_validation` are unrelated." }, { "left": "issue:44961", "right": "issue:45468", "accept": false, - "reason": "These titles describe unrelated topics; no shared underlying bug." + "reason": "A placeholder issue and Gemma-4 audio positional encoding bug are not duplicates." }, { "left": "issue:44351", "right": "issue:46617", "accept": false, - "reason": "HybridCache import failure and trust_remote_code cache FileNotFoundError are different failure modes." + "reason": "Missing `HybridCache` export and cache FileNotFoundError are both loading-related but not the same defect." }, { "left": "issue:44485", "right": "issue:44961", "accept": false, - "reason": "RoPE implementation discussion is unrelated to the racoon issue." + "reason": "GLM-5 RoPE implementation discussion is unrelated to the placeholder issue." + }, + { + "left": "issue:43576", + "right": "issue:44938", + "accept": false, + "reason": "Broken env CLI and Python 3.14 load failure are separate regressions." }, { "left": "issue:44908", "right": "issue:44961", "accept": false, - "reason": "Scheduler kwargs bug vs an unrelated issue." + "reason": "Scheduler kwargs handling and the placeholder issue are unrelated." }, { "left": "issue:36296", "right": "issue:43901", "accept": false, - "reason": "Tensor-parallel training bug and TextClassificationPipeline docs mismatch are unrelated." + "reason": "Tensor-parallel training and pipeline docs mention are different problems." }, { "left": "issue:44222", "right": "issue:44230", "accept": false, - "reason": "Both are FP8/MoE-related, but one is save_pretrained and the other is inference support; too different to merge as one bug." + "reason": "FP8 save_pretrained and FP8 inference support for Qwen3-VL/Qwen3.5 MoE are related only at a broad feature level, not the same bug." }, { "left": "issue:46143", "right": "issue:46144", - "accept": false, - "reason": "One is about kwargs not propagating through RoFormer methods; the other is attention-interface integration. Related area, but not the same concrete bug." + "accept": true, + "reason": "Both describe the same RoFormer plumbing gap: kwargs are not propagated and the attention implementation does not use the attention interface, which points to one underlying integration defect." }, { "left": "issue:36296", "right": "issue:43526", "accept": false, - "reason": "Tensor-parallel training and BeitImageProcessorFast label reduction are unrelated." + "reason": "Tensor parallel training and BeitImageProcessorFast label reduction are unrelated." }, { "left": "issue:43475", "right": "issue:46617", "accept": false, - "reason": "SAM3 video encoder attribute mismatch and trust_remote_code cache loading are unrelated." + "reason": "SAM3 vision encoder attribute bug and trust_remote_code cache loading failure are unrelated." }, { "left": "issue:41628", "right": "issue:42548", "accept": false, - "reason": "Both are import errors, but for different public symbols; no evidence they share the same root cause." + "reason": "Different import errors for different symbols; no shared underlying fix is evident." }, { "left": "issue:44315", "right": "issue:46209", "accept": false, - "reason": "Liger Kernel application with model_init and FP8 inference on B200 are separate issues." + "reason": "Liger Kernel application and FP8 inference on B200 are different feature/regression areas." }, { "left": "issue:44485", "right": "issue:46664", "accept": false, - "reason": "GLM-5 RoPE implementation and SmolVLM num_frames semantics are unrelated." - }, - { - "left": "issue:45362", - "right": "issue:45850", - "accept": false, - "reason": "A specific Qwen3.5-35B chat crash is not the same concrete issue as a broad meta-report that v5 breaks too many models." + "reason": "GLM-5 RoPE implementation and SmolVLM video-processor semantics are unrelated." } ] }, - "evaluator_result": null, - "evaluator_used": false, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet: it correctly treats the set as mostly heterogeneous and only flags the RoFormer pair as a plausible near-match. The individual soft-edge rejections are conservative overall, and the one accepted pair is reasonably supported by the issue titles." + }, + "evaluator_used": true, "retried": false, - "accepted_nontrivial_soft_edge": false, + "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null }, @@ -16920,6 +16787,7 @@ "issue:45230", "issue:45278", "issue:45341", + "issue:45362", "issue:45381", "issue:45478", "issue:45542", @@ -16931,6 +16799,7 @@ "issue:46664" ], "soft_pairs": [ + "issue:45362|issue:45850", "issue:43673|issue:44492", "issue:44623|issue:45230", "issue:44360|issue:46664", @@ -16972,304 +16841,280 @@ "issue:44315|issue:44829", "issue:43576|issue:44273" ], - "prepared_review_unit_hash": "1018b334709d17c89197d3f086faf435029a63a11e4babe16bb6ccdcfa8d8056", + "prepared_review_unit_hash": "fbc70cc6aacc7423d1fd46b044c96a9fec0b76e3eced96a0a38607a177313e54", "component_budget": { "node_count": 473, "item_count": 473, "soft_pair_count": 1738, - "serialized_chars": 901383, - "estimated_input_tokens": 225346, - "estimated_eval_tokens": 450948 + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 }, "budget": { - "node_count": 47, - "item_count": 47, - "soft_pair_count": 40, - "serialized_chars": 38597, - "estimated_input_tokens": 9650, - "estimated_eval_tokens": 19556 + "node_count": 48, + "item_count": 48, + "soft_pair_count": 41, + "serialized_chars": 39442, + "estimated_input_tokens": 9861, + "estimated_eval_tokens": 19978 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is a loose mix of Transformers v5 regression reports, import/lazy-loading complaints, model-specific breakages, and a few unrelated bugs. None of the soft pairs look like the same underlying issue closely enough to merge as duplicates.", - "confidence": 0.58, - "canonical_issue_reason": "issue:45850 is the broadest umbrella in the set and can represent the general 'v5 breaks many models' regression theme across the cluster.", + "summary": "The soft pairs are mostly loose thematic matches, not true duplicates. They span unrelated regressions across model configs, import/lazy-loading, docs, checkpoints, and scheduler/API changes. No pull requests are present.", + "confidence": 0.42, + "canonical_issue_reason": "issue:45478 is the most concrete, actionable report in the set: it names the model family, version threshold, and failure mode clearly, while most other issues are broader or more ambiguous.", "canonical_pr_reason": null, - "best_issue_reason": "issue:45850 is the best global representative because it is open-ended and captures the widest range of v5 regression symptoms, even though it is not very actionable.", + "best_issue_reason": "issue:45478 is the strongest standalone issue to anchor triage because it is specific and reproducible; it is more useful than the broad upgrade-breakage complaints.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:45362", + "right": "issue:45850", + "accept": false, + "reason": "A specific Qwen3.5 crash and a broad \"many models break\" complaint are only thematically similar, not the same bug." + }, { "left": "issue:43673", "right": "issue:44492", "accept": false, - "reason": "Both mention cache-related areas, but one is a GenerationMixin chunked_prefill cache bug and the other is a typo in cache strategy text; not the same code-path." + "reason": "Chunked-prefill cache regression vs a typo in cache strategy text/labels; different code paths." }, { "left": "issue:44623", "right": "issue:45230", "accept": false, - "reason": "45230 is too generic to establish the same processor.save_pretrained missing-files bug." + "reason": "Missing files in processor.save_pretrained is unrelated to a generic bug report." }, { "left": "issue:44360", "right": "issue:46664", "accept": false, - "reason": "Completely different topics: DSA indexing discussion vs video num_frames semantics." + "reason": "DSA/ReLU discussion and SmolVLM video num_frames semantics are unrelated." }, { "left": "issue:44273", "right": "issue:45542", "accept": false, - "reason": "Lazy-loading/import behavior versus TensorFlow backend detection; different failures." + "reason": "Lazy loading/import behavior is not the same as the TensorFlow backend error from tensorboard-only installs." }, { "left": "issue:44038", "right": "issue:45125", "accept": false, - "reason": "Different Qwen3-VL/MoE issues: one is a general bug report, the other is missing tensor-parallel plan." + "reason": "Both mention Qwen3.5 MoE, but one is a general bug/support report and the other is a tensor-parallel _tp_plan issue." }, { "left": "issue:44273", "right": "issue:45003", "accept": false, - "reason": "Both touch import/lazy loading, but one is a generic lazy-loading failure and the other is unsafe sys.modules access; not clearly the same bug." + "reason": "Lazy loading malfunction and unsafe sys.modules access are different root causes." }, { "left": "issue:44393", "right": "issue:45753", "accept": false, - "reason": "Both are Qwen-related vision bugs, but one is 2D bbox output errors and the other is an infinite loop on a specific image." + "reason": "Qwen3-VL 2D bbox hallucination is not the same as an infinite loop on a specific image." }, { "left": "issue:44393", "right": "issue:45071", "accept": false, - "reason": "Different layers of failure: model output hallucination/error versus PretrainedConfig type-checking regression." + "reason": "A model-specific vision bug and a PretrainedConfig type-checking regression are unrelated." }, { "left": "issue:45381", "right": "issue:46664", "accept": false, - "reason": "Both involve video processing, but one is wrong vision_position_ids and the other is inconsistent num_frames semantics." + "reason": "Both are video-related, but qwen2.5-vl vision_position_ids and SmolVLM num_frames semantics are separate issues." }, { "left": "issue:43576", "right": "issue:44556", "accept": false, - "reason": "One is the `transformers env` command; the other is checkpoint reload compatibility after upgrade." + "reason": "Broken env command and checkpoint reload incompatibility are unrelated." }, { "left": "issue:41669", "right": "issue:43673", "accept": false, - "reason": "Import-time performance from star imports is not the same as the v5 chunked_prefill cache bug." + "reason": "Import-star slowdown is a different problem from cache missing during chunked prefill." }, { "left": "issue:43525", "right": "issue:45850", "accept": false, - "reason": "Specific missing pad_token_id on Llama4Config versus a broad v5 regressions umbrella." + "reason": "Missing pad_token_id in Llama4Config is not the same as a broad v5 breakage complaint." }, { "left": "issue:43572", "right": "issue:45850", "accept": false, - "reason": "Specific StableLmConfig pad_token_idx regression versus a generic broad complaint." + "reason": "StableLmConfig pad_token_idx regression is distinct from generic version breakage." }, { "left": "issue:44230", "right": "issue:46095", "accept": false, - "reason": "FP8 inference support request versus a weight conversion question; not a duplicate bug." + "reason": "FP8 support request for Qwen3-VL/Qwen3.5 MoE is unrelated to a DeepSeekV4 weight-conversion question." }, { "left": "issue:44556", "right": "issue:45092", "accept": false, - "reason": "Both are upgrade/reload compatibility problems, but one is generic checkpoint reload failure and the other is remote-code meta-init incompatibility." + "reason": "Both concern loading older checkpoints after v5 upgrades, but the concrete failure modes are different." }, { "left": "issue:45278", "right": "issue:45850", "accept": false, - "reason": "General import errors after upgrade versus a broad 'many models break' complaint; related theme but not same issue." + "reason": "Many import errors after upgrade is a broad complaint, not the same bug as a generic \"too many models break\" issue." }, { "left": "issue:44492", "right": "issue:45850", "accept": false, - "reason": "A typo in cache strategy docs/API is not the same as the broad v5 regression umbrella." + "reason": "A cache-strategy typo is not the same as a general version-breakage report." }, { "left": "issue:42371", "right": "issue:44492", "accept": false, - "reason": "TF32 API guidance versus cache-strategy typo; unrelated." + "reason": "TF32 API settings guidance and cache-strategy typo are unrelated." }, { "left": "issue:41669", "right": "issue:46218", "accept": false, - "reason": "Both are import-related, but one is star-import bloat and the other is missing Qwen3_5 module export." + "reason": "Import-star slowdown and Qwen3_5 module import failure are different failures." }, { "left": "issue:43408", "right": "issue:44315", "accept": false, - "reason": "Sam3 tracker class mismatch versus Liger Kernel not applied with model_init; unrelated." + "reason": "sam3_video/sam3_tracker warning and Liger Kernel application are unrelated." }, { "left": "issue:42890", "right": "issue:44908", "accept": false, - "reason": "Missing set_seed in tests versus scheduler kwargs propagation; different bugs." + "reason": "Missing set_seed in tests is unrelated to inverse_sqrt scheduler kwargs handling." }, { "left": "issue:44246", "right": "issue:44273", "accept": false, - "reason": "Import slowdown and lazy-loading failure may be adjacent, but the reports are not specific enough to confirm the same defect." + "reason": "Both touch imports/lazy loading, but one is intermittent import slowness and the other is a general lazy-loading bug; too broad to merge." }, { "left": "issue:43352", "right": "issue:44393", "accept": false, - "reason": "Flash Attention 2 support limitation versus Qwen3-VL 2D bbox output issue; unrelated." + "reason": "Nemotron Flash Attention 2 support and Qwen3-VL bbox hallucination are unrelated." }, { "left": "issue:43576", "right": "issue:44908", "accept": false, - "reason": "`transformers env` command versus scheduler kwargs bug; unrelated." + "reason": "transformers env command and scheduler kwargs are unrelated." }, { "left": "issue:43976", "right": "issue:44393", "accept": false, - "reason": "Python version support regression versus a Qwen3-VL output bug; unrelated." + "reason": "Python version compatibility and Qwen3-VL bbox output are unrelated." }, { "left": "issue:44704", "right": "issue:44908", "accept": false, - "reason": "Processor kwargs forwarding bug versus learning-rate scheduler kwargs bug; different code paths." + "reason": "AutoProcessor kwargs passing and inverse_sqrt scheduler kwargs are unrelated." }, { "left": "issue:44908", "right": "issue:45071", "accept": false, - "reason": "Scheduler kwargs propagation versus config type-checking regression; unrelated." + "reason": "Scheduler kwargs handling and PretrainedConfig type checking are different bugs." }, { "left": "issue:44295", "right": "issue:45230", "accept": false, - "reason": "Position_ids buffer access error versus a generic bug report; no duplicate evidence." + "reason": "position_ids buffer access error is unrelated to a generic bug report." }, { "left": "issue:44297", "right": "issue:45230", "accept": false, - "reason": "Tokenizer class mismatch on save versus a generic bug report; no match." - }, - { - "left": "issue:44315", - "right": "issue:44829", - "accept": false, - "reason": "Liger Kernel application issue versus flash_attention_3 training degeneration; different symptoms and code paths." + "reason": "tokenizer_class mismatch on save is unrelated to a generic bug report." }, { - "left": "issue:44351", - "right": "issue:45850", - "accept": false, - "reason": "Missing HybridCache export versus a broad v5 breakage complaint." - }, - { - "left": "issue:44360", - "right": "issue:46664", + "left": "issue:41669", + "right": "issue:44351", "accept": false, - "reason": "DSA discussion versus video processor frame-count semantics; unrelated." + "reason": "Import-star slowdown and missing HybridCache import are different import-path problems." }, { - "left": "issue:44908", - "right": "issue:45071", + "left": "issue:44492", + "right": "issue:46328", "accept": false, - "reason": "Scheduler kwargs bug versus PretrainedConfig type checking; unrelated." + "reason": "Cache-strategy typo and SequenceFeatureExtractor.pad performance bug are unrelated." }, { - "left": "issue:45003", + "left": "issue:41669", "right": "issue:45341", "accept": false, - "reason": "Unsafe sys.modules access in modeling_utils versus a small testing_utils bug; unrelated." + "reason": "Model import-star slowdown and a small testing_utils bug are unrelated." }, { - "left": "issue:45092", + "left": "issue:43976", "right": "issue:45850", "accept": false, - "reason": "Specific remote-code/meta-init incompatibility versus a generic broad regression umbrella." + "reason": "Python-version support regression and generic v5 breakage are not the same issue." }, { - "left": "issue:45125", + "left": "issue:44393", "right": "issue:45478", "accept": false, - "reason": "Both are Qwen3.5 MoE-related, but one is missing `_tp_plan` and the other is a generic from_pretrained error; not enough to treat as the same bug." - }, - { - "left": "issue:45278", - "right": "issue:45850", - "accept": false, - "reason": "Generic import errors after upgrade and broad v5 breakage are related in theme but not the same issue." + "reason": "Qwen3-VL bbox hallucination and Qwen3.5 MoE from_pretrained failure involve different models and code paths." }, { - "left": "issue:45341", - "right": "issue:41669", - "accept": false, - "reason": "Testing utility bug versus import star-usage performance issue; unrelated." - }, - { - "left": "issue:45381", - "right": "issue:46664", - "accept": false, - "reason": "Both are video-related, but the concrete bugs differ: vision_position_ids vs num_frames meaning." - }, - { - "left": "issue:45478", - "right": "issue:45850", + "left": "issue:43976", + "right": "issue:46218", "accept": false, - "reason": "Qwen3.5 MoE from_pretrained failure is a specific model bug, not the same as the broad v5 complaint." + "reason": "Python support regression and Qwen3_5 import failure are unrelated." }, { - "left": "issue:45542", - "right": "issue:45753", + "left": "issue:44077", + "right": "issue:44861", "accept": false, - "reason": "TensorFlow backend import error versus Qwen3_5 image-specific loop; unrelated." + "reason": "Optional post_init policy and tied-weights-key AttributeError are different code paths." }, { - "left": "issue:45753", - "right": "issue:45850", + "left": "issue:43827", + "right": "issue:44908", "accept": false, - "reason": "Specific infinite-loop image case versus a broad v5 regression umbrella." + "reason": "Docs still referencing pipeline() and scheduler kwargs handling are unrelated." }, { - "left": "issue:46095", - "right": "issue:46218", + "left": "issue:31795", + "right": "issue:33453", "accept": false, - "reason": "Weight conversion question versus module import failure; unrelated." + "reason": "Documentation confusion and tokenizer loading regression are unrelated." }, { - "left": "issue:46328", - "right": "issue:44492", + "left": "issue:44315", + "right": "issue:44829", "accept": false, - "reason": "SequenceFeatureExtractor.pad performance bug versus cache-strategy typo; unrelated." + "reason": "Liger Kernel not applied with model_init and flash_attention_3 degenerate training are unrelated." }, { - "left": "issue:46664", - "right": "issue:45850", + "left": "issue:43576", + "right": "issue:44273", "accept": false, - "reason": "Video processor num_frames inconsistency is specific and not the same as the broad v5 breakage report." + "reason": "Broken env command and lazy-loading malfunction are only broadly related to tooling, not duplicate bugs." } ] }, @@ -17386,286 +17231,262 @@ "issue:31515|issue:33453", "issue:43298|issue:44230" ], - "prepared_review_unit_hash": "e141f4403daaaf2627267e747425096fb48cf0349b71e2c118b2e64472c0c826", + "prepared_review_unit_hash": "acf7e1d6e2e52ec0a26e21eed2e16a5ae5ea6bd5a9a4f8754f7266293a6a377e", "component_budget": { "node_count": 473, "item_count": 473, "soft_pair_count": 1738, - "serialized_chars": 901383, - "estimated_input_tokens": 225346, - "estimated_eval_tokens": 450948 + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 43, - "serialized_chars": 39011, - "estimated_input_tokens": 9753, - "estimated_eval_tokens": 19762 + "serialized_chars": 39014, + "estimated_input_tokens": 9754, + "estimated_eval_tokens": 19764 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "These items are mostly unrelated issue reports from different models, commands, and code paths. The soft-similarity links look like false positives, so none should be merged as duplicates.", - "confidence": 0.92, - "canonical_issue_reason": "No single canonical issue stands out: the set is heterogeneous, and the soft-similarity pairs do not indicate the same underlying bug.", + "summary": "This set is not a true duplicate cluster; it mixes many unrelated Transformers issues across model loading, tokenizers, processors, configs, scheduling, and feature requests. No soft pair looks like the same underlying bug/change, so all are rejected.", + "confidence": 0.96, + "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "issue:43519 is the closest thing to a representative anchor because it is a concrete, well-scoped bug report, but it is not a true duplicate hub.", + "best_issue_reason": null, "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43519", "right": "issue:46655", "accept": false, - "reason": "Both are processor-related, but they describe different bugs in different models: timestamp math vs video padding." + "reason": "Both are processor calculation bugs, but they affect different models and different logic (timestamp vs padding)." }, { "left": "issue:44360", "right": "issue:45468", "accept": false, - "reason": "DSA indexer ReLU and Gemma4 audio positional encoding are unrelated subsystems and failures." + "reason": "Different models and different defects; no shared code-path bug is evident." }, { "left": "issue:44655", "right": "issue:44861", "accept": false, - "reason": "Save-pretrained file loss and tied-weight key crashes are different code paths." + "reason": "Both touch save/load internals, but one is pipeline save_pretrained and the other is tied-weight key handling." }, { "left": "issue:44273", "right": "issue:44908", "accept": false, - "reason": "Lazy loading and inverse_sqrt scheduler kwargs are unrelated issues." + "reason": "Lazy loading and scheduler kwargs are unrelated subsystems." }, { "left": "issue:45230", "right": "issue:45310", "accept": false, - "reason": "A generic bug report is not the same as a specific Qwen3.5 from_pretrained failure." + "reason": "Generic bug report versus a specific Qwen3.5 MOE from_pretrained regression; not the same issue." }, { "left": "issue:43576", "right": "issue:46218", "accept": false, - "reason": "Broken env command and missing Qwen3_5 import are different API surface failures." + "reason": "Broken env command and Qwen import failure are unrelated." }, { "left": "issue:43519", "right": "issue:44485", "accept": false, - "reason": "Timestamp calculation in Qwen3VL and GLM-5 RoPE implementation are different bugs." + "reason": "Different model processors; one is timestamp math, the other is RoPE implementation." }, { "left": "issue:43519", "right": "issue:44908", "accept": false, - "reason": "Processor timestamp logic and scheduler kwargs handling do not overlap." + "reason": "Processor timestamp bug and scheduler kwargs bug are unrelated." }, { "left": "issue:41669", "right": "issue:43704", "accept": false, - "reason": "Import-star performance cleanup and a VRAM leak in Qwen3ForCausalLM are unrelated." + "reason": "Import-star cleanup and VRAM leak are different code changes/problems." }, { "left": "issue:46018", "right": "issue:46172", "accept": false, - "reason": "DeepSeek-V4 gating and a tensor transpose/dtype issue are different defects." + "reason": "Both are DeepSeek-V4 related, but they describe different concrete bugs and code paths." }, { "left": "issue:31795", "right": "issue:34689", "accept": false, - "reason": "Documentation confusion about forward args is not the same as a model-loading regression." + "reason": "Documentation confusion and model loading regression are not the same underlying bug." }, { "left": "issue:31515", "right": "issue:31795", "accept": false, - "reason": "Slow checkpoint loading and confusing forward docs are different problems." + "reason": "Slow checkpoint loading and forward-doc confusion are unrelated." }, { "left": "issue:43704", "right": "issue:44485", "accept": false, - "reason": "VRAM leakage and RoPE implementation are unrelated bug classes." + "reason": "VRAM leak in Qwen3 differs from GLM-5 RoPE implementation bug." }, { "left": "issue:44485", "right": "issue:45468", "accept": false, - "reason": "These are both position/encoding themed, but they target different models and different failures." + "reason": "Different model families and different failure modes." }, { "left": "issue:45850", "right": "issue:46218", "accept": false, - "reason": "A broad 'v5 breaks models' report and a specific import error are not the same bug." + "reason": "Broad v5 breakage report is not the same as a specific Qwen import error." }, { "left": "issue:43502", "right": "issue:43519", "accept": false, - "reason": "Local-files-only network leakage and timestamp calculation are unrelated." + "reason": "Local-files-only network leak and timestamp calculation are unrelated." }, { "left": "issue:42371", "right": "issue:43704", "accept": false, - "reason": "TF32 configuration guidance and a VRAM leak do not share a concrete code path." + "reason": "TF32 settings guidance and VRAM leak are different issues." }, { "left": "issue:43519", "right": "issue:44623", "accept": false, - "reason": "Processor timestamp math and missing files in save_pretrained are different issues." + "reason": "Processor timestamp math and processor.save_pretrained missing files are different bugs." }, { "left": "issue:44263", "right": "issue:45468", "accept": false, - "reason": "torch.split return values and audio positional encoding are unrelated implementation bugs." + "reason": "Torch.split indexing bug and Gemma audio positional encoding bug are unrelated." }, { "left": "issue:43606", "right": "issue:45468", "accept": false, - "reason": "CPU offload device mismatch is not the same as Gemma4 audio positional encoding." + "reason": "CPU offload device mismatch is unrelated to Gemma audio encoding." }, { "left": "issue:43298", "right": "issue:46095", "accept": false, - "reason": "Continuous batching for audio models and a DeepSeek-V4 weight conversion question are different topics." + "reason": "Continuous batching audio support and DeepSeek-V4 weight conversion are different topics." }, { "left": "issue:44261", "right": "issue:45468", "accept": false, - "reason": "Missing rms_norm_eps in MLA layernorm and Gemma4 audio positional encoding are unrelated." + "reason": "RMS norm epsilon precision issue and Gemma audio positional encoding are unrelated." + }, + { + "left": "issue:43519", + "right": "issue:43704", + "accept": false, + "reason": "Processor timestamp bug and Qwen VRAM leak do not share the same code-path." }, { "left": "issue:44351", "right": "issue:46218", "accept": false, - "reason": "Both are import errors, but they involve different missing symbols and likely different fixes." + "reason": "Both are import failures, but for different modules and causes." }, { "left": "issue:43352", "right": "issue:46218", "accept": false, - "reason": "Flash Attention support for NemotronH and a Qwen3_5 import failure are unrelated." + "reason": "Flash-attention support gap and Qwen import failure are unrelated." }, { "left": "issue:42913", "right": "issue:43066", "accept": false, - "reason": "Both concern tokenizer v5 behavior, but one is a regression report and the other is a decoder-type bug." + "reason": "Both concern tokenizer v5 behavior, but the titles point to different concrete failures." }, { "left": "issue:28348", "right": "issue:45250", "accept": false, - "reason": "Specific GPT2 Flash Attention support and a generic Flash Attention 2.0 request are not the same issue." + "reason": "Specific GPT2 support request versus a generic Flash Attention 2.0 issue; not the same change." }, { "left": "issue:44908", "right": "issue:44998", "accept": false, - "reason": "Scheduler kwargs handling and the placeholder 'Unemployment' issue are obviously unrelated." + "reason": "Scheduler kwarg bug and unrelated placeholder issue." }, { "left": "issue:44492", "right": "issue:44998", "accept": false, - "reason": "A cache-strategy typo and an unrelated placeholder issue are not duplicates." + "reason": "Cache strategy typo and unrelated placeholder issue." }, { "left": "issue:43618", "right": "issue:44998", "accept": false, - "reason": "CLIPOutput attentions regression has nothing in common with the placeholder issue." + "reason": "CLIPOutput attentions regression is unrelated to the placeholder issue." }, { "left": "issue:44998", "right": "issue:45850", "accept": false, - "reason": "The placeholder issue does not match the broad v5 breakage report." + "reason": "Placeholder issue versus broad v5 breakage report; no duplicate relation." }, { "left": "issue:44938", "right": "issue:44998", "accept": false, - "reason": "Python 3.14 loading failure and the placeholder issue are unrelated." + "reason": "Python 3.14 load failure and unrelated placeholder issue." }, { "left": "issue:44861", "right": "issue:45290", "accept": false, - "reason": "Tied-weight key crashes and chat-template tool-call crashes are different code paths." + "reason": "Two different crash sites: tied-weights handling versus chat template/tool-call handling." }, { "left": "issue:43525", "right": "issue:44861", "accept": false, - "reason": "Missing pad_token_id on Llama4Config is unrelated to tied-weight key handling." + "reason": "Missing pad_token_id on Llama4Config and tied-weights list crash are distinct bugs." }, { "left": "issue:31515", "right": "issue:38617", "accept": false, - "reason": "Slow from_pretrained loading and a missing import symbol are unrelated problems." + "reason": "Slow from_pretrained loading and layer_type_validation import error are different failures." }, { "left": "issue:43572", "right": "issue:44861", "accept": false, - "reason": "Missing pad_token_idx in StableLmConfig is not the same as tied-weight key crashes." + "reason": "StableLm pad token config regression and tied-weights crash are unrelated." }, { "left": "issue:44230", "right": "issue:44623", "accept": false, - "reason": "FP8 support for Qwen3-VL and missing processor save files are different features/issues." + "reason": "FP8 support request and missing processor save files are not the same issue." }, { "left": "issue:28348", "right": "issue:43352", "accept": false, - "reason": "Both mention Flash Attention 2.0, but they concern different models and separate support requests." - }, - { - "left": "issue:30333", - "right": "issue:31515", - "accept": false, - "reason": "MLflow reporting on failed jobs and checkpoint load slowness are unrelated." - }, - { - "left": "issue:31515", - "right": "issue:36296", - "accept": false, - "reason": "Checkpoint loading performance and tensor-parallel training bugs are different failures." - }, - { - "left": "issue:44230", - "right": "issue:45310", - "accept": false, - "reason": "FP8 support for Qwen3-VL and a Qwen3.5 MoE from_pretrained error are not the same bug." - }, - { - "left": "issue:31515", - "right": "issue:33453", - "accept": false, - "reason": "Slow checkpoint loading and tokenizer loading regression are different loading problems." - }, - { - "left": "issue:43298", - "right": "issue:44230", - "accept": false, - "reason": "Continuous batching for audio models and FP8 inference support are unrelated." + "reason": "Both mention Flash Attention 2.0, but they target different models and separate support requests." } ] }, @@ -17713,106 +17534,106 @@ "issue:37934|issue:43668", "issue:42936|issue:43668" ], - "prepared_review_unit_hash": "ba705fc553224067c5e16fcae7e5a07f0f918f3b09f2ee904ea0dd8006c4a432", + "prepared_review_unit_hash": "0abe26ce51b0176b9f4df243a9bbef12bc99d966f3469b311486a37cb88b6fda", "component_budget": { "node_count": 473, "item_count": 473, "soft_pair_count": 1738, - "serialized_chars": 901383, - "estimated_input_tokens": 225346, - "estimated_eval_tokens": 450948 + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 }, "budget": { "node_count": 9, "item_count": 9, "soft_pair_count": 12, - "serialized_chars": 8607, - "estimated_input_tokens": 2152, - "estimated_eval_tokens": 4560 + "serialized_chars": 8613, + "estimated_input_tokens": 2154, + "estimated_eval_tokens": 4564 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is heterogeneous: the issues span unrelated models, configs, warnings, and attention/loss bugs. None of the soft pairs look like the same underlying defect, so all soft edges should be rejected.", - "confidence": 0.86, - "canonical_issue_reason": "Issue 42936 is the most concrete, user-facing bug report with a clear affected model and an actionable symptom, making it a reasonable representative for the cluster even though the cluster itself is not truly duplicate-heavy.", + "summary": "These issues are a loose grab bag of unrelated bug reports across different models, configs, tokenizers, and utilities. None of the soft-similarity pairs looks like the same underlying bug or change, so there is no real duplicate cluster here.", + "confidence": 0.94, + "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "Issue 42936 is the strongest standalone issue candidate: it describes a specific incorrect behavior in Mask2former with enough detail to guide a fix.", + "best_issue_reason": "If a representative issue must be chosen, issue 44031 is the broadest user-facing regression report in the set; the others are narrower, unrelated correctness bugs.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43717", "right": "issue:46144", "accept": false, - "reason": "Different models and code paths: Mamba init/weight handling versus RoFormer attention interface usage. No indication they describe the same bug." + "reason": "Different components and bug classes: Mamba init_weights behavior vs RoFormer attention interface usage." }, { "left": "issue:42936", "right": "issue:46018", "accept": false, - "reason": "Mask2former ignore_value handling is unrelated to DeepSeek-V4 shared expert gating." + "reason": "Mask2Former ignore_value handling is unrelated to DeepSeek-V4 shared expert gating." }, { "left": "issue:43668", "right": "issue:44031", "accept": false, - "reason": "ModernBERTConfig type-hint correctness and tokenizer regex warnings are distinct configuration versus runtime warning issues." + "reason": "A config type-hint problem is not the same as a tokenizer regex warning regression." }, { "left": "issue:42936", "right": "issue:43240", "accept": false, - "reason": "One is a model-specific ignore_value bug; the other is about kwargs not being forwarded to a loss function. Different failure modes and code paths." + "reason": "One is about an unused model field; the other is about kwargs not reaching a loss function." }, { "left": "issue:43240", "right": "issue:46144", "accept": false, - "reason": "Loss-function kwargs forwarding and RoFormer attention interface usage are unrelated implementation bugs." + "reason": "Loss-function kwargs propagation and RoFormer attention interface are distinct code paths." }, { "left": "issue:42936", "right": "issue:44031", "accept": false, - "reason": "Mask2former ignore_value handling does not overlap with tokenizer regex pattern warnings." + "reason": "Mask2Former ignore_value misuse does not match a tokenizer warning regression." }, { "left": "issue:42936", "right": "issue:46144", "accept": false, - "reason": "These concern different architectures and defects; no shared concrete bug." + "reason": "Different model families and different defects; no shared underlying change." }, { "left": "issue:42936", "right": "issue:43813", "accept": false, - "reason": "Mask2former ignore_value and a typo in quantization_operation are unrelated." + "reason": "Unused ignore_value field and a quantization-typo string are unrelated." }, { "left": "issue:43240", "right": "issue:43668", "accept": false, - "reason": "Passing kwargs to a loss function is unrelated to an incorrect type hint in a config class." + "reason": "Passing kwargs to loss calculation is unrelated to an incorrect type annotation." }, { "left": "issue:43668", "right": "issue:43813", "accept": false, - "reason": "A type-hint error in ModernBERTConfig is not the same as a typo in quantization operation naming." + "reason": "Type hint correction vs typo in quantization operation name are not the same bug." }, { "left": "issue:37934", "right": "issue:43668", "accept": false, - "reason": "RMSNorm semantics in Llama4TextL2Norm and a ModernBERT norm_eps type hint are different issues." + "reason": "RMS-norm semantics question for Llama4TextL2Norm is unrelated to ModernBERTConfig norm_eps typing." }, { "left": "issue:42936", "right": "issue:43668", "accept": false, - "reason": "These are unrelated: model ignore_value usage versus config type-hint correctness." + "reason": "Mask2Former ignore_value usage and ModernBERT norm_eps typing concern different subsystems and fixes." } ] }, @@ -17845,52 +17666,52 @@ "issue:10519|issue:11005", "issue:11005|issue:11307" ], - "prepared_review_unit_hash": "ddf203159f6fe60295b13bf7b09dc717dd034adbfd9b7b8d9160d1884985558f", + "prepared_review_unit_hash": "49f022f3accb9978b93e355050abc8a4c23288524c26015e82980cef2c718b1e", "component_budget": { "node_count": 3, "item_count": 3, "soft_pair_count": 3, - "serialized_chars": 2877, - "estimated_input_tokens": 720, - "estimated_eval_tokens": 1696 + "serialized_chars": 2876, + "estimated_input_tokens": 719, + "estimated_eval_tokens": 1694 }, "budget": { "node_count": 3, "item_count": 3, "soft_pair_count": 3, - "serialized_chars": 2877, - "estimated_input_tokens": 720, - "estimated_eval_tokens": 1696 + "serialized_chars": 2876, + "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 clearly unrelated: one is about tokenizer truncation behavior, one asks for ReduceLROnPlateau-like scheduler functionality, and one concerns Wav2Vec2 word boundary timestamps. No duplicate cluster is present.", - "confidence": 0.99, - "canonical_issue_reason": "No canonical issue: the topics are different subsystems and different user problems, so any one issue would be a poor representative.", + "summary": "These three issues are not duplicates: they cover unrelated topics (tokenizer truncation behavior, a ReduceLROnPlateau-like training feature, and Wav2Vec2 word time offsets).", + "confidence": 0.98, + "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "None of these is a good global representative because the cluster is heterogeneous rather than a duplicate set.", + "best_issue_reason": "No issue is a good cluster representative because the issues are about different features/problems in different parts of the library.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:10519", "right": "issue:11307", "accept": false, - "reason": "Tokenizer truncation options and Wav2Vec2 word timestamps are unrelated features with no shared bug or code path." + "reason": "Different domains: truncation strategy for tokenization vs. Wav2Vec2 word timestamp extraction." }, { "left": "issue:10519", "right": "issue:11005", "accept": false, - "reason": "Truncation strategy and learning-rate scheduler behavior are different areas of the library and do not describe the same underlying issue." + "reason": "No overlap in underlying bug or feature request; one is about input truncation, the other about learning-rate scheduling behavior." }, { "left": "issue:11005", "right": "issue:11307", "accept": false, - "reason": "A scheduler feature request and Wav2Vec2 time-offset extraction are unrelated; they are not plausible duplicates." + "reason": "Unrelated requests: ReduceLROnPlateau-like functionality vs. audio model time offsets." } ] }, @@ -17922,12 +17743,12 @@ "issue:13244|issue:16618", "issue:15980|issue:16618" ], - "prepared_review_unit_hash": "e3afaa49fbf7141006770e9e74e1c7df14c0e97ab7d09300849980210873be3c", + "prepared_review_unit_hash": "92ad73f507ae42e1ab92a3660aef5c824e898d022e8cfe8a0fb61c0b544eaf59", "component_budget": { "node_count": 3, "item_count": 3, "soft_pair_count": 2, - "serialized_chars": 2479, + "serialized_chars": 2478, "estimated_input_tokens": 620, "estimated_eval_tokens": 1496 }, @@ -17935,7 +17756,7 @@ "node_count": 3, "item_count": 3, "soft_pair_count": 2, - "serialized_chars": 2479, + "serialized_chars": 2478, "estimated_input_tokens": 620, "estimated_eval_tokens": 1496 }, @@ -17944,24 +17765,24 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "The cluster appears to mix unrelated issues: one about TAPAS tokenization parity and two about model-loading / download failures. The two model-loading issues are similar at a high level, but the available titles are not enough to confirm they are the same underlying bug.", - "confidence": 0.83, - "canonical_issue_reason": "issue:16618 is the broadest and most representative of the model-loading failure side of the cluster, with a user-facing title that likely captures the common symptom.", + "summary": "The cluster appears to contain three distinct issues: one about Tapas tokenization mismatch, and two about model-loading/download failures that are not clearly the same bug. The model-loading issues share a broad theme, but the available titles point to different scenarios (private-model auth vs. a public model load failure), so I would not merge them as duplicates.", + "confidence": 0.46, + "canonical_issue_reason": "Issue 16618 is the broadest user-facing model-loading report in the set, so it is the best representative if one issue must be chosen.", "canonical_pr_reason": null, - "best_issue_reason": "issue:16618 is the best overall issue to keep because it is the clearest, most generic representation of the likely shared model-loading problem; issue:13244 is clearly a separate tokenization bug.", + "best_issue_reason": "16618 is the most general of the issues and the closest to a core runtime failure, while 13244 is clearly unrelated and 15980 is a narrower auth/error-message case.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:13244", "right": "issue:16618", "accept": false, - "reason": "Different problem domains: TAPAS tokenization parity versus a model load failure. The titles do not indicate the same bug or a mergeable shared fix." + "reason": "Tapas tokenization mismatch is unrelated to failing to load bert-base-uncased; they describe different subsystems and different failure modes." }, { "left": "issue:15980", "right": "issue:16618", "accept": false, - "reason": "Both involve model loading/downloads, but one is specifically about a private-model auth error while the other is a generic 'can't load bert-base-uncased' report. Not enough evidence they are the same underlying bug." + "reason": "15980 is specifically about the error message when accessing a private model without login, while 16618 is a public-model load failure; the underlying bug is not clearly the same." } ] }, @@ -17991,7 +17812,7 @@ "soft_pairs": [ "issue:26350|issue:33260" ], - "prepared_review_unit_hash": "d0cbe2514e12d7d7e2314508b12103695db092447d7904b281d103a09110e5d7", + "prepared_review_unit_hash": "2fdf89b6782c445044a49c54c955a694160d42ae2a79d9c9ff9dd0f67312693b", "component_budget": { "node_count": 2, "item_count": 2, @@ -18013,18 +17834,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These are two unrelated community-contribution feature requests: one for Flash Attention 2 architecture support and one for GGUF architecture support. They share only a broad \"more architectures\" theme, not the same underlying bug or change.", + "summary": "The two issues request unrelated community contributions: Flash Attention 2 support for more architectures vs GGUF support for more architectures. They share a broad theme (architecture support) but not the same underlying feature or bug.", "confidence": 0.98, - "canonical_issue_reason": "No clear duplicate/canonical relationship: the issues request different features for different subsystems and architecture support paths.", + "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "Neither issue is a global canonical for the other; if forced to pick a representative, either would only represent its own feature area, not the cluster.", + "best_issue_reason": null, "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:26350", "right": "issue:33260", "accept": false, - "reason": "Reject: they concern different enhancements in different code paths (Flash Attention 2 vs GGUF). Similar wording about architectures is too broad to indicate the same underlying change." + "reason": "Different requested capabilities for different backends/code paths; this is not the same underlying issue and should not be merged as duplicates." } ] }, @@ -18083,184 +17904,184 @@ "issue:44370|issue:46226", "issue:46222|issue:46226" ], - "prepared_review_unit_hash": "826d98b6a886234a2c92d1c01b8987dd361312eed67fbcea6ab57631d30fc55c", + "prepared_review_unit_hash": "8949938030c2c6a074259a1bcff9ec2da2cdb35774889e731348bd9451eccf41", "component_budget": { "node_count": 8, "item_count": 8, "soft_pair_count": 24, - "serialized_chars": 13530, - "estimated_input_tokens": 3383, - "estimated_eval_tokens": 7022 + "serialized_chars": 13526, + "estimated_input_tokens": 3382, + "estimated_eval_tokens": 7020 }, "budget": { "node_count": 8, "item_count": 8, "soft_pair_count": 24, - "serialized_chars": 13530, - "estimated_input_tokens": 3383, - "estimated_eval_tokens": 7022 + "serialized_chars": 13526, + "estimated_input_tokens": 3382, + "estimated_eval_tokens": 7020 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "This cluster is mostly i18n docs-translation requests that look similar because they share the same boilerplate, but they are not the same underlying change when the target language differs. I only consider the two identical Thai-titled issues a likely duplicate pair.", - "confidence": 0.79, - "canonical_issue_reason": null, + "summary": "Most similarities are boilerplate from the same translation-request template, not true duplicates. I\u2019d only treat the exact same-title twins as duplicate-worthy; the different language requests are separate issues.", + "confidence": 0.82, + "canonical_issue_reason": "issue:43834 is the best anchor for the repeated placeholder-titled translation requests: it is the earliest of the identical generic items and sits in the densest duplicate-like subgroup.", "canonical_pr_reason": null, - "best_issue_reason": "issue:43834 is the most central representative of the generic i18n template subgroup, but the cluster overall is mixed across different language requests so it should not be treated as a true canonical duplicate.", + "best_issue_reason": "issue:43834 is the strongest overall issue representative for this cluster because it is central to the repeated template-style items and is the cleanest canonical anchor among the issue set.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:27088", "right": "issue:44370", "accept": false, - "reason": "Different translation requests; shared boilerplate is not enough to make them the same issue." + "reason": "Different language targets (Turkish vs a separate translation request); the overlap is just the shared translation template." }, { "left": "issue:27088", "right": "issue:44365", "accept": false, - "reason": "Different translation requests; not the same target language or concrete change." + "reason": "Same boilerplate theme, but not the same concrete translation request." }, { "left": "issue:27088", "right": "issue:43834", "accept": false, - "reason": "Looks like the same i18n template family, but not the same underlying language-specific request." + "reason": "Turkish translation request is distinct from the generic placeholder-titled request." }, { "left": "issue:27088", "right": "issue:38435", "accept": false, - "reason": "Turkish vs Romanian translation requests are distinct." + "reason": "Turkish and Romanian are separate translation tasks, not one underlying issue." }, { "left": "issue:27088", "right": "issue:28936", "accept": false, - "reason": "Turkish vs Spanish translation requests are distinct." + "reason": "Different target languages, so these are not duplicates." }, { "left": "issue:28936", "right": "issue:43834", "accept": false, - "reason": "Different language request; similarity comes from shared template text." + "reason": "Spanish request vs generic placeholder title; too broad to be the same issue." }, { "left": "issue:28936", "right": "issue:44365", "accept": false, - "reason": "Different language request; not the same underlying issue." + "reason": "Shared template language only; the translation target is different or unspecified." }, { "left": "issue:28936", "right": "issue:38435", "accept": false, - "reason": "Spanish vs Romanian translation requests are distinct." + "reason": "Spanish and Romanian translation requests are distinct." }, { "left": "issue:28936", "right": "issue:44370", "accept": false, - "reason": "Different language request; too much boilerplate overlap." + "reason": "Different concrete translation request, despite similar boilerplate." }, { "left": "issue:38435", "right": "issue:44370", "accept": false, - "reason": "Romanian request and generic template issue are not clearly the same concrete change." + "reason": "Romanian request is separate from the generic placeholder issue." }, { "left": "issue:38435", "right": "issue:44365", "accept": false, - "reason": "Romanian request and generic template issue are not the same underlying issue." + "reason": "They look templated, but not clearly the same underlying translation task." }, { "left": "issue:38435", "right": "issue:43834", "accept": false, - "reason": "High text similarity appears to be template reuse, not the same language request." + "reason": "Different language target; similarity is only from shared issue template." }, { "left": "issue:38435", "right": "issue:46222", "accept": false, - "reason": "Different language/request content; not the same issue." + "reason": "Romanian translation request and the Thai-titled issue are not the same concrete issue." }, { "left": "issue:38435", "right": "issue:46226", "accept": false, - "reason": "Different language/request content; not the same issue." + "reason": "Same boilerplate family, but no evidence they are the same request." }, { "left": "issue:43834", "right": "issue:44370", - "accept": false, - "reason": "Near-identical boilerplate, but no evidence they target the same language or change." + "accept": true, + "reason": "Same exact generic title and near-identical body shape; this looks like a duplicate translation-request issue." }, { "left": "issue:43834", "right": "issue:44365", - "accept": false, - "reason": "Near-identical boilerplate, but the requests are not confirmed identical." + "accept": true, + "reason": "Same exact generic title and highly similar content; likely duplicate reports of the same issue." + }, + { + "left": "issue:44365", + "right": "issue:44370", + "accept": true, + "reason": "Same exact placeholder title and near-identical body; strong duplicate signal." }, { "left": "issue:43834", "right": "issue:46222", "accept": false, - "reason": "Different request text; similarity is too generic to merge." + "reason": "The titles do not match and one is a generic placeholder while the other is a concrete Thai-titled request." }, { "left": "issue:43834", "right": "issue:46226", "accept": false, - "reason": "Different request text; not enough to conclude same underlying issue." - }, - { - "left": "issue:44365", - "right": "issue:44370", - "accept": false, - "reason": "Same template family, but not proven to be the same specific translation request." + "reason": "Different titles and no clear evidence of the same underlying translation request." }, { "left": "issue:44365", "right": "issue:46222", "accept": false, - "reason": "Different request text; no clear duplicate evidence." + "reason": "Not the same concrete issue; the match is only superficial template similarity." }, { "left": "issue:44365", "right": "issue:46226", "accept": false, - "reason": "Different request text; no clear duplicate evidence." + "reason": "Different request titles and no explicit common target." }, { "left": "issue:44370", "right": "issue:46222", "accept": false, - "reason": "Different request text; no clear duplicate evidence." + "reason": "Shared boilerplate, but not the same underlying issue." }, { "left": "issue:44370", "right": "issue:46226", "accept": false, - "reason": "Different request text; no clear duplicate evidence." + "reason": "Different titles and no concrete duplicate evidence." }, { "left": "issue:46222", "right": "issue:46226", "accept": true, - "reason": "Same Thai-titled request with identical body length and near-identical metadata, so these look like the same issue duplicated." + "reason": "Same exact Thai title and matching body shape; these look like duplicate copies of the same issue." } ] }, "evaluator_result": { "accept": true, - "feedback": "The analysis is grounded in the packet and stays conservative. It correctly treats the Turkish/Spanish/Romanian/template issues as similar boilerplate but not confirmed duplicates, and the Thai pair is the only one with direct strong evidence of duplication (identical title, identical body length, and perfect soft similarity)." + "feedback": "The summary is grounded in the packet and stays conservative: it treats the identical-title pairs as the only duplicate-worthy matches, while rejecting cross-language/template similarities as mere boilerplate. The canonical anchor choice for issue:43834 is supported by the shared title pattern and creation order among the generic placeholder items." }, "evaluator_used": true, "retried": false, @@ -18342,256 +18163,274 @@ "issue:42503|issue:46653", "issue:45027|issue:46653" ], - "prepared_review_unit_hash": "47b90230f3ba12d7ddd337df15b2b15bf09498dd7baccad0eba5071dc4fdea78", + "prepared_review_unit_hash": "db4056249a32a337fbb3fae9a034c405a9cf3f05fceeec4d20ad204617e3eb16", "component_budget": { "node_count": 18, "item_count": 18, "soft_pair_count": 40, - "serialized_chars": 24150, - "estimated_input_tokens": 6038, - "estimated_eval_tokens": 12332 + "serialized_chars": 24144, + "estimated_input_tokens": 6036, + "estimated_eval_tokens": 12328 }, "budget": { "node_count": 18, "item_count": 18, "soft_pair_count": 40, - "serialized_chars": 24150, - "estimated_input_tokens": 6038, - "estimated_eval_tokens": 12332 + "serialized_chars": 24144, + "estimated_input_tokens": 6036, + "estimated_eval_tokens": 12328 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "This cluster is mostly a loose set of model-request issues with some thematic overlap (BERT-family models, audio/ASR/TTS models, and a few vision-language models), but they do not describe the same concrete change. I would not merge any of the soft edges as duplicates.", - "confidence": 0.74, - "canonical_issue_reason": "Issue 42738 is the broadest umbrella-style request in the set (generic BERT-like models with RoPE) and is the closest representative of the recurring BERT-family subset, though the overall cluster is heterogeneous.", + "summary": "This cluster is dominated by unrelated model-add/support requests that only share generic wording. They span different model families and modalities (text, vision, audio, ASR/TTS, OCR, architecture tweaks), so none of the soft edges look like true duplicates.", + "confidence": 0.35, + "canonical_issue_reason": "Issue 42738 is the broadest and most umbrella-like request in the set ('BERT-like models with RoPE'), so it is the best single representative of this mostly model-request cluster, though it should not subsume the other specific requests as duplicates.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 42738 is the best single representative issue because it is the most general model-request thread among otherwise disparate model additions; it is not a true duplicate hub, just the least specific item.", + "best_issue_reason": "Issue 42738 is the best representative of the cluster because it is the most general feature request among otherwise model-specific asks.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:36817", "right": "issue:41211", "accept": false, - "reason": "Different model requests: EuroBert config vs DEIMv2. Same broad \u201cadd a model\u201d theme, but not the same change." + "reason": "EuroBert (text encoder) vs DEIMv2 (vision detector); same request pattern only." }, { "left": "issue:36817", "right": "issue:43671", "accept": false, - "reason": "EuroBert model config and Qwen3-TTS support are unrelated model families and tasks." + "reason": "EuroBert vs Qwen3-TTS; different model families and modality." }, { "left": "issue:36817", "right": "issue:36879", "accept": false, - "reason": "EuroBert config vs RF-DETR model; different architectures and code paths." + "reason": "EuroBert vs RF-DETR; unrelated model domains." }, { "left": "issue:36879", "right": "issue:41211", "accept": false, - "reason": "RF-DETR and DEIMv2 are different models, not one shared implementation change." + "reason": "RF-DETR vs DEIMv2; both are detectors, but they are different concrete models and changes." }, { "left": "issue:36879", "right": "issue:43909", "accept": false, - "reason": "RF-DETR is a detector; LFM2.5 Audio is an audio model request." + "reason": "RF-DETR vs LFM2.5 Audio; vision detector vs audio model." }, { "left": "issue:36879", "right": "issue:44121", "accept": false, - "reason": "RF-DETR model add vs weight-sparse transformer request; different underlying changes." + "reason": "RF-DETR vs sparse transformer LM; different code paths and tasks." }, { "left": "issue:36879", "right": "issue:43837", "accept": false, - "reason": "RF-DETR and Qwen3-ASR target different model families and tasks." + "reason": "RF-DETR vs ASR model; different task and implementation." }, { "left": "issue:36879", "right": "issue:43671", "accept": false, - "reason": "RF-DETR and Qwen3-TTS are unrelated model requests." + "reason": "RF-DETR vs TTS model; different task and implementation." }, { "left": "issue:37015", "right": "issue:43837", "accept": false, - "reason": "NeoBERT vs Qwen3-ASR; both are model requests, but not the same model or fix." + "reason": "NeoBERT vs Qwen3-ASR; text encoder vs speech model." }, { "left": "issue:37015", "right": "issue:44121", "accept": false, - "reason": "NeoBERT model support and sparse-transformer support are distinct changes." + "reason": "NeoBERT vs weight-sparse transformer; distinct model families." }, { "left": "issue:37015", "right": "issue:43909", "accept": false, - "reason": "NeoBERT and LFM2.5 Audio are different model families." + "reason": "NeoBERT vs audio model; different modality." }, { "left": "issue:36879", "right": "issue:37015", "accept": false, - "reason": "RF-DETR and NeoBERT are unrelated models; no shared concrete bug/change." + "reason": "RF-DETR vs NeoBERT; unrelated families despite similar 'add model' wording." }, { "left": "issue:37015", "right": "issue:42503", "accept": false, - "reason": "NeoBERT and ModernVBERT are both BERT-like, but they are separate model additions, not the same change." + "reason": "NeoBERT vs ModernVBERT; both BERT-like names, but separate model additions." }, { "left": "issue:41115", "right": "issue:43837", "accept": false, - "reason": "MiniCPM3 architecture support and Qwen3-ASR support are different model requests." + "reason": "MiniCPM3 vs Qwen3-ASR; different model families and task." }, { "left": "issue:41115", "right": "issue:44121", "accept": false, - "reason": "MiniCPM3 model architecture vs OpenAI sparse transformer; no shared implementation target." + "reason": "MiniCPM3 vs weight-sparse transformer; unrelated requests." }, { "left": "issue:41115", "right": "issue:42503", "accept": false, - "reason": "MiniCPM3 and ModernVBERT are distinct model families; similar request shape only." + "reason": "MiniCPM3 vs ModernVBERT; separate model-specific requests." }, { "left": "issue:41115", "right": "issue:45027", "accept": false, - "reason": "MiniCPM3 architecture and Voxtral TTS support are unrelated." + "reason": "MiniCPM3 vs Voxtral TTS; multimodal LLM vs speech synthesis support." }, { "left": "issue:41211", "right": "issue:43671", "accept": false, - "reason": "DEIMv2 and Qwen3-TTS are different models and different tasks." + "reason": "DEIMv2 vs Qwen3-TTS; detector vs TTS." }, { "left": "issue:42503", "right": "issue:43837", "accept": false, - "reason": "ModernVBERT models vs Qwen3-ASR; same general domain of model support, but not the same concrete change." + "reason": "ModernVBERT vs Qwen3-ASR; different modalities and code paths." }, { "left": "issue:42503", "right": "issue:43671", "accept": false, - "reason": "ModernVBERT support and Qwen3-TTS support are separate model additions." + "reason": "ModernVBERT vs Qwen3-TTS; different tasks and model families." }, { "left": "issue:42503", "right": "issue:42738", "accept": false, - "reason": "Both are BERT-family themed, but one is generic RoPE support and the other is ModernVBERT-specific; not a duplicate." + "reason": "ModernVBERT vs generic BERT-with-RoPE request; related wording, but not the same concrete change." }, { "left": "issue:42738", "right": "issue:45027", "accept": false, - "reason": "BERT-like RoPE support and Voxtral TTS support are unrelated requests." + "reason": "BERT/RoPE architectural request vs Voxtral TTS; unrelated." }, { "left": "issue:42738", "right": "issue:46653", "accept": false, - "reason": "BERT-like model support vs an unrelated personal-name issue; no overlap." + "reason": "BERT/RoPE request vs unrelated person-name issue; not the same kind of artifact." }, { "left": "issue:43671", "right": "issue:43837", "accept": false, - "reason": "TTS support vs ASR support are adjacent audio tasks, but they are different model integrations." + "reason": "Qwen3-TTS vs Qwen3-ASR; same family name, but different concrete support and code paths." }, { "left": "issue:43671", "right": "issue:45027", "accept": false, - "reason": "Both are TTS-related, but Qwen3-TTS and Voxtral-4B-TTS are different models with different code paths." + "reason": "Qwen3-TTS vs Voxtral TTS; both TTS, but different models and likely different integrations." }, { "left": "issue:43671", "right": "issue:45177", "accept": false, - "reason": "Qwen3-TTS support and DeepSeek-OCR2 are unrelated model requests." + "reason": "TTS vs OCR; unrelated modalities." }, { "left": "issue:43671", "right": "issue:44121", "accept": false, - "reason": "TTS support and sparse-transformer support do not describe the same underlying change." + "reason": "TTS model vs sparse transformer LM; unrelated." }, { "left": "issue:43671", "right": "issue:43909", "accept": false, - "reason": "Qwen3-TTS and LFM2.5 Audio are both audio-adjacent, but they are distinct model requests." + "reason": "Qwen3-TTS vs LFM2.5 Audio; related audio domain, but distinct model additions." }, { "left": "issue:43837", "right": "issue:43909", "accept": false, - "reason": "ASR support vs audio model support; related domain, different implementation." + "reason": "ASR vs audio model; related domain, but different concrete requests." }, { "left": "issue:43837", "right": "issue:45322", "accept": false, - "reason": "ASR support and EUPE vision-language encoder support are unrelated." + "reason": "ASR vs vision-language encoder; different task and architecture." }, { "left": "issue:43837", "right": "issue:44121", "accept": false, - "reason": "ASR support and weight-sparse transformer support are different changes." + "reason": "ASR vs sparse transformer; unrelated." + }, + { + "left": "issue:43909", + "right": "issue:44205", + "accept": false, + "reason": "Audio model vs SAM3-LiteText; different model families and use cases." + }, + { + "left": "issue:43909", + "right": "issue:46645", + "accept": false, + "reason": "Audio model request vs Granite SWA+sink option; one adds a model, the other changes a model option." + }, + { + "left": "issue:44121", + "right": "issue:45322", + "accept": false, + "reason": "Sparse transformer vs EUPE encoder; distinct architectures." }, { "left": "issue:43837", "right": "issue:45027", "accept": false, - "reason": "ASR and TTS are both speech tasks, but these are different model integrations." + "reason": "ASR vs TTS; same broad area, but not the same underlying change." }, { "left": "issue:37015", "right": "issue:45027", "accept": false, - "reason": "NeoBERT and Voxtral TTS are unrelated model requests." + "reason": "NeoBERT vs Voxtral TTS; unrelated model families." }, { "left": "issue:43837", "right": "issue:45177", "accept": false, - "reason": "ASR support and OCR2 support are different tasks and models." + "reason": "ASR vs OCR; different modalities and implementations." }, { "left": "issue:45322", "right": "issue:46133", "accept": false, - "reason": "EUPE and TIPSv2 are both vision-language style requests, but they are separate models with different changes." + "reason": "EUPE vs TIPSv2; both vision-language related, but different model requests." }, { "left": "issue:42503", "right": "issue:46653", "accept": false, - "reason": "ModernVBERT models and the unrelated closed issue by name do not describe the same change." + "reason": "ModernVBERT vs unrelated person-name issue; no shared underlying change." }, { "left": "issue:45027", "right": "issue:46653", "accept": false, - "reason": "Voxtral TTS support and the unrelated closed issue by name are not duplicates." + "reason": "Voxtral TTS vs unrelated person-name issue; no shared underlying change." } ] }, @@ -18621,7 +18460,7 @@ "soft_pairs": [ "issue:38740|issue:46443" ], - "prepared_review_unit_hash": "8571e182f4a76469a0812f8a86ea7497fbc7ab6086917f949be1f8c2e37e5206", + "prepared_review_unit_hash": "b6e543f1c4fe0896dfc058515a8be783ae9a1e2445148ca2871ce58e44d6e3f5", "component_budget": { "node_count": 2, "item_count": 2, @@ -18643,18 +18482,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "The two issues are not duplicates: one is a docs request about adding Pruna as an optimization framework, the other is a Vision RoPE refactor. They concern different subsystems and different kinds of work.", - "confidence": 0.98, - "canonical_issue_reason": "No canonical duplicate issue: the titles and implied scopes are unrelated.", + "summary": "These issues are unrelated: one is a docs request to add Pruna, the other is a Vision RoPE refactor.", + "confidence": 0.97, + "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "Issue 46443 is the more substantive engineering task, but neither issue is a good duplicate canonical for the other.", + "best_issue_reason": "No single issue is a good canonical representative because the topics do not overlap.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:38740", "right": "issue:46443", "accept": false, - "reason": "Different domains and intent: documentation/integration request versus a Vision RoPE code refactor. No shared code-path or concrete bug/change." + "reason": "Different work items and no shared code-path or explicit link; a docs/framework addition is not the same as a Vision RoPE refactor." } ] }, @@ -18684,7 +18523,7 @@ "soft_pairs": [ "issue:41084|issue:43037" ], - "prepared_review_unit_hash": "d66c09a91be15d5c366f18d7e67bafcdddc086dcae72236382be7e38272d8193", + "prepared_review_unit_hash": "e2d80f8083cbcfca53a6bdf34fb9f0f2935a62ce50af90e86411b81b9fe6cc40", "component_budget": { "node_count": 2, "item_count": 2, @@ -18706,18 +18545,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These two issues do not look like duplicates: one is a generic request around block decoding, the other is a model-specific request for DeepSeek v3.2 support.", - "confidence": 0.96, - "canonical_issue_reason": "If a representative issue is needed, #43037 is the better one because it is more specific and actionable; #41084 is too terse and broad.", + "summary": "The two issues are not duplicates: one requests block decoding support in general, while the other asks for DeepSeek v3.2 support. They refer to different feature areas and no concrete overlap is shown.", + "confidence": 0.94, + "canonical_issue_reason": "Issue #43037 is the clearer representative because it names a specific model-support request, making it more actionable and easier to track than the broader \"Set Block Decoding\" title.", "canonical_pr_reason": null, - "best_issue_reason": "#43037 is the strongest standalone issue due to its clearer feature scope and target, making it more useful as a canonical tracker than #41084.", + "best_issue_reason": "Issue #43037 is the best issue to keep as the cluster representative since it is specific, self-contained, and less ambiguous than #41084.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:41084", "right": "issue:43037", "accept": false, - "reason": "Different underlying requests: generic block decoding behavior vs. support for a specific model version. No shared filenames, targets, or code-path evidence suggests the same bug/change." + "reason": "The titles indicate different underlying requests: block decoding functionality versus support for a specific DeepSeek model version. They do not appear to describe the same bug or change." } ] }, @@ -18747,7 +18586,7 @@ "soft_pairs": [ "issue:43598|issue:43600" ], - "prepared_review_unit_hash": "eceebc70dd926bd1f802a05eb7fdcb6c361117acc47b59885008a3689082a603", + "prepared_review_unit_hash": "1ed683f09599ef818f29dbfdd63ae7ce4e6336f6f57a896b328f63394f87d05f", "component_budget": { "node_count": 2, "item_count": 2, @@ -18769,18 +18608,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These two issues are related to Trainer customization, but they describe different bugs: one about `remove_unused_column` behavior, the other about FLOPS estimation for custom trainers. They should not be treated as duplicates.", - "confidence": 0.95, - "canonical_issue_reason": "Issue 43598 is the more concrete and narrowly scoped report, centered on `remove_unused_column` behavior in `Trainer`.", + "summary": "These are two separate enhancement issues in Trainer: one about making `remove_unused_column` more customizable, and another about handling custom trainers when estimating FLOPS. They share only a broad Trainer context, not the same underlying bug or change.", + "confidence": 0.96, + "canonical_issue_reason": "Issue #43598 is the more general Trainer customizability request and is slightly broader/more central than #43600, but the two are not duplicates.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 43598 is a better representative for this cluster because it states a specific Trainer customization pain point rather than a broader estimation concern.", + "best_issue_reason": "#43598 is the better representative issue because it frames a reusable Trainer behavior change, whereas #43600 is a distinct FLOPS-estimation concern.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43598", "right": "issue:43600", "accept": false, - "reason": "They concern different code paths and fixes: unused-column removal/custom trainer behavior vs FLOPS estimation for custom trainers. Shared Trainer context is not enough to make them the same underlying bug." + "reason": "Different concrete problems: one targets `remove_unused_column` behavior, the other targets FLOPS estimation for custom trainers. Same subsystem, but not the same fix." } ] }, @@ -18810,12 +18649,12 @@ "soft_pairs": [ "issue:43684|issue:44405" ], - "prepared_review_unit_hash": "9b89e93c84470f8ef64f724815fcf513269594a67947037a47289ea3e0723573", + "prepared_review_unit_hash": "536fa9775f8cdc376504d2ec45e511a8dd2b8f7c310e2d12f0f57530d8c46b54", "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 }, @@ -18823,7 +18662,7 @@ "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 1612, + "serialized_chars": 1611, "estimated_input_tokens": 403, "estimated_eval_tokens": 1062 }, @@ -18832,18 +18671,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These are two different feature requests: one asks for Qwen3-Omni registration in AutoModel/AutoModelForConditionalGeneration, while the other asks for AutoModelForSequenceClassification support for Qwen3.5. They target different model families and different auto classes, so they are not duplicates.", - "confidence": 0.96, - "canonical_issue_reason": "Issue 43684 is the broader, more central model-registration request and is the better representative of this cluster.", + "summary": "These two issues are both about Hugging Face model-class registration, but they target different model families and different auto-classes: Qwen3-Omni registration for AutoModel/AutoModelForConditionalGeneration vs. Qwen3.5 sequence-classification support for AutoModelForSequenceClassification. They are not the same underlying bug/request.", + "confidence": 0.95, + "canonical_issue_reason": "Issue #43684 is the broader registration request and is more representative of the cluster topic, while #44405 is a separate, narrower model-class support request.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 43684 best captures the cluster\u2019s main theme of adding auto-class support for a Qwen model; issue 44405 is a separate, narrower head-support request for a different config.", + "best_issue_reason": "#43684 is the better representative issue because it is the more general model-registration request and not tied to a specific downstream task class.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43684", "right": "issue:44405", "accept": false, - "reason": "Different Qwen variants and different auto-model targets; this does not look like the same underlying bug or change." + "reason": "Reject: different model families (Qwen3-Omni vs Qwen3.5/Qwen3_5Config) and different target auto-classes, so they do not describe the same code-path change." } ] }, @@ -18880,76 +18719,79 @@ "issue:43845|issue:43859", "issue:43846|issue:43859" ], - "prepared_review_unit_hash": "d369d97bee38b2f0ef34783f7e7d2fbfc434b364b8d717178f24601584408532", + "prepared_review_unit_hash": "4ee97d3a0666028b1bd04b37cb4389a241103091f927e865f9cdd972d989ad54", "component_budget": { "node_count": 4, "item_count": 4, "soft_pair_count": 6, - "serialized_chars": 4264, - "estimated_input_tokens": 1066, - "estimated_eval_tokens": 2388 + "serialized_chars": 4260, + "estimated_input_tokens": 1065, + "estimated_eval_tokens": 2386 }, "budget": { "node_count": 4, "item_count": 4, "soft_pair_count": 6, - "serialized_chars": 4264, - "estimated_input_tokens": 1066, - "estimated_eval_tokens": 2388 + "serialized_chars": 4260, + "estimated_input_tokens": 1065, + "estimated_eval_tokens": 2386 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These are too underspecified to safely deduplicate: one issue has a different title ('Fett') and the three 'huggingface' issues only share a generic title with no supporting body, targets, or code references. I would not merge any pair on the available evidence.", - "confidence": 0.17, - "canonical_issue_reason": "issue:43845 is the least-bad representative only because it is the earliest of the repeated 'huggingface' reports and is closer to the cluster\u2019s shared wording than 'Fett'; confidence is very low due to missing body details.", + "summary": "The cluster is too thinly evidenced to confidently call the three \u201chuggingface\u201d issues duplicates. They share a title, but the packet only shows soft similarity and timing/body-length metadata, so the safer call is to leave the pairwise links unresolved. The \u201cFett\u201d issue does not appear related.", + "confidence": 0.44, + "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "issue:43845 is the best issue candidate among the set for the same reason: it is the earliest repeated 'huggingface' title and has no evidence that the others describe a different concrete problem; still not a strong canonical choice.", + "best_issue_reason": "43845 is the best provisional anchor only because it is one of the repeated \u201chuggingface\u201d reports and is earliest among them, but the evidence is too weak to treat it as a confirmed canonical duplicate.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43835", "right": "issue:43845", "accept": false, - "reason": "Only weak title similarity; 'Fett' and 'huggingface' do not provide enough evidence that they are the same underlying issue." + "reason": "Different titles and no concrete shared issue content; soft similarity alone is not enough." }, { "left": "issue:43835", "right": "issue:43846", "accept": false, - "reason": "No substantive overlap beyond generic soft similarity; no body, target, or reference evidence ties these together." + "reason": "\u201cFett\u201d does not have enough evidence to match the huggingface reports." }, { "left": "issue:43835", "right": "issue:43859", "accept": false, - "reason": "Different titles with no supporting details, so there is no basis to treat them as duplicates." + "reason": "No concrete indication that the short \u201cFett\u201d report is the same as the huggingface reports." }, { "left": "issue:43845", "right": "issue:43846", "accept": false, - "reason": "The titles are identical, but there is still no body or reference evidence confirming the same bug or request; that is not enough to merge confidently." + "reason": "Same title and close timestamps are not enough on their own to confirm a duplicate bug report." }, { "left": "issue:43845", "right": "issue:43859", "accept": false, - "reason": "Generic repeated title alone is insufficient; no concrete shared problem is described." + "reason": "The evidence is limited to title similarity and metadata, which is too weak to accept as the same underlying issue." }, { "left": "issue:43846", "right": "issue:43859", "accept": false, - "reason": "Identical generic titles are not enough on their own without any supporting description or code references." + "reason": "These look similar, but there is no concrete content-level evidence that they describe the same bug." } ] }, - "evaluator_result": null, - "evaluator_used": false, - "retried": false, + "evaluator_result": { + "accept": false, + "feedback": "The cluster summary is a bit too strong for the packet. The three \u201chuggingface\u201d issues do share a title, but the evidence shown is limited to soft similarity plus timestamps/body lengths; the packet does not ground a confident claim that they are duplicates. In particular, the accepted soft-edge verdicts overstate the evidence by saying the reports \u201cstrongly suggest\u201d the same underlying issue. I would keep those pairwise verdicts conservative or unresolved unless there is more concrete content-level evidence." + }, + "evaluator_used": true, + "retried": true, "accepted_nontrivial_soft_edge": false, "error_kind": null, "error_message": null @@ -18984,98 +18826,95 @@ "issue:44572|issue:44937", "issue:44573|issue:44937" ], - "prepared_review_unit_hash": "6b310da2d13f802db7188046234da6f2639c539fe334de87bbfd9bbd9ceeb6ca", + "prepared_review_unit_hash": "4e579894b7a209a4e6fe42e78ac5daac04f5c5a40c26b5f7847506c5f7c47862", "component_budget": { "node_count": 5, "item_count": 5, "soft_pair_count": 9, - "serialized_chars": 5794, - "estimated_input_tokens": 1449, - "estimated_eval_tokens": 3154 + "serialized_chars": 5791, + "estimated_input_tokens": 1448, + "estimated_eval_tokens": 3152 }, "budget": { "node_count": 5, "item_count": 5, "soft_pair_count": 9, - "serialized_chars": 5794, - "estimated_input_tokens": 1449, - "estimated_eval_tokens": 3154 + "serialized_chars": 5791, + "estimated_input_tokens": 1448, + "estimated_eval_tokens": 3152 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "This cluster is dominated by low-signal spam-like issues; the only clear duplication is among the identical \"\" reports, but there is no substantive bug/change to unify across the cluster.", - "confidence": 0.92, - "canonical_issue_reason": "Issue 44371 is the earliest of the identical \"\" issues and is the cleanest representative of the only obvious duplicate subgroup.", + "summary": "These are all low-signal issue reports with spammy or generic titles and no bodies, links, or activity to establish a shared underlying bug. None of the pairs are strong enough to merge as duplicates.", + "confidence": 0.18, + "canonical_issue_reason": "issue:44418 is the least spam-like and most plausibly a real report, but it is still very generic and lacks any concrete problem description.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 44371 is the best issue anchor because it matches the other identical spam placeholder reports exactly and appears earliest.", + "best_issue_reason": "issue:44418 is the best available issue candidate only because it has a non-spam title; there is still no substantive evidence that it represents the cluster\u2019s true topic.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:44371", "right": "issue:44572", - "accept": true, - "reason": "Both are identical \"\" placeholder issues, so they look like the same underlying spam report." + "accept": false, + "reason": "Both are spam-like with no body detail; identical tokenless titles do not demonstrate the same underlying issue." }, { "left": "issue:44371", "right": "issue:44573", - "accept": true, - "reason": "Both are identical \"\" placeholder issues, so they look like the same underlying spam report." + "accept": false, + "reason": "Both are spam-like with no body detail; no concrete bug or request is visible to justify duplicate status." }, { "left": "issue:44371", "right": "issue:44418", "accept": false, - "reason": "Titles differ and one is a documentation request while the other is a spam placeholder; no shared underlying problem." + "reason": "A spam title and a generic documentation suggestion are not enough to show the same issue." }, { "left": "issue:44371", "right": "issue:44937", "accept": false, - "reason": "The titles are unrelated; \"Check out \\\"Google LLC\\\"\" is spammy but not the same report as \"\"." + "reason": "The titles are unrelated on their face; no evidence they refer to the same report." }, { "left": "issue:44418", "right": "issue:44572", "accept": false, - "reason": "A documentation enhancement and a spam placeholder are not the same issue." + "reason": "Generic documentation wording and spam are not the same underlying bug/change." }, { "left": "issue:44418", "right": "issue:44573", "accept": false, - "reason": "A documentation enhancement and a spam placeholder are not the same issue." + "reason": "Generic documentation wording and spam are not the same underlying bug/change." }, { "left": "issue:44572", "right": "issue:44573", - "accept": true, - "reason": "Both are identical \"\" placeholder issues, so they look like the same underlying spam report." + "accept": false, + "reason": "Both are spam-like, but there is still no concrete description showing a shared problem." }, { "left": "issue:44572", "right": "issue:44937", "accept": false, - "reason": "The titles are unrelated; the Google LLC phrasing is a different spam submission from the generic \"\" placeholder." + "reason": "Spam-like title versus a specific quoted-name title; no shared bug or change is evident." }, { "left": "issue:44573", "right": "issue:44937", "accept": false, - "reason": "The titles are unrelated; the Google LLC phrasing is a different spam submission from the generic \"\" placeholder." + "reason": "Spam-like title versus a specific quoted-name title; no shared bug or change is evident." } ] }, - "evaluator_result": { - "accept": true, - "feedback": "The summary is grounded in the packet: the only clearly duplicated subgroup is the three identical \"\" issues, and the rejected pairs are conservative given the differing titles. No PR merge claim is being made, so the analysis is appropriately scoped." - }, - "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 }, @@ -19098,7 +18937,7 @@ "soft_pairs": [ "issue:45145|issue:45151" ], - "prepared_review_unit_hash": "0388e7a58f9adf8d999b58e551ad4b013432c09fc79d5a24947ee4da7c388d2b", + "prepared_review_unit_hash": "6d372d41016017783d2b8a9158187f5c3fdc1fe875533ab5178d236cf3352954", "component_budget": { "node_count": 2, "item_count": 2, @@ -19120,24 +18959,24 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These two closed issues appear to be duplicate reports with the same title and very similar content. Issue #45145 is the better canonical issue because it was opened first; #45151 looks like a near-identical duplicate.", + "summary": "The two closed issues appear to be duplicates of the same proposal/report: identical titles, highly similar bodies, and no distinguishing links or files. Issue #45145 is the better canonical choice because it was created earlier.", "confidence": 0.99, - "canonical_issue_reason": "#45145 is the earlier report and has the same title/content pattern as #45151, so it is the best canonical issue for this duplicate pair.", + "canonical_issue_reason": "Issue #45145 was created earlier and appears to be the same report as #45151, with an identical title and highly similar content.", "canonical_pr_reason": null, - "best_issue_reason": "#45145 is the strongest single issue to keep as the representative record: it predates #45151 and describes the same underlying request/report.", + "best_issue_reason": "#45145 is the best issue representative because it is the earlier of the two near-identical issues and there is no evidence of a distinct scope between them.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:45145", "right": "issue:45151", "accept": true, - "reason": "Same exact title and highly similar bodies; these look like duplicate reports of the same underlying issue." + "reason": "Accept: the issues have the same title and substantially overlapping content, with no evidence of different underlying problems; they look like duplicates of the same issue." } ] }, "evaluator_result": { "accept": true, - "feedback": "The summary is grounded in the packet: both closed issues have the same title, similar body lengths, and soft-similarity evidence supports a duplicate relationship. Treating #45145 as the canonical earlier report is conservative and supported by the creation timestamps." + "feedback": "The judgment is grounded in the packet: the two issues share an identical title, similar body sizes, and a soft-similarity signal, with no explicit links or filenames suggesting separate scopes. The canonical choice of the earlier issue is reasonable, and the soft-edge accept is conservative enough here." }, "evaluator_used": true, "retried": false, @@ -19218,7 +19057,7 @@ "soft_pairs": [ "issue:45672|issue:45674" ], - "prepared_review_unit_hash": "9f790a242dc34c44583a4e665dc9603475c8fb59043684d9f489db9eccb2c692", + "prepared_review_unit_hash": "abddce1de611a6aa333499c6ca10379d70e5a77cc8a2de3f2d6a3539b0c325ad", "component_budget": { "node_count": 2, "item_count": 2, @@ -19240,18 +19079,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Both artifacts report 4-bit inference failures, but they appear to be different bugs: one is a Gemma4/audio-module uint8 `torch.finfo()` TypeError, the other is a BitsAndBytesConfig `llm_int8_skip_modules` regression that removes the default `lm_head` exclusion and triggers an AssertionError. They should not be merged.", - "confidence": 0.95, - "canonical_issue_reason": "Issue #45674 is the better canonical issue because it describes a broader configuration regression in 4-bit inference, with a clear mechanism and likely wider impact than the Gemma4-specific failure in #45672.", + "summary": "These are two separate 4-bit inference issues: one is Gemma4/audio-module specific and triggered by `torch.finfo()` on `uint8` weights, while the other is a `BitsAndBytesConfig` regression where `llm_int8_skip_modules` drops the default `lm_head` exclusion and causes an assertion failure. They share the broad 4-bit inference context but not the same bug or code path.", + "confidence": 0.93, + "canonical_issue_reason": "Issue 45674 is the better canonical issue because it describes a clearer, more generally applicable 4-bit inference regression in `BitsAndBytesConfig`, whereas 45672 is a narrower Gemma4/audio-module-specific failure.", "canonical_pr_reason": null, - "best_issue_reason": "Issue #45674 is the strongest representative of the cluster because it is the more general, actionable bug report and is not tied to one model or module layout.", + "best_issue_reason": "Issue 45674 is the strongest representative for the cluster since it is broader, more actionable, and tied to a concrete config behavior change rather than a model-specific symptom.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:45672", "right": "issue:45674", "accept": false, - "reason": "Different root causes and failure modes: one is a Gemma4 audio-module dtype/finfo error on uint8 weights, the other is a BitsAndBytesConfig exclusion-list regression causing an AssertionError. Same broad area (4-bit inference) is not enough to treat them as duplicates." + "reason": "Both concern 4-bit inference, but the failure modes differ: `torch.finfo()` on `uint8` weights in Gemma4 audio modules versus `llm_int8_skip_modules` removing the default `lm_head` exclusion. Different triggering code paths and different underlying bugs." } ] }, @@ -19281,7 +19120,7 @@ "soft_pairs": [ "issue:45762|issue:45763" ], - "prepared_review_unit_hash": "a5ac76416e24d4582ed954edbe7b04470cca6be704b16d35c9e35469cbac6652", + "prepared_review_unit_hash": "b03dab2f223092cd93760f27ff442cf0bb03b76aa7cd607cd1da611d4f15c15c", "component_budget": { "node_count": 2, "item_count": 2, @@ -19303,18 +19142,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These are two empty, closed issues with different personal-name/date titles and no shared content or explicit linkage; they do not look like duplicates of the same underlying bug or change.", - "confidence": 0.97, - "canonical_issue_reason": null, + "summary": "These two closed issues do not look like duplicates of the same underlying problem; they have different names and dates and no body/links to connect them. The similarity is only superficial.", + "confidence": 0.98, + "canonical_issue_reason": "issue:45762 is the better placeholder canonical issue only because it is the earlier report; there is no evidence these two issues describe the same thing.", "canonical_pr_reason": null, - "best_issue_reason": null, + "best_issue_reason": "issue:45762 is marginally the best representative for triage since it is older, but neither issue appears to be a true duplicate candidate.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:45762", "right": "issue:45763", "accept": false, - "reason": "Only soft title similarity is present; the names and dates differ, and there is no body, discussion, target, or other evidence that they describe the same issue." + "reason": "Different people and dates in the titles, with no shared body text, links, or supporting evidence of the same bug/request." } ] }, @@ -19344,7 +19183,7 @@ "soft_pairs": [ "issue:45764|issue:45765" ], - "prepared_review_unit_hash": "f2d8387fa8a0b747e63916b97a0e93fb02139cd2da0ad5b30bda2d852ef23fb1", + "prepared_review_unit_hash": "806e20052bcb6ccfdbf43539fee2eec4b972598836afaf3712a2816bd54eac0c", "component_budget": { "node_count": 2, "item_count": 2, @@ -19366,18 +19205,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "The two issues are both closed, empty, and appear to be unrelated personal-name/date entries rather than the same bug report. No duplicate relationship is supported beyond a weak soft similarity signal.", - "confidence": 0.97, - "canonical_issue_reason": "Neither issue contains any substantive description or shared technical context; if one must be kept as a representative, issue 45764 is marginally earlier, but there is no strong canonical choice.", + "summary": "These are two closed, content-less issue stubs with unrelated personal-name/date titles. They do not describe the same bug or request, so they should not be clustered as duplicates beyond weak textual similarity.", + "confidence": 0.98, + "canonical_issue_reason": "Issue 45764 is marginally earlier, but neither issue contains substantive problem details; there is no meaningful canonical issue here.", "canonical_pr_reason": null, - "best_issue_reason": "Neither issue is a good canonical duplicate target because both are content-free and unrelated in apparent topic.", + "best_issue_reason": "Issue 45764 is the best placeholder only because it was created first; both issues are effectively empty and not good representative reports.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:45764", "right": "issue:45765", "accept": false, - "reason": "The titles are different names with different dates and there is no body text, code context, or explicit linkage indicating the same underlying bug or request." + "reason": "The titles are different personal names with different dates and no body content or bug description; this is not the same underlying issue." } ] }, @@ -19407,7 +19246,7 @@ "soft_pairs": [ "issue:45810|issue:46295" ], - "prepared_review_unit_hash": "93949c8aaf68fa43975b48aa8f8c35d5050895b4a0a3e4d5183c79795f65d830", + "prepared_review_unit_hash": "18927839abe21d5d41bdb41331aa755587c55eaac150995cd97d173b0006fc93", "component_budget": { "node_count": 2, "item_count": 2, @@ -19429,18 +19268,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These two open issues are unrelated: one requests adding Qwen3_5ForTokenClassification, while the other asks for FP8 training support. They do not describe the same underlying bug or change.", + "summary": "The two issues are unrelated: one requests adding a Qwen3_5 token-classification model, while the other asks for FP8 training support. They should not be clustered as duplicates.", "confidence": 0.98, - "canonical_issue_reason": "No duplicate canonical issue: the requests target different features and code paths.", + "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "issue:45810 is the more specific, self-contained feature request; issue:46295 is broader and about training infrastructure.", + "best_issue_reason": null, "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:45810", "right": "issue:46295", "accept": false, - "reason": "Different topics and implementation areas: model class support vs FP8 training support. They are not the same underlying issue." + "reason": "Different feature requests for different code paths: model-class support for token classification vs training precision support (FP8). They do not describe the same underlying bug or change." } ] }, @@ -19480,91 +19319,91 @@ "issue:45871|issue:45873", "issue:45872|issue:45873" ], - "prepared_review_unit_hash": "91c7fe3ec84de48f145ce98a71e4cc21abd68f893ed67fc82041b991bd0df3c9", + "prepared_review_unit_hash": "c7f33cf8dd2037965622ec8c310b8a495799561512db95fb0941eefc3d5a86a9", "component_budget": { "node_count": 5, "item_count": 5, "soft_pair_count": 8, - "serialized_chars": 5722, - "estimated_input_tokens": 1431, - "estimated_eval_tokens": 3118 + "serialized_chars": 5728, + "estimated_input_tokens": 1432, + "estimated_eval_tokens": 3120 }, "budget": { "node_count": 5, "item_count": 5, "soft_pair_count": 8, - "serialized_chars": 5722, - "estimated_input_tokens": 1431, - "estimated_eval_tokens": 3118 + "serialized_chars": 5728, + "estimated_input_tokens": 1432, + "estimated_eval_tokens": 3120 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "The set contains two separate exact-duplicate issue pairs: 45870/45871 and 45872/45873. Issue 45869 is a different financial-data sales/sample post and is not the same underlying request as the others.", - "confidence": 0.98, - "canonical_issue_reason": null, + "summary": "There are two strong near-duplicate issue pairs: 45870/45871 and 45872/45873. Issue 45869 is related only by topic and should remain separate.", + "confidence": 0.89, + "canonical_issue_reason": "45870 is a clean representative for the first near-duplicate pair: it has the same title and body length as 45871, and it is the earlier of the two. With only metadata available, treat this as a likely duplicate rather than a proven exact match.", "canonical_pr_reason": null, - "best_issue_reason": "No single canonical issue spans the whole set; the closest representatives are 45870 for the 'Example: Financial time series prediction with DataForge' pair and 45872 for the 'DataForge: Financial time series + scraping service' pair.", + "best_issue_reason": "45870 is the best global issue representative because it anchors one of the clearest near-duplicate pairs and is the earliest instance. 45869 is a separate dataset request, while 45872/45873 appear to be a different, separate pair.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:45869", "right": "issue:45870", "accept": false, - "reason": "Different artifact types/content: dataset sample promotion versus a DataForge example issue; same financial domain but not the same underlying bug or change." + "reason": "Related financial domain, but this is a dataset bundle request versus a prediction example; not the same underlying issue." }, { "left": "issue:45869", "right": "issue:45871", "accept": false, - "reason": "Different artifact types/content: dataset sample promotion versus a DataForge example issue; not the same underlying request." + "reason": "Related financial domain, but this is a dataset bundle request versus a prediction example; not the same underlying issue." }, { "left": "issue:45870", "right": "issue:45871", "accept": true, - "reason": "Exact duplicate titles and matching metadata/body size indicate the same example issue." + "reason": "Same title, same body length, and very close timestamps; this is a strong near-duplicate pair, though the packet only supports a likely-duplicate judgment." }, { "left": "issue:45870", "right": "issue:45872", "accept": false, - "reason": "Both mention financial time series, but one is an example issue and the other is a scraping-service request; these are different underlying topics." + "reason": "Both mention financial time series, but the titles indicate different requests: an example prediction issue versus a scraping-service request." }, { "left": "issue:45870", "right": "issue:45873", "accept": false, - "reason": "Different concrete request: example prediction walkthrough versus financial time series + scraping service." + "reason": "Both mention financial time series, but the titles indicate different requests: an example prediction issue versus a scraping-service request." }, { "left": "issue:45871", "right": "issue:45872", "accept": false, - "reason": "Different concrete request: example prediction walkthrough versus financial time series + scraping service." + "reason": "Both mention financial time series, but the titles indicate different requests: an example prediction issue versus a scraping-service request." }, { "left": "issue:45871", "right": "issue:45873", "accept": false, - "reason": "Different concrete request: example prediction walkthrough versus financial time series + scraping service." + "reason": "Both mention financial time series, but the titles indicate different requests: an example prediction issue versus a scraping-service request." }, { "left": "issue:45872", "right": "issue:45873", "accept": true, - "reason": "Exact duplicate titles and matching metadata/body size indicate the same service-related issue." + "reason": "Same title, same body length, and very close timestamps; this is a strong near-duplicate pair, though the packet only supports a likely-duplicate judgment." } ] }, "evaluator_result": { - "accept": true, - "feedback": "Grounded overall: the packet clearly supports two exact-duplicate pairs (45870/45871 and 45872/45873) via identical titles and matching body lengths, while 45869 is a different dataset-sample post. The soft-edge rejections between the two topic clusters are conservative and consistent with the metadata. No overstatement detected." + "accept": false, + "feedback": "The clustering is a bit too strong for the packet. The items only provide titles, lengths, timestamps, and soft-similarity signals; there is no body text to justify claims like \u201cfully duplicated,\u201d \u201ceffectively identical content,\u201d or \u201cdirect duplicates.\u201d The 45870/45871 and 45872/45873 pairs may be strong matches, but the verdicts should stay conservative (e.g. likely duplicates / near-identical) unless the packet shows exact content equivalence. The 45869 separation is fine." }, "evaluator_used": true, - "retried": false, + "retried": true, "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null @@ -19588,7 +19427,7 @@ "soft_pairs": [ "issue:46439|issue:46441" ], - "prepared_review_unit_hash": "795733179f69a92f98358ae5b1924989b0803663b24ecc3fd8e3e0b09da5af48", + "prepared_review_unit_hash": "3d6842affd6c26258b98074fb04761a37ee3430bb06b6c88cc06e4a211f17538", "component_budget": { "node_count": 2, "item_count": 2, @@ -19610,24 +19449,27 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Both issues concern static-cache failures around `linear_attention`, but they describe different concrete breakages: one is shape corruption during `early_initialization`, the other is a `KeyError` in Qwen3.5. Related, but not clearly the same bug.", - "confidence": 0.7, - "canonical_issue_reason": "Issue 46439 is the broader, more framework-level report: it names the offending `early_initialization` path and the resulting shape/runtime failure in linear-attention layers.", + "summary": "The two issues are in the same static-cache / linear-attention area, but the packet only shows title-level similarity and one soft similarity edge. 46439 is the broader root-cause report, while 46441 is a model-specific crash that could be related but is not proven to be the same bug.", + "confidence": 0.58, + "canonical_issue_reason": "Issue 46439 is the best canonical issue because it describes the underlying mechanism more directly: static-cache `early_initialization` corrupting linear-attention shapes and causing runtime failure.", "canonical_pr_reason": null, - "best_issue_reason": "46439 is the best representative issue because it is more general and actionable, while 46441 reads like a model-specific symptom of a static-cache/linear-attention lookup failure.", + "best_issue_reason": "46439 is the strongest anchor for the cluster since it captures the broader bug mechanism rather than one specific symptom.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:46439", "right": "issue:46441", "accept": false, - "reason": "They overlap on static cache + linear-attention, but one reports wrong-shape corruption during initialization and the other reports a `KeyError: 'linear_attention'`; different failure modes, so not safe to merge as duplicates." + "reason": "Do not merge as duplicates on the current evidence: there are no shared targets/files or body-level links, and the packet only supports a possible relation in the same subsystem, not the same underlying defect." } ] }, - "evaluator_result": null, - "evaluator_used": false, - "retried": false, + "evaluator_result": { + "accept": false, + "feedback": "The cluster summary is mostly grounded, but the soft-edge verdict is too strong for the evidence available. The packet only gives title-level similarity and one `soft_similarity` edge with no shared targets/files; that supports a possible relation, not a confident accept. Issue 46441 could be related to the same static-cache/linear-attention area, but the packet does not establish that its `KeyError` is the same underlying bug as 46439\u2019s shape corruption. Keep the relationship tentative unless there is stronger issue-body evidence." + }, + "evaluator_used": true, + "retried": true, "accepted_nontrivial_soft_edge": false, "error_kind": null, "error_message": null @@ -19651,12 +19493,12 @@ "soft_pairs": [ "issue:46551|issue:46552" ], - "prepared_review_unit_hash": "0b9e436821dac0a6a07337fefe04da9f82ec11ee1c158be3fac4afa3c9e5c389", + "prepared_review_unit_hash": "7e8761665fef41f111d8f662998438e7ba60455dcd7170e1b2056c5b40050a62", "component_budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 1707, + "serialized_chars": 1706, "estimated_input_tokens": 427, "estimated_eval_tokens": 1110 }, @@ -19664,7 +19506,7 @@ "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 1707, + "serialized_chars": 1706, "estimated_input_tokens": 427, "estimated_eval_tokens": 1110 }, @@ -19673,24 +19515,24 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These two issues are effectively identical duplicates: same title, same body length, and no distinguishing evidence beyond one being closed and the other open with references.", - "confidence": 0.99, - "canonical_issue_reason": "Issue 46552 is the better canonical target because it is the open copy and already has inbound references, making it the more active tracker.", + "summary": "These two issues are effectively exact duplicates: same title, same body length, and identical bug description about Wav2Vec2CTCTokenizer v5 leaking the word delimiter as a special token. The open issue is the better representative for ongoing tracking.", + "confidence": 0.98, + "canonical_issue_reason": "Issue #46552 is the better canonical issue because it is open and already has inbound references, so it is the more active tracker for the same bug.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 46552 is the best issue to keep because it is open, actively referenced, and appears to be the primary thread for the duplicate report.", + "best_issue_reason": "Issue #46552 is the strongest issue candidate overall: same problem statement, open state, and existing references make it the most suitable single issue to keep.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:46551", "right": "issue:46552", "accept": true, - "reason": "Same title and same described bug; no evidence of distinct scope or behavior, so they are the same underlying issue." + "reason": "Exact same title and identical described bug; both refer to the same tokenizer special-token leak, so this is a true duplicate pair." } ] }, "evaluator_result": { "accept": true, - "feedback": "The summary is grounded in the packet: the titles are identical, the body lengths match, and the open issue has inbound references while the closed one does not. The soft-edge verdict is appropriately conservative given the available evidence, and there are no PR mergeability concerns here." + "feedback": "Grounded in the packet: the two issues have identical titles, identical body lengths, and the same soft-similarity evidence, so treating them as duplicates is conservative. The canonical choice of the open issue with inbound references is also supported." }, "evaluator_used": true, "retried": false, @@ -19709,7 +19551,7 @@ "source": "llm", "cache_hit": false, "model": "gpt-5.4-mini?service_tier=flex", - "cluster_id": "cluster-39895-15", + "cluster_id": "cluster-39895-14", "nodes": [ "pull_request:39895", "pull_request:41855", @@ -19724,22 +19566,19 @@ "pull_request:45213", "pull_request:45655", "pull_request:46048", - "pull_request:46267", - "pull_request:46662" + "pull_request:46267" ], "soft_pairs": [ "pull_request:39895|pull_request:43424", - "pull_request:41855|pull_request:43424", "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:44827", "pull_request:43424|pull_request:43823", - "pull_request:43823|pull_request:46662", "pull_request:43823|pull_request:45655", "pull_request:43823|pull_request:43995", - "pull_request:43995|pull_request:46662", "pull_request:43995|pull_request:45213", "pull_request:43995|pull_request:44827", "pull_request:44827|pull_request:45170", @@ -19748,189 +19587,156 @@ "pull_request:44827|pull_request:45655", "pull_request:44827|pull_request:45213", "pull_request:43823|pull_request:45213", - "pull_request:45213|pull_request:46662", - "pull_request:45655|pull_request:46662", - "pull_request:44827|pull_request:46662", "pull_request:43251|pull_request:43254" ], - "prepared_review_unit_hash": "f5348bb5922399d36eb2a3d4a27792c95166968b297f5434b5cdde4fde54074a", + "prepared_review_unit_hash": "861040f44251a5a02dc79c15e9ec16927d6d5035fe2654cc74d6d8a1b11e9e03", "component_budget": { - "node_count": 15, - "item_count": 15, - "soft_pair_count": 24, - "serialized_chars": 41945, - "estimated_input_tokens": 10487, - "estimated_eval_tokens": 21230 + "node_count": 14, + "item_count": 14, + "soft_pair_count": 19, + "serialized_chars": 38084, + "estimated_input_tokens": 9521, + "estimated_eval_tokens": 19298 }, "budget": { - "node_count": 15, - "item_count": 15, - "soft_pair_count": 24, - "serialized_chars": 41945, - "estimated_input_tokens": 10487, - "estimated_eval_tokens": 21230 + "node_count": 14, + "item_count": 14, + "soft_pair_count": 19, + "serialized_chars": 38084, + "estimated_input_tokens": 9521, + "estimated_eval_tokens": 19298 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "This cluster is mostly a mix of unrelated feature/model PRs linked only by loose topical similarity. The only true duplicate pair is the two fixed_cross_entropy patches in loss_utils.py; everything else should stay separate.", - "confidence": 0.97, + "summary": "Only one pair looks like a true duplicate: the two `fixed_cross_entropy` kwargs PRs (#43251 and #43254). The rest are unrelated model additions, refactors, or test/documentation changes that do not fix the same concrete bug/change.", + "confidence": 0.95, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 43251 is the clearest canonical representative of the only real duplicate pair: it directly targets issue 43240 and implements the fixed_cross_entropy kwargs passthrough in the shared file.", + "canonical_pr_reason": "PR #43251 is the better canonical choice: it explicitly targets issue #43240 and implements the broader `cross_entropy` kwargs passthrough in `fixed_cross_entropy`, while #43254 is just an alternate phrasing of the same fix.", "best_issue_reason": null, - "best_pr_reason": "PR 43251 is the best standalone representative for the shared loss-utils fix because it has the explicit issue link and the broader, more direct implementation.", + "best_pr_reason": "PR #43251 is the strongest representative for the cluster because it has the explicit issue link, a clear bug title, and the same underlying code-path change as #43254.", "soft_edge_verdicts": [ { "left": "pull_request:39895", "right": "pull_request:43424", "accept": false, - "reason": "Unrelated PRs: VideoPrism model addition vs executorch dynamic-shape test; different code paths and goals." + "reason": "Both are PRs, but they are unrelated feature/test additions in different areas (VideoPrism model support vs ExecuTorch dynamic-shape export tests)." }, { "left": "pull_request:41855", - "right": "pull_request:43424", + "right": "pull_request:43823", "accept": false, - "reason": "Mistral tokenizer methods vs executorch export tests; same broad library, but not the same underlying bug/change." + "reason": "Mistral tokenizer helper methods and adding MobileLLM are unrelated changes; they do not fix the same code path." }, { "left": "pull_request:41855", - "right": "pull_request:43823", + "right": "pull_request:43424", "accept": false, - "reason": "Tokenizer API fix vs new MobileLLM model addition; clearly separate changes." + "reason": "Tokenizer method additions and ExecuTorch export test coverage are different changes with no shared underlying bug." }, { "left": "pull_request:42668", "right": "pull_request:43823", "accept": false, - "reason": "Processor auto-loading robustness vs adding MobileLLM; different models and different failure modes." + "reason": "These are different processor/model robustness additions for different models; they are not the same bug or change." }, { "left": "pull_request:43251", "right": "pull_request:43823", "accept": false, - "reason": "Loss function kwargs fix and MobileLLM model addition are unrelated." + "reason": "A cross-entropy kwargs fix and adding MobileLLM are unrelated." }, { "left": "pull_request:43251", "right": "pull_request:44827", "accept": false, - "reason": "Loss-utils keyword passthrough vs Mistral4 test/model work; not the same code-path problem." + "reason": "The kwargs passthrough fix in loss utils is unrelated to Mistral4 test/model updates." }, { "left": "pull_request:43424", "right": "pull_request:44827", "accept": false, - "reason": "Executorch dynamic-shape test vs Mistral4 fixes; unrelated changes." + "reason": "Dynamic-shape export tests and Mistral4 test/model changes are not the same concrete issue." }, { "left": "pull_request:43424", "right": "pull_request:43823", "accept": false, - "reason": "Executorch export test and MobileLLM model addition are unrelated." - }, - { - "left": "pull_request:43823", - "right": "pull_request:46662", - "accept": false, - "reason": "MobileLLM model addition vs SmolVLM video padding fix; different models and subsystems." + "reason": "Different model features/tests; no evidence they address the same underlying problem." }, { "left": "pull_request:43823", "right": "pull_request:45655", "accept": false, - "reason": "New MobileLLM model vs config resolution order fix; unrelated." + "reason": "Adding MobileLLM and fixing `cls.config` resolution are unrelated code changes." }, { "left": "pull_request:43823", "right": "pull_request:43995", "accept": false, - "reason": "New MobileLLM model vs Falcon output-collection refactor; different features." - }, - { - "left": "pull_request:43995", - "right": "pull_request:46662", - "accept": false, - "reason": "Falcon refactor and SmolVLM video padding are separate model-specific changes." + "reason": "MobileLLM model addition and Falcon output-collection refactor are different work items." }, { "left": "pull_request:43995", "right": "pull_request:45213", "accept": false, - "reason": "Falcon refactor vs model-creation skill draft; no shared bug/change." + "reason": "Falcon refactoring and the model-creation skill PR are unrelated." }, { "left": "pull_request:43995", "right": "pull_request:44827", "accept": false, - "reason": "Falcon standardized-output refactor and Mistral4 tests are different code paths." + "reason": "Falcon interface refactoring and Mistral4 test fixes are not the same underlying bug/change." }, { "left": "pull_request:44827", "right": "pull_request:45170", "accept": false, - "reason": "Mistral4 test fixes vs layrnorm->layernorm renaming; different underlying issues." + "reason": "Mistral4 test/model changes and `layrnorm -> layernorm` renaming across CLIP-like models are unrelated." }, { "left": "pull_request:44827", "right": "pull_request:46267", "accept": false, - "reason": "Mistral4 fixes vs chat parsing update; unrelated changes." + "reason": "Mistral4 fixes and chat response parsing updates target different subsystems and problems." }, { "left": "pull_request:44827", "right": "pull_request:46048", "accept": false, - "reason": "Mistral4 fixes vs TDT loss kernel addition; separate features." + "reason": "Mistral4 test/model updates and adding the TDT loss kernel are unrelated." }, { "left": "pull_request:44827", "right": "pull_request:45655", "accept": false, - "reason": "Mistral4 test fixes vs config resolution order fix; unrelated." + "reason": "These PRs address different issues: Mistral4 test/model changes vs config resolution order in `modeling_utils`." }, { "left": "pull_request:44827", "right": "pull_request:45213", "accept": false, - "reason": "Mistral4 changes vs model-creation skill draft; not a duplicate change." + "reason": "A model/test fix and a model-creation skill/documentation PR are not the same change." }, { "left": "pull_request:43823", "right": "pull_request:45213", "accept": false, - "reason": "New MobileLLM model vs model-creation skill draft; no same underlying bug/change." - }, - { - "left": "pull_request:45213", - "right": "pull_request:46662", - "accept": false, - "reason": "Model-creation skill draft vs SmolVLM video padding fix; unrelated." - }, - { - "left": "pull_request:45655", - "right": "pull_request:46662", - "accept": false, - "reason": "Config resolution order fix vs SmolVLM video padding; different problems." - }, - { - "left": "pull_request:44827", - "right": "pull_request:46662", - "accept": false, - "reason": "Mistral4 fixes vs SmolVLM video padding are unrelated." + "reason": "Adding MobileLLM and the model-creation skill are not duplicate fixes; one adds a model, the other adds workflow tooling." }, { "left": "pull_request:43251", "right": "pull_request:43254", "accept": true, - "reason": "Both patch src/transformers/loss/loss_utils.py to forward weight/label_smoothing through fixed_cross_entropy for the same issue 43240; they could plausibly be merged into one PR." + "reason": "Both PRs implement the same underlying fix in `fixed_cross_entropy`: forwarding supported kwargs such as `weight` and `label_smoothing` to `nn.functional.cross_entropy`. They are plausible mergeable duplicates." } ] }, "evaluator_result": { "accept": true, - "feedback": "The summary is grounded in the packet: only PR 43251/43254 share an explicit issue target and the same loss_utils.py fix, while the other pairings are conservatively rejected as different features/models/subsystems. The mergeability call for 43251 and 43254 is plausible for the same bugfix, and no soft-edge verdict appears overstated." + "feedback": "Grounded overall. The only supported duplicate/mergeable pair in the packet is #43251/#43254: both touch `src/transformers/loss/loss_utils.py`, share issue target #43240, and implement the same `fixed_cross_entropy` kwargs passthrough fix. The other soft-edge pairs are conservatively rejected as unrelated model additions, refactors, or test/docs changes." }, "evaluator_used": true, "retried": false, @@ -19957,12 +19763,12 @@ "soft_pairs": [ "pull_request:40546|pull_request:42781" ], - "prepared_review_unit_hash": "f28c5e08339ab19143904787db5f864dc0d48c3ac56f0d91c8e7b573ece6c64d", + "prepared_review_unit_hash": "e655b8ef7072f36c856a1f76bbb1afd37d88060fc8a99af76d047e897bbb6fda", "component_budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 7225, + "serialized_chars": 7226, "estimated_input_tokens": 1807, "estimated_eval_tokens": 3870 }, @@ -19970,7 +19776,7 @@ "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 7225, + "serialized_chars": 7226, "estimated_input_tokens": 1807, "estimated_eval_tokens": 3870 }, @@ -19979,18 +19785,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These are related VibeVoice PRs, but they target different scopes: a core VibeVoice implementation versus a later realtime extension with additional tokenizer/generation work. They are not duplicates.", - "confidence": 0.88, + "summary": "Both PRs are about the VibeVoice family, but they are not the same change: #40546 is the core VibeVoice implementation, while #42781 adds VibeVoice Realtime plus the acoustic tokenizer and generation-related follow-ups. They overlap by model family and shared auto-registration/docs, not by the same concrete fix.", + "confidence": 0.89, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 40546 is the broader base implementation (\u201cImplement VibeVoice\u201d), is not a draft, and has much higher review/activity. It looks like the primary upstream PR for the model family.", + "canonical_pr_reason": "#40546 is the better canonical PR for the cluster because it is the primary, more focused VibeVoice implementation and appears more mature/reviewed; #42781 is a later, broader realtime follow-up with draft status.", "best_issue_reason": null, - "best_pr_reason": "PR 40546 is the best canonical PR because it represents the main VibeVoice integration path, while 42781 is a draft add-on for realtime support and depends on a different feature slice.", + "best_pr_reason": "#40546 is the best single representative of the cluster topic: it directly matches the base VibeVoice implementation and is less ambiguous than the broader realtime extension in #42781.", "soft_edge_verdicts": [ { "left": "pull_request:40546", "right": "pull_request:42781", "accept": false, - "reason": "Do not merge: 40546 implements the base VibeVoice model, while 42781 adds VibeVoice Realtime plus acoustic tokenizer/generation changes. They share filenames because they touch the same package, but they are different concrete changes and would not plausibly be merged as one PR." + "reason": "Reject: these are related VibeVoice workstreams, but not the same underlying change. #40546 implements core VibeVoice, while #42781 extends the family with realtime support and the acoustic tokenizer; they are not plausibly one merged PR for the same code-path bug/change." } ] }, @@ -20020,12 +19826,12 @@ "soft_pairs": [ "pull_request:40755|pull_request:41224" ], - "prepared_review_unit_hash": "24067a35803181395a7e20bc55cc841e4a5c5f74c7eaebcb69bb6c8974a8a261", + "prepared_review_unit_hash": "2b88ca252ede4e1b2f42554c4aa0195d81a99a30fd65e01a1b60e973600fec09", "component_budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 4676, + "serialized_chars": 4675, "estimated_input_tokens": 1169, "estimated_eval_tokens": 2594 }, @@ -20033,7 +19839,7 @@ "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 4676, + "serialized_chars": 4675, "estimated_input_tokens": 1169, "estimated_eval_tokens": 2594 }, @@ -20042,18 +19848,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These are unrelated pull requests: one adds TimesFM covariate forecasting, the other adds DINOv3 image classification support. The soft similarity is superficial and not a duplicate signal.", + "summary": "These PRs are unrelated: one adds TimesFM covariate forecasting, the other adds DINOv3 image classification support. They do not fix the same bug or implement the same change.", "confidence": 0.99, "canonical_issue_reason": null, - "canonical_pr_reason": "No single canonical PR for a duplicate cluster here; the two PRs target different models and different code paths, so they should not be grouped as the same change.", + "canonical_pr_reason": null, "best_issue_reason": null, - "best_pr_reason": "If one PR must anchor the set, 40755 is the more substantial feature addition with broader implementation context, but it is not a duplicate of 41224.", + "best_pr_reason": "pull_request:41224 is the more self-contained and specific change set, with a clear model-classification feature addition and supporting code/tests/docs. It is not a duplicate of the TimesFM PR, just the stronger standalone PR in this cluster.", "soft_edge_verdicts": [ { "left": "pull_request:40755", "right": "pull_request:41224", "accept": false, - "reason": "Different models, different functionality, and different code paths; they do not look like the same underlying bug or change." + "reason": "Different models, different tasks, and different code paths: TimesFM forecasting with covariates versus DINOv3 image classification. They are not the same underlying change and should not be merged as duplicates." } ] }, @@ -20146,7 +19952,7 @@ "soft_pairs": [ "pull_request:42975|pull_request:42976" ], - "prepared_review_unit_hash": "52276b481baad2f2facc8d9d36881072ad99764c9272723f4232c49fe60c7bbf", + "prepared_review_unit_hash": "e0358ba1d19df304f352eb5529a44a95de7ae716cdcf1b6c565e9fec499a5c43", "component_budget": { "node_count": 2, "item_count": 2, @@ -20168,18 +19974,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Both PRs are GitHub Actions maintenance updates, but they address different toolchains and code paths: PR 42975 is a broad Node 24 compatibility refresh for general workflow actions, while PR 42976 is focused on Docker build/publish action bumps. Similar theme, not the same underlying change.", - "confidence": 0.93, + "summary": "Both PRs are broad GitHub Actions dependency bumps, but they target different action families and different workflow subsets. They are related maintenance, not the same underlying fix/change.", + "confidence": 0.94, "canonical_issue_reason": null, - "canonical_pr_reason": null, + "canonical_pr_reason": "PR 42975 is the cleaner canonical representative: it has a clearer, single purpose\u2014updating workflow actions for Node 24 compatibility across many files.", "best_issue_reason": null, - "best_pr_reason": "PR 42975 is the broader and more representative workflow-infrastructure upgrade, but it should not be treated as a duplicate of PR 42976 because the latter is specifically about docker/build-push-action version bumps.", + "best_pr_reason": "PR 42975 is the best standalone PR in the cluster because its intent is more explicit and it covers the broader workflow-action compatibility update.", "soft_edge_verdicts": [ { "left": "pull_request:42975", "right": "pull_request:42976", "accept": false, - "reason": "Different concrete changes: 42975 updates checkout/cache/upload-artifact across many workflows for Node 24 compatibility, while 42976 upgrades docker/build-push-action in Docker-related workflows. Same general maintenance area, but not the same underlying bug/change and not mergeable as one PR." + "reason": "Do not merge: 42975 upgrades core workflow actions like checkout/cache/upload-artifact for Node 24 compatibility, while 42976 bumps docker/build-push-action and a few other workflow actions. They touch overlapping workflow files but fix different dependency upgrades and are not the same concrete change." } ] }, @@ -20220,7 +20026,7 @@ "pull_request:45490|pull_request:46562", "pull_request:46562|pull_request:46658" ], - "prepared_review_unit_hash": "63fd53606e9a10f6803c80e09c176393fbc0b5b8ce2408a1f11fa2a3e813de24", + "prepared_review_unit_hash": "b51af974b91ce1dbdfa2a6b0329c972665671169b8c14a9ac3541cbe5378a61e", "component_budget": { "node_count": 6, "item_count": 6, @@ -20242,60 +20048,60 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "This cluster is mostly a set of unrelated model-addition PRs. The only plausible duplicate-like pair is the Step3p7 implementation PR and its mini-model/assets companion PR; the rest are different models entirely.", - "confidence": 0.91, + "summary": "This cluster is not a duplicate set: it contains several unrelated model-addition PRs that only share generic boilerplate (docs/auto mappings). The only near-match is the Step3p7 Flash pair, but the WIP mini-model/script PR is support material, not the same change as the model integration PR.", + "confidence": 0.97, "canonical_issue_reason": null, - "canonical_pr_reason": "No single canonical duplicate across the whole cluster: ViT NEPA, Molmo, Molmo2, CTSM, and Step3p7 are different model additions. If one PR must anchor the cluster, 46562 is the closest because it has the only near-duplicate companion (46658) and contains the core Step3p7 implementation.", + "canonical_pr_reason": "pull_request:46562 is the most central substantive change in the only near-overlap: it is the actual Step3p7 Flash model implementation with docs/tests, whereas the other PRs are separate model additions and pull_request:46658 is just a mini-model/debug/script bundle.", "best_issue_reason": null, - "best_pr_reason": "46562 is the best representative only because it is the core Step3p7 model PR; the other PRs are unrelated model adds, so none is globally suitable as a canonical duplicate.", + "best_pr_reason": "pull_request:46562 is the best representative PR for this cluster because it contains a complete concrete code-path change; the rest are different models or ancillary assets and should not be merged together as duplicates.", "soft_edge_verdicts": [ { "left": "pull_request:42978", "right": "pull_request:43448", "accept": false, - "reason": "Different model additions (ViT NEPA vs Molmo); shared docs/auto-registration files are just boilerplate, not the same change." + "reason": "Different models (ViT NEPA vs Molmo). Shared docs and auto-registration files are generic model-onboarding boilerplate, not the same underlying change." }, { "left": "pull_request:42978", "right": "pull_request:43451", "accept": false, - "reason": "Different model additions (ViT NEPA vs Molmo2); overlap is generic model-registration scaffolding only." + "reason": "Different models (ViT NEPA vs Molmo2). Overlap is only in shared framework registration/docs, which is too broad to treat as a duplicate PR." }, { "left": "pull_request:42978", "right": "pull_request:45490", "accept": false, - "reason": "Different model additions (ViT NEPA vs CTSM); same broad transformers repo area, but not the same underlying bug/change." + "reason": "Different models (ViT NEPA vs CTSM). They touch common onboarding files, but they implement unrelated model additions." }, { "left": "pull_request:43448", "right": "pull_request:43451", "accept": false, - "reason": "Molmo and Molmo2 are related by name but are distinct model families with different files and APIs; not a single duplicate PR." + "reason": "Molmo and Molmo2 are distinct model integrations with different files, configs, and tests; same naming family is not enough to merge them." }, { "left": "pull_request:43448", "right": "pull_request:45490", "accept": false, - "reason": "Molmo and CTSM are separate model integrations; shared auto/docs files are incidental." + "reason": "Molmo vs CTSM are unrelated model additions; shared docs/auto files are standard repo plumbing, not duplicate evidence." }, { "left": "pull_request:43451", "right": "pull_request:45490", "accept": false, - "reason": "Molmo2 and CTSM are different model additions, not the same concrete change." + "reason": "Molmo2 and CTSM are different model families and code paths; the overlap is only in generic registration/configuration files." }, { "left": "pull_request:45490", "right": "pull_request:46562", "accept": false, - "reason": "CTSM and Step3p7 are unrelated model integrations; they only share generic auto/docs touchpoints." + "reason": "CTSM and Step3p7 Flash are different models; similar broad model-addition structure does not indicate the same bug or change." }, { "left": "pull_request:46562", "right": "pull_request:46658", - "accept": true, - "reason": "Both PRs are about Step3p7 Flash; 46658 looks like the companion mini-model/assets/demo setup for the same model integration and could plausibly be merged into the main Step3p7 PR." + "accept": false, + "reason": "Both mention Step 3.7 Flash, but pull_request:46562 is the actual model integration while pull_request:46658 adds a mini model plus scripts/debug assets; they are not the same concrete change and would not plausibly be one PR." } ] }, @@ -20344,118 +20150,118 @@ "pull_request:46380|pull_request:46539", "pull_request:46380|pull_request:46579" ], - "prepared_review_unit_hash": "8fc8cbfabd70704e2aa184248b0831a26c273a925eacd175de2ac07ae6b96085", + "prepared_review_unit_hash": "76836826012d67fa074b38d3e13e125dd54c618b34be40d861123f894139890a", "component_budget": { "node_count": 8, "item_count": 8, "soft_pair_count": 14, - "serialized_chars": 23164, - "estimated_input_tokens": 5791, - "estimated_eval_tokens": 11838 + "serialized_chars": 23159, + "estimated_input_tokens": 5790, + "estimated_eval_tokens": 11836 }, "budget": { "node_count": 8, "item_count": 8, "soft_pair_count": 14, - "serialized_chars": 23164, - "estimated_input_tokens": 5791, - "estimated_eval_tokens": 11838 + "serialized_chars": 23159, + "estimated_input_tokens": 5790, + "estimated_eval_tokens": 11836 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These PRs are not duplicates: they span unrelated areas like Whisper decoding, model test files, FP8 dequantization, generation validation, CI infrastructure, and typo/doc cleanups. The soft similarity links are mostly superficial filename or wording overlap, so none of the proposed pairs should be merged as the same change.", + "summary": "The cluster is heterogeneous: it mixes unrelated PRs for Whisper tokenization/tests, tie_weights backward-compatibility, FP8 dequantization, doctest cleanup, generation watermarking validation, CI/runtime setup, and typo fixes. None of the soft edges look like the same underlying bug or change, so this is not a duplicate cluster.", "confidence": 0.97, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 46579 is the most coherent standalone maintenance change in the set: a broad but single-purpose typo cleanup across code paths. The other PRs are unrelated bug fixes or test/CI changes.", + "canonical_pr_reason": "No canonical PR: the items target different subsystems and code paths, with only superficial \"fix\"/cleanup similarity.", "best_issue_reason": null, - "best_pr_reason": "PR 46579 is the best representative PR for the cluster because it is self-contained, clearly scoped, and not dependent on a narrow runtime bug or flaky test setup.", + "best_pr_reason": "No best PR for deduplication: none of these PRs is a plausible merged-into-one representative for the others.", "soft_edge_verdicts": [ { "left": "pull_request:43291", "right": "pull_request:43665", "accept": false, - "reason": "Whisper tokenizer/test fix vs unrelated CLIP/ViT test edits; different models and different failure modes." + "reason": "Both touch tests, but they are for different models/code paths (Whisper tokenization vs CLIP/ViT test edits)." }, { "left": "pull_request:43291", "right": "pull_request:44815", "accept": false, - "reason": "Whisper decoding logic vs FP8 dequant/loading logic; no shared code path or underlying bug." + "reason": "Whisper decoding/test changes are unrelated to the FP8 dequant/loading fix." }, { "left": "pull_request:43291", "right": "pull_request:46380", "accept": false, - "reason": "Whisper tokenizer fix vs generation watermarking validation fix; unrelated subsystems." + "reason": "Whisper tokenizer logic and generation watermarking validation are different bugs in different subsystems." }, { "left": "pull_request:43498", "right": "pull_request:46579", "accept": false, - "reason": "Backward-compatibility shim for tie_weights vs a typo cleanup PR; not the same change." + "reason": "Tie-weights backward compatibility is unrelated to generic typo cleanup across many files." }, { "left": "pull_request:43665", - "right": "pull_request:44815", + "right": "pull_request:46539", "accept": false, - "reason": "Bogus test-file edits in CLIP/ViT vs FP8 dequantization fix; completely different bugs." + "reason": "Synthetic test assertion edits in model tests are unrelated to Phimoe CI memory-limit/setup changes." }, { "left": "pull_request:43665", - "right": "pull_request:46380", + "right": "pull_request:44815", "accept": false, - "reason": "Bogus test edits vs generation config/watermarking validation; unrelated code paths." + "reason": "These are different fixes: test-file edits versus FP8/model-loading behavior." }, { "left": "pull_request:43665", - "right": "pull_request:46539", + "right": "pull_request:46380", "accept": false, - "reason": "Bogus test edits vs Phimoe CI integration fix; not the same underlying issue." + "reason": "Only broad test-related similarity; one is test noise, the other is a generation config bug fix." }, { "left": "pull_request:43665", "right": "pull_request:46579", "accept": false, - "reason": "Bogus test edits vs code typo cleanup; different intent and scope." + "reason": "Model test edits do not match a codebase-wide typo cleanup." }, { "left": "pull_request:44815", - "right": "pull_request:46380", + "right": "pull_request:46539", "accept": false, - "reason": "FP8 dequantization/loading fix vs generation watermarking validation fix; unrelated functionality." + "reason": "FP8 dequantization logic and Phimoe integration CI changes do not share the same underlying bug." }, { "left": "pull_request:44815", - "right": "pull_request:46539", + "right": "pull_request:46380", "accept": false, - "reason": "FP8 kernel/load-path fix vs CI memory/integration fix; no common bug." + "reason": "Different concrete problems: FP8 padding/dequant versus watermarking config validation." }, { "left": "pull_request:44815", "right": "pull_request:46579", "accept": false, - "reason": "FP8 dequantization fix vs typo cleanup; only broad code churn overlaps." + "reason": "Shared filenames are incidental; the actual changes address unrelated code cleanup vs FP8/model-loading behavior." }, { "left": "pull_request:45114", "right": "pull_request:46579", "accept": false, - "reason": "Doctest markdown cleanup vs code typo fixes; similar wording theme but not the same change." + "reason": "Doctest formatting fixes are unrelated to generic typo cleanup in code files." }, { "left": "pull_request:46380", "right": "pull_request:46539", "accept": false, - "reason": "Generation watermarking validation fix vs Phimoe CI integration fix; unrelated issues." + "reason": "Generation watermarking validation is unrelated to Phimoe integration test CI/environment fixes." }, { "left": "pull_request:46380", "right": "pull_request:46579", "accept": false, - "reason": "Generation config/runtime fix vs typo cleanup; not a duplicate." + "reason": "A validation fix in generation config is not the same change as broad typo corrections." } ] }, @@ -20485,12 +20291,12 @@ "soft_pairs": [ "pull_request:43488|pull_request:43532" ], - "prepared_review_unit_hash": "213fe2a92020ff5f2251164d9168e6c0be3e8c6e095c12d112b273f519834e85", + "prepared_review_unit_hash": "f9c7c721b4328687472aacf50de261a9f55741a1aaa9c935edfbf2308bc7f62b", "component_budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 4562, + "serialized_chars": 4563, "estimated_input_tokens": 1141, "estimated_eval_tokens": 2538 }, @@ -20498,7 +20304,7 @@ "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 4562, + "serialized_chars": 4563, "estimated_input_tokens": 1141, "estimated_eval_tokens": 2538 }, @@ -20507,18 +20313,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These two pull requests are not duplicates: one is a repo-bot/check formatting change around setup/dependency metadata, while the other rewires comment/CI workflows and several testing/notification paths. They only overlap on a Vit test file, which is incidental.", + "summary": "The two pull requests are not duplicates: one is a setup/version/dependency bot-check change, while the other is a CI/workflow and testing utility change. They only share a test file and broadly related repo context, not the same underlying fix.", "confidence": 0.95, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 43532 is the better canonical representative only because it is the broader, more discussion-heavy change; PR 43488 is a narrow bot/check tweak with different intent and code paths.", + "canonical_pr_reason": "PR #43532 is the better representative of the cluster because it is the broader, more substantive repository-change PR; #43488 is a narrow bot-format check focused on setup/version metadata.", "best_issue_reason": null, - "best_pr_reason": "PR 43532 is the stronger standalone PR to keep as the representative artifact: it touches multiple workflows and testing utilities and has more substantive review/discussion, whereas 43488 is a small targeted setup/versioning change.", + "best_pr_reason": "PR #43532 is the most suitable representative among the PRs, but only as a loose cluster anchor, not as a duplicate target; it changes CI/workflow behavior and related testing utilities rather than the same concrete code-path as #43488.", "soft_edge_verdicts": [ { "left": "pull_request:43488", "right": "pull_request:43532", "accept": false, - "reason": "Different underlying changes: 43488 adjusts setup/versioning and test dependency metadata, while 43532 changes CI/comment workflow behavior plus testing/notification code. Shared Vit test filename is not enough to make them the same fix." + "reason": "Different changes: #43488 alters setup/version/dependency metadata for a bot-format check, while #43532 rewires CI/workflow jobs and testing utilities. Shared test-file touchpoint is incidental, not the same underlying bug or fix." } ] }, @@ -20548,7 +20354,7 @@ "soft_pairs": [ "pull_request:43612|pull_request:43613" ], - "prepared_review_unit_hash": "8167ef48c8a3dd24161ec44fdfd27dd29d33336a427b2369077a477d02f923ce", + "prepared_review_unit_hash": "c12590a45f886b39a2937557609572271bf846c9325b9520f5e09e7b7b239d8d", "component_budget": { "node_count": 2, "item_count": 2, @@ -20570,18 +20376,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These are two similar but distinct PRs: one adds a Promptable Concept Segmentation pipeline, the other adds a Promptable Visual Segmentation pipeline. They share scaffolding files and documentation patterns, but target different tasks/models and are not the same underlying change.", - "confidence": 0.94, + "summary": "Two open PRs add different segmentation pipelines: one for Promptable Concept Segmentation and one for Promptable Visual Segmentation. They share repository scaffolding and documentation patterns, but they target different features and model integrations, so they are not duplicates.", + "confidence": 0.95, "canonical_issue_reason": null, - "canonical_pr_reason": null, + "canonical_pr_reason": "PR #43613 is the stronger standalone candidate: it adds the visual segmentation pipeline with broader model/test coverage and a more complete implementation footprint. PR #43612 is a separate concept-segmentation feature, not the same change.", "best_issue_reason": null, - "best_pr_reason": "PR 43613 is the slightly broader and more complete pipeline addition, so it is the better representative if one PR must stand in for the cluster.", + "best_pr_reason": "PR #43613 is the best representative of the cluster because it is the more fully scoped pipeline addition and appears more production-ready on its own.", "soft_edge_verdicts": [ { "left": "pull_request:43612", "right": "pull_request:43613", "accept": false, - "reason": "Do not merge: the PRs implement different segmentation pipelines with different task scope and model/test integrations. Shared docs/registry plumbing is not enough to make them the same concrete change." + "reason": "Different underlying features. One adds Promptable Concept Segmentation, the other Promptable Visual Segmentation; they touch shared docs and pipeline registration, but the implementation targets and model/test files differ, so they should not be merged as duplicates." } ] }, @@ -20611,40 +20417,40 @@ "soft_pairs": [ "pull_request:43651|pull_request:43663" ], - "prepared_review_unit_hash": "3028328d3afc1e26d9f5941fd098f0860f53131626d22c745e76be63de887cbd", + "prepared_review_unit_hash": "14d039ea58d99dae80e8813aac965ddebad09b562a7d35f8121de09338f0aa57", "component_budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 4292, - "estimated_input_tokens": 1073, - "estimated_eval_tokens": 2402 + "serialized_chars": 4293, + "estimated_input_tokens": 1074, + "estimated_eval_tokens": 2404 }, "budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 4292, - "estimated_input_tokens": 1073, - "estimated_eval_tokens": 2402 + "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": "These two PRs both touch `trainer.py` but fix different Trainer extension points: one changes gradient-accumulation loss scaling, the other refactors signature-column filtering for PEFT/custom trainers. They are not the same underlying change and should not be merged as duplicates.", - "confidence": 0.96, + "summary": "Two open Trainer PRs touch the same file but fix different Trainer behaviors: gradient-accumulation loss scaling vs. input signature/column filtering. They are not duplicates.", + "confidence": 0.97, "canonical_issue_reason": null, - "canonical_pr_reason": null, + "canonical_pr_reason": "No issue is present; between the two PRs, #43651 is the more self-contained behavior change and a slightly better representative of the cluster.", "best_issue_reason": null, - "best_pr_reason": null, + "best_pr_reason": "#43651 is the stronger representative because it introduces a focused trainer behavior change in the training step, whereas #43663 is a separate refactor/override point for signature-column filtering.", "soft_edge_verdicts": [ { "left": "pull_request:43651", "right": "pull_request:43663", "accept": false, - "reason": "Different Trainer code paths and different behaviors: 43651 adds `_loss_is_scaled_for_ga` for loss normalization during training, while 43663 adds `_get_signature_columns` to control input-column filtering. Shared file/subsystem alone is not enough to treat them as the same PR." + "reason": "Different underlying code paths and goals: #43651 changes loss normalization during training, while #43663 adds a helper for signature-based column filtering. They do not appear to fix the same bug or be mergeable as one PR." } ] }, @@ -20674,12 +20480,12 @@ "soft_pairs": [ "pull_request:43656|pull_request:43842" ], - "prepared_review_unit_hash": "55ca7dae452b1c4201bd468ec49c1682367bbb4d4811a6e98faf8b1bd1f45706", + "prepared_review_unit_hash": "8779d23c85ff12a4aa208a41b5d1ab751369c4cc89be9bfa956a5e2192be2878", "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 }, @@ -20687,7 +20493,7 @@ "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 3663, + "serialized_chars": 3664, "estimated_input_tokens": 916, "estimated_eval_tokens": 2088 }, @@ -20696,18 +20502,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Two open PRs touch the same CLI file and both mention TypeAdapter-related NameErrors, but they fix different causes: one is a typing/forward-reference cleanup, the other is a pydantic-availability import issue.", - "confidence": 0.89, + "summary": "Both PRs touch the same CLI module and both deal with annotation-related NameErrors, but they fix different failure modes: one is about forward-reference typing in `serve.py`, the other is about `TypeAdapter` being unavailable when `pydantic` is missing. They are related, but not the same underlying bug.", + "confidence": 0.92, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 43842 is the cleaner representative because it targets a specific runtime failure mode: avoiding a TypeAdapter NameError when pydantic is not installed.", + "canonical_pr_reason": "PR 43842 is the more specific and concrete fix: it explicitly targets the `TypeAdapter` NameError when `pydantic` is not installed and makes the annotation safe at runtime. It has a clear issue link and a narrowly scoped change.", "best_issue_reason": null, - "best_pr_reason": "PR 43842 has the tighter, more concrete fix scope and a direct issue link; it is the better standalone representative of this cluster.", + "best_pr_reason": "PR 43842 is the best representative of the cluster because it addresses a clearly stated runtime bug with an explicit target issue and a focused patch in the affected code path.", "soft_edge_verdicts": [ { "left": "pull_request:43656", "right": "pull_request:43842", "accept": false, - "reason": "Same file and related symptom, but different underlying bugs/fixes: 43656 changes annotation handling with __future__ imports, while 43842 fixes importing/typing TypeAdapter when pydantic is absent. They would not naturally merge into one PR." + "reason": "Related but not duplicate: 43656 changes future-annotation handling and local model/processor type hints, while 43842 fixes `TypeAdapter` access when `pydantic` is absent. They address different NameError sources and would not naturally be merged as one bug fix." } ] }, @@ -20844,8 +20650,8 @@ "pull_request:44018|pull_request:44019", "pull_request:44019|pull_request:44071", "pull_request:44068|pull_request:44116", - "pull_request:44018|pull_request:44116", "pull_request:44068|pull_request:44722", + "pull_request:44018|pull_request:44116", "pull_request:44071|pull_request:44722", "pull_request:44018|pull_request:44085", "pull_request:44018|pull_request:44086", @@ -20865,8 +20671,8 @@ "pull_request:44017|pull_request:44071", "pull_request:44073|pull_request:44101", "pull_request:44073|pull_request:44161", - "pull_request:44056|pull_request:44101", "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", @@ -20899,12 +20705,12 @@ "pull_request:44044|pull_request:44129", "pull_request:44044|pull_request:44154" ], - "prepared_review_unit_hash": "0ad177b50909b7fa87c227d69a5b2146556b6b97bf1dd66434a35ef22c31f93b", + "prepared_review_unit_hash": "4356ea22d971a2dafecb4a0de8628b1638539cf48dda1db9fa8df759cf17a76d", "component_budget": { "node_count": 28, "item_count": 28, "soft_pair_count": 78, - "serialized_chars": 84895, + "serialized_chars": 84893, "estimated_input_tokens": 21224, "estimated_eval_tokens": 42704 }, @@ -20921,396 +20727,360 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a family of model-specific refactors for standardized output tracing, not a set of true duplicates. The only strong duplicate-style overlaps are within the GPT-Neo and GPT-J lines; the rest are parallel refactors in different models sharing the same umbrella issue.", - "confidence": 0.87, + "summary": "This cluster is mostly a set of independent model-specific output-tracing refactors under the same umbrella issue (#43979), not one duplicate bug/change. The only strong duplicate-looking edges are within the GPT-Neo pair and the GPT-J/CodeGen pair(s).", + "confidence": 0.91, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 44722 is the broadest and most complete GPT-J-family refactor here, with explicit issue linkage and coverage of both GPT-J and CodeGen output-tracing changes.", + "canonical_pr_reason": "PR 44066 is the clearest representative of the shared output-tracing refactor theme: it has an explicit issue link, touches the core GPT-J/CodeGen path, and shows the standardized decorator-based migration pattern. The other PRs are mostly separate model ports rather than duplicates of one another.", "best_issue_reason": null, - "best_pr_reason": "PR 44722 best represents the cluster because it is the most mature concrete implementation of the standardized output-tracing refactor and appears to subsume the narrower GPT-J-only variants.", + "best_pr_reason": "PR 44066 is the best overall cluster representative because it most clearly captures the umbrella refactor pattern while still being concrete and issue-linked; most other PRs are narrower, model-specific variants.", "soft_edge_verdicts": [ { "left": "pull_request:43996", "right": "pull_request:44085", "accept": false, - "reason": "Same umbrella topic, but different model families and code paths." + "reason": "Different model families and code paths (FNet/CVT vs GPT-J); same umbrella theme is not enough for duplicate PR triage." }, { "left": "pull_request:43996", "right": "pull_request:44044", "accept": false, - "reason": "Different models; both refactor output tracing but are not the same change." + "reason": "Different models (FNet/CVT vs DeBERTa-v2); both are output-tracing refactors, but not the same underlying change." }, { "left": "pull_request:44066", "right": "pull_request:44085", "accept": true, - "reason": "Both are GPT-J output-tracing refactors in the same code path; one is a narrower variant of the same patch family." + "reason": "Same GPT-J output-tracing refactor family on the same code path, with overlapping decorator migration and shared file-level changes." }, { "left": "pull_request:44007", "right": "pull_request:44072", "accept": false, - "reason": "Vision backbones only share the umbrella refactor theme, not the same concrete change." + "reason": "Different architectures and refactors (ResNet/RT-DETR/RegNet vs EfficientNet); broad standardized-output work is too generic to merge as duplicates." }, { "left": "pull_request:44072", "right": "pull_request:44722", "accept": false, - "reason": "Different model families and unrelated implementation details." + "reason": "Different models (EfficientNet vs GPT-J/CodeGen); same issue tag does not make these the same concrete change." }, { "left": "pull_request:44066", "right": "pull_request:44072", "accept": false, - "reason": "Different model families; not mergeable as one concrete fix." + "reason": "GPT-J/CodeGen output tracing vs EfficientNet output tracing are separate model-specific refactors." }, { "left": "pull_request:44013", "right": "pull_request:44044", "accept": false, - "reason": "MobileNetV2 vs DeBERTa; same refactor pattern, different code paths." + "reason": "MobileNetV2 and DeBERTa-v2 are unrelated code paths; only the refactor pattern is shared." }, { "left": "pull_request:44066", "right": "pull_request:44086", "accept": false, - "reason": "GPT-J/CodeGen refactor versus MGP-STR; not the same bug or change." + "reason": "GPT-J/CodeGen refactor and MGP-STR refactor are different model implementations and not the same bug/change." }, { "left": "pull_request:44066", "right": "pull_request:44071", "accept": false, - "reason": "GPT-J/CodeGen and MPT are separate implementations, not one duplicate change." + "reason": "GPT-J/CodeGen and MPT are distinct model refactors; same decorator transition is too broad to duplicate." }, { "left": "pull_request:44018", "right": "pull_request:44068", "accept": true, - "reason": "Same GPT-Neo output-tracing refactor against the same file and tests; plausible single PR." + "reason": "Both target GPT-Neo output tracing on the same model/file and appear to be the same standardized-refactor change." }, { "left": "pull_request:44066", "right": "pull_request:44068", "accept": false, - "reason": "Different model families; only the refactor pattern is shared." + "reason": "GPT-J/CodeGen vs GPT-Neo are different model code paths; no duplicate concrete change." }, { "left": "pull_request:44018", "right": "pull_request:44066", "accept": false, - "reason": "GPT-Neo and GPT-J/CodeGen are distinct code paths." + "reason": "Different models (GPT-Neo vs GPT-J/CodeGen); shared issue target is not enough to merge as duplicates." }, { "left": "pull_request:44019", "right": "pull_request:44722", "accept": false, - "reason": "ResNet output tracing and GPT-J/CodeGen output tracing are unrelated implementations." + "reason": "ResNet output-tracing refactor is unrelated to GPT-J/CodeGen." }, { "left": "pull_request:44019", "right": "pull_request:44085", "accept": false, - "reason": "ResNet and GPT-J are different model families and fixes." + "reason": "ResNet and GPT-J are unrelated model implementations, so these are not the same underlying change." }, { "left": "pull_request:44086", "right": "pull_request:44722", "accept": false, - "reason": "MGP-STR and GPT-J/CodeGen do not share the same concrete code path." + "reason": "MGP-STR and GPT-J/CodeGen are different code paths and model families." }, { "left": "pull_request:44018", "right": "pull_request:44019", "accept": false, - "reason": "GPT-Neo vs ResNet; same umbrella theme only." + "reason": "Both are output-tracing refactors, but GPT-Neo and ResNet are different concrete changes." }, { "left": "pull_request:44019", "right": "pull_request:44071", "accept": false, - "reason": "ResNet and MPT are unrelated model implementations." + "reason": "ResNet vs MPT; same general migration style, but not the same bug or patch." }, { "left": "pull_request:44068", "right": "pull_request:44116", "accept": false, - "reason": "GPT-Neo and Flaubert/XLM are different code paths." + "reason": "GPT-Neo is unrelated to Flaubert; shared refactor theme does not make them duplicates." }, { - "left": "pull_request:44018", - "right": "pull_request:44116", + "left": "pull_request:44068", + "right": "pull_request:44722", "accept": false, - "reason": "GPT-Neo and Flaubert are not the same underlying change." + "reason": "GPT-Neo and GPT-J/CodeGen are separate model-specific refactors." }, { - "left": "pull_request:44068", - "right": "pull_request:44722", + "left": "pull_request:44018", + "right": "pull_request:44116", "accept": false, - "reason": "Different model families; not a duplicate fix." + "reason": "GPT-Neo vs Flaubert; different concrete code paths." }, { "left": "pull_request:44071", "right": "pull_request:44722", "accept": false, - "reason": "MPT and GPT-J/CodeGen are separate refactors." + "reason": "MPT and GPT-J/CodeGen are unrelated model implementations." }, { "left": "pull_request:44018", "right": "pull_request:44085", "accept": false, - "reason": "GPT-Neo and GPT-J are distinct model code paths." + "reason": "GPT-Neo and GPT-J are different model refactors, despite both using the same output-tracing pattern." }, { "left": "pull_request:44018", "right": "pull_request:44086", "accept": false, - "reason": "GPT-Neo and MGP-STR are unrelated implementations." + "reason": "GPT-Neo vs MGP-STR; not the same underlying change." }, { "left": "pull_request:44068", "right": "pull_request:44086", "accept": false, - "reason": "GPT-Neo vs MGP-STR; same pattern, different fix." - }, - { - "left": "pull_request:44098", - "right": "pull_request:44101", - "accept": false, - "reason": "ViLT and Flaubert/XLM do not fix the same concrete bug." + "reason": "GPT-Neo and MGP-STR are distinct code paths." }, { - "left": "pull_request:44018", - "right": "pull_request:44071", + "left": "pull_request:44071", + "right": "pull_request:44086", "accept": false, - "reason": "GPT-Neo and MPT are different code paths." + "reason": "MPT and MGP-STR are unrelated models; no shared concrete fix." }, { "left": "pull_request:44059", "right": "pull_request:44086", "accept": false, - "reason": "GPT-2 and MGP-STR are unrelated model implementations." + "reason": "GPT-2 and MGP-STR are different models, so these are not duplicates." }, { "left": "pull_request:44085", "right": "pull_request:44086", "accept": false, - "reason": "GPT-J and MGP-STR are different changes, not duplicates." + "reason": "GPT-J and MGP-STR are unrelated code paths." }, { - "left": "pull_request:44017", - "right": "pull_request:44161", - "accept": false, - "reason": "SegFormer and LongT5 are separate model refactors." - }, - { - "left": "pull_request:44017", - "right": "pull_request:44073", + "left": "pull_request:44098", + "right": "pull_request:44101", "accept": false, - "reason": "SegFormer and VisualBERT differ in model and code path." + "reason": "ViLT vs Flaubert/XLM are different model families and changes." }, { - "left": "pull_request:44017", - "right": "pull_request:44722", + "left": "pull_request:44018", + "right": "pull_request:44071", "accept": false, - "reason": "Different model families; only the umbrella tracing work overlaps." + "reason": "GPT-Neo and MPT are different model-specific output-tracing refactors." }, { - "left": "pull_request:44068", + "left": "pull_request:44059", "right": "pull_request:44071", "accept": false, - "reason": "GPT-Neo and MPT are not the same underlying change." + "reason": "GPT-2 and MPT are separate model implementations." }, { - "left": "pull_request:43997", - "right": "pull_request:44073", + "left": "pull_request:44056", + "right": "pull_request:44101", "accept": false, - "reason": "RegNet and VisualBERT are unrelated fixes." + "reason": "MPNet vs Flaubert/XLM are unrelated model refactors." }, { - "left": "pull_request:44071", - "right": "pull_request:44086", + "left": "pull_request:44018", + "right": "pull_request:44059", "accept": false, - "reason": "MPT and MGP-STR are separate model implementations." + "reason": "GPT-Neo and GPT-2 are different code paths, so not the same change." }, { - "left": "pull_request:44073", - "right": "pull_request:44101", + "left": "pull_request:44059", + "right": "pull_request:44068", "accept": false, - "reason": "VisualBERT and Flaubert/XLM are different code paths." + "reason": "GPT-2 and GPT-Neo are distinct model refactors." }, { - "left": "pull_request:44056", - "right": "pull_request:44101", + "left": "pull_request:44101", + "right": "pull_request:44161", "accept": false, - "reason": "MPNet and Flaubert/XLM do not share the same concrete fix." + "reason": "Flaubert/XLM vs LongT5 are unrelated model-specific refactors." }, { "left": "pull_request:43997", "right": "pull_request:44017", "accept": false, - "reason": "RegNet/ResNet-family versus SegFormer; same theme, different models." + "reason": "RegNet and Segformer are different architectures; same standardized-output theme is too broad." }, { "left": "pull_request:43997", "right": "pull_request:44056", "accept": false, - "reason": "RegNet and MPNet are unrelated implementations." + "reason": "RegNet and MPNet are unrelated model refactors." }, { "left": "pull_request:44071", "right": "pull_request:44076", "accept": false, - "reason": "MPT and ImageGPT are separate model refactors." + "reason": "MPT vs ImageGPT are different model families and code paths." }, { "left": "pull_request:44074", "right": "pull_request:44101", "accept": false, - "reason": "TextNet and Flaubert/XLM are not the same change." + "reason": "TextNet and Flaubert/XLM are different concrete changes." }, { "left": "pull_request:44056", "right": "pull_request:44161", "accept": false, - "reason": "MPNet and LongT5 differ in model and code path." + "reason": "MPNet and LongT5 are unrelated model-specific patches." }, { "left": "pull_request:43997", "right": "pull_request:44018", "accept": false, - "reason": "RegNet and GPT-Neo are unrelated model fixes." - }, - { - "left": "pull_request:43997", - "right": "pull_request:44161", - "accept": false, - "reason": "RegNet and LongT5 are separate refactors." - }, - { - "left": "pull_request:44074", - "right": "pull_request:44161", - "accept": false, - "reason": "TextNet and LongT5 are different code paths." - }, - { - "left": "pull_request:44018", - "right": "pull_request:44161", - "accept": false, - "reason": "GPT-Neo and LongT5 are not duplicates." - }, - { - "left": "pull_request:43997", - "right": "pull_request:44071", - "accept": false, - "reason": "RegNet and MPT are separate model implementations." + "reason": "RegNet and GPT-Neo are different model implementations." }, { "left": "pull_request:44010", "right": "pull_request:44071", "accept": false, - "reason": "SqueezeBERT and MPT are unrelated fixes." + "reason": "SqueezeBert and MPT are unrelated code paths." }, { "left": "pull_request:44010", "right": "pull_request:44073", "accept": false, - "reason": "SqueezeBERT and VisualBERT are different models/code paths." + "reason": "SqueezeBert and VisualBert are different models; not the same concrete bug." }, { "left": "pull_request:44018", "right": "pull_request:44129", "accept": false, - "reason": "GPT-Neo and SpeechT5 are unrelated model refactors." + "reason": "GPT-Neo and SpeechT5 are unrelated refactors." }, { "left": "pull_request:43997", "right": "pull_request:44074", "accept": false, - "reason": "RegNet and TextNet are different implementations." + "reason": "RegNet and TextNet are different model families." }, { "left": "pull_request:43997", "right": "pull_request:44129", "accept": false, - "reason": "RegNet and SpeechT5 are not the same concrete change." + "reason": "RegNet and SpeechT5 are unrelated code paths." }, { "left": "pull_request:44071", "right": "pull_request:44129", "accept": false, - "reason": "MPT and SpeechT5 are separate model fixes." + "reason": "MPT and SpeechT5 are separate model-specific changes." }, { "left": "pull_request:43996", "right": "pull_request:44161", "accept": false, - "reason": "CVT/FNet and LongT5 are unrelated model code paths." + "reason": "FNet/CVT and LongT5 are unrelated models." }, { "left": "pull_request:43996", "right": "pull_request:44017", "accept": false, - "reason": "CVT/FNet and SegFormer are different model refactors." + "reason": "FNet/CVT and Segformer are different architectures." }, { "left": "pull_request:44010", "right": "pull_request:44044", "accept": false, - "reason": "SqueezeBERT and DeBERTa are distinct implementations." + "reason": "SqueezeBert and DeBERTa-v2 are unrelated code paths." }, { "left": "pull_request:44044", "right": "pull_request:44074", "accept": false, - "reason": "DeBERTa and TextNet are unrelated fixes." + "reason": "DeBERTa-v2 and TextNet are different model-specific refactors." }, { "left": "pull_request:44066", "right": "pull_request:44722", "accept": true, - "reason": "Same GPT-J family output-tracing refactor; 44722 appears to be the broader follow-up/superset." + "reason": "Both are GPT-J/CodeGen output-tracing refactors on the same files and same underlying migration." }, { "left": "pull_request:44085", "right": "pull_request:44722", "accept": true, - "reason": "Both target the same GPT-J output-tracing code path and could be merged into one PR." + "reason": "Both are GPT-J output-tracing refactors on the same model/file, with the same standardized decorator migration." }, { "left": "pull_request:44085", "right": "pull_request:44129", "accept": false, - "reason": "GPT-J and SpeechT5 are different model families." + "reason": "GPT-J and SpeechT5 are unrelated model implementations." }, { "left": "pull_request:43996", "right": "pull_request:44007", "accept": false, - "reason": "CVT/FNet and ResNet are separate model refactors." + "reason": "FNet/CVT and ResNet/RT-DETR/RegNet are different model families." }, { "left": "pull_request:43996", "right": "pull_request:44129", "accept": false, - "reason": "CVT/FNet and SpeechT5 do not share the same underlying fix." + "reason": "FNet/CVT and SpeechT5 are unrelated concrete changes." }, { "left": "pull_request:44044", "right": "pull_request:44129", "accept": false, - "reason": "DeBERTa and SpeechT5 are unrelated code paths." + "reason": "DeBERTa-v2 and SpeechT5 are different model-specific refactors." }, { "left": "pull_request:44044", "right": "pull_request:44154", "accept": false, - "reason": "DeBERTa and Vits are different model implementations." + "reason": "DeBERTa-v2 and ViTS are unrelated model code paths." } ] }, "evaluator_result": { "accept": true, - "feedback": "The summary is grounded in the packet: it correctly characterizes this as mostly model-specific output-tracing refactors rather than broad duplicates, and the highlighted GPT-Neo/GPT-J overlaps are the only ones that look plausibly mergeable into single PRs. The soft-edge accepts are conservative enough, especially for the GPT-Neo and GPT-J family pairs, and the rejection reasons for cross-family pairs stay within the evidence shown. " + "feedback": "The summary is grounded in the packet: it treats the cluster as mostly model-specific output-tracing refactors under the shared umbrella issue, and it only calls out duplicate-like edges where the same model/file is involved (GPT-Neo, GPT-J/CodeGen). The false-vs-true soft-edge verdicts are generally conservative and do not overstate cross-model similarity." }, "evaluator_used": true, "retried": false, @@ -21348,12 +21118,12 @@ "pull_request:43997|pull_request:44010", "pull_request:44010|pull_request:44129" ], - "prepared_review_unit_hash": "969adea0db5e1f2fd0e5d12503c05fbe90a2503ba03c23d0879e8fbe281b869a", + "prepared_review_unit_hash": "83677dbdca63af4822d45b4020420db3ac0f94cf94d7ddc0bef5bbeaf7f3c53d", "component_budget": { "node_count": 28, "item_count": 28, "soft_pair_count": 78, - "serialized_chars": 84895, + "serialized_chars": 84893, "estimated_input_tokens": 21224, "estimated_eval_tokens": 42704 }, @@ -21361,57 +21131,57 @@ "node_count": 8, "item_count": 8, "soft_pair_count": 6, - "serialized_chars": 17126, - "estimated_input_tokens": 4282, - "estimated_eval_tokens": 8820 + "serialized_chars": 17124, + "estimated_input_tokens": 4281, + "estimated_eval_tokens": 8818 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "All items are model-specific PRs that migrate output tracing/capture to a standardized decorator or hook pattern. They share the same refactor theme, but each targets a different architecture and code path, so they should not be treated as duplicates.", - "confidence": 0.93, + "summary": "All items are model-specific PRs in the same broad refactor wave toward standardized output tracing/capture_outputs, but they do not fix the same concrete bug or change the same code path. The cluster is best treated as related-by-theme, not duplicates.", + "confidence": 0.91, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 44129 is the strongest representative: it has an explicit issue link, a concrete SpeechT5 path, and a more clearly scoped output-tracing refactor than the others.", + "canonical_pr_reason": "PR 44129 is the most representative anchor: it has an explicit issue target and a concrete output-tracing refactor in SpeechT5, making it the clearest single PR to stand for this cluster theme.", "best_issue_reason": null, - "best_pr_reason": "PR 44129 is the best single representative of the cluster because it is the most fully specified and most directly tied to a tracked change, while still fitting the common output-capture migration theme.", + "best_pr_reason": "PR 44129 is the strongest standalone representative because it is tied to issue 43979 and has the most explicit problem framing; the others are similar refactors but on different models.", "soft_edge_verdicts": [ { "left": "pull_request:44024", "right": "pull_request:44076", "accept": false, - "reason": "Both are output-tracing refactors, but they apply to different model implementations (FocalNet vs ImageGPT) and different internal hooks; this is a shared pattern, not the same underlying change." + "reason": "Both are standardized-output refactors, but they touch different models (FocalNet vs ImageGPT) and distinct forward paths, so they are not the same fix." }, { "left": "pull_request:44073", "right": "pull_request:44074", "accept": false, - "reason": "VisualBert and TextNet are different models with different forward signatures and output plumbing; they are too model-specific to be one mergeable PR." + "reason": "VisualBert and TextNet are separate model implementations; this is the same refactor pattern, not one duplicate code-path change." }, { "left": "pull_request:44056", "right": "pull_request:44073", "accept": false, - "reason": "MPNet and VisualBert are separate code paths with different output-recording targets, so these are not the same bug or change." + "reason": "MPNet and VisualBert share output-tracing cleanup theme, but they are unrelated model-specific changes and would not be merged as one PR." }, { "left": "pull_request:44056", "right": "pull_request:44074", "accept": false, - "reason": "MPNet and TextNet only share the broad migration idea; the concrete implementations and affected model logic are different." + "reason": "Same broad decorator/capture_outputs migration, but different models and code paths; not a single underlying bug or change." }, { "left": "pull_request:43997", "right": "pull_request:44010", "accept": false, - "reason": "RegNet and SqueezeBert are unrelated model families; the diffs are similar refactors, but not the same underlying code-path fix." + "reason": "RegNet and SqueezeBert are separate model stacks; the PRs are similar refactors, not duplicates of the same fix." }, { "left": "pull_request:44010", "right": "pull_request:44129", "accept": false, - "reason": "SqueezeBert and SpeechT5 change different model internals and output behaviors; they are not plausibly mergeable into one PR as a single fix." + "reason": "SqueezeBert and SpeechT5 both adopt standardized output capture, but they address different model code paths and cannot plausibly be merged into one PR." } ] }, @@ -21491,268 +21261,274 @@ "pull_request:44028|pull_request:44030", "pull_request:44029|pull_request:44030" ], - "prepared_review_unit_hash": "6f735a9d100d3912199557bc41e908d1f283703086fb6296b962886e5b426b6a", + "prepared_review_unit_hash": "abcb4b2e2798401144f28fe2555c361dbe9291b4293758ad0b7031e813414cfb", "component_budget": { "node_count": 13, "item_count": 13, "soft_pair_count": 40, - "serialized_chars": 40174, - "estimated_input_tokens": 10044, - "estimated_eval_tokens": 20344 + "serialized_chars": 40168, + "estimated_input_tokens": 10042, + "estimated_eval_tokens": 20340 }, "budget": { "node_count": 13, "item_count": 13, "soft_pair_count": 40, - "serialized_chars": 40174, - "estimated_input_tokens": 10044, - "estimated_eval_tokens": 20344 + "serialized_chars": 40168, + "estimated_input_tokens": 10042, + "estimated_eval_tokens": 20340 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These PRs are all separate model-specific output-tracing refactors, not duplicates of one another. They share a naming/theme pattern but target different code paths and would not plausibly be merged into one PR.", - "confidence": 0.84, + "summary": "These are not duplicates; they are separate model-specific PRs all doing the same broad output-tracing refactor pattern in different code paths. I would not merge any pair as the same change.", + "confidence": 0.96, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 44025 is a good representative anchor for the cluster because it is a concrete output-tracing refactor affecting two closely related Depth Anything files, but it is still not a duplicate of the others.", + "canonical_pr_reason": "PR 44025 is the best representative among the set because it is a broader, clearly scoped output-tracing refactor across two related depth-anything files; there is no single true canonical PR for the whole cluster.", "best_issue_reason": null, - "best_pr_reason": "PR 44025 is the best representative of the theme, but only as an exemplar; it is not a cluster duplicate because the underlying model/code-path changes differ across PRs.", + "best_pr_reason": "PR 44025 is the strongest representative candidate: it is self-contained, touches a cohesive model family, and follows the cluster\u2019s main decorator-based output-tracing pattern.", "soft_edge_verdicts": [ { "left": "pull_request:43998", "right": "pull_request:44002", "accept": false, - "reason": "Different models and forward paths; same refactor pattern only." + "reason": "Same refactor theme, but different model code paths (`timm_backbone` vs `upernet`); not the same underlying change and not mergeable into one PR." }, { "left": "pull_request:43998", "right": "pull_request:44000", "accept": false, - "reason": "Distinct model code paths; shared output-tracing cleanup is too broad to merge as duplicates." + "reason": "Both are output-tracing refactors, but they target unrelated models (`timm_backbone` vs `vision_text_dual_encoder`) with different forward signatures." }, { "left": "pull_request:43998", "right": "pull_request:44001", "accept": false, - "reason": "Separate model implementation, not the same bug or change." + "reason": "Different models and different implementation details; shared refactor style is not enough to treat them as duplicates." }, { "left": "pull_request:43998", "right": "pull_request:44003", "accept": false, - "reason": "Different architecture and files; only a generic refactor theme overlaps." + "reason": "These touch separate model families (`timm_backbone` vs `mamba/falcon_mamba`) and are distinct code-path changes." }, { "left": "pull_request:43998", "right": "pull_request:44029", "accept": false, - "reason": "Different model family and code path; not plausibly one PR." + "reason": "Both are output-tracing cleanups, but they are in different architectures (`timm_backbone` vs `rwkv`) and not the same fix." }, { "left": "pull_request:43999", "right": "pull_request:44001", "accept": false, - "reason": "Both are output-tracing refactors, but for unrelated models." + "reason": "`mobilenet_v1` and `univnet` are unrelated model implementations; same refactor pattern, but not the same underlying bug/change." }, { "left": "pull_request:43999", "right": "pull_request:44002", "accept": false, - "reason": "Same theme, different model-specific implementation details." + "reason": "Different model families and distinct forward/output handling changes; not plausibly one merged PR." }, { "left": "pull_request:43999", "right": "pull_request:44025", "accept": false, - "reason": "Different model targets; not the same underlying fix." + "reason": "These are separate model-specific output-tracing refactors in different files and subsystems." }, { "left": "pull_request:43999", "right": "pull_request:44026", "accept": false, - "reason": "Encoder-decoder wrapper vs MobileNet V1 are unrelated code paths." + "reason": "`mobilenet_v1` and `vision_encoder_decoder` are unrelated; no shared concrete code-path problem." }, { "left": "pull_request:43999", "right": "pull_request:44027", "accept": false, - "reason": "Separate model refactors; no shared concrete bug." + "reason": "Same general refactor idea, but different encoder-decoder implementation and different change scope." }, { "left": "pull_request:44000", "right": "pull_request:44001", "accept": false, - "reason": "Different models; similarity is only in the refactor pattern." + "reason": "`vision_text_dual_encoder` and `univnet` are unrelated models; these are not duplicates." }, { "left": "pull_request:44000", "right": "pull_request:44002", "accept": false, - "reason": "Different forward implementations; not the same change." + "reason": "Different models and different forward signatures; too broad a similarity to merge as one PR." }, { "left": "pull_request:44000", "right": "pull_request:44003", "accept": false, - "reason": "Distinct subsystems and model logic; too broad to deduplicate." + "reason": "Separate model families (`vision_text_dual_encoder` vs `mamba/falcon_mamba`) with different output-capturing behavior." }, { "left": "pull_request:44000", "right": "pull_request:44025", "accept": false, - "reason": "Separate model families with different tracing updates." + "reason": "Both are output-tracing refactors, but they modify different model stacks and different concrete code paths." }, { "left": "pull_request:44001", "right": "pull_request:44002", "accept": false, - "reason": "UnivNet and UperNet are unrelated model paths." + "reason": "`univnet` and `upernet` are unrelated implementations; not the same bug or change." }, { "left": "pull_request:44001", "right": "pull_request:44003", "accept": false, - "reason": "Different architectures; not the same concrete bug or fix." + "reason": "Different architectures and different output-capture integrations; not mergeable into a single PR." }, { "left": "pull_request:44001", "right": "pull_request:44025", "accept": false, - "reason": "Different model-specific refactors; no single merged PR candidate." + "reason": "The refactor style matches, but the target models and call paths are different." }, { "left": "pull_request:44002", "right": "pull_request:44029", "accept": false, - "reason": "Different models and implementation surfaces; only pattern-level similarity." + "reason": "`upernet` and `rwkv` are unrelated code paths; the shared output-tracing theme is too broad to be a duplicate." }, { "left": "pull_request:44002", "right": "pull_request:44003", "accept": false, - "reason": "No shared underlying code-path issue; only generic output-tracing cleanup." + "reason": "Different model families and different refactor mechanics; not the same concrete fix." }, { "left": "pull_request:44003", "right": "pull_request:44028", "accept": false, - "reason": "Falcon/Mamba vs SuperPoint are unrelated targets." + "reason": "`mamba/falcon_mamba` and `superpoint` are unrelated; these are separate model-specific changes." }, { "left": "pull_request:44003", "right": "pull_request:44004", "accept": false, - "reason": "Different model families and concrete changes; not duplicates." + "reason": "`mamba` and `codegen` are different code paths with different API changes, so they are not duplicates." }, { "left": "pull_request:43998", "right": "pull_request:44004", "accept": false, - "reason": "Different model code paths; shared refactor theme is insufficient." + "reason": "`timm_backbone` and `codegen` change unrelated forward/output contracts; same refactor language, different underlying work." }, { "left": "pull_request:44001", "right": "pull_request:44004", "accept": false, - "reason": "UnivNet and CodeGen have unrelated forward paths." + "reason": "Separate models and separate output behavior; not the same underlying code-path problem." }, { "left": "pull_request:44002", "right": "pull_request:44004", "accept": false, - "reason": "Different model implementations; not one underlying bug." + "reason": "`upernet` and `codegen` are unrelated implementations, so they should not be merged as duplicates." + }, + { + "left": "pull_request:44000", + "right": "pull_request:44004", + "accept": false, + "reason": "Different model families and different signatures; this is only thematic similarity." }, { "left": "pull_request:44025", "right": "pull_request:44026", "accept": false, - "reason": "Depth Anything refactor vs vision-encoder-decoder wrapper are different changes." + "reason": "`depth_anything` and `vision_encoder_decoder` are different subsystems; not the same concrete change." }, { "left": "pull_request:44025", "right": "pull_request:44027", "accept": false, - "reason": "Different model families and code paths." + "reason": "These are separate encoder-decoder/depth model refactors, not one shared code-path fix." }, { "left": "pull_request:44025", "right": "pull_request:44028", "accept": false, - "reason": "Separate model implementations; only the refactor style matches." + "reason": "Different model families and different output-capture implementations; not duplicates." }, { "left": "pull_request:44025", "right": "pull_request:44029", "accept": false, - "reason": "Different model-specific output tracing changes." + "reason": "`depth_anything` and `rwkv` have distinct internals and cannot be treated as one merged PR." }, { "left": "pull_request:44025", "right": "pull_request:44030", "accept": false, - "reason": "Depth Anything and DPR are unrelated concrete changes." + "reason": "Different models (`depth_anything` vs `dpr`) and different output-return behavior; not the same change." }, { "left": "pull_request:44026", - "right": "pull_request:44027", + "right": "pull_request:44029", "accept": false, - "reason": "Vision encoder-decoder vs speech encoder-decoder are distinct code paths." + "reason": "`vision_encoder_decoder` and `rwkv` are unrelated model implementations with no shared concrete bug." }, { "left": "pull_request:44026", - "right": "pull_request:44028", + "right": "pull_request:44027", "accept": false, - "reason": "Different models and different implementations." + "reason": "Different wrapper models and different API refactors; not a duplicate pair." }, { "left": "pull_request:44026", - "right": "pull_request:44029", + "right": "pull_request:44028", "accept": false, - "reason": "No shared concrete bug; only generic tracing refactor overlap." + "reason": "These touch unrelated models and should not be collapsed together." }, { "left": "pull_request:44026", "right": "pull_request:44030", "accept": false, - "reason": "Unrelated model code paths; not mergeable as one PR." + "reason": "`vision_encoder_decoder` and `dpr` are separate code paths; only the broad refactor theme is shared." }, { "left": "pull_request:44027", "right": "pull_request:44028", "accept": false, - "reason": "Different model families and different output semantics." + "reason": "`speech_encoder_decoder` and `superpoint` are unrelated and not the same underlying issue." }, { "left": "pull_request:44027", "right": "pull_request:44029", "accept": false, - "reason": "Speech encoder-decoder and RWKV are unrelated changes." + "reason": "Different model families and different output handling changes; not mergeable as one PR." }, { "left": "pull_request:44027", "right": "pull_request:44030", "accept": false, - "reason": "Different models; same broad refactor theme only." + "reason": "`speech_encoder_decoder` and `dpr` are distinct code paths, so this is not a duplicate." }, { "left": "pull_request:44028", "right": "pull_request:44029", "accept": false, - "reason": "SuperPoint and RWKV touch unrelated code paths." + "reason": "`superpoint` and `rwkv` are unrelated models; same refactor style is not enough." }, { "left": "pull_request:44028", "right": "pull_request:44030", "accept": false, - "reason": "Different model implementations; not the same duplicate fix." + "reason": "Different implementations and different forward/output semantics; not the same change." }, { "left": "pull_request:44029", "right": "pull_request:44030", "accept": false, - "reason": "RWKV and DPR are separate model-specific refactors." + "reason": "`rwkv` and `dpr` are separate model-specific changes with no shared concrete code path." } ] }, @@ -21782,12 +21558,12 @@ "soft_pairs": [ "pull_request:44662|pull_request:45401" ], - "prepared_review_unit_hash": "8e3e4ca2b60c3809835ee555952ce15095009dc6455a52922830ac05d14215d3", + "prepared_review_unit_hash": "cf2c20a68a5b137faad39befa23698943f80a84fcf037d6da05c529902dfa11c", "component_budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 6552, + "serialized_chars": 6551, "estimated_input_tokens": 1638, "estimated_eval_tokens": 3532 }, @@ -21795,7 +21571,7 @@ "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 6552, + "serialized_chars": 6551, "estimated_input_tokens": 1638, "estimated_eval_tokens": 3532 }, @@ -21804,18 +21580,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Both items are independent model-integration pull requests, one for PenguinVL and one for Voxtral TTS. They share only broad Transformers scaffolding files and docs/AutoMap wiring, but they target different models, different code paths, and different release goals.", - "confidence": 0.97, + "summary": "Both items are separate model-addition PRs for different architectures: PenguinVL and Voxtral TTS. They share Transformers boilerplate files, but the code paths, filenames, and targets are distinct, so they should not be clustered as duplicates.", + "confidence": 0.96, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 45401 is the stronger canonical choice because it is more tightly scoped and already linked to an explicit tracking issue, making it the clearer representative for this cluster of model-addition work.", + "canonical_pr_reason": "PR #44662 is the more canonical representative because it is a mature, heavily reviewed, non-draft implementation PR for a complete new model integration.", "best_issue_reason": null, - "best_pr_reason": "PR 45401 is the best PR representative: it has an explicit issue target, a focused TTS-specific implementation, and a narrower change set than the larger PenguinVL addition.", + "best_pr_reason": "PR #44662 is the strongest standalone PR in the set: it has substantially more review activity and a fuller implementation footprint, so it is the better representative even though it is not a duplicate of #45401.", "soft_edge_verdicts": [ { "left": "pull_request:44662", "right": "pull_request:45401", "accept": false, - "reason": "Different underlying models and integrations (PenguinVL vs Voxtral TTS). The overlap is only in generic Transformers registration/docs files, which is not enough to treat them as the same change." + "reason": "These are different model integrations (PenguinVL vs Voxtral TTS) with different model directories, configs, processors, and tests. Shared auto-registration and docs boilerplate is not enough to treat them as the same underlying change." } ] }, @@ -21845,12 +21621,12 @@ "soft_pairs": [ "pull_request:44979|pull_request:45363" ], - "prepared_review_unit_hash": "0faf3c700e88bd0f1a65f9e35a27bc9531bc39550bd7e806491715211e0c05f5", + "prepared_review_unit_hash": "aaa7499296e442d078007fdc909efa5e857a0c46bb8e6df8d7bea4c3a034f6c8", "component_budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 4406, + "serialized_chars": 4407, "estimated_input_tokens": 1102, "estimated_eval_tokens": 2460 }, @@ -21858,7 +21634,7 @@ "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 4406, + "serialized_chars": 4407, "estimated_input_tokens": 1102, "estimated_eval_tokens": 2460 }, @@ -21867,18 +21643,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These two PRs are in the same broad area (module/kernel fusion) but they target different layers: one introduces a general Module Fusion API, while the other adds a specific n-to-1 kernel fusion path via KernelConfig and supporting integration code. They do not look like the same change and should not be merged as duplicates.", - "confidence": 0.94, + "summary": "These two pull requests are related to fusion, but they implement different layers of the feature: one introduces a generic Module Fusion API, while the other adds KernelConfig-based n-to-1 kernel fusion wiring and usage. They are conceptually adjacent, not duplicates.", + "confidence": 0.92, "canonical_issue_reason": null, - "canonical_pr_reason": "PR #44979 is the better canonical PR because it introduces the broader foundational module fusion API that the other PR appears to build on.", + "canonical_pr_reason": "No clear canonical PR: PR #44979 adds the underlying Module Fusion API, while PR #45363 adds a separate KernelConfig-driven fusion integration path. They overlap in theme but are different deliverables.", "best_issue_reason": null, - "best_pr_reason": "PR #44979 is the best representative of the cluster since it is the more general, reusable change rather than a specific KernelConfig-based application.", + "best_pr_reason": "PR #45363 is the better representative of the cluster because it is the more end-to-end, user-facing feature implementation, touching loading/integration code and including an example. PR #44979 is more of a lower-level API foundation.", "soft_edge_verdicts": [ { "left": "pull_request:44979", "right": "pull_request:45363", "accept": false, - "reason": "Related fusion work, but not the same underlying change: #44979 adds a general module fusion API, whereas #45363 implements a specific n-to-1 kernel fusion mechanism with KernelConfig and example/integration updates. They would not plausibly be merged into one PR as duplicates." + "reason": "Not the same concrete change: #44979 introduces a new generic module-fusion API and tests, while #45363 implements KernelConfig-based n-to-1 kernel fusion integration across model loading and utilities. Related, but not mergeable as one duplicate PR." } ] }, @@ -21908,7 +21684,7 @@ "soft_pairs": [ "pull_request:45086|pull_request:45317" ], - "prepared_review_unit_hash": "380e447c900dd43e3275a829a838ff5d5c6938798127c751820742551d39c438", + "prepared_review_unit_hash": "eb3ba161859ec2dd84e2525c04c840573238e671bcaac2436e4ec1dac6826a37", "component_budget": { "node_count": 2, "item_count": 2, @@ -21930,18 +21706,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Both PRs fix the same Mistral regex patch AttributeError in the same code path, with nearly identical code changes; 45317 is the more complete candidate because it adds tests.", - "confidence": 0.97, + "summary": "Both pull requests address the same AttributeError in `_patch_mistral_regex` by switching from `backend_tokenizer.pre_tokenizer` to `tokenizer.pre_tokenizer` in the same file and targeting the same issue. They look like duplicate fixes for the same underlying bug; `45317` is the stronger PR because it includes tests.", + "confidence": 0.98, "canonical_issue_reason": null, - "canonical_pr_reason": "pull_request:45317 is the stronger canonical PR: it addresses the same bug as 45086 but includes accompanying tests and a clearer title, making it the more complete fix.", + "canonical_pr_reason": "`pull_request:45317` is the better canonical PR: it fixes the same code path as `45086` but also adds test coverage and appears to be the more complete treatment of the bug.", "best_issue_reason": null, - "best_pr_reason": "pull_request:45317 is the best PR overall because it appears to be the same fix as 45086 plus test coverage, so it is the most merge-ready representation of the change.", + "best_pr_reason": "`pull_request:45317` is the best overall PR choice because it addresses the same concrete AttributeError and includes validation via tests, making it the most suitable representative fix.", "soft_edge_verdicts": [ { "left": "pull_request:45086", "right": "pull_request:45317", "accept": true, - "reason": "Same underlying AttributeError fix in _patch_mistral_regex, same target issue, same file, and identical code-path change; 45317 just adds tests." + "reason": "Same underlying bug and same fix in the same tokenizer code path; the second PR is essentially a more complete version of the first." } ] }, @@ -21975,52 +21751,52 @@ "pull_request:45569|pull_request:46565", "pull_request:46332|pull_request:46565" ], - "prepared_review_unit_hash": "0ba8647714ba2371c1a6265456ada7eb325705873e7d95df8b10665f0cbe74f8", + "prepared_review_unit_hash": "1a42cdc71c27141ca0eed215963803f524aff935582de3c63d02dd6e74ebfed2", "component_budget": { "node_count": 4, "item_count": 4, "soft_pair_count": 3, - "serialized_chars": 13473, - "estimated_input_tokens": 3369, - "estimated_eval_tokens": 6994 + "serialized_chars": 13471, + "estimated_input_tokens": 3368, + "estimated_eval_tokens": 6992 }, "budget": { "node_count": 4, "item_count": 4, "soft_pair_count": 3, - "serialized_chars": 13473, - "estimated_input_tokens": 3369, - "estimated_eval_tokens": 6994 + "serialized_chars": 13471, + "estimated_input_tokens": 3368, + "estimated_eval_tokens": 6992 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These are all Nemotron-related PRs, but they target different model families and change sets: NemotronH refactor, NemotronH dense/sparse split, Nemotron ASR streaming, and Nemotron 3.5 ASR streaming. They are related by naming and shared auto/docs plumbing, but not duplicates of the same underlying fix or feature.", - "confidence": 0.91, + "summary": "The cluster is heterogeneous: one NemotronH refactor PR, one NemotronH dense/sparse architecture/docs PR, and two separate ASR streaming feature PRs for different Nemotron variants. None of the soft pairs look like the same concrete change, so all duplicate links should be rejected.", + "confidence": 0.94, "canonical_issue_reason": null, - "canonical_pr_reason": null, + "canonical_pr_reason": "No single canonical PR: the items cover distinct changes (NemotronH refactor, NemotronH dense/sparse split, Nemotron ASR streaming, and Nemotron 3.5 ASR streaming) rather than one duplicate fix or feature.", "best_issue_reason": null, - "best_pr_reason": "PR 45569 is the broadest and most central NemotronH change: it introduces the dense/sparse split, updates docs and auto mappings, and includes the corresponding model/config files. If a single PR must anchor this cluster, it is the most representative, but it is not a duplicate of the others.", + "best_pr_reason": "PR 46332 is the most self-contained, broad feature PR in the cluster, but it is still distinct from the others; it is the closest thing to a representative standalone change.", "soft_edge_verdicts": [ { "left": "pull_request:45115", "right": "pull_request:45569", "accept": false, - "reason": "Both touch NemotronH internals, but 45115 is a refactor/inheritance change while 45569 adds the dense/sparse model split and docs. Overlap is superficial; they are not the same concrete change." + "reason": "Related NemotronH work, but not the same change: 45115 is a refactor/inheritance and conversion-mapping cleanup, while 45569 adds the dense/sparse split, docs, and broader model plumbing." }, { "left": "pull_request:45569", "right": "pull_request:46565", "accept": false, - "reason": "They share Nemotron naming and auto/docs files, but 45569 is about NemotronH dense/sparse modeling, while 46565 is Nemotron 3.5 ASR streaming. Different model families and code paths." + "reason": "Different model families and code paths: 45569 is NemotronH architecture/refactor work, while 46565 adds Nemotron 3.5 ASR streaming." }, { "left": "pull_request:46332", "right": "pull_request:46565", "accept": false, - "reason": "Both are ASR streaming PRs, but they target different model versions/families (Nemotron ASR vs Nemotron 3.5 ASR) with distinct implementation scope. Not mergeable as one duplicate change." + "reason": "Both are ASR streaming features, but for different concrete models (Nemotron ASR vs Nemotron 3.5 ASR) with overlapping infrastructure rather than the same underlying bug/change." } ] }, @@ -22050,7 +21826,7 @@ "soft_pairs": [ "pull_request:45332|pull_request:45333" ], - "prepared_review_unit_hash": "37c0c5d419de29a4dabbc573f78f1bd43d89df4c5f01db104ec4d2d4ad7571ad", + "prepared_review_unit_hash": "d9781ea5705a6e32105acf04c26c2c6f18292e9162c2081cda3a9b3cc295e5f3", "component_budget": { "node_count": 2, "item_count": 2, @@ -22072,18 +21848,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Related but not duplicates: one PR adds per-layer heterogeneous config plumbing, the other adds heterogeneous model/cache integration built on top of that config support.", - "confidence": 0.89, + "summary": "Both PRs are part of the same heterogeneous-modeling effort, but they are not the same change: #45333 is config-only foundation work, while #45332 is a much broader modeling/cache/masking implementation. Shared files reflect common infrastructure, not a single duplicate fix.", + "confidence": 0.9, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 45332 is the broader end-to-end feature change, covering model/cache integration in addition to heterogeneous support plumbing; PR 45333 is the narrower config foundation.", + "canonical_pr_reason": "#45333 is the more focused PR and best represents the shared per-layer configuration work; #45332 is a larger umbrella change that builds on that foundation with additional modeling behavior.", "best_issue_reason": null, - "best_pr_reason": "PR 45332 is the best representative of the cluster because it implements the larger user-facing heterogeneous model support, while 45333 is supporting configuration infrastructure.", + "best_pr_reason": "#45333 is the best representative for the cluster because it is the narrower, cleaner config-support change, whereas #45332 spans extra code paths beyond the shared configuration layer.", "soft_edge_verdicts": [ { "left": "pull_request:45332", "right": "pull_request:45333", "accept": false, - "reason": "Related feature work, but not the same concrete change: 45333 introduces per-layer config support, while 45332 consumes that support to change cache/modeling behavior. They are dependent pieces, not duplicate PRs." + "reason": "Reject: they overlap on heterogeneous-config infrastructure, but #45332 adds broader model/cache/masking changes and is not the same concrete code-path fix as the config-only #45333. They would not be merged as one duplicate PR." } ] }, @@ -22113,7 +21889,7 @@ "soft_pairs": [ "pull_request:45415|pull_request:45425" ], - "prepared_review_unit_hash": "531c4c0851a61efb44016f59e377e303250ae8b95fffc19720e1028b661645d5", + "prepared_review_unit_hash": "02b0e53f0813d936c310d1e527f7d511f333f361248c9d6d115b99cd5b185ab4", "component_budget": { "node_count": 2, "item_count": 2, @@ -22135,18 +21911,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Both items are typing-focused PRs in transformers, but they do not look like the same change. PR 45415 is a broad repository-wide type-checking sweep, while PR 45425 adds specific typing aliases/protocols for model/PEFT integration paths. Their overlap in a few files looks incidental, not like a single mergeable fix.", - "confidence": 0.93, + "summary": "Both pull requests are typing-related and overlap in a few files, but they address different changes: PR 45415 is a broad repository-wide type-checking sweep with config/ignore adjustments, while PR 45425 adds specific typing protocols and compatibility shims for model utilities. They do not look like the same underlying bug or change.", + "confidence": 0.91, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 45425 is the more focused and cohesive change: it adds explicit typing support needed by specific code paths, rather than a broad type-checking pass across many unrelated modules.", + "canonical_pr_reason": "PR 45425 is the better representative artifact because it is a concrete, narrowly scoped typing fix with specific code-level compatibility changes, rather than a broad type-checking cleanup pass.", "best_issue_reason": null, - "best_pr_reason": "PR 45425 is the best representative of the cluster because it is narrowly scoped and centered on one typing problem domain; PR 45415 is too broad and heterogeneous to serve as a clean canonical duplicate candidate.", + "best_pr_reason": "PR 45425 is the most useful standalone PR in the cluster: it has a clear technical goal, focused implementation, and is easier to understand as a single change set.", "soft_edge_verdicts": [ { "left": "pull_request:45415", "right": "pull_request:45425", "accept": false, - "reason": "Reject: they share a typing theme and a few filenames, but 45415 is a wide type-checking cleanup across many files, whereas 45425 adds targeted typing helpers for specific model/integration paths. They do not appear to fix the same concrete bug or be mergeable into one PR." + "reason": "Shared typing-related files are incidental; 45415 is a broad type-checking enablement/ignore update, while 45425 adds specific typing protocols and compatibility helpers. They are not the same concrete fix and would not naturally merge into one PR." } ] }, @@ -22176,7 +21952,7 @@ "soft_pairs": [ "pull_request:45513|pull_request:46084" ], - "prepared_review_unit_hash": "2e18457e5868398df64d6f33280435b5dd81f326f96a66b24790c9244ad8f3a1", + "prepared_review_unit_hash": "f77adfa46a4f3839b175af507a62f86460ade023c5bed60aa03ce5b9099271f4", "component_budget": { "node_count": 2, "item_count": 2, @@ -22198,18 +21974,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These are two unrelated pull requests fixing different cached-forward bugs in different model families: Qwen3.5/GDN linear attention vs. Mamba2 use_cache with seq_len > 1. They should not be clustered as duplicates.", - "confidence": 0.97, + "summary": "These PRs are only superficially similar: both mention cached multi-token forward paths, but they target different model families and different concrete bugs. #45513 fixes Qwen3.5/Olmo hybrid linear-attention caching behavior, while #46084 fixes Mamba2 use_cache behavior when seq_len > 1.", + "confidence": 0.96, "canonical_issue_reason": null, - "canonical_pr_reason": "pull_request:46084 is the cleaner representative: it has an explicit issue target, a narrowly scoped fix for a specific Mamba2 cached-forward seq_len>1 bug, and tests that directly cover that path.", + "canonical_pr_reason": "No single canonical PR: the two PRs address different code paths in different model implementations, so they are not duplicates of the same fix.", "best_issue_reason": null, - "best_pr_reason": "pull_request:46084 is the best standalone representative of its bug class because the change is focused, explicit about the failing condition, and easier to map to a single underlying defect than the broader multi-model PR.", + "best_pr_reason": "#45513 is the broader representative of the cached multi-token forward bug class because it spans several related model variants and includes direct test coverage, but it still should not be merged with #46084 as a duplicate.", "soft_edge_verdicts": [ { "left": "pull_request:45513", "right": "pull_request:46084", "accept": false, - "reason": "Different underlying bugs and code paths: 45513 fixes Qwen3.5/GDN linear-attention cached multi-token forward, while 46084 fixes Mamba2 use_cache when seq_len > 1. They do not look mergeable into one PR." + "reason": "Reject: they touch different architectures and fix different failure modes; the overlap is only at the high level of cached seq_len>1 inference." } ] }, @@ -22239,7 +22015,7 @@ "soft_pairs": [ "pull_request:45548|pull_request:45662" ], - "prepared_review_unit_hash": "2e8fa0695abeeb5ebe34614985e110a80c68942376131ff5f5c646d19d11efc1", + "prepared_review_unit_hash": "90dfe60dd2ec0085a9e7898353351a567d23e348be05be7db201a305b4b2b6a5", "component_budget": { "node_count": 2, "item_count": 2, @@ -22261,18 +22037,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Both PRs touch expert-parallel loading in Transformers, but they fix different failure modes: one preserves distributed config/EP loading state, the other prevents expert parameters from being overwritten after sharding. They are related in subsystem, not duplicates.", - "confidence": 0.93, + "summary": "These are related to expert-parallel loading, but they fix different bugs in different code paths: 45548 addresses serialization/loading of `distributed_config` and the `has_ep` check for DeepSpeed ZeRO-3/accelerate, while 45662 fixes FSDP2/rank-0 broadcast overwriting expert parameters after sharding. They are not duplicates.", + "confidence": 0.96, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 45548 is the broader and earlier loading-path fix, centered on serialized distributed_config handling and EP detection during model load, so it is the best representative of the cluster.", + "canonical_pr_reason": "No canonical PR: the two PRs target different expert-parallel loading failures and should not be collapsed as the same change.", "best_issue_reason": null, - "best_pr_reason": "PR 45548 is the clearest canonical representative because its title and diff directly address the core EP loading state issue, while 45662 targets a separate post-shard overwrite bug.", + "best_pr_reason": "45662 is the more concrete self-contained fix for a specific corruption bug (rank-0 broadcast overwriting experts), but it is still distinct from 45548's serialization/loading fix.", "soft_edge_verdicts": [ { "left": "pull_request:45548", "right": "pull_request:45662", "accept": false, - "reason": "Reject: the PRs address different concrete bugs. 45548 fixes EP/ZeRO-3 loading state and config serialization; 45662 fixes experts being overwritten after rank-0 broadcast/post-shard wrapping. Same area, not the same change." + "reason": "Different underlying bugs and code paths: 45548 is about `distributed_config` serialization and `has_ep` for ZeRO-3/accelerate, while 45662 prevents post-shard expert tensors from being overwritten during FSDP2 loading. They are related, not duplicates." } ] }, @@ -22302,40 +22078,40 @@ "soft_pairs": [ "pull_request:45639|pull_request:45645" ], - "prepared_review_unit_hash": "df020006cbef5e9ab8b43c86580983bd6f1f62212796b813e05a8523d8f230dc", + "prepared_review_unit_hash": "2e5216b16e218040cf8dd58b78b50653a3ec9e971d035c167a69404252c866f6", "component_budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 4588, - "estimated_input_tokens": 1147, - "estimated_eval_tokens": 2550 + "serialized_chars": 4589, + "estimated_input_tokens": 1148, + "estimated_eval_tokens": 2552 }, "budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 4588, - "estimated_input_tokens": 1147, - "estimated_eval_tokens": 2550 + "serialized_chars": 4589, + "estimated_input_tokens": 1148, + "estimated_eval_tokens": 2552 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Both PRs address the same xdist-related captured_info logging collision in Transformers testing utilities; PR 45645 appears to be the broader, more complete fix while 45639 is a narrower subset.", - "confidence": 0.94, - "canonical_issue_reason": null, - "canonical_pr_reason": "PR 45645 is the better canonical PR because it covers the core xdist-safe file handling in testing_utils and also updates CI/workflow log collection so all captured_info shards are preserved.", - "best_issue_reason": null, - "best_pr_reason": "PR 45645 is the strongest representative of the change set: it addresses the same underlying bug as 45639 but includes the broader operational fix needed to actually collect the worker-specific artifacts.", + "summary": "Both PRs address the same xdist collision problem around captured_info debug artifacts. PR 45639 fixes the output-file write path in testing_utils; PR 45645 includes that fix plus CI/workflow changes to read multiple per-worker files, making it the more complete implementation of the same change.", + "confidence": 0.91, + "canonical_issue_reason": "Issue 45561 appears to be the shared underlying xdist-safe captured_info artifact problem referenced by both PRs.", + "canonical_pr_reason": "PR 45645 is the better canonical PR because it contains the core xdist-safe file-write fix and the necessary CI/workflow follow-up to consume multiple captured_info files.", + "best_issue_reason": "No standalone issue is present in the cluster; the shared target 45561 is the relevant tracking issue.", + "best_pr_reason": "PR 45645 is the best representative because it covers the full end-to-end fix, not just the writer-side change.", "soft_edge_verdicts": [ { "left": "pull_request:45639", "right": "pull_request:45645", "accept": true, - "reason": "Same underlying bug: xdist collisions for captured_info artifacts in testing_utils. 45645 extends the same fix with workflow/log-collection updates, so these are plausibly mergeable into one PR." + "reason": "Both PRs fix the same concrete captured_info/xdist collision bug in testing_utils, and 45645 is an extension of the same fix with CI log aggregation." } ] }, @@ -22368,52 +22144,52 @@ "pull_request:45670|pull_request:46401", "pull_request:46320|pull_request:46401" ], - "prepared_review_unit_hash": "9217cf8d2e3386c62b55461f192f052b4e8cce1e3812b2e045a8a3c0d168a070", + "prepared_review_unit_hash": "463fe5ab393f1cc3120a174248214125cff5e55fd390dcef745720e3a013b3ae", "component_budget": { "node_count": 3, "item_count": 3, "soft_pair_count": 3, - "serialized_chars": 7524, - "estimated_input_tokens": 1881, - "estimated_eval_tokens": 4018 + "serialized_chars": 7525, + "estimated_input_tokens": 1882, + "estimated_eval_tokens": 4020 }, "budget": { "node_count": 3, "item_count": 3, "soft_pair_count": 3, - "serialized_chars": 7524, - "estimated_input_tokens": 1881, - "estimated_eval_tokens": 4018 + "serialized_chars": 7525, + "estimated_input_tokens": 1882, + "estimated_eval_tokens": 4020 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These are three open PRs: one unrelated AutoModel registration fix for GLM-ASR, and two separate attention-dispatch feature PRs for different model families. The two attn_implementation PRs are similar in wording but target different code paths and model stacks, so they should not be deduplicated.", - "confidence": 0.89, + "summary": "These are three separate PRs: one auto-model registration tweak and two model-specific attn_implementation dispatch additions. The latter two are related at a high level but target different model families and code paths, so they should not be merged as duplicates.", + "confidence": 0.95, "canonical_issue_reason": null, - "canonical_pr_reason": "PR #46401 is the strongest representative of the attn_implementation-dispatch theme because it covers multiple BLIP-family models and includes broader implementation/test/documentation changes.", + "canonical_pr_reason": "No clear duplicate canonical PR: 45670 is unrelated, and 46320/46401 are distinct model-specific dispatch implementations rather than the same concrete change.", "best_issue_reason": null, - "best_pr_reason": "PR #46401 is the best overall representative in this cluster: it is the most substantial and clearly scoped change among the similar PRs, while #45670 is unrelated and #46320 is narrower.", + "best_pr_reason": "46401 is the broadest/most complete of the attn-dispatch PRs, but it is still not a duplicate of 46320; it best represents its own change set rather than a cluster canonical.", "soft_edge_verdicts": [ { "left": "pull_request:45670", "right": "pull_request:46320", "accept": false, - "reason": "Different changes: #45670 adds GLM-ASR to AutoModel mappings, while #46320 implements attn_implementation dispatch for RecurrentGemma. They do not fix the same code-path problem." + "reason": "Unrelated changes: 45670 only updates AutoModel registration for glmasr/dia, while 46320 adds attn_implementation dispatch for RecurrentGemma in model code, docs, and tests." }, { "left": "pull_request:45670", "right": "pull_request:46401", "accept": false, - "reason": "Unrelated changes: #45670 is an AutoModel registry update for a new model, whereas #46401 adds attention-dispatch support for BLIP-family models." + "reason": "Unrelated changes: 45670 is an AutoModel mapping tweak, while 46401 adds attn_implementation dispatch for BLIP-2/InstructBLIP/InstructBLIPVideo plus docs/tests." }, { "left": "pull_request:46320", "right": "pull_request:46401", "accept": false, - "reason": "Both are attn_implementation PRs, but they target different model families and separate implementation paths (RecurrentGemma vs BLIP-2/InstructBLIP/InstructBLIPVideo). Similar feature area is not enough for duplicate PRs." + "reason": "Both are attn_implementation feature work, but they fix different model families with different code paths; they could not plausibly be merged into one duplicate PR." } ] }, @@ -22443,7 +22219,7 @@ "soft_pairs": [ "pull_request:45720|pull_request:46259" ], - "prepared_review_unit_hash": "fc89c6b6e013323f10d3c3b7196a00114cb4bc3c8084a34febc0dad8aa1978d8", + "prepared_review_unit_hash": "5900c32868f19603d2891623b9f0cb01b1e2c85f3fec422a621b8702a1a5f518", "component_budget": { "node_count": 2, "item_count": 2, @@ -22465,18 +22241,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "The two PRs are both linter-driven refactors in Transformers, but they target different rules and different code paths: TRF003 changes forward/return_dict behavior across model forwards, while TRF018 changes weight initialization logic across model classes. They are not the same underlying change, so the soft duplicate edge should be rejected.", + "summary": "These two PRs are both broad transformer model lint-rule changes, but they target different rules and different code paths: TRF003 tuple-return handling vs TRF018 model initialization cleanup. The overlap in model filenames looks incidental to the shared linter scope, not the same fix.", "confidence": 0.96, "canonical_issue_reason": null, - "canonical_pr_reason": null, + "canonical_pr_reason": "No canonical PR: the PRs implement different lint rules and fix different underlying behaviors, so they are not duplicates.", "best_issue_reason": null, - "best_pr_reason": null, + "best_pr_reason": "Neither PR is a strong global representative for a duplicate cluster, but 45720 is slightly more representative of the shared-file overlap because it touches more of the same model forward methods; still, it should not be treated as the same change as 46259.", "soft_edge_verdicts": [ { "left": "pull_request:45720", "right": "pull_request:46259", "accept": false, - "reason": "Different lint rules and unrelated fixes: PR 45720 implements TRF003 by adjusting model forward signatures/tuple returns, while PR 46259 implements TRF018 by refactoring _init_weights behavior. Overlap in touched model files is only due to broad refactoring scope, not the same bug or change." + "reason": "Different linter rules and different concrete code changes: one adjusts return-tuple handling in forwards, the other changes weight initialization logic. Shared model files alone are not enough to treat them as the same PR." } ] }, @@ -22572,7 +22348,7 @@ "soft_pairs": [ "pull_request:45932|pull_request:46100" ], - "prepared_review_unit_hash": "66d28c46e5b840c8dd97e43d484d103b8918a37e2c1b1e89d4e9a7e771ea0a49", + "prepared_review_unit_hash": "e6a1df079b92efa81ef68892c2bca1504cc8cdd271d8fa6d29ee733fa318f303", "component_budget": { "node_count": 2, "item_count": 2, @@ -22594,18 +22370,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "The two PRs are not duplicates: one is a CI/model code change, the other is a test migration. They share broad test-related context but not the same underlying bug or fix.", + "summary": "These two PRs are both test-related in a broad sense, but they target different changes: one is a noisy multi-file model/workflow update, the other is a focused ViT test migration. They do not look like the same underlying bug or change.", "confidence": 0.94, "canonical_issue_reason": null, - "canonical_pr_reason": "PR #45932 is the broader, more substantive change set and is the better cluster anchor; PR #46100 is mostly a test migration and is clearly separate.", + "canonical_pr_reason": "PR #46100 is the better representative because it is a single, coherent test-migration change with tightly scoped test-file edits, whereas PR #45932 is a broad mixed-change PR touching workflow plumbing and multiple model files.", "best_issue_reason": null, - "best_pr_reason": "PR #45932 is the better standalone representative because it contains the actual code/CI changes, while PR #46100 is a distinct testing-only migration.", + "best_pr_reason": "PR #46100 is the most self-contained and interpretable artifact in the set; #45932 appears to be a mixed-purpose or experimental PR and is not a good canonical anchor.", "soft_edge_verdicts": [ { "left": "pull_request:45932", "right": "pull_request:46100", "accept": false, - "reason": "Different work items: #45932 changes CI workflow and model/config code, while #46100 adds/migrates ViT tests. They do not appear to fix the same concrete bug or implement the same change, so they should not be merged as duplicates." + "reason": "Both mention tests, but they are not the same concrete change: #45932 is a broad model/workflow update, while #46100 is a focused ViT test migration. They could not plausibly be merged as one PR for the same bug or feature." } ] }, @@ -22638,12 +22414,12 @@ "pull_request:46073|pull_request:46196", "pull_request:46154|pull_request:46196" ], - "prepared_review_unit_hash": "267451dacbbc5d35c88363338ad3496040dc2db087d0eaa50abf53c54c150708", + "prepared_review_unit_hash": "812777e820c55ba2c24a48095256e334cb4885dabf84e92d50ca19bfafe79d87", "component_budget": { "node_count": 3, "item_count": 3, "soft_pair_count": 3, - "serialized_chars": 8398, + "serialized_chars": 8400, "estimated_input_tokens": 2100, "estimated_eval_tokens": 4456 }, @@ -22651,7 +22427,7 @@ "node_count": 3, "item_count": 3, "soft_pair_count": 3, - "serialized_chars": 8398, + "serialized_chars": 8400, "estimated_input_tokens": 2100, "estimated_eval_tokens": 4456 }, @@ -22660,30 +22436,30 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "All three PRs are the same feature family (attention-implementation dispatch / SDPA support) but they target different model code paths, so they should not be merged as duplicates.", - "confidence": 0.94, + "summary": "Three model-specific PRs all add SDPA/attn_implementation dispatch support, but they target different concrete model code paths (CTRL, T5, and wav2vec2/audio family). They are thematically related, not true duplicates.", + "confidence": 0.9, "canonical_issue_reason": null, - "canonical_pr_reason": "No canonical PR: CTRL, T5, and wav2vec2/seamless audio are separate model implementations with different affected files and code paths.", + "canonical_pr_reason": "PR 46196 is the best representative of the cluster because it is the broadest and most substantial implementation, spanning multiple audio model files and the main code-path changes for SDPA dispatch.", "best_issue_reason": null, - "best_pr_reason": "PR #46196 is the broadest representative of the family, touching many related audio model implementations and showing the most complete SDPA dispatch rollout, but it is still not a duplicate of the others.", + "best_pr_reason": "PR 46196 is the strongest cluster exemplar: it covers the largest set of affected models and contains the most complete attn_implementation dispatch work, while the others are narrower model-specific variants.", "soft_edge_verdicts": [ { "left": "pull_request:46073", "right": "pull_request:46154", "accept": false, - "reason": "Both add attn_implementation support, but for different architectures (CTRL vs T5) and different forward paths; not the same concrete fix." + "reason": "Both add attention-dispatch support, but for different model implementations (CTRL vs T5) with different code paths; this is a pattern match, not the same change." }, { "left": "pull_request:46073", "right": "pull_request:46196", "accept": false, - "reason": "Same general feature theme, but CTRL and wav2vec2/audio models are unrelated code paths and the changes are model-specific." + "reason": "CTRL-specific dispatch support and wav2vec2/audio-family dispatch support are distinct model code paths and not plausibly one merged PR for the same bug/change." }, { "left": "pull_request:46154", "right": "pull_request:46196", "accept": false, - "reason": "Both concern SDPA dispatch, but they patch different model families with different attention implementations and file sets; not mergeable as one PR." + "reason": "T5 and wav2vec2/audio-family SDPA dispatch patches target different architectures and different implementation details, so they are not duplicates." } ] }, @@ -22713,12 +22489,12 @@ "soft_pairs": [ "pull_request:46126|pull_request:46269" ], - "prepared_review_unit_hash": "e36dbdd05407d41960c1953271866fe3efda6fb363a4666d12669f47fcbe7969", + "prepared_review_unit_hash": "154028899b0b30f3ad9f821dd76ff7ebef7ac0a0a790eb65c74b04624304b270", "component_budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 5524, + "serialized_chars": 5523, "estimated_input_tokens": 1381, "estimated_eval_tokens": 3018 }, @@ -22726,7 +22502,7 @@ "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 5524, + "serialized_chars": 5523, "estimated_input_tokens": 1381, "estimated_eval_tokens": 3018 }, @@ -22735,18 +22511,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Both PRs touch distributed tensor-parallel/MoE files, but #46126 is a focused EP+Trainer integration change while #46269 is a broad distributed branch-base bundle. The overlap looks incidental, not the same concrete fix.", - "confidence": 0.85, + "summary": "Two open PRs in the distributed area, but they are not duplicates: one is a focused EP/Trainer integration, the other is a broad distributed-branch base sweep. Shared files look like overlap from a common branch, not the same concrete change.", + "confidence": 0.88, "canonical_issue_reason": null, - "canonical_pr_reason": "If one PR must anchor the cluster, #46126 is the better canonical representative because it has a single, concrete EP/Trainer scope; #46269 is too broad and mixes many unrelated distributed changes.", + "canonical_pr_reason": "PR #46126 is the more focused, concrete change: EP + Trainer integration with targeted edits in the distributed/tensor-parallel path and trainer/config plumbing.", "best_issue_reason": null, - "best_pr_reason": "#46126 is the best representative of the shared area because it targets one clear change path and is easier to reason about than the large, mixed-scope #46269.", + "best_pr_reason": "PR #46126 is the best representative of a single actionable change. PR #46269 is too broad and spans many unrelated distributed subsystems, so it is not a good duplicate/merge target.", "soft_edge_verdicts": [ { "left": "pull_request:46126", "right": "pull_request:46269", "accept": false, - "reason": "Shared distributed filenames are not enough here: #46126 is a specific EP/Trainer integration, while #46269 is a large branch-base PR spanning many unrelated distributed edits. They do not look like the same underlying change and would not plausibly merge into one PR." + "reason": "Reject: the overlap is only in shared distributed files; #46126 is a specific EP/Trainer integration, while #46269 is a large distributed base PR touching many unrelated areas. They do not appear to fix the same concrete code-path problem and would not plausibly merge into one PR." } ] }, @@ -22808,190 +22584,190 @@ "pull_request:46504|pull_request:46563", "pull_request:46432|pull_request:46563" ], - "prepared_review_unit_hash": "340087a4601f66081d5d1b8e7644d3e7a21ee74f8f135fae318c9df93600a096", + "prepared_review_unit_hash": "84bd9339541c09efceec87820aa81785499bb750413d45c8535023cecd1ea5cd", "component_budget": { "node_count": 9, "item_count": 9, "soft_pair_count": 26, - "serialized_chars": 23359, - "estimated_input_tokens": 5840, - "estimated_eval_tokens": 11936 + "serialized_chars": 23352, + "estimated_input_tokens": 5838, + "estimated_eval_tokens": 11932 }, "budget": { "node_count": 9, "item_count": 9, "soft_pair_count": 26, - "serialized_chars": 23359, - "estimated_input_tokens": 5840, - "estimated_eval_tokens": 11936 + "serialized_chars": 23352, + "estimated_input_tokens": 5838, + "estimated_eval_tokens": 11932 }, "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 because of their titles and `.github/workflows/*` scope. The only same-file group is `pr-ci-security-gate.yml`, but the three PRs there change different behaviors, so they should not be deduplicated.", - "confidence": 0.92, + "summary": "This cluster is mostly a set of unrelated workflow-maintenance PRs touching different YAML files. The only near-duplicate subgroup is the pr-ci-security-gate hardening work, where one pair (46554/46558) looks mergeable into a single PR; the reusable-workflow rewrite (46563) is a different change path.", + "confidence": 0.64, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 46554 is the most substantive representative of the set: it makes a concrete security-gate behavior change in `pr-ci-security-gate.yml` and is the clearest central example among the similar-looking workflow updates.", + "canonical_pr_reason": "PR 46554 is the best representative anchor because it makes the most substantive, concrete change to the pr-ci-security-gate workflow by adding actual security checks and checkout behavior, whereas the rest of the cluster is either unrelated workflow churn or smaller follow-up hardening.", "best_issue_reason": null, - "best_pr_reason": "PR 46554 is the strongest standalone PR here because it has the clearest functional intent and the most concrete code-path change; the other PRs are either unrelated workflow tweaks or different changes in the same workflow file.", + "best_pr_reason": "PR 46554 is the strongest standalone representative of the only related subgroup in the cluster: it addresses the pr-ci-security-gate workflow with direct, code-path-level security checks and is broader than the nearby follow-up edits.", "soft_edge_verdicts": [ { "left": "pull_request:46138", "right": "pull_request:46554", "accept": false, - "reason": "Different workflow files and different behavior; only broad CI-maintenance similarity, not the same underlying change." + "reason": "Different workflow files and different behavior: self-comment CI input handling vs PR security-gate hardening." }, { "left": "pull_request:46138", "right": "pull_request:46558", "accept": false, - "reason": "Unrelated workflow update in a different file; no shared code-path problem." + "reason": "Different workflow files and different behavior; only broad CI/workflow similarity, not the same underlying change." }, { "left": "pull_request:46138", "right": "pull_request:46563", "accept": false, - "reason": "Unrelated workflow update in a different file; only superficial title similarity." + "reason": "Self-comment CI stdin handling is unrelated to the pr-ci-security-gate workflow rewrite." }, { "left": "pull_request:46138", "right": "pull_request:46504", "accept": false, - "reason": "Different CI workflow and different shell/input handling fix; not the same bug or change." + "reason": "These touch different workflows and different scripts/variables; no shared concrete bug or change." }, { "left": "pull_request:46138", "right": "pull_request:46432", "accept": false, - "reason": "Different workflow and different operational hardening change; not duplicate content." + "reason": "Different workflow files and unrelated fixes: comment parsing vs SSH runner security/runtime tweaks." }, { "left": "pull_request:46432", "right": "pull_request:46503", "accept": false, - "reason": "Different workflows and different logic; both are CI changes but not the same underlying fix." + "reason": "SSH runner hardening and benchmark trigger logic are unrelated workflow changes." }, { "left": "pull_request:46432", "right": "pull_request:46501", "accept": false, - "reason": "Both touch workflows, but one is anti-slop permissions and the other is SSH runner hardening; distinct changes." + "reason": "Different workflow files and different intent; no same underlying bug." }, { "left": "pull_request:46432", "right": "pull_request:46502", "accept": false, - "reason": "Different workflow behavior: reviewer checkout ref vs. SSH runner hardening; not the same issue." + "reason": "SSH runner security/runtime changes do not match assign-reviewers checkout behavior." }, { "left": "pull_request:46432", "right": "pull_request:46554", "accept": false, - "reason": "Different workflows and different security concerns; no single concrete change to merge." + "reason": "Different workflows and different problem domains; not the same concrete fix." }, { "left": "pull_request:46432", "right": "pull_request:46558", "accept": false, - "reason": "Different workflows and different fixes; only broad CI/security similarity." + "reason": "Different workflow files and different behavior; only broad CI maintenance overlap." }, { "left": "pull_request:46501", "right": "pull_request:46503", "accept": false, - "reason": "Different workflow files and unrelated permission/scheduling changes." + "reason": "Anti-slop permissions cleanup and benchmark trigger support are unrelated changes." }, { "left": "pull_request:46501", "right": "pull_request:46502", "accept": false, - "reason": "Different workflow logic in different files; no shared bug or code-path." + "reason": "Different workflows and different fix targets; no common code-path issue." }, { "left": "pull_request:46501", "right": "pull_request:46504", "accept": false, - "reason": "Different workflow files and unrelated fixes." + "reason": "Unrelated workflow updates in different files." }, { "left": "pull_request:46501", "right": "pull_request:46554", "accept": false, - "reason": "Different workflows; same general CI area is too broad to dedupe." + "reason": "Anti-slop workflow permissions and pr-ci-security-gate hardening are different changes." }, { "left": "pull_request:46502", "right": "pull_request:46503", "accept": false, - "reason": "Different workflow behaviors in different files; not the same underlying change." + "reason": "Assign-reviewers checkout ref change and benchmark workflow_dispatch support are unrelated." }, { "left": "pull_request:46502", "right": "pull_request:46504", "accept": false, - "reason": "Unrelated workflow hardening vs. commit-message parsing fix; different code paths." + "reason": "Different workflow files and different behavior; no same underlying bug." }, { "left": "pull_request:46502", "right": "pull_request:46554", "accept": false, - "reason": "Different workflow files and distinct fixes; not mergeable as one duplicate PR." + "reason": "Assign-reviewers and pr-ci-security-gate affect different workflows and code paths." }, { "left": "pull_request:46503", "right": "pull_request:46504", "accept": false, - "reason": "Different workflows and different functionality; only superficial CI similarity." + "reason": "Benchmark trigger logic and build-ci docker tag detection are unrelated changes." }, { "left": "pull_request:46503", "right": "pull_request:46554", "accept": false, - "reason": "Different workflow behavior; not the same bug/change." + "reason": "Different workflows and different functional changes." }, { "left": "pull_request:46504", "right": "pull_request:46554", "accept": false, - "reason": "Different files and different fixes; no duplicate code-path problem." + "reason": "Build-ci docker image tagging and pr-ci-security-gate hardening are unrelated." }, { "left": "pull_request:46504", "right": "pull_request:46558", "accept": false, - "reason": "Different workflow areas and unrelated logic changes." + "reason": "Different workflows and different code paths; only generic YAML churn overlaps." }, { "left": "pull_request:46554", "right": "pull_request:46558", - "accept": false, - "reason": "Same workflow file, but they change different steps for different purposes (security checks vs. permissions/approval gating); not the same concrete change." + "accept": true, + "reason": "Same pr-ci-security-gate workflow hardening path; both changes target the fork-PR security gate and could plausibly be merged into one PR." }, { "left": "pull_request:46554", "right": "pull_request:46563", "accept": false, - "reason": "Same file, but one changes executable security-gate logic while the other adds explanatory comments around a reusable workflow call; not duplicate." + "reason": "46554 adds local security checks, while 46563 switches to a reusable workflow and adds notes; different implementation paths." }, { "left": "pull_request:46558", "right": "pull_request:46563", "accept": false, - "reason": "Same workflow file, but the patches address different concerns and code paths; not the same underlying fix." + "reason": "Permission/approval tightening in the local workflow is not the same change as replacing it with a reusable workflow." }, { "left": "pull_request:46504", "right": "pull_request:46563", "accept": false, - "reason": "Different workflow files and unrelated changes; no duplicate relationship." + "reason": "Build-ci docker image workflow changes are unrelated to the pr-ci-security-gate workflow." }, { "left": "pull_request:46432", "right": "pull_request:46563", "accept": false, - "reason": "Different workflows and different security hardening goals; not the same PR change." + "reason": "SSH runner workflow hardening is unrelated to the pr-ci-security-gate reusable-workflow migration." } ] }, @@ -23021,12 +22797,12 @@ "soft_pairs": [ "pull_request:46152|pull_request:46207" ], - "prepared_review_unit_hash": "0fc885f82b7a895b1cf9a82e98848b9a2540f9810e4a0086e32e0788f0d675d8", + "prepared_review_unit_hash": "ee22f6e2763b52b5469bfd8adb2b41320a7f392143f99a8b050f9d4dbafc9e34", "component_budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 4275, + "serialized_chars": 4276, "estimated_input_tokens": 1069, "estimated_eval_tokens": 2394 }, @@ -23034,7 +22810,7 @@ "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 4275, + "serialized_chars": 4276, "estimated_input_tokens": 1069, "estimated_eval_tokens": 2394 }, @@ -23043,18 +22819,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These are related only at a very high level: both replace brittle file-based source access with safer source/introspection mechanisms, but they touch different code paths and would not plausibly be merged as one PR.", - "confidence": 0.94, + "summary": "These are two separate pull requests that both replace source-file opening with source inspection, but they apply to different code paths and problems: one in import_utils for resolving package resources, the other in modeling_utils for attention-implementation detection.", + "confidence": 0.88, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 46152 is the better representative for the cluster theme because it directly centralizes source-file access in import_utils and introduces a reusable traversal helper, making it the broader infrastructure change.", + "canonical_pr_reason": "PR 46152 is the cleaner canonical representative: it introduces a reusable source/file-resolution fix in import_utils, while 46207 is a separate function-specific change in modeling_utils.", "best_issue_reason": null, - "best_pr_reason": "PR 46152 is the best overall representative because it is the more general source-reading refactor; PR 46207 is a narrower, separate fix in modeling_utils.", + "best_pr_reason": "PR 46152 is the best single representative for the cluster because it is the broader, more self-contained source-reading fix and is not dependent on the unrelated modeling_utils change.", "soft_edge_verdicts": [ { "left": "pull_request:46152", "right": "pull_request:46207", "accept": false, - "reason": "Both avoid open()-based source reads, but they affect different modules and distinct code paths (import_utils vs modeling_utils). They are similar in approach, not the same underlying bug or change, so they should not be merged as duplicates." + "reason": "Both PRs modernize source access, but they fix different concrete paths in different modules with different behavior changes. They are not the same underlying bug and would not plausibly merge into one PR as duplicates." } ] }, @@ -23086,7 +22862,7 @@ "pull_request:46436|pull_request:46474", "pull_request:46173|pull_request:46474" ], - "prepared_review_unit_hash": "ae2300dab4ff26c80629e26b6b7e31ed36f0d000f0e07e76dbc94baeb62d8141", + "prepared_review_unit_hash": "dabde6cd411e1011f9e3eb17a6894b97d8023dc5b980ff3585b2e01529c14b91", "component_budget": { "node_count": 3, "item_count": 3, @@ -23108,30 +22884,30 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Two PRs are near-duplicates of the same StaticLayer.get_seq_length type-hint fix, while the cache max-len config PR is unrelated and should stay separate.", + "summary": "PRs #46173 and #46474 are duplicates: both make the same StaticLayer.get_seq_length type-hint change in src/transformers/cache_utils.py for issue #45987. PR #46436 is separate, changing static cache max length handling in generation/utils.py.", "confidence": 0.97, "canonical_issue_reason": null, - "canonical_pr_reason": "PR #46173 is the strongest canonical representative for the duplicated type-hint fix because it has the explicit linked issue target and matches the same cache_utils change as #46474.", + "canonical_pr_reason": "#46173 is the strongest canonical representative because it has the same cache_utils type-hint fix, an explicit link to issue #45987, and a clear, specific title.", "best_issue_reason": null, - "best_pr_reason": "PR #46173 is the best overall representative of the duplicate pair; #46474 is effectively the same patch, while #46436 addresses a different generation cache-length bug.", + "best_pr_reason": "#46173 is the best representative of the duplicate pair: it directly targets the underlying StaticLayer.get_seq_length annotation fix and is explicitly tied to the tracking issue.", "soft_edge_verdicts": [ { - "left": "pull_request:46173", + "left": "pull_request:46436", "right": "pull_request:46474", - "accept": true, - "reason": "Same file, same line-level change, same explicit issue target, and the diffs implement the same StaticLayer.get_seq_length return-type annotation fix." + "accept": false, + "reason": "Different changes and code paths: one adjusts static cache max_len selection in generation/utils.py, the other only changes a return type hint in cache_utils.py." }, { - "left": "pull_request:46436", + "left": "pull_request:46173", "right": "pull_request:46474", - "accept": false, - "reason": "Different bugs and code paths: one changes static cache max length handling in generation, the other only adjusts a type hint in cache_utils." + "accept": true, + "reason": "Same underlying bug/change: both update StaticLayer.get_seq_length return type annotation in cache_utils.py, with identical intent and nearly identical diffs." } ] }, "evaluator_result": { "accept": true, - "feedback": "The analysis is grounded in the packet: PRs 46173 and 46474 both make the same cache_utils type-hint change and share issue target 45987, while 46436 targets a different issue and modifies generation cache-length behavior. The soft-edge verdicts are conservative and the summary does not overstate the evidence." + "feedback": "The summary is grounded in the packet. The 46173/46474 pair is supported by the shared issue target, same filename, and matching type-hint diff, while 46436 is clearly a different change in a different code path. The soft-edge verdicts are conservative enough, and I don\u2019t see overstatement beyond what the diffs support." }, "evaluator_used": true, "retried": false, @@ -23158,12 +22934,12 @@ "soft_pairs": [ "pull_request:46310|pull_request:46546" ], - "prepared_review_unit_hash": "b0ad33899c064619daf3961fc98218a157517a4a5122ac7246b98a242a0a9067", + "prepared_review_unit_hash": "db1633bb030374040929897c0f2d071190f03fc5e089fd500803a2c5e4d621a1", "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 }, @@ -23171,7 +22947,7 @@ "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 4850, + "serialized_chars": 4851, "estimated_input_tokens": 1213, "estimated_eval_tokens": 2682 }, @@ -23180,18 +22956,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Two open PRs fix the same sliding-window right-boundary off-by-one in Moonshine streaming masking; the second PR is broader and also covers Gemma4 plus tests, so it is the better representative.", - "confidence": 0.93, + "summary": "Both PRs fix the same Moonshine streaming sliding-window boundary bug (`-dist < right_window_size` vs `<=`), with 46546 appearing to be the fuller follow-up/superset change. The pair should be clustered together as duplicate fixes for the same underlying issue.", + "confidence": 0.97, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 46546 is the more complete fix: it includes the Moonshine streaming boundary correction, adds Gemma4 coverage for the same helper bug pattern, and includes test updates.", + "canonical_pr_reason": "PR 46546 is the better canonical PR: it fixes the same boundary bug in Moonshine streaming, includes the related test update, and appears to be the more complete version of the change.", "best_issue_reason": null, - "best_pr_reason": "PR 46546 is the best representative because it is broader, includes validation, and addresses the same concrete boundary condition more completely than PR 46310.", + "best_pr_reason": "PR 46546 is the strongest representative because it addresses the concrete bug in the target code path and adds tests; PR 46310 is the narrower earlier version of the same fix.", "soft_edge_verdicts": [ { "left": "pull_request:46310", "right": "pull_request:46546", "accept": true, - "reason": "Both PRs change the same right-window mask boundary from < to <= in the same sliding-window helper, so they address the same underlying off-by-one bug and could be merged into one combined fix." + "reason": "Same underlying Moonshine streaming right-window boundary bug in the same mask function; 46546 is effectively a broader version of the same fix and could plausibly subsume 46310." } ] }, @@ -23221,7 +22997,7 @@ "soft_pairs": [ "pull_request:46376|pull_request:46417" ], - "prepared_review_unit_hash": "175e374d61af0e2f321e6a437f52bfc980359401e4216d0029242d7b57d86ee7", + "prepared_review_unit_hash": "d30254547471604ad9340068dcd2954eacd2d40a42f72757687df0f7fb9f7fc8", "component_budget": { "node_count": 2, "item_count": 2, @@ -23243,18 +23019,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These pull requests are not duplicates: one adds multimodal support to continuous batching internals, while the other adds a new HunYuan VL model package, docs, and tests.", - "confidence": 0.98, + "summary": "The two pull requests are unrelated: one adds multimodal support to continuous batching infrastructure, while the other adds the HunYuan VL model and its docs/tests. They do not look like the same underlying change and should not be merged as duplicates.", + "confidence": 0.97, "canonical_issue_reason": null, - "canonical_pr_reason": "No single canonical PR is appropriate here because the PRs target different features and code paths; if a representative is required, pull_request:46417 is the more self-contained addition with broader implementation and review activity.", + "canonical_pr_reason": "PR #46417 is the better representative because it is a coherent, self-contained feature addition with extensive model, processor, docs, and tests coverage.", "best_issue_reason": null, - "best_pr_reason": "pull_request:46417 is the better standalone representative only in the sense that it is a complete feature addition with clear scope; however, it is not a duplicate of pull_request:46376 and should not be clustered with it.", + "best_pr_reason": "PR #46417 is the strongest standalone PR in the set: it is a complete model integration rather than an in-progress infrastructure refactor.", "soft_edge_verdicts": [ { "left": "pull_request:46376", "right": "pull_request:46417", "accept": false, - "reason": "Different underlying changes: 46376 modifies continuous batching cache/scheduler/runtime for multimodal support, while 46417 adds a new HunYuan VL model, processing, docs, and tests. They do not fix the same code-path problem and could not plausibly be merged as one PR." + "reason": "Different change families and code paths: continuous batching/multimodal runtime support versus adding a new HunYuan VL model package, docs, and tests. Shared mention of multimodality is too broad to treat as a duplicate." } ] }, @@ -23284,7 +23060,7 @@ "soft_pairs": [ "pull_request:46574|pull_request:46580" ], - "prepared_review_unit_hash": "7a7adc5bdf57dbefbc595ce0c81534f03a49ba8972aceafb057f4e1c27f18d41", + "prepared_review_unit_hash": "e47f1ab560366d9e11fda24625d2c0e3e6b50b20841e84ad7cc4e5cb4d91c2a0", "component_budget": { "node_count": 2, "item_count": 2, @@ -23306,18 +23082,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These are two unrelated pull requests: one fixes CodeLlama decode behavior that strips real leading spaces, while the other fixes DeepSeek-Coder/DeepSeek-LLM loading into the wrong tokenizer class. They target different code paths and should not be deduplicated.", - "confidence": 0.98, + "summary": "These PRs are unrelated tokenizer fixes: one changes CodeLlama decoding to preserve real leading spaces, while the other changes AutoTokenizer checkpoint routing for specific DeepSeek models that load the wrong tokenizer class.", + "confidence": 0.99, "canonical_issue_reason": null, "canonical_pr_reason": null, "best_issue_reason": null, - "best_pr_reason": "Neither PR is a good canonical representative for the other; they address different tokenizer bugs in different subsystems and are not mergeable as one change.", + "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "pull_request:46574", "right": "pull_request:46580", "accept": false, - "reason": "Different underlying bugs and fixes: CodeLlama decoder stripping vs. AutoTokenizer checkpoint-class override for DeepSeek checkpoints. They do not share the same code path or change set." + "reason": "Different bug classes and code paths: CodeLlama decoder stripping vs. DeepSeek model/tokenizer-class misrouting. They are not the same change and would not plausibly merge into one PR." } ] }, @@ -23347,7 +23123,7 @@ "soft_pairs": [ "pull_request:46603|pull_request:46604" ], - "prepared_review_unit_hash": "707342ee986c262326321981026c84c1535696405342836282f2f3eb97ff5321", + "prepared_review_unit_hash": "608abd69ddb2d867b72e1ac4feb9f089536abefbeafc9f17ede5e49729d99743", "component_budget": { "node_count": 2, "item_count": 2, @@ -23369,18 +23145,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Two open Mistral PRs are related but distinct: one refactors/moves `MistralConverter` into the `integrations/mistral` package, while the other adds tekken tokenizer detection, conversion, and save utilities. They share files because the broader tokenizer work likely builds on the refactor, but they are not the same change and should not be merged as duplicates.", - "confidence": 0.91, + "summary": "These two Mistral PRs touch the same files but implement different changes: #46603 is a refactor that moves MistralConverter into the integrations package, while #46604 adds tekken tokenizer detection/conversion/save utilities. They are related, but not the same underlying change, so they should not be clustered as duplicates.", + "confidence": 0.96, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 46604 is the better canonical PR for the cluster because it is the broader, more substantive Mistral tokenizer change and appears to encompass the converter relocation as part of the feature work.", + "canonical_pr_reason": "#46604 is the more substantive feature PR and best represents the cluster\u2019s Mistral tokenizer work; #46603 is primarily a code move/refactor.", "best_issue_reason": null, - "best_pr_reason": "PR 46604 is the best representative PR: it addresses the main functional tokenizer work, whereas PR 46603 is only the narrower code move/refactor.", + "best_pr_reason": "#46604 is the best representative PR because it adds the actual new tokenizer functionality, whereas #46603 is mostly an organizational extraction.", "soft_edge_verdicts": [ { "left": "pull_request:46603", "right": "pull_request:46604", "accept": false, - "reason": "Not the same underlying change. 46603 is a package refactor moving `MistralConverter`; 46604 adds tekken tokenizer detection/conversion/save utilities. Shared files reflect dependency, not duplicate code-path fixes." + "reason": "Different code-path changes: #46603 relocates MistralConverter, while #46604 implements tekken detection/conversion/save support. Shared files alone are not enough to treat them as the same PR." } ] }, diff --git a/analysis/current/manifest.json b/analysis/current/manifest.json index af29eb949745e764a4646f04d5780661aa5635e7..a1ef5cc516611cd8296d39c03d34039d7e0dcea5 100644 --- a/analysis/current/manifest.json +++ b/analysis/current/manifest.json @@ -1,8 +1,8 @@ { - "analysis_id": "hybrid-model-20260616t060022z", + "analysis_id": "hybrid-model-20260616t120046z", "archived_artifacts": { - "hybrid": "snapshots/20260616T060022Z/analysis-runs/hybrid-model-20260616t060022z/analysis-report-hybrid.json", - "hybrid_reviews": "snapshots/20260616T060022Z/analysis-runs/hybrid-model-20260616t060022z/analysis-report-hybrid.llm-reviews.json" + "hybrid": "snapshots/20260616T120046Z/analysis-runs/hybrid-model-20260616t120046z/analysis-report-hybrid.json", + "hybrid_reviews": "snapshots/20260616T120046Z/analysis-runs/hybrid-model-20260616t120046z/analysis-report-hybrid.llm-reviews.json" }, "artifacts": { "hybrid": "analysis/current/analysis-report-hybrid.json", @@ -10,9 +10,9 @@ }, "channel": "canonical", "model": null, - "published_at": "2026-06-16T06:09:52Z", + "published_at": "2026-06-16T12:10:43Z", "repo": "huggingface/transformers", "schema_version": 1, - "snapshot_id": "20260616T060022Z", + "snapshot_id": "20260616T120046Z", "variant": "hybrid" } diff --git a/snapshots/20260616T120046Z/analysis-runs/hybrid-model-20260616t120046z/analysis-report-hybrid.json b/snapshots/20260616T120046Z/analysis-runs/hybrid-model-20260616t120046z/analysis-report-hybrid.json new file mode 100644 index 0000000000000000000000000000000000000000..859fda2246fb29c427c38b731e0c8273a3d23f82 --- /dev/null +++ b/snapshots/20260616T120046Z/analysis-runs/hybrid-model-20260616t120046z/analysis-report-hybrid.json @@ -0,0 +1,2200 @@ +{ + "schema_version": "1.0", + "repo": "huggingface/transformers", + "snapshot_id": "20260616T120046Z", + "generated_at": "2026-06-16T12:10:41Z", + "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-41115-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #41115.", + "status": "open", + "confidence": 0.55, + "canonical_issue_number": 41115, + "canonical_pr_number": 41116, + "issue_numbers": [ + 41115 + ], + "pr_numbers": [ + 41116, + 45613 + ], + "evidence_types": [ + "closing_reference" + ], + "pr_comparisons": [ + { + "left_pr_number": 41116, + "right_pr_number": 45613, + "code_similarity": 0.797, + "size_similarity": 0.885, + "file_overlap": 0.75, + "area_overlap": 0.831, + "patch_similarity": 0.693, + "shared_filenames": [ + "docs/source/en/_toctree.yml", + "docs/source/en/model_doc/minicpm3.md", + "src/transformers/models/auto/modeling_auto.py", + "src/transformers/models/minicpm3/__init__.py", + "src/transformers/models/minicpm3/configuration_minicpm3.py", + "src/transformers/models/minicpm3/modeling_minicpm3.py", + "src/transformers/models/minicpm3/modular_minicpm3.py", + "tests/models/minicpm3/__init__.py", + "tests/models/minicpm3/test_modeling_minicpm3.py" + ], + "shared_file_areas": [ + { + "filename": "docs/source/en/_toctree.yml", + "left_ranges": [ + [ + 713, + 720 + ] + ], + "right_ranges": [ + [ + 730, + 737 + ] + ] + }, + { + "filename": "docs/source/en/model_doc/minicpm3.md", + "left_ranges": [ + [ + 1, + 67 + ] + ], + "right_ranges": [ + [ + 1, + 45 + ] + ] + }, + { + "filename": "src/transformers/models/auto/modeling_auto.py", + "left_ranges": [ + [ + 285, + 291 + ], + [ + 700, + 706 + ], + [ + 1298, + 1304 + ] + ], + "right_ranges": [ + [ + 720, + 726 + ], + [ + 1331, + 1337 + ] + ] + }, + { + "filename": "src/transformers/models/minicpm3/__init__.py", + "left_ranges": [ + [ + 1, + 29 + ] + ], + "right_ranges": [ + [ + 1, + 29 + ] + ] + }, + { + "filename": "src/transformers/models/minicpm3/configuration_minicpm3.py", + "left_ranges": [ + [ + 1, + 141 + ] + ], + "right_ranges": [ + [ + 1, + 126 + ] + ] + }, + { + "filename": "src/transformers/models/minicpm3/modeling_minicpm3.py", + "left_ranges": [ + [ + 1, + 544 + ] + ], + "right_ranges": [ + [ + 1, + 522 + ] + ] + }, + { + "filename": "src/transformers/models/minicpm3/modular_minicpm3.py", + "left_ranges": [ + [ + 1, + 444 + ] + ], + "right_ranges": [ + [ + 1, + 342 + ] + ] + }, + { + "filename": "tests/models/minicpm3/test_modeling_minicpm3.py", + "left_ranges": [ + [ + 1, + 109 + ] + ], + "right_ranges": [ + [ + 1, + 136 + ] + ] + } + ] + } + ] + }, + { + "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-45987-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #45987.", + "status": "open", + "confidence": 0.75, + "canonical_issue_number": 45987, + "canonical_pr_number": 46173, + "issue_numbers": [ + 45987 + ], + "pr_numbers": [ + 46173, + 46474 + ], + "evidence_types": [ + "closing_reference", + "shared_issue_target" + ], + "pr_comparisons": [ + { + "left_pr_number": 46173, + "right_pr_number": 46474, + "code_similarity": 0.65, + "size_similarity": 1.0, + "file_overlap": 1.0, + "area_overlap": 0.0, + "patch_similarity": 1.0, + "shared_filenames": [ + "src/transformers/cache_utils.py" + ], + "shared_file_areas": [ + { + "filename": "src/transformers/cache_utils.py", + "left_ranges": [ + [ + 69, + 75 + ], + [ + 375, + 381 + ] + ], + "right_ranges": [ + [ + 79, + 85 + ], + [ + 461, + 467 + ] + ] + } + ] + } + ] + }, + { + "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": 0.494, + "size_similarity": 0.619, + "file_overlap": 0.5, + "area_overlap": 0.434, + "patch_similarity": 0.451, + "shared_filenames": [ + "src/transformers/testing_utils.py", + "tests/utils/test_testing_utils.py" + ], + "shared_file_areas": [ + { + "filename": "src/transformers/testing_utils.py", + "left_ranges": [ + [ + 3595, + 3602 + ], + [ + 3819, + 3853 + ] + ], + "right_ranges": [ + [ + 3525, + 3558 + ], + [ + 3782, + 3788 + ] + ] + }, + { + "filename": "tests/utils/test_testing_utils.py", + "left_ranges": [ + [ + 1, + 86 + ] + ], + "right_ranges": [ + [ + 1, + 114 + ] + ] + } + ] + } + ] + }, + { + "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-46554-2", + "summary": "Cluster of 2 related pull requests linked by soft_similarity.", + "status": "open", + "confidence": 0.5, + "canonical_issue_number": null, + "canonical_pr_number": 46558, + "issue_numbers": [], + "pr_numbers": [ + 46554, + 46558 + ], + "evidence_types": [ + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 46554, + "right_pr_number": 46558, + "code_similarity": 0.433, + "size_similarity": 0.429, + "file_overlap": 1.0, + "area_overlap": 0.1, + "patch_similarity": 0.081, + "shared_filenames": [ + ".github/workflows/pr-ci-security-gate.yml" + ], + "shared_file_areas": [ + { + "filename": ".github/workflows/pr-ci-security-gate.yml", + "left_ranges": [ + [ + 16, + 45 + ] + ], + "right_ranges": [ + [ + 13, + 19 + ], + [ + 57, + 63 + ] + ] + } + ] + } + ] + }, + { + "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-41115-3", + "canonical_pr_number": 41116, + "duplicate_pr_numbers": [ + 45613 + ], + "target_issue_number": 41115, + "reason": "PRs in cluster-41115-3 are treated as duplicates because they converge on issue #41115 with closing_reference evidence." + }, + { + "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-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-45987-3", + "canonical_pr_number": 46173, + "duplicate_pr_numbers": [ + 46474 + ], + "target_issue_number": 45987, + "reason": "PRs in cluster-45987-3 are treated as duplicates because they converge on issue #45987 with closing_reference, shared_issue_target 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-46554-2", + "canonical_pr_number": 46558, + "duplicate_pr_numbers": [ + 46554 + ], + "target_issue_number": null, + "reason": "PRs in cluster-46554-2 are treated as duplicates because they share soft_similarity evidence." + } + ], + "best_issue": { + "cluster_id": "cluster-41211-3", + "issue_number": 41211, + "reason": "Issue #41211 is the strongest global issue candidate because it is open, belongs to a cluster with 3 artifacts, and carries 6 discussion comments plus 6 inbound references.", + "score": 148.0 + }, + "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/20260616T120046Z/analysis-runs/hybrid-model-20260616t120046z/analysis-report-hybrid.llm-reviews.json b/snapshots/20260616T120046Z/analysis-runs/hybrid-model-20260616t120046z/analysis-report-hybrid.llm-reviews.json new file mode 100644 index 0000000000000000000000000000000000000000..b1632b8116de79f556308e968f5f9834e9067490 --- /dev/null +++ b/snapshots/20260616T120046Z/analysis-runs/hybrid-model-20260616t120046z/analysis-report-hybrid.llm-reviews.json @@ -0,0 +1,23171 @@ +{ + "schema_version": "1.0", + "repo": "huggingface/transformers", + "snapshot_id": "20260616T120046Z", + "generated_at": "2026-06-16T12:10:41Z", + "model": "gpt-5.4-mini?service_tier=flex", + "reviews": [ + { + "label": "issue", + "component_index": 1, + "component_count": 21, + "review_unit_index": 1, + "review_unit_count": 56, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-473", + "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" + ], + "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: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:43596|issue:46165", + "issue:34634|issue:35545", + "issue:30277|issue:43638", + "issue:43596|issue:43638", + "issue:43039|issue:45161", + "issue:43975|issue:44457", + "issue:44292|issue:44457", + "issue:44841|issue:46661", + "issue:34928|issue:44805", + "issue:24643|issue:30277", + "issue:45859|issue:45902", + "issue:43638|issue:45161", + "issue:34928|issue:36331", + "issue:43596|issue:45161", + "issue:43716|issue:45507", + "issue:45874|issue:46294", + "issue:44292|issue:45237", + "issue:39290|issue:45081", + "issue:45823|issue:45874", + "issue:45800|issue:46515", + "issue:43425|issue:44292", + "issue:34928|issue:41720", + "issue:43828|issue:45161", + "issue:43828|issue:44292", + "issue:43541|issue:43716", + "issue:43716|issue:46468", + "issue:43716|issue:45685", + "issue:44292|issue:44918", + "issue:43716|issue:44292", + "issue:45507|issue:45685", + "issue:44918|issue:45161", + "issue:43039|issue:44918", + "issue:41762|issue:43638", + "issue:43827|issue:44512", + "issue:43828|issue:45685", + "issue:44918|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": "88e6572aee9f47660b4bbf1c0e5799069bce465f5790a2a11f0e63a28694b8f0", + "component_budget": { + "node_count": 473, + "item_count": 473, + "soft_pair_count": 1738, + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 62, + "serialized_chars": 45736, + "estimated_input_tokens": 11434, + "estimated_eval_tokens": 23124 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly heterogeneous. I only see a few true duplicate mini-clusters: the DeepSpeed ZeRO-3 Bert IndexError pair, the Qwen3.5Moe rotary-embedding/offload pair, and the torch 2.4.1 incompatibility pair. Most other soft edges connect broadly similar but clearly different bugs, models, or docs issues.", + "confidence": 0.81, + "canonical_issue_reason": "issue:43596 is the cleanest concrete bug report among the clearly duplicated subgroup: it describes the same ZeRO-3 init IndexError (size 0 / index 0 out of bounds) as issue:43638, with the older issue being the better representative.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:43596 is the best issue representative for the only strong duplicate subgroup in this set because it is specific, reproducible, and directly matches the paired ZeRO-3/Bert failure. The rest of the cluster is too mixed to justify a single global canonical across all items.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44322", + "right": "issue:45464", + "accept": false, + "reason": "Different problems: missing Qwen3.5 config field vs streaming chat/completions failure." + }, + { + "left": "issue:45237", + "right": "issue:45685", + "accept": false, + "reason": "Different backends and failures: AMD GPU support vs MPS histogram not implemented for Int." + }, + { + "left": "issue:43716", + "right": "issue:45237", + "accept": false, + "reason": "Different root causes: dtype mismatch in image preprocessing vs AMD GPU execution failure." + }, + { + "left": "issue:43828", + "right": "issue:45237", + "accept": false, + "reason": "Different issues: autocast dtype mismatch vs AMD GPU-specific model failure." + }, + { + "left": "issue:45237", + "right": "issue:46468", + "accept": false, + "reason": "Different artifacts: AMD GPU incompatibility vs broken Nemotron example." + }, + { + "left": "issue:45237", + "right": "issue:45507", + "accept": false, + "reason": "Different bugs: AMD GPU failure vs invalid method call in GraniteMoEHybrid." + }, + { + "left": "issue:30277", + "right": "issue:35545", + "accept": false, + "reason": "Different code paths: DeepSpeed collective mismatch vs ONNX export error." + }, + { + "left": "issue:35545", + "right": "issue:42915", + "accept": false, + "reason": "Different model/export issue vs FP8 config failure." + }, + { + "left": "issue:35545", + "right": "issue:41720", + "accept": false, + "reason": "Different failures: ONNX export vs auto device mapping cuda assert." + }, + { + "left": "issue:45901", + "right": "issue:46015", + "accept": false, + "reason": "Different topics: table-question-answering crash vs outdated RAG examples." + }, + { + "left": "issue:44509", + "right": "issue:44512", + "accept": false, + "reason": "Both are docs cleanup, but they target different removed items and pages." + }, + { + "left": "issue:35545", + "right": "issue:36010", + "accept": false, + "reason": "Different import/export issues with no shared concrete bug." + }, + { + "left": "issue:44918", + "right": "issue:46165", + "accept": false, + "reason": "Different failures: Qwen3.5 input-embedding unpacking vs resume/trust_remote_code issue." + }, + { + "left": "issue:44805", + "right": "issue:45823", + "accept": false, + "reason": "Different problems: mask shape IndexError vs Gemma4 PLE device mismatch." + }, + { + "left": "issue:43638", + "right": "issue:44805", + "accept": false, + "reason": "Different error signatures and model paths: ZeRO-3 Bert init vs generic mask shape mismatch." + }, + { + "left": "issue:43638", + "right": "issue:46165", + "accept": false, + "reason": "Different training/resume issue vs ZeRO-3 Bert IndexError." + }, + { + "left": "issue:43596", + "right": "issue:44805", + "accept": false, + "reason": "Different bugs: ZeRO-3 Bert IndexError vs mask shape mismatch." + }, + { + "left": "issue:44805", + "right": "issue:46294", + "accept": false, + "reason": "Different multimodal/video failure vs generic mask shape mismatch." + }, + { + "left": "issue:44805", + "right": "issue:45161", + "accept": false, + "reason": "Different problems: mask shape IndexError vs GPT-OSS MoE tensor-parallel issue." + }, + { + "left": "issue:46165", + "right": "issue:46468", + "accept": false, + "reason": "Different artifacts: resume/trust_remote_code issue vs broken Nemotron example." + }, + { + "left": "issue:44560", + "right": "issue:46165", + "accept": false, + "reason": "Different failures: video StopIteration vs training resume problem." + }, + { + "left": "issue:43596", + "right": "issue:46165", + "accept": false, + "reason": "Different failures: ZeRO-3 Bert IndexError vs resume/trust_remote_code issue." + }, + { + "left": "issue:34634", + "right": "issue:35545", + "accept": false, + "reason": "Different model subsystem bugs: BarkProcessor voice preset vs ONNX export." + }, + { + "left": "issue:30277", + "right": "issue:43638", + "accept": false, + "reason": "Both involve DeepSpeed/ZeRO, but the concrete failures and models differ." + }, + { + "left": "issue:43596", + "right": "issue:43638", + "accept": true, + "reason": "Same underlying DeepSpeed ZeRO-3 Bert IndexError; only the model variant differs." + }, + { + "left": "issue:43039", + "right": "issue:45161", + "accept": false, + "reason": "Different issues: Liger Kernel cross-entropy path vs GPT-OSS tensor-parallel failure." + }, + { + "left": "issue:43975", + "right": "issue:44457", + "accept": false, + "reason": "Different bugs: detokenization regression vs LoRA merge/save/load mismatch." + }, + { + "left": "issue:44292", + "right": "issue:44457", + "accept": false, + "reason": "Different model/runtime error vs LoRA result inconsistency." + }, + { + "left": "issue:44841", + "right": "issue:46661", + "accept": false, + "reason": "Different failures: Voxtral processor bug vs remote-code AutoModel config_class error." + }, + { + "left": "issue:34928", + "right": "issue:44805", + "accept": false, + "reason": "Both are tensor/shape errors, but they come from different mechanisms and code paths." + }, + { + "left": "issue:24643", + "right": "issue:30277", + "accept": false, + "reason": "Different DeepSpeed errors: weight dimensionality vs collective mismatch." + }, + { + "left": "issue:45859", + "right": "issue:45902", + "accept": true, + "reason": "Same Qwen3.5Moe rotary-embedding/materialization bug under CPU offload/meta-to-empty handling." + }, + { + "left": "issue:43827", + "right": "issue:44512", + "accept": false, + "reason": "Both are docs removals, but they reference different removed commands/tasks." + }, + { + "left": "issue:45800", + "right": "issue:46515", + "accept": true, + "reason": "Near-identical torch 2.4.1 incompatibility reports." + }, + { + "left": "issue:43425", + "right": "issue:44292", + "accept": false, + "reason": "Different issues: Torch version incompatibility vs Qwen-3 NVFP4 runtime failure." + }, + { + "left": "issue:34928", + "right": "issue:41720", + "accept": false, + "reason": "Different bugs: activation-checkpointing/FSDP recompute mismatch vs Qwen3 auto device-map assert." + }, + { + "left": "issue:43828", + "right": "issue:45161", + "accept": false, + "reason": "Different model/runtime problems: autocast dtype mismatch vs GPT-OSS MoE tensor-parallel failure." + }, + { + "left": "issue:43541", + "right": "issue:43716", + "accept": false, + "reason": "Different failures: torch dynamo/grouped_mm tracing vs image preprocessor/model dtype mismatch." + }, + { + "left": "issue:43716", + "right": "issue:46468", + "accept": false, + "reason": "Different bugs: Mistral dtype mismatch vs Nemotron example breakage." + }, + { + "left": "issue:43716", + "right": "issue:45685", + "accept": false, + "reason": "Different issues: image/model dtype mismatch vs MPS histogram kernel limitation." + }, + { + "left": "issue:44292", + "right": "issue:44918", + "accept": false, + "reason": "Different bugs: Qwen-3 NVFP4 runtime failure vs TRL SFT embedding unpacking failure." + }, + { + "left": "issue:43716", + "right": "issue:44292", + "accept": false, + "reason": "Different models and failure modes; not the same bug." + }, + { + "left": "issue:45507", + "right": "issue:45685", + "accept": false, + "reason": "Different code paths: invalid method call vs missing MPS kernel support." + }, + { + "left": "issue:44918", + "right": "issue:45161", + "accept": false, + "reason": "Different problems: TRL input embedding handling vs GPT-OSS MoE tensor-parallel issue." + }, + { + "left": "issue:43039", + "right": "issue:44918", + "accept": false, + "reason": "Different issues with no shared concrete failure path." + }, + { + "left": "issue:41762", + "right": "issue:43638", + "accept": false, + "reason": "Both are ZeRO-3 IndexErrors, but the models and triggering paths differ enough to avoid merging." + }, + { + "left": "issue:43827", + "right": "issue:44512", + "accept": false, + "reason": "Different docs omissions; not the same underlying change." + }, + { + "left": "issue:43828", + "right": "issue:45685", + "accept": false, + "reason": "Different failures: autocast dtype mismatch vs MPS Int histogram limitation." + }, + { + "left": "issue:44918", + "right": "issue:45685", + "accept": false, + "reason": "Different bugs: SFT embedding unpacking vs MPS kernel limitation." + }, + { + "left": "issue:43039", + "right": "issue:43828", + "accept": false, + "reason": "Different failures in different models/subsystems." + }, + { + "left": "issue:30277", + "right": "issue:36331", + "accept": false, + "reason": "Different problems: DeepSpeed collective mismatch vs trainer API signature change." + }, + { + "left": "issue:34928", + "right": "issue:36010", + "accept": false, + "reason": "Different issues: FSDP/checkpointing size mismatch vs GenerationMixin import error." + }, + { + "left": "issue:45237", + "right": "issue:46015", + "accept": false, + "reason": "Different artifacts: AMD GPU model failure vs docs/examples issue." + }, + { + "left": "issue:43039", + "right": "issue:44805", + "accept": false, + "reason": "Different code paths: Liger Kernel cross-entropy vs generic mask shape mismatch." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays conservative. It correctly treats the cluster as mostly heterogeneous while identifying the three plausible duplicate mini-clusters (43596/43638, 45859/45902, 45800/46515). The soft-edge rejections are appropriately cautious, and the accepted pairs look mergeable as the same bugfix." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 21, + "review_unit_index": 2, + "review_unit_count": 56, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-473", + "nodes": [ + "issue:36010", + "issue:39290", + "issue:41093", + "issue:41720", + "issue:41762", + "issue:43039", + "issue:43296", + "issue:43366", + "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:44805", + "issue:44841", + "issue:44863", + "issue:44918", + "issue:45020", + "issue:45070", + "issue:45084", + "issue:45237", + "issue:45362", + "issue:45399", + "issue:45464", + "issue:45685", + "issue:45800", + "issue:45812", + "issue:45823", + "issue:45901", + "issue:45920", + "issue:46015", + "issue:46294", + "issue:46304", + "issue:46402", + "issue:46661" + ], + "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:43039|issue:44560", + "issue:45823|issue:46294", + "issue:43975|issue:46015", + "issue:44841|issue:45823", + "issue:44863|issue:46015", + "issue:45020|issue:46304", + "issue:45812|issue:45920", + "issue:43866|issue:44863", + "issue:44292|issue:46015", + "issue:43872|issue:46304", + "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:43596|issue:45800", + "issue:43792|issue:45070", + "issue:45362|issue:45464", + "issue:44292|issue:45901", + "issue:39290|issue:45823", + "issue:45464|issue:45685", + "issue:43541|issue:45685", + "issue:43541|issue:43828", + "issue:36010|issue:39290", + "issue:43404|issue:43866", + "issue:44841|issue:45084", + "issue:41093|issue:41762", + "issue:43296|issue:43366", + "issue:45399|issue:46661", + "issue:43572|issue:45070", + "issue:45800|issue:45823", + "issue:44805|issue:44918" + ], + "prepared_review_unit_hash": "60435c75a71799a810b1478b5782fe4db6503a1900e99f765f97c39a310f0463", + "component_budget": { + "node_count": 473, + "item_count": 473, + "soft_pair_count": 1738, + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 41, + "serialized_chars": 39813, + "estimated_input_tokens": 9954, + "estimated_eval_tokens": 20164 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is mostly heterogeneous. There are two accepted duplicate subclusters: the ZeRO-3 load/init out-of-bounds reports (41762\u219443596) and the v5 AutoTokenizer token-ID regression reports (45812\u219445920). So 45920 is a strong representative for one cluster, but not the only clear duplicate anchor in the packet.", + "confidence": 0.86, + "canonical_issue_reason": "Use 45920 as the canonical issue for the broad v5 AutoTokenizer token-ID regression cluster: it is open and spans multiple model families. The packet also supports a separate accepted duplicate subcluster around 41762\u219443596, so 45920 should be treated as one representative cluster anchor, not the only one.", + "canonical_pr_reason": null, + "best_issue_reason": "45920 is the best overall issue representative because it is open, broad, and central to one of the two clear duplicate clusters. 41762 is also a valid duplicate-cluster anchor, but it is narrower and closed.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43866", + "right": "issue:46402", + "accept": false, + "reason": "Different failures: corrupted checkpoint / missing weights loading, not the same code path." + }, + { + "left": "issue:45800", + "right": "issue:46304", + "accept": false, + "reason": "Both are compatibility complaints, but they point to different breakages and fixes." + }, + { + "left": "issue:43296", + "right": "issue:46304", + "accept": false, + "reason": "Unrelated: model load failure in vLLM vs a PyTorch distributed-support regression." + }, + { + "left": "issue:41762", + "right": "issue:43596", + "accept": true, + "reason": "Same underlying ZeRO-3 loading bug: both report an out-of-bounds index 0 on size 0 during model init/load, with different model classes only." + }, + { + "left": "issue:39290", + "right": "issue:41720", + "accept": false, + "reason": "Different model/config and different runtime error; only both mention vLLM/auto mapping loosely." + }, + { + "left": "issue:43901", + "right": "issue:44512", + "accept": false, + "reason": "Both are docs issues, but they concern different API/documentation changes." + }, + { + "left": "issue:44589", + "right": "issue:46304", + "accept": false, + "reason": "Completely different failures: missing storage type vs distributed build/runtime support." + }, + { + "left": "issue:45237", + "right": "issue:45901", + "accept": false, + "reason": "Different subsystems and error classes; no shared bug path." + }, + { + "left": "issue:43039", + "right": "issue:44560", + "accept": false, + "reason": "Liger-kernel cross-entropy dispatch vs video chat-template StopIteration are unrelated." + }, + { + "left": "issue:45823", + "right": "issue:46294", + "accept": false, + "reason": "Different models and different device/template failures; not the same bug." + }, + { + "left": "issue:43975", + "right": "issue:46015", + "accept": false, + "reason": "Tokenizer detokenization bug vs RAG docs drift are unrelated." + }, + { + "left": "issue:44841", + "right": "issue:45823", + "accept": false, + "reason": "Different models and distinct forward/device issues." + }, + { + "left": "issue:44863", + "right": "issue:46015", + "accept": false, + "reason": "Checkpoint-loading failure vs outdated docs; unrelated." + }, + { + "left": "issue:45020", + "right": "issue:46304", + "accept": false, + "reason": "Remote-code regressions and PyTorch build support are different problem areas." + }, + { + "left": "issue:45812", + "right": "issue:45920", + "accept": true, + "reason": "Same v5 AutoTokenizer regression: wrong token IDs across Granite and other model families, so they describe the same underlying bug." + }, + { + "left": "issue:43866", + "right": "issue:44863", + "accept": false, + "reason": "Both mention loading, but one is a corrupted checkpoint and the other is a model implementation/load issue." + }, + { + "left": "issue:44292", + "right": "issue:46015", + "accept": false, + "reason": "Qwen NVFP4 runtime error and RAG docs issues are unrelated." + }, + { + "left": "issue:43872", + "right": "issue:46304", + "accept": false, + "reason": "bitsandbytes parameter-construction failure is unrelated to distributed build support." + }, + { + "left": "issue:39290", + "right": "issue:43531", + "accept": false, + "reason": "Both mention sliding_window, but the models and concrete errors differ." + }, + { + "left": "issue:41093", + "right": "issue:43638", + "accept": false, + "reason": "Both are IndexErrors, but one is a mask/tensor length mismatch and the other is a zero-size out-of-bounds load error." + }, + { + "left": "issue:43792", + "right": "issue:45901", + "accept": false, + "reason": "Whisper load failure vs table-question-answering crash are unrelated." + }, + { + "left": "issue:43854", + "right": "issue:43866", + "accept": false, + "reason": "Different model load problems with no shared failure mechanism." + }, + { + "left": "issue:43425", + "right": "issue:45070", + "accept": false, + "reason": "Torch-version incompatibility vs pydantic PretrainedConfig field break are separate regressions." + }, + { + "left": "issue:43975", + "right": "issue:45901", + "accept": false, + "reason": "Different tasks and error modes; no shared underlying defect." + }, + { + "left": "issue:43596", + "right": "issue:45800", + "accept": false, + "reason": "ZeRO-3 loading bug and torch/transformers compatibility issue are unrelated." + }, + { + "left": "issue:43792", + "right": "issue:45070", + "accept": false, + "reason": "Whisper runtime failure is unrelated to the config/pydantic regression." + }, + { + "left": "issue:45362", + "right": "issue:45464", + "accept": false, + "reason": "Both are Qwen3.5 chat-related, but the reported failures and call paths differ." + }, + { + "left": "issue:44292", + "right": "issue:45901", + "accept": false, + "reason": "Different model/runtime issue vs table QA crash; no duplicate evidence." + }, + { + "left": "issue:39290", + "right": "issue:45823", + "accept": false, + "reason": "Different models and different failure surfaces." + }, + { + "left": "issue:45464", + "right": "issue:45685", + "accept": false, + "reason": "Chat streaming failure and MPS histogram kernel error are unrelated." + }, + { + "left": "issue:43541", + "right": "issue:45685", + "accept": false, + "reason": "Different op/runtime failures; no sign of the same bug." + }, + { + "left": "issue:43541", + "right": "issue:43828", + "accept": false, + "reason": "Both involve numerical/tracing issues, but the concrete failures and affected models differ." + }, + { + "left": "issue:36010", + "right": "issue:39290", + "accept": false, + "reason": "ImportError for GenerationMixin and Gemma config attribute errors are unrelated." + }, + { + "left": "issue:43404", + "right": "issue:43866", + "accept": false, + "reason": "Weight-tying bug is unrelated to a corrupted checkpoint/load failure." + }, + { + "left": "issue:44841", + "right": "issue:45084", + "accept": false, + "reason": "Processor failure for Voxtral and template compilation errors do not share a concrete bug." + }, + { + "left": "issue:41093", + "right": "issue:41762", + "accept": false, + "reason": "Different IndexErrors from different code paths: mask-shape mismatch vs ZeRO-3 load-time out-of-bounds." + }, + { + "left": "issue:43296", + "right": "issue:43366", + "accept": false, + "reason": "Different model-support problems and different runtimes." + }, + { + "left": "issue:45399", + "right": "issue:46661", + "accept": false, + "reason": "Flash-attn fallback gating and remote-code config_class errors are unrelated." + }, + { + "left": "issue:43572", + "right": "issue:45070", + "accept": false, + "reason": "Missing pad_token_idx and pydantic PretrainedConfig-field regression are different config issues." + }, + { + "left": "issue:45800", + "right": "issue:45823", + "accept": false, + "reason": "Torch/version compatibility vs device-map mismatch are unrelated." + }, + { + "left": "issue:44805", + "right": "issue:44918", + "accept": false, + "reason": "Same broad area of model inputs, but different errors and different code paths." + } + ] + }, + "evaluator_result": { + "accept": false, + "feedback": "The 45920 anchor is plausible, but the write-up overstates uniqueness. The packet supports two accepted duplicate subclusters: 41762\u219443596 and 45812\u219445920. So saying 45920 anchors the \"only clear duplicate subcluster\" is not grounded. Please revise the summary to acknowledge both clusters and keep the duplicate claims more conservative." + }, + "evaluator_used": true, + "retried": true, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 21, + "review_unit_index": 3, + "review_unit_count": 56, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-473", + "nodes": [ + "issue:39186", + "issue:40990", + "issue:41093", + "issue:42915", + "issue:43054", + "issue:43366", + "issue:43404", + "issue:43531", + "issue:43541", + "issue:43596", + "issue:43645", + "issue:43716", + "issue:43828", + "issue:43873", + "issue:43975", + "issue:43994", + "issue:44291", + "issue:44292", + "issue:44360", + "issue:44387", + "issue:44479", + "issue:44492", + "issue:44512", + "issue:44530", + "issue:44589", + "issue:44863", + "issue:44918", + "issue:44960", + "issue:45005", + "issue:45020", + "issue:45070", + "issue:45161", + "issue:45183", + "issue:45399", + "issue:45464", + "issue:45507", + "issue:45632", + "issue:45698", + "issue:45759", + "issue:45800", + "issue:45902", + "issue:46294", + "issue:46463", + "issue:46489", + "issue:46491", + "issue:46640", + "issue:46661" + ], + "soft_pairs": [ + "issue:44387|issue:45005", + "issue:43541|issue:45161", + "issue:43828|issue:45464", + "issue:43366|issue:45464", + "issue:44291|issue:45070", + "issue:43531|issue:45070", + "issue:43716|issue:46294", + "issue:43541|issue:44589", + "issue:45020|issue:45800", + "issue:41093|issue:43596", + "issue:44479|issue:46294", + "issue:43054|issue:43994", + "issue:45759|issue:45902", + "issue:46640|issue:46661", + "issue:39186|issue:44918", + "issue:43716|issue:44918", + "issue:40990|issue:44960", + "issue:45632|issue:45698", + "issue:43645|issue:45399", + "issue:43404|issue:45005", + "issue:42915|issue:45399", + "issue:44530|issue:45005", + "issue:43873|issue:45005", + "issue:44360|issue:44512", + "issue:44863|issue:45005", + "issue:45183|issue:46463", + "issue:44492|issue:44512", + "issue:46489|issue:46491", + "issue:44918|issue:45507", + "issue:43975|issue:44292", + "issue:44292|issue:45507" + ], + "prepared_review_unit_hash": "9061d7c7ce21e8e9fb4d9b7d5543dec825dff52b3e47925708c87c7312e239d8", + "component_budget": { + "node_count": 473, + "item_count": 473, + "soft_pair_count": 1738, + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 31, + "serialized_chars": 36232, + "estimated_input_tokens": 9058, + "estimated_eval_tokens": 18372 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a heterogeneous mix of unrelated Transformers bug reports. The only plausible duplicate pair is the remote-code/custom-module path-collision reports (45632/45698); the rest should stay separate.", + "confidence": 0.74, + "canonical_issue_reason": "issue:45632 is the clearest root-cause report: it explicitly describes a trust_remote_code cache-path collision for local models sharing a leaf directory name, which matches the symptom in 45698.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:45632 is the best anchor for this cluster because it states the underlying loading/cache collision directly and is the most likely canonical duplicate target for 45698.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44387", + "right": "issue:45005", + "accept": false, + "reason": "Both concern training/quantization quirks, but one is int4 OOM memory behavior and the other is tied-weights handling; different bugs." + }, + { + "left": "issue:43541", + "right": "issue:45161", + "accept": false, + "reason": "Grouped_mm tracing errors in Mixtral are unrelated to GPT-OSS tensor-parallel support." + }, + { + "left": "issue:43828", + "right": "issue:45464", + "accept": false, + "reason": "Phi MoE autocast dtype mismatch is unrelated to Qwen3.5 streaming/chat-completions video inference." + }, + { + "left": "issue:43366", + "right": "issue:45464", + "accept": false, + "reason": "GGUF gpt-oss architecture support is a different feature gap from Qwen3.5 streaming/video failures." + }, + { + "left": "issue:44291", + "right": "issue:45070", + "accept": false, + "reason": "Unexpected _is_hf_initialized argument in init_empty_weights is a different loading bug than the pydantic PretrainedConfig field regression." + }, + { + "left": "issue:43531", + "right": "issue:45070", + "accept": false, + "reason": "Sliding-window behavior in Qwen3-MoE is unrelated to the PretrainedConfig/pydantic breakage." + }, + { + "left": "issue:43716", + "right": "issue:46294", + "accept": false, + "reason": "Both involve multimodal dtype/input issues, but one is image-preprocessor dtype mismatch and the other is video token/feature mismatch; not the same bug." + }, + { + "left": "issue:43541", + "right": "issue:44589", + "accept": false, + "reason": "Float32 grouped_mm tracing failure and missing Float8 storage type are distinct runtime issues." + }, + { + "left": "issue:45020", + "right": "issue:45800", + "accept": false, + "reason": "Generic remote_code breakage is broader than the specific torch 2.4.1 vs transformers 5.8.0 incompatibility report." + }, + { + "left": "issue:41093", + "right": "issue:43596", + "accept": false, + "reason": "Mask-length mismatch and zero-sized tensor index error are different failure modes with no clear shared code path." + }, + { + "left": "issue:44479", + "right": "issue:46294", + "accept": false, + "reason": "Both mention Qwen video input regressions, but 44479 is a broader v5.3.0 regression across several models while 46294 is a specific chat-template token/feature mismatch." + }, + { + "left": "issue:43054", + "right": "issue:43994", + "accept": false, + "reason": "Siglip2 embedding quality and nonsensical AutoModel/pipeline output are related in topic, but not clearly the same concrete bug." + }, + { + "left": "issue:45759", + "right": "issue:45902", + "accept": false, + "reason": "Text_config unwrapping in from_config and uninitialized inv_freq after meta\u2192to_empty(cuda) are separate multimodal initialization issues." + }, + { + "left": "issue:46640", + "right": "issue:46661", + "accept": false, + "reason": "Flash Attention 2 integration failure is unrelated to remote-code models missing config_class." + }, + { + "left": "issue:39186", + "right": "issue:44918", + "accept": false, + "reason": "FSDP 2-D weight error and Qwen3.5 embedding unpacking in TRL are different model/training issues." + }, + { + "left": "issue:43716", + "right": "issue:44918", + "accept": false, + "reason": "Image-preprocessor dtype mismatch is not the same as TRL SFT unpacking of Qwen3.5 input embeddings." + }, + { + "left": "issue:40990", + "right": "issue:44960", + "accept": false, + "reason": "High perplexity on gpt-oss-20b and GLM5 are unrelated model-specific reports." + }, + { + "left": "issue:45632", + "right": "issue:45698", + "accept": true, + "reason": "Both point to the same remote-code/custom-module loading problem: path collisions cause the wrong module to be loaded after save/load or when models share a leaf directory name." + }, + { + "left": "issue:43645", + "right": "issue:45399", + "accept": false, + "reason": "Jupyter custom-model initialization regression is unrelated to flash-attn2 fallback gating." + }, + { + "left": "issue:43404", + "right": "issue:45005", + "accept": false, + "reason": "A specific untied lm_head in Mistral3ForConditionalGeneration is not the same bug as translation-model tied-weights issues in v5." + }, + { + "left": "issue:42915", + "right": "issue:45399", + "accept": false, + "reason": "FineGrainedFP8Config failure on Qwen3Moe is unrelated to flash-attn2 fallback checks." + }, + { + "left": "issue:44530", + "right": "issue:45005", + "accept": false, + "reason": "PagedAttentionCache group-type crash on Qwen3.5 is unrelated to tied-weight regressions." + }, + { + "left": "issue:43873", + "right": "issue:45005", + "accept": false, + "reason": "Quantization/offloading memory behavior is a different issue from tied weights." + }, + { + "left": "issue:44360", + "right": "issue:44512", + "accept": false, + "reason": "A DSA indexer ReLU discussion is unrelated to a docs typo." + }, + { + "left": "issue:44863", + "right": "issue:45005", + "accept": false, + "reason": "NemotronH checkpoint loading failure is unrelated to translation-model tied weights." + }, + { + "left": "issue:45183", + "right": "issue:46463", + "accept": false, + "reason": "Generic XOR validation errors across models are broader than the T5 decoder-input error message bug; not clearly the same code-path defect." + }, + { + "left": "issue:44492", + "right": "issue:44512", + "accept": false, + "reason": "Cache-strategy typo and removed run-command docs are unrelated documentation issues." + }, + { + "left": "issue:46489", + "right": "issue:46491", + "accept": false, + "reason": "Both are tokenizer regressions, but they affect different model families and likely different code paths." + }, + { + "left": "issue:44918", + "right": "issue:45507", + "accept": false, + "reason": "Qwen3.5 SFT embedding unpacking and GraniteMoEHybrid invalid method calls are different model-specific bugs." + }, + { + "left": "issue:43975", + "right": "issue:44292", + "accept": false, + "reason": "DeepSeek-Coder tokenizer detokenization regression is unrelated to init_empty_weights argument handling." + }, + { + "left": "issue:44292", + "right": "issue:45507", + "accept": false, + "reason": "Unexpected _is_hf_initialized handling and GraniteMoEHybrid invalid method calls are unrelated." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The cluster summary is grounded in the packet: most items are clearly unrelated, and the 45632/45698 pair is a plausible canonical duplicate around trust_remote_code/custom-module loading path collisions. The soft-edge rejections are conservative and do not overstate evidence." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 21, + "review_unit_index": 4, + "review_unit_count": 56, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-473", + "nodes": [ + "issue:24643", + "issue:28530", + "issue:36331", + "issue:38175", + "issue:39692", + "issue:41553", + "issue:41762", + "issue:42491", + "issue:42915", + "issue:42947", + "issue:43054", + "issue:43493", + "issue:43643", + "issue:43644", + "issue:43782", + "issue:43824", + "issue:43828", + "issue:43856", + "issue:43866", + "issue:43872", + "issue:43881", + "issue:43883", + "issue:43950", + "issue:43975", + "issue:44368", + "issue:44387", + "issue:44403", + "issue:44451", + "issue:44488", + "issue:44534", + "issue:44589", + "issue:44661", + "issue:44910", + "issue:44928", + "issue:44960", + "issue:44991", + "issue:45020", + "issue:45081", + "issue:45399", + "issue:45520", + "issue:45663", + "issue:45685", + "issue:45701", + "issue:45823", + "issue:45859", + "issue:45902", + "issue:45920", + "issue:46489" + ], + "soft_pairs": [ + "issue:43950|issue:44534", + "issue:39692|issue:43054", + "issue:43644|issue:43950", + "issue:36331|issue:41762", + "issue:43828|issue:44928", + "issue:43872|issue:44589", + "issue:45920|issue:46489", + "issue:43054|issue:45701", + "issue:43866|issue:44403", + "issue:43950|issue:45902", + "issue:38175|issue:43054", + "issue:41553|issue:43883", + "issue:44910|issue:45859", + "issue:42947|issue:44387", + "issue:43054|issue:43493", + "issue:45663|issue:45902", + "issue:41553|issue:43881", + "issue:44403|issue:44451", + "issue:43975|issue:45520", + "issue:24643|issue:28530", + "issue:42947|issue:43856", + "issue:44403|issue:44488", + "issue:43643|issue:45020", + "issue:41762|issue:42915", + "issue:44991|issue:45399", + "issue:45685|issue:45823", + "issue:43782|issue:44960", + "issue:43824|issue:43872", + "issue:43881|issue:43883", + "issue:45020|issue:45823", + "issue:44661|issue:45020", + "issue:45081|issue:45399", + "issue:42491|issue:44368", + "issue:43828|issue:45823" + ], + "prepared_review_unit_hash": "e81250a261248f6fe31b197efe0932332f48daff0cd262373d4fc0848c75de96", + "component_budget": { + "node_count": 473, + "item_count": 473, + "soft_pair_count": 1738, + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 34, + "serialized_chars": 38115, + "estimated_input_tokens": 9529, + "estimated_eval_tokens": 19314 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Most pairs are unrelated. The strongest duplicate clusters are the non-persistent-buffer corruption regression, the SigLIP2 implementation/discrepancy reports, and the remote_code loading regression. No pull requests were present.", + "confidence": 0.79, + "canonical_issue_reason": "issue:45920 is the best umbrella issue: it is open, broad, and explicitly covers a multi-model tokenizer regression affecting several model families, so it is the most useful canonical anchor for deduping nearby reports.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:45920 has the broadest scope and strongest canonical value for grouping related tokenizer-regression reports, and it remains open with inbound references.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43950", + "right": "issue:44534", + "accept": true, + "reason": "Same non-persistent-buffer corruption regression in Transformers v5; the titles describe the same defect with different wording." + }, + { + "left": "issue:39692", + "right": "issue:43054", + "accept": false, + "reason": "Docs/example errors and quantization mismatch are not the same bug as SigLIP2 embedding quality degradation." + }, + { + "left": "issue:43644", + "right": "issue:43950", + "accept": true, + "reason": "Both report the same v5 regression where non-persistent buffers are corrupted or junk-filled on load." + }, + { + "left": "issue:36331", + "right": "issue:41762", + "accept": false, + "reason": "Unexpected compute_loss kwarg and ZeRO-3 loading IndexError are unrelated bugs." + }, + { + "left": "issue:43828", + "right": "issue:44928", + "accept": false, + "reason": "Different Qwen/Phi failure modes and different root causes; not the same code-path bug." + }, + { + "left": "issue:43872", + "right": "issue:44589", + "accept": false, + "reason": "Both are low-level quantization issues, but they affect different storage/parameter paths." + }, + { + "left": "issue:45920", + "right": "issue:46489", + "accept": false, + "reason": "Related tokenizer regressions, but 46489 is a DeepSeek-Coder-specific detokenization gap rather than the same concrete issue." + }, + { + "left": "issue:43054", + "right": "issue:45701", + "accept": false, + "reason": "SigLIP2 embedding quality and a generic tokenization-version change are not the same bug." + }, + { + "left": "issue:43866", + "right": "issue:44403", + "accept": false, + "reason": "Checkpoint corruption and loading noise are different problems." + }, + { + "left": "issue:43950", + "right": "issue:45902", + "accept": false, + "reason": "43950 is a generic non-persistent-buffer corruption regression, while 45902 is a specific Qwen3_5MoeTextRotaryEmbedding uninitialized-memory/materialization bug; the evidence does not show the same underlying defect." + }, + { + "left": "issue:38175", + "right": "issue:43054", + "accept": false, + "reason": "Both involve SigLIP2, but zero probabilities and degraded text embeddings are not clearly the same underlying defect." + }, + { + "left": "issue:41553", + "right": "issue:43883", + "accept": false, + "reason": "Bad AutoTokenizer error messaging and a missing attribute on Molmo are unrelated." + }, + { + "left": "issue:44910", + "right": "issue:45859", + "accept": false, + "reason": "Flash-Attn packed-sequence crash and CPU-offload incompatibility are different Qwen3.5 issues." + }, + { + "left": "issue:42947", + "right": "issue:44387", + "accept": false, + "reason": "Gradient checkpointing ineffective and int4 quantization OOM are distinct memory bugs." + }, + { + "left": "issue:43054", + "right": "issue:43493", + "accept": true, + "reason": "Both point to the same SigLIP2 implementation discrepancy, which plausibly explains the degraded embeddings." + }, + { + "left": "issue:45663", + "right": "issue:45902", + "accept": false, + "reason": "Gemma4 FSDP2 KeyError is unrelated to non-persistent-buffer materialization corruption." + }, + { + "left": "issue:41553", + "right": "issue:43881", + "accept": false, + "reason": "Different models and different loading failures; no shared underlying bug." + }, + { + "left": "issue:44403", + "right": "issue:44451", + "accept": false, + "reason": "Generic load noise vs inability to load a specific model are not the same defect." + }, + { + "left": "issue:43975", + "right": "issue:45520", + "accept": false, + "reason": "Tokenizer detokenization bug and a Python 3.13 flash_attn import error are unrelated." + }, + { + "left": "issue:24643", + "right": "issue:28530", + "accept": false, + "reason": "DeepSpeed training RuntimeError and early-stopping metric handling are unrelated." + }, + { + "left": "issue:42947", + "right": "issue:43856", + "accept": false, + "reason": "LoRA checkpointing and Qwen3 MoE memory usage are different performance/memory issues." + }, + { + "left": "issue:44403", + "right": "issue:44488", + "accept": false, + "reason": "Load noise and a specific model load failure are not the same bug." + }, + { + "left": "issue:43643", + "right": "issue:45020", + "accept": true, + "reason": "Same remote_code regression surface: AutoConfig missing fields is a concrete manifestation of models using remote_code breaking in recent versions." + }, + { + "left": "issue:41762", + "right": "issue:42915", + "accept": false, + "reason": "Gemma3 ZeRO-3 load failure and Qwen3Moe FineGrainedFP8 failure are unrelated." + }, + { + "left": "issue:44991", + "right": "issue:45399", + "accept": false, + "reason": "Tokenizer loading failure and flash-attn fallback gating are unrelated." + }, + { + "left": "issue:45685", + "right": "issue:45823", + "accept": false, + "reason": "MPS histogram-on-int error and Gemma4 device mismatch are different device-placement bugs." + }, + { + "left": "issue:43782", + "right": "issue:44960", + "accept": false, + "reason": "Qwen3VL weight_only loading and GLM5 are unrelated." + }, + { + "left": "issue:43824", + "right": "issue:43872", + "accept": false, + "reason": "Importing a missing model class and bitsandbytes Int8Params construction are unrelated." + }, + { + "left": "issue:43881", + "right": "issue:43883", + "accept": false, + "reason": "Different model load failures with different symptoms and causes." + }, + { + "left": "issue:45020", + "right": "issue:45823", + "accept": false, + "reason": "Remote_code regression and Gemma4 PLE device mismatch are unrelated." + }, + { + "left": "issue:44661", + "right": "issue:45020", + "accept": false, + "reason": "Tokenizer-mapping edge case and remote_code regression are different code paths." + }, + { + "left": "issue:45081", + "right": "issue:45399", + "accept": false, + "reason": "Mistral regex patch crash and flash-attn fallback gating are unrelated." + }, + { + "left": "issue:42491", + "right": "issue:44368", + "accept": false, + "reason": "LoRA model incompatibility across branches and a tied-embeddings warning are not the same bug." + }, + { + "left": "issue:43828", + "right": "issue:45823", + "accept": false, + "reason": "Autocast dtype mismatch and Gemma4 device mismatch are different forward-pass failures." + } + ] + }, + "evaluator_result": { + "accept": false, + "feedback": "The overall grouping is mostly grounded, but at least one soft-edge accept is too strong for the evidence in the packet. In particular, issue:43950 \u2194 issue:45902 is not clearly the same bug from the titles alone: 43950 is a generic non-persistent-buffer corruption regression, while 45902 is a specific Qwen3_5MoeTextRotaryEmbedding uninitialized-memory/materialization issue. The SigLIP2 and remote_code links are more plausible, but the verdict set is not uniformly conservative enough to accept." + }, + "evaluator_used": true, + "retried": true, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 21, + "review_unit_index": 5, + "review_unit_count": 56, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-473", + "nodes": [ + "issue:24643", + "issue:30064", + "issue:36331", + "issue:41093", + "issue:41553", + "issue:42915", + "issue:42947", + "issue:43257", + "issue:43366", + "issue:43404", + "issue:43531", + "issue:43596", + "issue:43792", + "issue:43827", + "issue:43828", + "issue:43854", + "issue:43872", + "issue:43940", + "issue:43950", + "issue:44220", + "issue:44291", + "issue:44479", + "issue:44488", + "issue:44509", + "issue:44530", + "issue:44560", + "issue:44805", + "issue:44863", + "issue:44928", + "issue:44991", + "issue:45020", + "issue:45305", + "issue:45313", + "issue:45325", + "issue:45356", + "issue:45488", + "issue:45588", + "issue:45663", + "issue:45715", + "issue:45759", + "issue:45823", + "issue:45902", + "issue:45923", + "issue:46315", + "issue:46402", + "issue:46489", + "issue:46515", + "issue:46661" + ], + "soft_pairs": [ + "issue:45020|issue:45356", + "issue:45715|issue:46315", + "issue:45488|issue:46489", + "issue:43827|issue:44509", + "issue:43404|issue:43950", + "issue:45759|issue:46661", + "issue:43872|issue:44291", + "issue:42915|issue:43872", + "issue:45020|issue:46661", + "issue:43792|issue:44220", + "issue:43257|issue:45305", + "issue:42947|issue:45923", + "issue:41553|issue:44991", + "issue:36331|issue:41093", + "issue:43940|issue:45313", + "issue:43854|issue:44863", + "issue:44928|issue:45588", + "issue:44479|issue:44560", + "issue:41093|issue:44805", + "issue:41553|issue:44488", + "issue:24643|issue:30064", + "issue:44863|issue:46402", + "issue:44530|issue:44863", + "issue:43366|issue:43828", + "issue:43596|issue:46515", + "issue:45325|issue:45902", + "issue:43366|issue:43531", + "issue:45663|issue:45823" + ], + "prepared_review_unit_hash": "4061509026ecae0360822f0e21a6a92947f2cd5ed1f7a384810c8548c9ecad62", + "component_budget": { + "node_count": 473, + "item_count": 473, + "soft_pair_count": 1738, + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 28, + "serialized_chars": 36625, + "estimated_input_tokens": 9157, + "estimated_eval_tokens": 18570 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly heterogeneous; there is one clear duplicate pair (docs mentioning removed pipeline tasks) and the rest look like separate regressions across different model/tokenizer/loading paths.", + "confidence": 0.79, + "canonical_issue_reason": "Issue 45020 is the broadest umbrella in the set: it frames a general v5 regression affecting remote-code model loading, which is the closest match to several of the nearby loading/tokenizer breakages.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 45020 is the most representative cluster anchor because it is broad enough to subsume related v5 loading regressions, even though most other items are still distinct.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:45020", + "right": "issue:45356", + "accept": false, + "reason": "Both are version-related loading/tokenizer regressions, but they target different concrete failures and model paths." + }, + { + "left": "issue:45715", + "right": "issue:46315", + "accept": false, + "reason": "Both are tokenizer performance regressions, but they affect different conversion directions and different hot paths." + }, + { + "left": "issue:45488", + "right": "issue:46489", + "accept": false, + "reason": "Both involve DeepSeek tokenization, but the reported failures and affected model families are not the same concrete bug." + }, + { + "left": "issue:43827", + "right": "issue:44509", + "accept": true, + "reason": "Same docs regression: both report stale pipeline-task documentation for removed v5 pipelines." + }, + { + "left": "issue:43404", + "right": "issue:43950", + "accept": false, + "reason": "Different failures: untied lm_head weights versus corrupted non-persistent buffers." + }, + { + "left": "issue:45759", + "right": "issue:46661", + "accept": false, + "reason": "Both involve model construction/loading, but they are different API failures and not the same code-path bug." + }, + { + "left": "issue:43872", + "right": "issue:44291", + "accept": false, + "reason": "Shared '_is_hf_initialized' wording, but the failures arise from different loading mechanisms and are not clearly one bug." + }, + { + "left": "issue:42915", + "right": "issue:43872", + "accept": false, + "reason": "Different incompatibilities: FP8 config handling versus bitsandbytes parameter construction." + }, + { + "left": "issue:45020", + "right": "issue:46661", + "accept": false, + "reason": "Both are remote-code loading problems, but one is a broad regression and the other is a specific missing config_class attribute issue." + }, + { + "left": "issue:43792", + "right": "issue:44220", + "accept": false, + "reason": "Both are whisper/audio related, but the titles do not indicate the same concrete failure." + }, + { + "left": "issue:43257", + "right": "issue:45305", + "accept": false, + "reason": "Both mention DeepSpeed, but one is weight conversion/loading and the other is gradient averaging behavior." + }, + { + "left": "issue:42947", + "right": "issue:45923", + "accept": false, + "reason": "Gradient-checkpointing problems, but one is ineffective LoRA checkpointing and the other is a missing capability flag." + }, + { + "left": "issue:41553", + "right": "issue:44991", + "accept": false, + "reason": "Both are tokenizer-loading issues, but they concern different models and different failure modes." + }, + { + "left": "issue:36331", + "right": "issue:41093", + "accept": false, + "reason": "Both are trainer/indexing errors, but the specific kwargs/mask mismatch bugs are different." + }, + { + "left": "issue:43940", + "right": "issue:45313", + "accept": false, + "reason": "Both are DeepSpeed ZeRO-3 loading problems, but one is missing all weights for Qwen3-Next and the other is language_model weight loading for Qwen3.5." + }, + { + "left": "issue:43854", + "right": "issue:44863", + "accept": false, + "reason": "Different model-loading failures for different architectures." + }, + { + "left": "issue:44928", + "right": "issue:45588", + "accept": false, + "reason": "Both touch attention/cache paths, but one is RLHF position-id instability and the other is a flash-attention auxiliary-state crash." + }, + { + "left": "issue:44479", + "right": "issue:44560", + "accept": false, + "reason": "Both are Qwen video regressions, but they are distinct symptoms and model variants." + }, + { + "left": "issue:41093", + "right": "issue:44805", + "accept": false, + "reason": "Similar error template, but the mismatched shapes are different and there is no evidence of one shared root cause." + }, + { + "left": "issue:41553", + "right": "issue:44488", + "accept": false, + "reason": "Both concern tokenizer loading, but they are different models and different error reports." + }, + { + "left": "issue:24643", + "right": "issue:30064", + "accept": false, + "reason": "Training DeepSpeed weight-shape error versus image-processor segmentation-map handling; unrelated bugs." + }, + { + "left": "issue:44863", + "right": "issue:46402", + "accept": false, + "reason": "Different loading failures for different model families." + }, + { + "left": "issue:44530", + "right": "issue:44863", + "accept": false, + "reason": "Different Qwen/GGUF loading issues with unrelated failure mechanisms." + }, + { + "left": "issue:43366", + "right": "issue:43828", + "accept": false, + "reason": "Unrelated: GGUF architecture support versus autocast dtype mismatch." + }, + { + "left": "issue:43596", + "right": "issue:46515", + "accept": false, + "reason": "Different incompatibilities: ZeRO-3/Bert index error versus torch version incompatibility." + }, + { + "left": "issue:45325", + "right": "issue:45902", + "accept": false, + "reason": "Both concern Qwen position/rotary internals, but the reported bugs are different and not the same code path." + }, + { + "left": "issue:43366", + "right": "issue:43531", + "accept": false, + "reason": "Different Qwen-related bugs: GGUF architecture support versus sliding-window handling." + }, + { + "left": "issue:45663", + "right": "issue:45823", + "accept": false, + "reason": "Same model family, but one is an FSDP2 KeyError and the other is a device-map mismatch." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The cluster summary is grounded in the packet and stays conservative. The one accepted soft pair (43827/44509) is well supported by the titles, and the remaining false verdicts avoid overclaiming shared root cause. I don\u2019t see evidence of overstated links or an unmergeable PR-pair issue here." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 21, + "review_unit_index": 6, + "review_unit_count": 56, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-473", + "nodes": [ + "issue:36010", + "issue:38175", + "issue:39692", + "issue:41720", + "issue:42491", + "issue:42617", + "issue:42915", + "issue:43317", + "issue:43366", + "issue:43550", + "issue:43596", + "issue:43632", + "issue:43638", + "issue:43644", + "issue:43645", + "issue:43646", + "issue:43716", + "issue:43720", + "issue:43824", + "issue:43950", + "issue:43957", + "issue:43994", + "issue:44155", + "issue:44451", + "issue:44488", + "issue:44534", + "issue:44560", + "issue:44589", + "issue:44661", + "issue:44683", + "issue:44756", + "issue:44805", + "issue:44877", + "issue:44910", + "issue:44918", + "issue:44960", + "issue:44977", + "issue:45030", + "issue:45276", + "issue:45584", + "issue:45685", + "issue:45812", + "issue:45823", + "issue:45834", + "issue:46459", + "issue:46489", + "issue:46512", + "issue:46515" + ], + "soft_pairs": [ + "issue:43638|issue:46515", + "issue:44155|issue:45584", + "issue:43716|issue:44560", + "issue:43950|issue:45834", + "issue:43366|issue:44589", + "issue:43824|issue:43957", + "issue:44560|issue:45685", + "issue:42915|issue:43957", + "issue:43645|issue:43646", + "issue:43550|issue:44910", + "issue:44560|issue:44918", + "issue:44683|issue:46515", + "issue:42617|issue:43366", + "issue:45812|issue:46489", + "issue:36010|issue:42915", + "issue:45823|issue:46515", + "issue:44661|issue:45823", + "issue:44560|issue:44805", + "issue:38175|issue:43994", + "issue:43632|issue:45834", + "issue:44877|issue:45030", + "issue:43638|issue:44661", + "issue:43596|issue:44661", + "issue:43317|issue:44756", + "issue:44661|issue:44805", + "issue:42491|issue:43720", + "issue:38175|issue:39692", + "issue:43644|issue:44534", + "issue:41720|issue:42915", + "issue:42491|issue:45276", + "issue:42915|issue:44756", + "issue:44451|issue:44488", + "issue:46459|issue:46512", + "issue:44960|issue:44977" + ], + "prepared_review_unit_hash": "071d11b9cc084d3556931ef8072de4e7244037addda7ce4319806b28ee1659a4", + "component_budget": { + "node_count": 473, + "item_count": 473, + "soft_pair_count": 1738, + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 34, + "serialized_chars": 37773, + "estimated_input_tokens": 9444, + "estimated_eval_tokens": 19144 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Mostly a noisy cluster of unrelated Transformers issues. The only clear duplicate pair is the Transformers v5 non-persistent-buffer corruption regression; there is also a likely duplicate pair about custom model initialization in v5.", + "confidence": 0.79, + "canonical_issue_reason": "Issue 43644 is the clearest and earliest report of the shared Transformers 5.0.0 non-persistent-buffer corruption bug; 44534 is the same regression restated later.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 43644 is the best anchor for the cluster because it describes the core bug most explicitly and is the earliest representative of the only strong duplicate group.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43638", + "right": "issue:46515", + "accept": false, + "reason": "Different failures: Zero3/Bert index error vs a torch 2.4.1 compatibility complaint." + }, + { + "left": "issue:44155", + "right": "issue:45584", + "accept": false, + "reason": "Both are audio/generation problems, but the symptoms and code paths are different." + }, + { + "left": "issue:43716", + "right": "issue:44560", + "accept": false, + "reason": "Different model families and failure modes; only both are multimodal regressions." + }, + { + "left": "issue:43950", + "right": "issue:45834", + "accept": false, + "reason": "Both mention loading/model behavior, but the bugs are not the same concrete failure." + }, + { + "left": "issue:43366", + "right": "issue:44589", + "accept": false, + "reason": "Unrelated issues: GGUF gpt-oss support vs a Float8 storage/load error." + }, + { + "left": "issue:43824", + "right": "issue:43957", + "accept": false, + "reason": "Different model-loading regressions; one is an import failure, the other is meta-device loading breakage." + }, + { + "left": "issue:44560", + "right": "issue:45685", + "accept": false, + "reason": "Both involve model internals, but they describe different bugs and platforms." + }, + { + "left": "issue:42915", + "right": "issue:43957", + "accept": false, + "reason": "Shared broad v5 regression theme, but the concrete failures differ." + }, + { + "left": "issue:43645", + "right": "issue:43646", + "accept": true, + "reason": "Same Transformers v5 custom-model initialization breakage; the Jupyter detail is just a usage context." + }, + { + "left": "issue:43550", + "right": "issue:44910", + "accept": false, + "reason": "Different model-specific attention/compile failures with no clear shared root cause." + }, + { + "left": "issue:44560", + "right": "issue:44918", + "accept": false, + "reason": "Both are Qwen3.5-related, but one is video embedding StopIteration and the other is SFT input-embedding unpacking." + }, + { + "left": "issue:44683", + "right": "issue:46515", + "accept": false, + "reason": "Different issues: compiled flex_attention vs a torch version incompatibility report." + }, + { + "left": "issue:42617", + "right": "issue:43366", + "accept": false, + "reason": "Not the same bug; one is a script/runtime failure, the other is a model-architecture support request." + }, + { + "left": "issue:45812", + "right": "issue:46489", + "accept": false, + "reason": "Related tokenizer regressions, but different models and different concrete failures." + }, + { + "left": "issue:36010", + "right": "issue:42915", + "accept": false, + "reason": "ImportError for GenerationMixin is unrelated to the Qwen3Moe FP8 failure." + }, + { + "left": "issue:45823", + "right": "issue:46515", + "accept": false, + "reason": "Device-mapping forward mismatch is not the same as a torch 2.4.1 incompatibility issue." + }, + { + "left": "issue:44661", + "right": "issue:45823", + "accept": false, + "reason": "Different bug classes: tokenizer mapping-name validation vs Gemma4 PLE device mismatch." + }, + { + "left": "issue:44560", + "right": "issue:44805", + "accept": false, + "reason": "Both are index/error reports, but the models and code paths differ." + }, + { + "left": "issue:38175", + "right": "issue:43994", + "accept": false, + "reason": "Same SigLIP2 model family, but the reports are too different to treat as the same bug confidently." + }, + { + "left": "issue:43632", + "right": "issue:45834", + "accept": false, + "reason": "Both have index-related symptoms, but the underlying failures are not clearly the same." + }, + { + "left": "issue:44877", + "right": "issue:45030", + "accept": false, + "reason": "Different config-validation failures with no clear shared root cause." + }, + { + "left": "issue:43638", + "right": "issue:44661", + "accept": false, + "reason": "Zero3/Bert index error is unrelated to tokenizer mapping-name handling." + }, + { + "left": "issue:43596", + "right": "issue:44661", + "accept": false, + "reason": "Different failures: Zero3/Bert initialization vs tokenizer mapping validation." + }, + { + "left": "issue:43317", + "right": "issue:44756", + "accept": false, + "reason": "Both are loading/offload related, but one is dequantized model offload and the other is mmap/OOM on Strix Halo." + }, + { + "left": "issue:44661", + "right": "issue:44805", + "accept": false, + "reason": "No evidence they are the same bug; just both are loader/config issues." + }, + { + "left": "issue:42491", + "right": "issue:43720", + "accept": false, + "reason": "Different model- and framework-specific loading/training issues." + }, + { + "left": "issue:38175", + "right": "issue:39692", + "accept": false, + "reason": "Both involve SigLIP2, but one is wrong outputs and the other is a documentation example with multiple errors." + }, + { + "left": "issue:43644", + "right": "issue:44534", + "accept": true, + "reason": "Exact same Transformers v5 regression: non-persistent buffers are filled with junk on load." + }, + { + "left": "issue:41720", + "right": "issue:42915", + "accept": false, + "reason": "Qwen3 auto-device-map CUDA assert is unrelated to Qwen3Moe FP8 config failure." + }, + { + "left": "issue:42491", + "right": "issue:45276", + "accept": false, + "reason": "Different bugs: qwen3_moe LoRA compatibility vs Gemma4 embedding resize behavior." + }, + { + "left": "issue:42915", + "right": "issue:44756", + "accept": false, + "reason": "Different subsystems and failure modes; only a broad loading-performance resemblance." + }, + { + "left": "issue:44451", + "right": "issue:44488", + "accept": false, + "reason": "Similar symptom of loading failure, but the affected models differ and no shared concrete bug is shown." + }, + { + "left": "issue:46459", + "right": "issue:46512", + "accept": false, + "reason": "Both mention dtype, but one is silent dtype ignore on checkpoint load and the other is stale config dtype after from_config." + }, + { + "left": "issue:44960", + "right": "issue:44977", + "accept": false, + "reason": "No clear duplicate relation; GLM5 and Qwen3.5 flash-attention issues are different." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The cluster summary is grounded in the packet: 43644/44534 are clearly the same non-persistent-buffer regression, and 43645/43646 are a plausible duplicate pair on custom model initialization. The rejected soft edges are generally conservative and do not overclaim duplicates from weak title similarity." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 21, + "review_unit_index": 7, + "review_unit_count": 56, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-473", + "nodes": [ + "issue:36010", + "issue:39404", + "issue:39692", + "issue:41720", + "issue:42491", + "issue:42947", + "issue:43257", + "issue:43404", + "issue:43425", + "issue:43454", + "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:44488", + "issue:44756", + "issue:44910", + "issue:44960", + "issue:44991", + "issue:45276", + "issue:45292", + "issue:45313", + "issue:45335", + "issue:45357", + "issue:45362", + "issue:45406", + "issue:45488", + "issue:45507", + "issue:45588", + "issue:45663", + "issue:45704", + "issue:45759", + "issue:45812", + "issue:45859", + "issue:46050", + "issue:46468" + ], + "soft_pairs": [ + "issue:45313|issue:45357", + "issue:45507|issue:46468", + "issue:43856|issue:43873", + "issue:44488|issue:44991", + "issue:43644|issue:43645", + "issue:43708|issue:44075", + "issue:39692|issue:43994", + "issue:43856|issue:44387", + "issue:43257|issue:43854", + "issue:43854|issue:43931", + "issue:42491|issue:45292", + "issue:42491|issue:43931", + "issue:44960|issue:46050", + "issue:45488|issue:45812", + "issue:45362|issue:45406", + "issue:45276|issue:45335", + "issue:44164|issue:44291", + "issue:44910|issue:45588", + "issue:44402|issue:44488", + "issue:43404|issue:43454", + "issue:42947|issue:45704", + "issue:41720|issue:44155", + "issue:43873|issue:44387", + "issue:39692|issue:44960", + "issue:43425|issue:43957", + "issue:45663|issue:45759", + "issue:44291|issue:44756", + "issue:36010|issue:39404", + "issue:45663|issue:45859", + "issue:44387|issue:44756" + ], + "prepared_review_unit_hash": "ed56ad1fb12d9f41b0d9cd1eff3c8ff0c070f5dd81112d6d76be3f6a9a750de4", + "component_budget": { + "node_count": 473, + "item_count": 473, + "soft_pair_count": 1738, + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 30, + "serialized_chars": 36207, + "estimated_input_tokens": 9052, + "estimated_eval_tokens": 18360 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This set is mostly a noisy cluster of Transformers issues that share broad themes (model loading, generation heads, tokenizer regressions, Qwen/Gemma variants) but usually point to different concrete code paths. I did not find a safe cross-issue duplicate merge; the closest pair is the resize_token_embeddings regression, but it still looks model-specific.", + "confidence": 0.64, + "canonical_issue_reason": "issue:45276 is the clearest representative of the only near-duplicate subtopic in the set: a resize_token_embeddings propagation bug affecting output-side embeddings. It is more general than its paired Gemma/T5Gemma variant, but the cluster overall is too heterogeneous to treat it as a true umbrella duplicate.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:45488 is the strongest standalone issue in the list: it is active, has discussion and inbound references, and describes a concrete user-facing tokenizer regression with broad impact on DeepSeek V3/R1-style models.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:45313", + "right": "issue:45357", + "accept": false, + "reason": "Both are Qwen3.5 loading/save regressions, but one is DeepSpeed ZeRO-3 loading and the other is save_pretrained visual-encoder key serialization; different failure modes and code paths." + }, + { + "left": "issue:45507", + "right": "issue:46468", + "accept": false, + "reason": "Generic model call failure vs a broken official Nemotron example; too little evidence they share the same root cause." + }, + { + "left": "issue:43856", + "right": "issue:43873", + "accept": false, + "reason": "Both mention Qwen3 MoE/offloading/memory, but one is training memory usage and the other is quantization offloading behavior; not the same bug." + }, + { + "left": "issue:44488", + "right": "issue:44991", + "accept": false, + "reason": "Both are tokenizer load failures on different models, but the affected models and reported symptoms differ enough that this looks like a shared regression class, not one duplicate bug." + }, + { + "left": "issue:43644", + "right": "issue:43645", + "accept": false, + "reason": "Both are Transformers 5.0 breakages, but one is non-persistent buffers and the other is custom model initialization in Jupyter; separate issues." + }, + { + "left": "issue:43708", + "right": "issue:44075", + "accept": false, + "reason": "Trainer checkpoint resume/max_steps logic is unrelated to optimizer SGD arguments being ignored." + }, + { + "left": "issue:39692", + "right": "issue:43994", + "accept": false, + "reason": "Both involve SigLIP2-ish model behavior, but one is documentation example errors and the other is bad outputs from AutoModel/pipeline; different artifacts and likely different fixes." + }, + { + "left": "issue:43856", + "right": "issue:44387", + "accept": false, + "reason": "Both concern memory, but one is Qwen3 MoE training memory usage and the other is int4 quantization reserved-memory/OOM regression; distinct underlying problems." + }, + { + "left": "issue:43257", + "right": "issue:43854", + "accept": false, + "reason": "Qwen3 MoE weight conversion under accelerate+deepspeed is not the same as a GLM-4.7 model-loading failure in unit tests." + }, + { + "left": "issue:43854", + "right": "issue:43931", + "accept": false, + "reason": "Both are model loading errors, but for different models and different mismatches (GLM-4.7 test load vs Qwen3-VL weight shapes)." + }, + { + "left": "issue:42491", + "right": "issue:45292", + "accept": false, + "reason": "LoRA qwen3_moe hf4.x/hf5.x compatibility is unrelated to a generic resize_token_embeddings/output_embeddings issue." + }, + { + "left": "issue:42491", + "right": "issue:43931", + "accept": false, + "reason": "The LoRA hf4.x vs hf5.x incompatibility is a training/compatibility regression, not a Qwen3-VL shape-mismatch loading bug." + }, + { + "left": "issue:44960", + "right": "issue:46050", + "accept": false, + "reason": "GLM5 is too vague to merge with a sinq quantization save/load corruption report; no clear shared code path." + }, + { + "left": "issue:45488", + "right": "issue:45812", + "accept": false, + "reason": "Both are tokenizer regressions, but one is LlamaTokenizer pre-tokenizer override and the other is Granite token IDs; different tokenizer logic and likely separate fixes." + }, + { + "left": "issue:45362", + "right": "issue:45406", + "accept": false, + "reason": "Qwen3.5 chat crash and Gemma4Processor missing _tokenizer are different runtime failures in different components." + }, + { + "left": "issue:45276", + "right": "issue:45335", + "accept": false, + "reason": "Both are resize_token_embeddings propagation bugs, but they hit different model families and different internal targets (Gemma4 output/embed_tokens_per_layer vs T5Gemma decoder.embed_tokens); not safe to merge as one duplicate." + }, + { + "left": "issue:44164", + "right": "issue:44291", + "accept": false, + "reason": "save/from_pretrained extra_state handling and init_empty_weights _is_hf_initialized TypeError are separate serialization/loading issues." + }, + { + "left": "issue:44910", + "right": "issue:45588", + "accept": false, + "reason": "Flash attention illegal memory access on Qwen3.5 position_ids is unrelated to a sink-less models AttributeError in integrations/flash_attention.py." + }, + { + "left": "issue:44402", + "right": "issue:44488", + "accept": false, + "reason": "Tokenizer vocab-size mismatch for a specific model is not the same underlying bug as a separate model load failure in cjvt/sleng-bert." + }, + { + "left": "issue:43404", + "right": "issue:43454", + "accept": false, + "reason": "Both involve untied lm_head weights and bad generation, but they are different model classes (Mistral3 multimodal vs AyaVision); likely model-specific omissions rather than one duplicate bug." + }, + { + "left": "issue:42947", + "right": "issue:45704", + "accept": false, + "reason": "Gradient checkpointing with PEFT LoRA is unrelated to T5 silently using apex.FusedRMSNorm and its memory leak." + }, + { + "left": "issue:41720", + "right": "issue:44155", + "accept": false, + "reason": "Qwen3 auto device mapping cudaErrorAssert and AudioFlamingo3 batched inference leakage are different subsystems and failure modes." + }, + { + "left": "issue:43873", + "right": "issue:44387", + "accept": false, + "reason": "Quantization/offloading not working as expected and int4 reserved-memory OOM are both memory-related, but not the same concrete bug." + }, + { + "left": "issue:39692", + "right": "issue:44960", + "accept": false, + "reason": "SigLIP2 docs errors and GLM5 are not clearly the same underlying issue; the titles do not point to one shared code path." + }, + { + "left": "issue:43425", + "right": "issue:43957", + "accept": false, + "reason": "Torch 2.10 incompatibility and meta-device model loading breakage are separate compatibility regressions." + }, + { + "left": "issue:45663", + "right": "issue:45759", + "accept": false, + "reason": "Gemma-4 FSDP2 shared_kv_states KeyError and Qwen 3.5/3.6 multimodal config unwrapping are different model/config bugs." + }, + { + "left": "issue:44291", + "right": "issue:44756", + "accept": false, + "reason": "init_empty_weights/_is_hf_initialized TypeError and mmap OOM on Strix Halo are unrelated loading/runtime issues." + }, + { + "left": "issue:36010", + "right": "issue:39404", + "accept": false, + "reason": "An ImportError for GenerationMixin and Whisper pipeline return_language regression are completely different problems." + }, + { + "left": "issue:45663", + "right": "issue:45859", + "accept": false, + "reason": "Gemma-4 FSDP2 attention state bug and Qwen3.5MoE CPU offload rotary embedding incompatibility are distinct code paths." + }, + { + "left": "issue:44387", + "right": "issue:44756", + "accept": false, + "reason": "Int4 quantization reserved-memory growth and mmap OOM on Strix Halo are not the same bug." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 21, + "review_unit_index": 8, + "review_unit_count": 56, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-473", + "nodes": [ + "issue:39404", + "issue:42796", + "issue:42947", + "issue:43335", + "issue:43381", + "issue:43425", + "issue:43611", + "issue:43645", + "issue:43646", + "issue:43708", + "issue:43716", + "issue:43828", + "issue:43854", + "issue:43906", + "issue:43957", + "issue:44155", + "issue:44164", + "issue:44292", + "issue:44402", + "issue:44451", + "issue:44479", + "issue:44530", + "issue:44589", + "issue:44683", + "issue:44756", + "issue:44805", + "issue:44863", + "issue:44877", + "issue:44912", + "issue:44928", + "issue:45084", + "issue:45216", + "issue:45276", + "issue:45292", + "issue:45335", + "issue:45588", + "issue:45663", + "issue:45684", + "issue:45685", + "issue:45698", + "issue:45701", + "issue:45812", + "issue:45823", + "issue:46315", + "issue:46326", + "issue:46468", + "issue:46491", + "issue:46665" + ], + "soft_pairs": [ + "issue:44292|issue:44912", + "issue:44155|issue:45823", + "issue:43335|issue:45663", + "issue:42947|issue:43381", + "issue:43957|issue:44756", + "issue:45276|issue:45292", + "issue:44863|issue:46468", + "issue:45685|issue:46468", + "issue:44479|issue:44877", + "issue:44164|issue:44756", + "issue:43381|issue:44928", + "issue:43854|issue:45216", + "issue:45588|issue:45663", + "issue:45684|issue:45698", + "issue:42796|issue:46326", + "issue:45292|issue:45335", + "issue:43646|issue:43708", + "issue:43828|issue:43957", + "issue:44402|issue:44451", + "issue:43645|issue:44164", + "issue:44589|issue:44756", + "issue:43425|issue:43828", + "issue:45812|issue:46315", + "issue:46315|issue:46491", + "issue:39404|issue:45701", + "issue:43611|issue:43645", + "issue:44683|issue:45823", + "issue:39404|issue:46665", + "issue:44530|issue:45084", + "issue:43425|issue:45823", + "issue:43425|issue:43716", + "issue:44479|issue:44805", + "issue:43906|issue:44530", + "issue:44589|issue:45084" + ], + "prepared_review_unit_hash": "b4fa6ca8ecd1046c008a9631a85d861f4eef3ae6e813cd37de3c0335c1cd7186", + "component_budget": { + "node_count": 473, + "item_count": 473, + "soft_pair_count": 1738, + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 34, + "serialized_chars": 37880, + "estimated_input_tokens": 9470, + "estimated_eval_tokens": 19196 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Most pairs are related only by broad theme, not the same bug. The only clear duplicate cluster is the `resize_token_embeddings` regression across related model variants (`45276`, `45292`, `45335`).", + "confidence": 0.93, + "canonical_issue_reason": "`issue:45292` is the cleanest core report of the shared regression: `resize_token_embeddings` fails to update output embeddings. The other accepted reports are model-specific manifestations of the same missing propagation behavior.", + "canonical_pr_reason": null, + "best_issue_reason": "`issue:45292` is the best global representative because it states the underlying API bug in the most generic, reusable way and avoids model-specific detail.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44292", + "right": "issue:44912", + "accept": false, + "reason": "Both are quantization/loading problems, but they affect different models and describe different failure modes." + }, + { + "left": "issue:44155", + "right": "issue:45823", + "accept": false, + "reason": "Different subsystems and symptoms: batched audio track leakage vs device-mapping mismatch." + }, + { + "left": "issue:43335", + "right": "issue:45663", + "accept": false, + "reason": "Both mention model configuration/training failures, but the actual bugs are unrelated." + }, + { + "left": "issue:42947", + "right": "issue:43381", + "accept": false, + "reason": "Different gradient-checkpointing issues: LoRA effectiveness vs eval-mode prohibition." + }, + { + "left": "issue:43957", + "right": "issue:44756", + "accept": false, + "reason": "Unrelated loading/OOM problems despite both involving model initialization." + }, + { + "left": "issue:45276", + "right": "issue:45292", + "accept": true, + "reason": "Same underlying bug: `resize_token_embeddings` does not propagate updated embeddings correctly; `45276` is a Gemma4-specific manifestation." + }, + { + "left": "issue:44863", + "right": "issue:46468", + "accept": false, + "reason": "Both are Nemotron-related, but one is checkpoint loading and the other is an example failure; not the same bug." + }, + { + "left": "issue:45685", + "right": "issue:46468", + "accept": false, + "reason": "MPS integer histogram support and Nemotron example breakage are unrelated." + }, + { + "left": "issue:44479", + "right": "issue:44877", + "accept": false, + "reason": "Different regressions in different model families and code paths." + }, + { + "left": "issue:44164", + "right": "issue:44756", + "accept": false, + "reason": "Save/load extra_state handling is unrelated to the Strix Halo mmap OOM issue." + }, + { + "left": "issue:43381", + "right": "issue:44928", + "accept": false, + "reason": "One is about checkpointing in eval mode; the other is an RLHF NaN/SDPA fallback regression." + }, + { + "left": "issue:43854", + "right": "issue:45216", + "accept": false, + "reason": "Different model-loading regressions with different failure points and triggers." + }, + { + "left": "issue:45588", + "right": "issue:45663", + "accept": false, + "reason": "Flash-attention sinkless-model crash and Gemma4 FSDP KeyError are distinct issues." + }, + { + "left": "issue:45684", + "right": "issue:45698", + "accept": false, + "reason": "Both concern save/load custom models, but one is permissions propagation and the other is loading the wrong module." + }, + { + "left": "issue:42796", + "right": "issue:46326", + "accept": false, + "reason": "Different tokenizer bugs with different APIs and symptoms." + }, + { + "left": "issue:45292", + "right": "issue:45335", + "accept": true, + "reason": "Same resize-token-embeddings regression: the updated embeddings are not reflected in model output/decoder embeddings." + }, + { + "left": "issue:43646", + "right": "issue:43708", + "accept": false, + "reason": "Both are Transformers 5.0 regressions, but one is custom model initialization and the other is checkpoint-resume step calculation." + }, + { + "left": "issue:43828", + "right": "issue:43957", + "accept": false, + "reason": "Different dtype/device-loading regressions with different repros and code paths." + }, + { + "left": "issue:44402", + "right": "issue:44451", + "accept": false, + "reason": "Both are model/tokenizer loading issues, but they concern different checkpoints and different root causes." + }, + { + "left": "issue:43645", + "right": "issue:44164", + "accept": false, + "reason": "Notebook custom-model initialization and extra_state save/load handling are not the same bug." + }, + { + "left": "issue:44589", + "right": "issue:44756", + "accept": false, + "reason": "Float8 storage lookup and mmap OOM are unrelated failures." + }, + { + "left": "issue:43425", + "right": "issue:43828", + "accept": false, + "reason": "Torch version incompatibility is broader and different from the autocast dtype mismatch bug." + }, + { + "left": "issue:45812", + "right": "issue:46315", + "accept": false, + "reason": "Tokenizer ID regression and slow `convert_tokens_to_ids` are distinct tokenizer issues." + }, + { + "left": "issue:46315", + "right": "issue:46491", + "accept": false, + "reason": "Performance regression in token lookup is unrelated to CodeLlama leading-space round-trip behavior." + }, + { + "left": "issue:39404", + "right": "issue:45701", + "accept": false, + "reason": "Whisper pipeline `return_language` regression and generic tokenization changes are not the same issue." + }, + { + "left": "issue:43611", + "right": "issue:43645", + "accept": false, + "reason": "Both involve custom models in Transformers 5.0, but one is base-model-prefix loading and the other is notebook initialization; too different to merge." + }, + { + "left": "issue:44683", + "right": "issue:45823", + "accept": false, + "reason": "Compiled flex_attention on torch>=2.9 and Gemma4 device mismatch are unrelated." + }, + { + "left": "issue:39404", + "right": "issue:46665", + "accept": false, + "reason": "Whisper pipeline language return and `from_pretrained` inside Python classes are unrelated behaviors." + }, + { + "left": "issue:44530", + "right": "issue:45084", + "accept": false, + "reason": "PagedAttentionCache group-type crash and template-node compilation error are different code paths." + }, + { + "left": "issue:43425", + "right": "issue:45823", + "accept": false, + "reason": "Torch compatibility issue vs Gemma4 device-mapping forward failure." + }, + { + "left": "issue:43425", + "right": "issue:43716", + "accept": false, + "reason": "Torch version incompatibility is not the same as the Mistral image dtype mismatch." + }, + { + "left": "issue:44479", + "right": "issue:44805", + "accept": false, + "reason": "Video-input regression and tensor mask shape mismatch are unrelated." + }, + { + "left": "issue:43906", + "right": "issue:44530", + "accept": false, + "reason": "Isolated reproduction of another issue and Qwen3.5 cache crash are different reports." + }, + { + "left": "issue:44589", + "right": "issue:45084", + "accept": false, + "reason": "Different failures: missing Float8 storage vs compilation of non-template nodes." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "Grounded and conservative overall. The only merged cluster that is clearly supported by the packet is the `resize_token_embeddings` regression across `45276`, `45292`, and `45335`; the other pairwise rejections stay appropriately narrow and do not overclaim duplicates." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 21, + "review_unit_index": 9, + "review_unit_count": 56, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-473", + "nodes": [ + "issue:36032", + "issue:39404", + "issue:42796", + "issue:42915", + "issue:43299", + "issue:43404", + "issue:43425", + "issue:43452", + "issue:43582", + "issue:43596", + "issue:43611", + "issue:43632", + "issue:43643", + "issue:43644", + "issue:43716", + "issue:43856", + "issue:43901", + "issue:43906", + "issue:43931", + "issue:43994", + "issue:44079", + "issue:44164", + "issue:44265", + "issue:44462", + "issue:44479", + "issue:44488", + "issue:44509", + "issue:44530", + "issue:44661", + "issue:44671", + "issue:44683", + "issue:44737", + "issue:44756", + "issue:44898", + "issue:44912", + "issue:44918", + "issue:44987", + "issue:44991", + "issue:45081", + "issue:45084", + "issue:45759", + "issue:45859", + "issue:46050", + "issue:46285", + "issue:46326", + "issue:46402", + "issue:46489" + ], + "soft_pairs": [ + "issue:43901|issue:44509", + "issue:44265|issue:44898", + "issue:39404|issue:44912", + "issue:43994|issue:44079", + "issue:43425|issue:44918", + "issue:44671|issue:44991", + "issue:36032|issue:43452", + "issue:46050|issue:46326", + "issue:43906|issue:46326", + "issue:44164|issue:45759", + "issue:43299|issue:43931", + "issue:44991|issue:45081", + "issue:43856|issue:44756", + "issue:43596|issue:44683", + "issue:44164|issue:44479", + "issue:44488|issue:45081", + "issue:44462|issue:46489", + "issue:44488|issue:44987", + "issue:42796|issue:43404", + "issue:43611|issue:43644", + "issue:39404|issue:44671", + "issue:44737|issue:45859", + "issue:46285|issue:46402", + "issue:43716|issue:45084", + "issue:36032|issue:42915", + "issue:44661|issue:44683", + "issue:43582|issue:43632", + "issue:43643|issue:44530" + ], + "prepared_review_unit_hash": "ab598da95210c25570707dac98d055422afb1764c223fd87cf927992e1e67e94", + "component_budget": { + "node_count": 473, + "item_count": 473, + "soft_pair_count": 1738, + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 28, + "serialized_chars": 35589, + "estimated_input_tokens": 8898, + "estimated_eval_tokens": 18052 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a loose grouping of mostly distinct Transformers v5 regression reports, centered on model/tokenizer loading and behavior changes. They read as related by theme, not as true duplicates, so the soft edges should mostly be rejected.", + "confidence": 0.82, + "canonical_issue_reason": "issue:43611 is the broadest representative of the cluster\u2019s main theme: v5 breaking model loading behavior. It is generic enough to stand in for the larger set of loading regressions without implying the specific tokenizer/model bugs are duplicates.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:43611 is the best single issue to anchor this cluster because it is a general, high-level regression report rather than one narrow model-specific symptom.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43901", + "right": "issue:44509", + "accept": false, + "reason": "Both are docs issues, but they cover different pipeline/documentation regressions: stale return_all_scores wording vs removed text2text/summarization/translation tasks." + }, + { + "left": "issue:44265", + "right": "issue:44898", + "accept": false, + "reason": "Both involve model/export behavior, but one is about torch.export and torch_compilable_check, while the other is a Perceiver image-classification resolution/interpolation bug." + }, + { + "left": "issue:39404", + "right": "issue:44912", + "accept": false, + "reason": "Whisper pipeline return_language and MXFP4 quantization fallback on git-oss are unrelated bugs in different subsystems." + }, + { + "left": "issue:43994", + "right": "issue:44079", + "accept": false, + "reason": "One is a SigLIP2 output quality/loading regression, the other is a ModelOutput key assignment bug; same broad area, different code-path problem." + }, + { + "left": "issue:43425", + "right": "issue:44918", + "accept": false, + "reason": "Torch 2.10 compatibility and Qwen3.5 embedding unpacking in TRL are unrelated failures." + }, + { + "left": "issue:44671", + "right": "issue:44991", + "accept": false, + "reason": "Both are tokenizer/model regressions in v5, but they affect different models and likely different loading paths." + }, + { + "left": "issue:36032", + "right": "issue:43452", + "accept": false, + "reason": "T5 tokenizer special-token conflict and gguf_file loading breakage are different tokenizer-loading issues." + }, + { + "left": "issue:46050", + "right": "issue:46326", + "accept": false, + "reason": "Quantized save/load corruption and mistral_common continue_final_message behavior are not the same bug." + }, + { + "left": "issue:43906", + "right": "issue:46326", + "accept": false, + "reason": "An isolated reproduction of an older issue is not the same underlying tokenizer behavior as mistral_common continue_final_message." + }, + { + "left": "issue:44164", + "right": "issue:45759", + "accept": false, + "reason": "extra_state save/load handling and text_config unwrapping in composite configs are distinct serialization/configuration bugs." + }, + { + "left": "issue:43299", + "right": "issue:43931", + "accept": true, + "reason": "Both point to the same Qwen3-VL MoE model-loading failure in v5, with the second looking like a concrete manifestation of the same shape-mismatch/load regression." + }, + { + "left": "issue:44991", + "right": "issue:45081", + "accept": false, + "reason": "Both are tokenizer regressions, but one is a generic tokenizer load failure for a specific model and the other is a mistral regex patch crash." + }, + { + "left": "issue:43856", + "right": "issue:44756", + "accept": false, + "reason": "Qwen3 MoE training memory usage and disabling mmap on Strix Halo are unrelated performance/oom reports." + }, + { + "left": "issue:43596", + "right": "issue:44683", + "accept": false, + "reason": "DeepSpeed ZeRO-3/BertModel indexing and compiled flex_attention on torch>=2.9 are different failure modes." + }, + { + "left": "issue:44164", + "right": "issue:44479", + "accept": false, + "reason": "save/from_pretrained extra_state handling is unrelated to the v5 video-input regression on Qwen VL models." + }, + { + "left": "issue:44488", + "right": "issue:45081", + "accept": false, + "reason": "Both involve tokenizer loading, but one is a missing tokenizer load for a model repo and the other is a specific mistral regex patch backend_tokenizer crash." + }, + { + "left": "issue:44462", + "right": "issue:46489", + "accept": false, + "reason": "AutoTokenizer ignoring tokenizer.json and DeepSeek-Coder v1 incorrect output after a prior fix are not clearly the same underlying bug." + }, + { + "left": "issue:44488", + "right": "issue:44987", + "accept": false, + "reason": "Both mention model/tokenizer loading failures, but they concern different repositories and different root causes." + }, + { + "left": "issue:42796", + "right": "issue:43404", + "accept": false, + "reason": "Ministral tokenizer BPE-marker handling and Mistral3 lm_head weight tying are separate model/tokenizer issues." + }, + { + "left": "issue:43611", + "right": "issue:43644", + "accept": false, + "reason": "Both are v5 loading regressions, but one is about base_model_prefix handling and the other about non-persistent buffers being initialized with junk." + }, + { + "left": "issue:39404", + "right": "issue:44671", + "accept": false, + "reason": "Whisper pipeline language return and CamemBERT masked-LM prediction regressions are unrelated model/pipeline bugs." + }, + { + "left": "issue:44737", + "right": "issue:45859", + "accept": false, + "reason": "Both are device/offload related, but XLNet CPU arange placement and Qwen3.5 MoE rotary embedding CPU offload incompatibility are different issues." + }, + { + "left": "issue:46285", + "right": "issue:46402", + "accept": false, + "reason": "CLIP structure inconsistency and Sam3Model text-encoder weights not loading are distinct architectural/load regressions." + }, + { + "left": "issue:43716", + "right": "issue:45084", + "accept": false, + "reason": "Mistral-3 image preprocessor dtype mismatch and non-template-node compilation failure are unrelated." + }, + { + "left": "issue:36032", + "right": "issue:42915", + "accept": false, + "reason": "T5 tokenizer special-token conflict and Qwen3 MoE FP8 config failure are different model-loading problems." + }, + { + "left": "issue:44661", + "right": "issue:44683", + "accept": false, + "reason": "add-new-model-like / TOKENIZER_MAPPING_NAMES and compiled flex_attention are unrelated." + }, + { + "left": "issue:43582", + "right": "issue:43632", + "accept": false, + "reason": "Apple Silicon caching_allocator_warmup TypeError and the _is_hf_initialized flag regression are different bugs." + }, + { + "left": "issue:43643", + "right": "issue:44530", + "accept": false, + "reason": "AutoConfig trust_remote_code field loss and PagedAttentionCache linear_attention crashes are unrelated." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary stays grounded in the packet and is appropriately conservative about the cluster being a loose theme rather than a duplicate set. The soft-edge verdicts generally avoid overclaiming, and the one accepted pair is plausibly the same Qwen3-VL family regression." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 21, + "review_unit_index": 10, + "review_unit_count": 56, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-473", + "nodes": [ + "issue:28530", + "issue:34567", + "issue:36032", + "issue:40444", + "issue:40990", + "issue:42175", + "issue:42915", + "issue:43262", + "issue:43317", + "issue:43322", + "issue:43377", + "issue:43404", + "issue:43421", + "issue:43452", + "issue:43482", + "issue:43540", + "issue:43541", + "issue:43644", + "issue:43645", + "issue:43646", + "issue:43746", + "issue:43819", + "issue:43883", + "issue:43994", + "issue:44206", + "issue:44220", + "issue:44265", + "issue:44291", + "issue:44403", + "issue:44451", + "issue:44464", + "issue:44466", + "issue:44488", + "issue:44530", + "issue:44534", + "issue:44610", + "issue:44811", + "issue:44849", + "issue:44987", + "issue:44991", + "issue:45081", + "issue:45084", + "issue:45325", + "issue:45507", + "issue:45584", + "issue:45987", + "issue:46178", + "issue:46326" + ], + "soft_pairs": [ + "issue:40990|issue:43421", + "issue:44451|issue:44987", + "issue:42915|issue:44530", + "issue:43541|issue:45084", + "issue:34567|issue:40444", + "issue:43644|issue:43646", + "issue:42175|issue:46178", + "issue:44811|issue:45584", + "issue:44206|issue:44220", + "issue:43883|issue:44488", + "issue:40444|issue:46326", + "issue:36032|issue:43645", + "issue:43452|issue:43482", + "issue:43317|issue:43482", + "issue:45084|issue:45507", + "issue:44987|issue:44991", + "issue:43262|issue:43377", + "issue:43746|issue:46326", + "issue:43883|issue:44991", + "issue:44464|issue:45987", + "issue:43646|issue:44534", + "issue:44403|issue:45081", + "issue:42915|issue:45084", + "issue:43645|issue:44534", + "issue:43404|issue:44466", + "issue:43377|issue:43819", + "issue:43994|issue:44610", + "issue:44849|issue:45325", + "issue:28530|issue:34567", + "issue:44451|issue:45081", + "issue:44987|issue:45081", + "issue:42915|issue:44291", + "issue:43322|issue:43540", + "issue:44265|issue:44610" + ], + "prepared_review_unit_hash": "00732948046e6ba93428485b2f85e9b96edcf5091da0e81a5e70df484fcfe318", + "component_budget": { + "node_count": 473, + "item_count": 473, + "soft_pair_count": 1738, + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 34, + "serialized_chars": 38077, + "estimated_input_tokens": 9520, + "estimated_eval_tokens": 19296 + }, + "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 only clear duplicate pair is the v5 non-persistent-buffer regression; the rest are superficially similar but point to different models, code paths, or failure modes.", + "confidence": 0.72, + "canonical_issue_reason": "issue:43644 is the cleanest canonical because it is the earlier, broader report of the same v5 non-persistent-buffer corruption bug, and issue:44534 reads like a near-verbatim follow-up.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:43644 is the best representative of the cluster because it captures the only strong duplicate relationship and states the underlying regression clearly.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:40990", + "right": "issue:43421", + "accept": false, + "reason": "Both concern training/Tokenizer behavior, but one is early stopping metric selection and the other is runtime special-token post-processing; different bugs." + }, + { + "left": "issue:44451", + "right": "issue:44987", + "accept": false, + "reason": "Both are model-loading failures, but they affect different models and likely different config/tokenizer paths; too broad to merge." + }, + { + "left": "issue:42915", + "right": "issue:44530", + "accept": false, + "reason": "Both are multimodal-model failures, but one is FP8 config on Qwen3Moe and the other is PagedAttention cache group-type handling on Qwen3.5; different code paths." + }, + { + "left": "issue:43541", + "right": "issue:45084", + "accept": false, + "reason": "A torch dynamo grouped_mm tracing error and a template-node compilation error are unrelated failures." + }, + { + "left": "issue:34567", + "right": "issue:40444", + "accept": false, + "reason": "Num-input-token accounting and Qwen2.5-VL iterable-dataset multi-image loading are unrelated issues." + }, + { + "left": "issue:43644", + "right": "issue:43646", + "accept": false, + "reason": "Both are v5 regressions, but one is buffer serialization/junk state and the other is custom model initialization; not the same bug." + }, + { + "left": "issue:42175", + "right": "issue:46178", + "accept": false, + "reason": "One is missing tensorflow in an extra install, the other is an unknown backend mapping error for tensorflow_text; different packaging/runtime issues." + }, + { + "left": "issue:44811", + "right": "issue:45584", + "accept": false, + "reason": "Both involve Whisper token handling, but batch_decode skip_special_tokens and empty-transcription handling after align_special_tokens are distinct defects." + }, + { + "left": "issue:44206", + "right": "issue:44220", + "accept": false, + "reason": "One is a LASRFeatureExtractor API regression, the other is a low-level fbank feature extraction issue; not the same code path." + }, + { + "left": "issue:43883", + "right": "issue:44488", + "accept": false, + "reason": "Both mention loading models, but one is an all_tied_weights_keys attribute error and the other is a specific model-loading regression; different failures." + }, + { + "left": "issue:40444", + "right": "issue:46326", + "accept": false, + "reason": "Multiple-image iterable dataset handling and mistral_common continue_final_message behavior are unrelated." + }, + { + "left": "issue:36032", + "right": "issue:43645", + "accept": false, + "reason": "Tokenizer class-method conflict and custom model initialization in notebooks are different problems." + }, + { + "left": "issue:43452", + "right": "issue:43482", + "accept": false, + "reason": "GGUF-related loader breakage and a specific Qwen2.5-GGUF v5 loading failure are not the same underlying issue." + }, + { + "left": "issue:43317", + "right": "issue:43482", + "accept": false, + "reason": "Dequantized model offload with device_map=auto and Qwen2.5-GGUF loading failure are separate loader paths." + }, + { + "left": "issue:45084", + "right": "issue:45507", + "accept": false, + "reason": "Template compilation failure and GraniteMoEHybrid invalid method call are unrelated." + }, + { + "left": "issue:44987", + "right": "issue:44991", + "accept": false, + "reason": "Both are model/tokenizer loading regressions on v5, but for different models and failure sites; not enough to unify." + }, + { + "left": "issue:43262", + "right": "issue:43377", + "accept": false, + "reason": "Audio processor sampling-rate defaulting and MIMI encoder batching/padding-mask behavior are unrelated." + }, + { + "left": "issue:43746", + "right": "issue:46326", + "accept": false, + "reason": "PEFT local-checkpoint loading for GraniteSpeech and mistral_common continue_final_message behavior are different bugs." + }, + { + "left": "issue:43883", + "right": "issue:44991", + "accept": false, + "reason": "An attribute error during weight tying and a tokenizer loading regression for a different model are not the same issue." + }, + { + "left": "issue:44464", + "right": "issue:45987", + "accept": false, + "reason": "Chunked generation inconsistency under compiled forward and StaticCache seq-length type mismatch are different runtime behaviors." + }, + { + "left": "issue:43646", + "right": "issue:44534", + "accept": false, + "reason": "Custom model initialization and non-persistent-buffer junk are different regressions, even if both are v5-related." + }, + { + "left": "issue:44403", + "right": "issue:45081", + "accept": false, + "reason": "Noise on loading a transformer and a Mistral regex patch crash are unrelated." + }, + { + "left": "issue:42915", + "right": "issue:45084", + "accept": false, + "reason": "FP8 config failure on Qwen3Moe and template-node compilation failure are different code paths." + }, + { + "left": "issue:43645", + "right": "issue:44534", + "accept": false, + "reason": "Custom notebook model initialization and non-persistent-buffer corruption are distinct v5 problems." + }, + { + "left": "issue:43404", + "right": "issue:44466", + "accept": false, + "reason": "lm_head tying in Mistral3ForConditionalGeneration and device-dependent lm_head serialization are related only at a high level; the concrete bugs differ." + }, + { + "left": "issue:43377", + "right": "issue:43819", + "accept": false, + "reason": "Batched-vs-single MIMI encoder outputs and DAC.from_latents missing STE are separate model-specific correctness bugs." + }, + { + "left": "issue:43994", + "right": "issue:44610", + "accept": false, + "reason": "Siglip2 nonsensical outputs and OmDet-Turbo processor size mismatch are different vision-model issues." + }, + { + "left": "issue:44849", + "right": "issue:45325", + "accept": false, + "reason": "Qwen3.5 hidden_states behavior and Qwen2.5-VL rope index scaling are separate model bugs." + }, + { + "left": "issue:28530", + "right": "issue:34567", + "accept": false, + "reason": "Early stopping metric lookup and token-count tracking are unrelated trainer issues." + }, + { + "left": "issue:44451", + "right": "issue:45081", + "accept": false, + "reason": "A model load failure for one checkpoint and a Mistral tokenizer regex crash are not the same underlying bug." + }, + { + "left": "issue:44987", + "right": "issue:45081", + "accept": false, + "reason": "Both involve loading regressions, but one is a model-load failure and the other is a tokenizer regex patch failure; different code paths." + }, + { + "left": "issue:42915", + "right": "issue:44291", + "accept": false, + "reason": "FineGrainedFP8Config on Qwen3Moe and init_empty_weights/_is_hf_initialized incompatibility are unrelated." + }, + { + "left": "issue:43322", + "right": "issue:43540", + "accept": false, + "reason": "Llava Next loading segfault and Qwen3OmniMoe video-processing ValueError are distinct multimodal issues." + }, + { + "left": "issue:44265", + "right": "issue:44610", + "accept": false, + "reason": "torch.export exportability with torch_compilable_check and an image processor input-size mismatch are unrelated." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 21, + "review_unit_index": 11, + "review_unit_count": 56, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-473", + "nodes": [ + "issue:33666", + "issue:35707", + "issue:36032", + "issue:40444", + "issue:40990", + "issue:41553", + "issue:41720", + "issue:42796", + "issue:42915", + "issue:43089", + "issue:43296", + "issue:43381", + "issue:43421", + "issue:43425", + "issue:43482", + "issue:43525", + "issue:43531", + "issue:43582", + "issue:43611", + "issue:43632", + "issue:43742", + "issue:43746", + "issue:43792", + "issue:44062", + "issue:44079", + "issue:44220", + "issue:44291", + "issue:44403", + "issue:44479", + "issue:44534", + "issue:44568", + "issue:44589", + "issue:44683", + "issue:44756", + "issue:44811", + "issue:44877", + "issue:44987", + "issue:44991", + "issue:45216", + "issue:45325", + "issue:45357", + "issue:45538", + "issue:45663", + "issue:45812", + "issue:45923", + "issue:45987", + "issue:46315", + "issue:46491" + ], + "soft_pairs": [ + "issue:43089|issue:45987", + "issue:36032|issue:43742", + "issue:44062|issue:44987", + "issue:42915|issue:44589", + "issue:44479|issue:45325", + "issue:45216|issue:45357", + "issue:44991|issue:46315", + "issue:36032|issue:41720", + "issue:43531|issue:44877", + "issue:44220|issue:44683", + "issue:43425|issue:44403", + "issue:43296|issue:43531", + "issue:40444|issue:45325", + "issue:43611|issue:44534", + "issue:43632|issue:44756", + "issue:45812|issue:46491", + "issue:42796|issue:43746", + "issue:44403|issue:44991", + "issue:43381|issue:45923", + "issue:33666|issue:40444", + "issue:43792|issue:44589", + "issue:40990|issue:44811", + "issue:44079|issue:45663", + "issue:44291|issue:44589", + "issue:43632|issue:44534", + "issue:43421|issue:44568", + "issue:43482|issue:43525", + "issue:35707|issue:40444", + "issue:43482|issue:45538", + "issue:41553|issue:43582" + ], + "prepared_review_unit_hash": "cc636a9ea0aae7f34a809876fb4091556f9005e5a5a3697f5b2a01a8b34c83b2", + "component_budget": { + "node_count": 473, + "item_count": 473, + "soft_pair_count": 1738, + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 30, + "serialized_chars": 36533, + "estimated_input_tokens": 9134, + "estimated_eval_tokens": 18524 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is heterogeneous: most items are unrelated model-specific bugs, tokenizer regressions, or environment/version incompatibilities. I don\u2019t see any soft pair that is clearly the same underlying bug/change, so I would not merge any of the candidate edges.", + "confidence": 0.91, + "canonical_issue_reason": "issue:45812 is the strongest anchor if one must be chosen: it is a broad, concrete v4\u2192v5 regression affecting all Granite models and has multiple inbound references. The rest of the cluster is mostly narrow, unrelated model/version-specific issues.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:45812 is the best representative issue overall because it has the broadest impact and clearest regression framing among an otherwise mixed set of mostly unrelated bugs.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43089", + "right": "issue:45987", + "accept": false, + "reason": "Both mention GPU-side generation behavior, but one is a generation performance overhead report and the other is a cache API contract bug; different code paths and symptoms." + }, + { + "left": "issue:36032", + "right": "issue:43742", + "accept": false, + "reason": "A tokenizer class-name conflict is unrelated to a KeyError while loading a specific model; no shared underlying bug." + }, + { + "left": "issue:44062", + "right": "issue:44987", + "accept": false, + "reason": "One is an AddedToken keyword collision, the other is a model load failure on transformers>=5.1.0; too vague to treat as the same bug." + }, + { + "left": "issue:42915", + "right": "issue:44589", + "accept": false, + "reason": "Both are FP8-related, but one is a Qwen3MoE FineGrainedFP8Config failure and the other is a missing Float8 storage type; different failure mechanisms." + }, + { + "left": "issue:44479", + "right": "issue:45325", + "accept": false, + "reason": "Both involve Qwen VL models, but one is a video-input regression and the other is still-image temporal position-id scaling; different bugs." + }, + { + "left": "issue:45216", + "right": "issue:45357", + "accept": false, + "reason": "Both are Qwen3.5 save_pretrained regressions, but one reports incorrect saved checkpoints broadly while the other is specifically about visual encoder keys; not clearly the same defect." + }, + { + "left": "issue:44991", + "right": "issue:46315", + "accept": false, + "reason": "Tokenizer loading failure vs performance regression in convert_tokens_to_ids; unrelated symptoms and code paths." + }, + { + "left": "issue:36032", + "right": "issue:41720", + "accept": false, + "reason": "A T5 tokenizer method-conflict bug is unrelated to a Qwen3 auto-device-map CUDA assert." + }, + { + "left": "issue:43531", + "right": "issue:44877", + "accept": false, + "reason": "Sliding-window behavior in Qwen3-MoE and strict config loading for granite_speech are different issues." + }, + { + "left": "issue:44220", + "right": "issue:44683", + "accept": false, + "reason": "Audio fbank extraction and compiled flex_attention are unrelated subsystems." + }, + { + "left": "issue:43425", + "right": "issue:44403", + "accept": false, + "reason": "Torch version incompatibility and noisy load logging are not the same underlying bug." + }, + { + "left": "issue:43296", + "right": "issue:43531", + "accept": false, + "reason": "PaddleOCR-VL model loading in vLLM and Qwen3-MoE sliding_window handling are unrelated." + }, + { + "left": "issue:40444", + "right": "issue:45325", + "accept": false, + "reason": "Both are Qwen2.5-VL-related, but one is finetuning with multiple images per prompt and the other is rope/position-id scaling for still images; different code paths." + }, + { + "left": "issue:43611", + "right": "issue:44534", + "accept": false, + "reason": "Base-model-prefix loading and non-persistent buffer initialization are distinct v5 regressions." + }, + { + "left": "issue:43632", + "right": "issue:44756", + "accept": false, + "reason": "An initialization flag regression and an mmap/OOM workaround are unrelated." + }, + { + "left": "issue:45812", + "right": "issue:46491", + "accept": false, + "reason": "Both are tokenizer regressions, but one is wrong token IDs for Granite and the other is a leading-space round-trip issue in CodeLlama; not the same bug." + }, + { + "left": "issue:42796", + "right": "issue:43746", + "accept": false, + "reason": "Tokenizer BPE-marker handling and GraniteSpeech PEFT checkpoint loading are unrelated." + }, + { + "left": "issue:44403", + "right": "issue:44991", + "accept": false, + "reason": "General load-noise logging is unrelated to a tokenizer loading failure for EMBEDDIA/est-roberta." + }, + { + "left": "issue:43381", + "right": "issue:45923", + "accept": false, + "reason": "Eval-mode gradient-checkpointing restriction and a missing supports_gradient_checkpointing flag are related in theme only, not the same concrete defect." + }, + { + "left": "issue:33666", + "right": "issue:40444", + "accept": false, + "reason": "Both are Qwen VL training issues, but one is multi-GPU training and the other is iterable-dataset/multi-image prompt failure; different bugs." + }, + { + "left": "issue:43792", + "right": "issue:44589", + "accept": false, + "reason": "Whisper runtime failure and Float8 storage lookup error are unrelated." + }, + { + "left": "issue:40990", + "right": "issue:44811", + "accept": false, + "reason": "High perplexity on a language-model benchmark and Whisper batch_decode behavior are unrelated." + }, + { + "left": "issue:44079", + "right": "issue:45663", + "accept": false, + "reason": "ModelOutput key assignment and Gemma4 FSDP2 shared_kv_states rebuilding are different code-path problems." + }, + { + "left": "issue:44291", + "right": "issue:44589", + "accept": false, + "reason": "init_empty_weights argument handling and Float8 storage lookup are unrelated regressions." + }, + { + "left": "issue:43632", + "right": "issue:44534", + "accept": false, + "reason": "_is_hf_initialized handling and non-persistent buffer junk values are different initialization bugs." + }, + { + "left": "issue:43421", + "right": "issue:44568", + "accept": false, + "reason": "Runtime post-processor updates for special tokens and add_special_tokens BOS/EOS behavior in mdeberta-v3 are not the same issue." + }, + { + "left": "issue:43482", + "right": "issue:43525", + "accept": false, + "reason": "Qwen2.5-GGUF loading failure and missing pad_token_id on Llama4Config are unrelated." + }, + { + "left": "issue:35707", + "right": "issue:40444", + "accept": false, + "reason": "Progressive generation with inputs_embeds/past_key_values is unrelated to Qwen2.5-VL multi-image finetuning failure." + }, + { + "left": "issue:43482", + "right": "issue:45538", + "accept": false, + "reason": "GGUF loading in transformers v5 and CLIPTokenizer model_max_length are unrelated tokenization/loading issues." + }, + { + "left": "issue:41553", + "right": "issue:43582", + "accept": false, + "reason": "A bad AutoTokenizer error message for Voxtral and an AppleSilicon allocator TypeError 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": 21, + "review_unit_index": 12, + "review_unit_count": 56, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-473", + "nodes": [ + "issue:40444", + "issue:41720", + "issue:42175", + "issue:42796", + "issue:42915", + "issue:43296", + "issue:43317", + "issue:43377", + "issue:43454", + "issue:43482", + "issue:43531", + "issue:43606", + "issue:43611", + "issue:43632", + "issue:43644", + "issue:43646", + "issue:43701", + "issue:43749", + "issue:43792", + "issue:43818", + "issue:43856", + "issue:43873", + "issue:43957", + "issue:43994", + "issue:44079", + "issue:44265", + "issue:44291", + "issue:44462", + "issue:44466", + "issue:44479", + "issue:44534", + "issue:44829", + "issue:44877", + "issue:44898", + "issue:45072", + "issue:45081", + "issue:45325", + "issue:45357", + "issue:45381", + "issue:45479", + "issue:45491", + "issue:45542", + "issue:45701", + "issue:45800", + "issue:45812", + "issue:45823", + "issue:45834", + "issue:46402" + ], + "soft_pairs": [ + "issue:43317|issue:43856", + "issue:42175|issue:45542", + "issue:43531|issue:44291", + "issue:44829|issue:45479", + "issue:44079|issue:44534", + "issue:44462|issue:45812", + "issue:41720|issue:45081", + "issue:43818|issue:46402", + "issue:43377|issue:45491", + "issue:43818|issue:45357", + "issue:43611|issue:43646", + "issue:43644|issue:44079", + "issue:44265|issue:45072", + "issue:43749|issue:44079", + "issue:42175|issue:43296", + "issue:40444|issue:45381", + "issue:45325|issue:45381", + "issue:43994|issue:44898", + "issue:45325|issue:45834", + "issue:43454|issue:44466", + "issue:43792|issue:43957", + "issue:42915|issue:43482", + "issue:43632|issue:45072", + "issue:42796|issue:45381", + "issue:43701|issue:45325", + "issue:44291|issue:44877", + "issue:43873|issue:45701", + "issue:43296|issue:45800", + "issue:43606|issue:44265", + "issue:44479|issue:45823" + ], + "prepared_review_unit_hash": "345a2181e9233e070ac21b38a7777d93e70991ac9ac774b01ad40d272ab80288", + "component_budget": { + "node_count": 473, + "item_count": 473, + "soft_pair_count": 1738, + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 30, + "serialized_chars": 36853, + "estimated_input_tokens": 9214, + "estimated_eval_tokens": 18684 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is still mostly a set of distinct Transformers v5 regression reports rather than a tight duplicate cluster. The `issue:43611` \u2194 `issue:43646` link should be treated as related but not duplicated: both concern v5 load/init behavior, but the packet does not show the same concrete code path or fix.", + "confidence": 0.71, + "canonical_issue_reason": "Issue 43646 is the broadest and most representative anchor for the cluster because it captures the general Transformers v5 custom-model initialization regression theme, but it should not be used to imply that nearby load/init reports are definite duplicates without stronger evidence.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 43646 is the best single issue anchor because it is broad, actionable, and closest to the shared v5 loading/initialization regression pattern.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43317", + "right": "issue:43856", + "accept": false, + "reason": "Different problems: dequantized model offload/device mapping vs MoE training memory usage." + }, + { + "left": "issue:42175", + "right": "issue:45542", + "accept": false, + "reason": "Both mention TensorFlow packaging, but one is missing TF in the `.[torch]` backend and the other is a tensorboard-only undefined-backend error." + }, + { + "left": "issue:43531", + "right": "issue:44291", + "accept": false, + "reason": "Unrelated bugs: Qwen3-MoE sliding-window behavior vs `init_empty_weights`/`_is_hf_initialized` argument handling." + }, + { + "left": "issue:44829", + "right": "issue:45479", + "accept": false, + "reason": "Different training regressions with different triggers and failure modes." + }, + { + "left": "issue:44079", + "right": "issue:44534", + "accept": false, + "reason": "`ModelOutput` key assignment bug vs non-persistent buffer initialization bug." + }, + { + "left": "issue:44462", + "right": "issue:45812", + "accept": false, + "reason": "Both are tokenizer-related, but one ignores repo `tokenizer.json` while the other miscomputes Granite token IDs." + }, + { + "left": "issue:41720", + "right": "issue:45081", + "accept": false, + "reason": "Different subsystems: Qwen3 device mapping crash vs Mistral tokenizer regex patch crash." + }, + { + "left": "issue:43818", + "right": "issue:46402", + "accept": false, + "reason": "Unrelated models and failures: Video-LLaVA tower/weights issue vs Sam3Model text-encoder loading issue." + }, + { + "left": "issue:43377", + "right": "issue:45491", + "accept": false, + "reason": "Both involve batching, but the underlying bugs differ: missing padding-mask support vs sliding-window/all-padding NaNs." + }, + { + "left": "issue:43818", + "right": "issue:45357", + "accept": false, + "reason": "Different model families and issues: Video-LLaVA architecture bug vs Qwen3.5 save_pretrained key regression." + }, + { + "left": "issue:43611", + "right": "issue:43646", + "accept": false, + "reason": "Related v5 initialization/loading regressions, but the packet only shows broad thematic overlap (`base_model_prefix` loading vs custom model init); not enough evidence for the same underlying bug or code path." + }, + { + "left": "issue:43644", + "right": "issue:44079", + "accept": false, + "reason": "Different core bugs: non-persistent buffer corruption vs ModelOutput key assignment." + }, + { + "left": "issue:44265", + "right": "issue:45072", + "accept": false, + "reason": "Different code paths: `torch.export`/`torch_compilable_check` vs dtype mismatch in inference." + }, + { + "left": "issue:43749", + "right": "issue:44079", + "accept": false, + "reason": "FSDP CPU-efficient loading breakage is unrelated to ModelOutput key assignment." + }, + { + "left": "issue:42175", + "right": "issue:43296", + "accept": false, + "reason": "Different packaging/model-loading issues with no clear shared code-path." + }, + { + "left": "issue:40444", + "right": "issue:45381", + "accept": false, + "reason": "Both are Qwen2.5-VL vision-position related, but one is finetuning with multiple images and the other is a video-input regression; not clearly the same bug." + }, + { + "left": "issue:45325", + "right": "issue:45381", + "accept": false, + "reason": "Closely related area, but one reports still-image temporal scaling and the other video `vision_position_ids`; not enough to treat as the same defect." + }, + { + "left": "issue:43994", + "right": "issue:44898", + "accept": false, + "reason": "Different vision-model failures on different architectures and inputs." + }, + { + "left": "issue:45325", + "right": "issue:45834", + "accept": false, + "reason": "Both concern Qwen2.5-VL position logic, but the reported failures are different: temporal scaling vs OCR long-input index error." + }, + { + "left": "issue:43454", + "right": "issue:44466", + "accept": false, + "reason": "Different symptoms around weight tying/serialization vs inconsistent `lm_head.weight` serialization." + }, + { + "left": "issue:43792", + "right": "issue:43957", + "accept": false, + "reason": "Whisper runtime failure vs generic `torch.device('meta')` model-loading regression." + }, + { + "left": "issue:42915", + "right": "issue:43482", + "accept": false, + "reason": "Unrelated: Qwen3-MoE FP8 config failure vs Qwen2.5 GGUF loading under v5." + }, + { + "left": "issue:43632", + "right": "issue:45072", + "accept": false, + "reason": "Different regressions: `_is_hf_initialized` flag handling vs bfloat16 dtype mismatches." + }, + { + "left": "issue:42796", + "right": "issue:45381", + "accept": false, + "reason": "Tokenizer BPE-marker handling is unrelated to Qwen2.5-VL video position IDs." + }, + { + "left": "issue:43701", + "right": "issue:45325", + "accept": false, + "reason": "Checkpoint key mismatch is unrelated to Qwen2.5-VL position-id scaling." + }, + { + "left": "issue:44291", + "right": "issue:44877", + "accept": false, + "reason": "`init_empty_weights` argument handling and strict config loading are different issues." + }, + { + "left": "issue:43873", + "right": "issue:45701", + "accept": false, + "reason": "Quantized offloading behavior and tokenization changes are unrelated." + }, + { + "left": "issue:43296", + "right": "issue:45800", + "accept": false, + "reason": "Different compatibility problems: PaddleOCR-VL/vLLM loading vs torch 2.4.1 and Transformers 5.8.0." + }, + { + "left": "issue:43606", + "right": "issue:44265", + "accept": false, + "reason": "CPU offload device mismatch is unrelated to `torch.export` failures." + }, + { + "left": "issue:44479", + "right": "issue:45823", + "accept": false, + "reason": "Different model families and failures: Qwen video-input regression vs Gemma4 PLE device-mapping mismatch." + } + ] + }, + "evaluator_result": { + "accept": false, + "feedback": "Most soft-edge rejections are appropriately conservative, but the accepted pair `issue:43611` \u2194 `issue:43646` is overstated. From the packet we only know one is about `base_model_prefix` loading and the other about custom model initialization; both are v5 load/init regressions, but there is not enough evidence here to call them the same underlying bug or a clear duplicate-like pair. The cluster summary should keep that linkage tentative unless the issue bodies show the same code path or fix." + }, + "evaluator_used": true, + "retried": true, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 21, + "review_unit_index": 13, + "review_unit_count": 56, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-473", + "nodes": [ + "issue:34634", + "issue:35707", + "issue:39692", + "issue:40444", + "issue:40990", + "issue:41720", + "issue:42617", + "issue:42796", + "issue:42915", + "issue:42947", + "issue:43257", + "issue:43262", + "issue:43317", + "issue:43322", + "issue:43377", + "issue:43454", + "issue:43526", + "issue:43550", + "issue:43572", + "issue:43606", + "issue:43720", + "issue:43746", + "issue:43749", + "issue:43819", + "issue:43873", + "issue:43881", + "issue:43883", + "issue:44186", + "issue:44265", + "issue:44387", + "issue:44488", + "issue:44534", + "issue:44568", + "issue:44610", + "issue:44898", + "issue:44977", + "issue:44991", + "issue:45072", + "issue:45584", + "issue:45704", + "issue:45759", + "issue:45823", + "issue:45920", + "issue:45923", + "issue:46050", + "issue:46209", + "issue:46305", + "issue:46315" + ], + "soft_pairs": [ + "issue:43262|issue:43746", + "issue:40990|issue:45584", + "issue:35707|issue:41720", + "issue:43881|issue:44488", + "issue:34634|issue:35707", + "issue:43454|issue:43550", + "issue:43606|issue:43819", + "issue:39692|issue:40444", + "issue:44186|issue:44265", + "issue:41720|issue:45823", + "issue:46305|issue:46315", + "issue:43377|issue:44610", + "issue:42796|issue:46209", + "issue:44568|issue:45920", + "issue:43257|issue:45923", + "issue:43454|issue:45072", + "issue:43317|issue:43873", + "issue:43377|issue:44898", + "issue:42796|issue:44977", + "issue:43454|issue:43720", + "issue:40990|issue:42947", + "issue:43322|issue:46050", + "issue:42617|issue:42915", + "issue:43749|issue:44534", + "issue:43572|issue:45759", + "issue:41720|issue:43322", + "issue:43526|issue:43883", + "issue:43881|issue:44991", + "issue:44387|issue:45704", + "issue:41720|issue:44977" + ], + "prepared_review_unit_hash": "c7f4fa6103f192407af89a1fe222a6bceb2bd9615f3af22b37e95ade36f02e80", + "component_budget": { + "node_count": 473, + "item_count": 473, + "soft_pair_count": 1738, + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 30, + "serialized_chars": 36797, + "estimated_input_tokens": 9200, + "estimated_eval_tokens": 18656 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is highly heterogeneous and does not contain any clear duplicate pairs. Several items share broad themes like tokenizer regressions, loading/offload problems, or device-map issues, but they point to different models and different failure modes.", + "confidence": 0.88, + "canonical_issue_reason": "No true duplicate cluster stands out; if a representative issue must be chosen, #45920 is the broadest open regression affecting multiple models and has the strongest evidence of being a general issue rather than a niche case.", + "canonical_pr_reason": null, + "best_issue_reason": "#45920 is the best single issue to represent this set because it is open, cross-model, and more broadly scoped than the other candidates, which are mostly model-specific or narrowly scoped bugs.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43262", + "right": "issue:43746", + "accept": false, + "reason": "Different subsystems: audio chat-template sampling rate vs PEFT checkpoint loading." + }, + { + "left": "issue:40990", + "right": "issue:45584", + "accept": false, + "reason": "Unrelated failures: perplexity evaluation on GPT-OSS vs Whisper empty-transcription generation." + }, + { + "left": "issue:35707", + "right": "issue:41720", + "accept": false, + "reason": "Different code paths: inputs_embeds/past_key_values generation vs auto device mapping on Qwen3." + }, + { + "left": "issue:43881", + "right": "issue:44488", + "accept": false, + "reason": "Both are loading failures, but for different models and likely different root causes." + }, + { + "left": "issue:34634", + "right": "issue:35707", + "accept": false, + "reason": "Bark voice_preset handling is unrelated to generation with inputs_embeds and past_key_values." + }, + { + "left": "issue:43454", + "right": "issue:43550", + "accept": false, + "reason": "Different problems: weight tying/generation corruption vs torch.compile SDPA failure." + }, + { + "left": "issue:43606", + "right": "issue:43819", + "accept": false, + "reason": "CPU offload device mismatch is unrelated to DAC latent reconstruction parity." + }, + { + "left": "issue:39692", + "right": "issue:40444", + "accept": false, + "reason": "Documentation example errors are unrelated to Qwen2.5-VL fine-tuning with iterable datasets." + }, + { + "left": "issue:44186", + "right": "issue:44265", + "accept": false, + "reason": "Tokenizer crash on NER/padding is unrelated to torch.export compatibility checks." + }, + { + "left": "issue:41720", + "right": "issue:45823", + "accept": false, + "reason": "Both mention auto device mapping, but they involve different models and distinct runtime failures." + }, + { + "left": "issue:46305", + "right": "issue:46315", + "accept": false, + "reason": "Sliding-window attention off-by-one is unrelated to tokenizer conversion performance." + }, + { + "left": "issue:43377", + "right": "issue:44610", + "accept": false, + "reason": "MIMI padding-mask batching bug is unrelated to OmDet-Turbo image-size mismatch." + }, + { + "left": "issue:42796", + "right": "issue:46209", + "accept": false, + "reason": "Tokenizer BPE marker handling and FP8 inference on B200 are different bugs." + }, + { + "left": "issue:44568", + "right": "issue:45920", + "accept": false, + "reason": "Specific BOS/EOS insertion failure for one tokenizer is not the same as broad wrong-token-id regressions." + }, + { + "left": "issue:43257", + "right": "issue:45923", + "accept": false, + "reason": "MOE weight conversion under accelerate+deepspeed is unrelated to a missing supports_gradient_checkpointing flag." + }, + { + "left": "issue:43454", + "right": "issue:45072", + "accept": false, + "reason": "Weight tying issue and bfloat16 dtype mismatch are separate model/runtime bugs." + }, + { + "left": "issue:43317", + "right": "issue:43873", + "accept": false, + "reason": "Dequantized model loading with cpu offload and generic quantization offloading are related only at a broad level, not the same bug." + }, + { + "left": "issue:43377", + "right": "issue:44898", + "accept": false, + "reason": "Batching/padding-mask output mismatch is unrelated to Perceiver non-default resolution failures." + }, + { + "left": "issue:42796", + "right": "issue:44977", + "accept": false, + "reason": "Tokenizer BPE marker bug is unrelated to flash-attention generation issues." + }, + { + "left": "issue:43454", + "right": "issue:43720", + "accept": false, + "reason": "Weight-tying failure is unrelated to packed-weight unpacking during accelerate loading." + }, + { + "left": "issue:40990", + "right": "issue:42947", + "accept": false, + "reason": "Perplexity anomaly and gradient-checkpointing ineffectiveness are different issues." + }, + { + "left": "issue:43322", + "right": "issue:46050", + "accept": false, + "reason": "Llava Next loading segfault and sinq save/load breakage are distinct model-loading bugs." + }, + { + "left": "issue:42617", + "right": "issue:42915", + "accept": false, + "reason": "3d_parallel.py execution failure is unrelated to Qwen3Moe FP8 config loading." + }, + { + "left": "issue:43749", + "right": "issue:44534", + "accept": false, + "reason": "FSDP CPU RAM efficient loading and non-persistent buffer junk are not the same code-path problem." + }, + { + "left": "issue:43572", + "right": "issue:45759", + "accept": false, + "reason": "Missing pad_token_idx and text_config unwrapping are different config bugs." + }, + { + "left": "issue:41720", + "right": "issue:43322", + "accept": false, + "reason": "Auto device-map assert on Qwen3 and Llava Next segfault are unrelated failures." + }, + { + "left": "issue:43526", + "right": "issue:43883", + "accept": false, + "reason": "BEiT label reduction bug is unrelated to missing all_tied_weights_keys on Molmo." + }, + { + "left": "issue:43881", + "right": "issue:44991", + "accept": false, + "reason": "Model loading failure for glm-4v-9b is unrelated to tokenizer loading regression for EMBEDDIA/est-roberta." + }, + { + "left": "issue:44387", + "right": "issue:45704", + "accept": false, + "reason": "Int4 quantization memory growth and apex RMSNorm memory leak are different resource issues." + }, + { + "left": "issue:41720", + "right": "issue:44977", + "accept": false, + "reason": "Both involve Qwen variants, but one is a device-map crash and the other is a flash-attention generation 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": 21, + "review_unit_index": 14, + "review_unit_count": 56, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-473", + "nodes": [ + "issue:26413", + "issue:28530", + "issue:33666", + "issue:36010", + "issue:36331", + "issue:40990", + "issue:41720", + "issue:42617", + "issue:42915", + "issue:43065", + "issue:43295", + "issue:43317", + "issue:43404", + "issue:43450", + "issue:43526", + "issue:43531", + "issue:43550", + "issue:43572", + "issue:43577", + "issue:43606", + "issue:43646", + "issue:43708", + "issue:43720", + "issue:43873", + "issue:43881", + "issue:44008", + "issue:44186", + "issue:44361", + "issue:44423", + "issue:44466", + "issue:44610", + "issue:44671", + "issue:44743", + "issue:44756", + "issue:44877", + "issue:44898", + "issue:44991", + "issue:45072", + "issue:45161", + "issue:45357", + "issue:45406", + "issue:45632", + "issue:45684", + "issue:45701", + "issue:45823", + "issue:46291", + "issue:46402", + "issue:46665" + ], + "soft_pairs": [ + "issue:44466|issue:45357", + "issue:36331|issue:42617", + "issue:26413|issue:28530", + "issue:36331|issue:42915", + "issue:43450|issue:45701", + "issue:44877|issue:44991", + "issue:40990|issue:45701", + "issue:44186|issue:45072", + "issue:43065|issue:46402", + "issue:43873|issue:44756", + "issue:43550|issue:43720", + "issue:41720|issue:42617", + "issue:43577|issue:43708", + "issue:43404|issue:45357", + "issue:43550|issue:43606", + "issue:43646|issue:46665", + "issue:43295|issue:45701", + "issue:44186|issue:44610", + "issue:44008|issue:45823", + "issue:44610|issue:44898", + "issue:42915|issue:43317", + "issue:44423|issue:45406", + "issue:36010|issue:42617", + "issue:45632|issue:45684", + "issue:41720|issue:45161", + "issue:28530|issue:33666", + "issue:43531|issue:43572", + "issue:44671|issue:44743", + "issue:42915|issue:46291", + "issue:44186|issue:44361", + "issue:36331|issue:41720", + "issue:43526|issue:43881" + ], + "prepared_review_unit_hash": "ff3c64188044eb1ba97f249d5f3d487db05a308a1ca8a4778a1ba13a2f5dfc36", + "component_budget": { + "node_count": 473, + "item_count": 473, + "soft_pair_count": 1738, + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 32, + "serialized_chars": 37369, + "estimated_input_tokens": 9343, + "estimated_eval_tokens": 18942 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This set is mostly a loose bag of unrelated issue reports; the soft pairs share broad subsystem vocabulary but not the same concrete bug or change. I would not merge any of the proposed pairs as duplicates.", + "confidence": 0.87, + "canonical_issue_reason": "issue:43873 is the broadest and most active issue in the set, centered on offloading/quantization behavior, so it is the best representative anchor if one must pick a canonical issue. It is not a duplicate of the other reports, just the most central among several adjacent device/offload problems.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:43873 has the strongest triage signal (discussion + inbound refs) and sits closest to the recurring device/offload theme seen across a few nearby reports, making it the best issue candidate overall.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44466", + "right": "issue:45357", + "accept": false, + "reason": "Different failures: lm_head serialization/device dependence vs incorrect visual encoder keys in save_pretrained." + }, + { + "left": "issue:36331", + "right": "issue:42617", + "accept": false, + "reason": "Unexpected compute_loss kwarg in CustomTrainer is unrelated to a 3d_parallel.py execution problem." + }, + { + "left": "issue:26413", + "right": "issue:28530", + "accept": false, + "reason": "Checkpoint resume/epoch_iterator failure is distinct from early-stopping metric resolution." + }, + { + "left": "issue:36331", + "right": "issue:42915", + "accept": false, + "reason": "Trainer signature regression and Qwen3Moe FP8 config failure are different code paths." + }, + { + "left": "issue:43450", + "right": "issue:45701", + "accept": false, + "reason": "Video batch-shape bug is unrelated to tokenization behavior changes across versions." + }, + { + "left": "issue:44877", + "right": "issue:44991", + "accept": false, + "reason": "Strict config loading for granite_speech is not the same as tokenizer loading failure for EMBEDDIA/est-roberta." + }, + { + "left": "issue:40990", + "right": "issue:45701", + "accept": false, + "reason": "High perplexity on gpt-oss is a model/eval-quality issue, not the tokenizer-version regression described in 45701." + }, + { + "left": "issue:44186", + "right": "issue:45072", + "accept": false, + "reason": "Tokenizer crash on NER/padding is unrelated to dtype mismatch in bfloat16 inference." + }, + { + "left": "issue:43065", + "right": "issue:46402", + "accept": false, + "reason": "Dummy Conv2d in Sam3PixelDecoder and missing Sam3Model text encoder weights are different SAM3 bugs." + }, + { + "left": "issue:43873", + "right": "issue:44756", + "accept": false, + "reason": "Quantization/offload malfunction is not the same as disabling mmap to avoid OOM on Strix Halo." + }, + { + "left": "issue:43550", + "right": "issue:43720", + "accept": false, + "reason": "torch.compile+SDPA failure for Bamba is unrelated to packed-weight unpacking during accelerate loading." + }, + { + "left": "issue:41720", + "right": "issue:42617", + "accept": false, + "reason": "Qwen3 auto device_map cuda assert is a different failure from the 3d_parallel.py runtime issue." + }, + { + "left": "issue:43577", + "right": "issue:43708", + "accept": false, + "reason": "Blip2 dtype staying float32 is unrelated to resume_from_checkpoint max_steps calculation." + }, + { + "left": "issue:43404", + "right": "issue:45357", + "accept": false, + "reason": "Tied lm_head weights in Mistral3 are not the same bug as incorrect visual encoder keys in Qwen3.5 save_pretrained." + }, + { + "left": "issue:43550", + "right": "issue:43606", + "accept": false, + "reason": "Both are runtime/model-loading bugs, but they affect different models and code paths; not one concrete duplicate." + }, + { + "left": "issue:43646", + "right": "issue:46665", + "accept": false, + "reason": "Both concern custom-model loading, but the titles point to different symptoms and not a clearly shared concrete bug." + }, + { + "left": "issue:43295", + "right": "issue:45701", + "accept": false, + "reason": "Processor.tokenizer/API regression is broader custom-code breakage, not the same as tokenization changes reported in 45701." + }, + { + "left": "issue:44186", + "right": "issue:44610", + "accept": false, + "reason": "LayoutLMv2Tokenizer crashing and OmDet-Turbo image-size mismatch are unrelated model-specific processor bugs." + }, + { + "left": "issue:44008", + "right": "issue:45823", + "accept": false, + "reason": "Gemma3n variable collision in forward is unrelated to Gemma4 PLE device mismatch under device_map=\"auto\"." + }, + { + "left": "issue:44610", + "right": "issue:44898", + "accept": false, + "reason": "Both involve image sizing, but they affect different models and different mechanisms (processor output vs positional encoding handling)." + }, + { + "left": "issue:42915", + "right": "issue:43317", + "accept": false, + "reason": "Qwen3Moe FP8 config failure is different from dequantized model loading with gpu+cpu offload." + }, + { + "left": "issue:44423", + "right": "issue:45406", + "accept": false, + "reason": "Continuous batching crash in multimodal serving and Gemma4Processor missing _tokenizer are separate server/processor issues." + }, + { + "left": "issue:36010", + "right": "issue:42617", + "accept": false, + "reason": "GenerationMixin import failure is unrelated to the 3d_parallel.py issue." + }, + { + "left": "issue:45632", + "right": "issue:45684", + "accept": false, + "reason": "Cache-path collision for local models is unrelated to read-only permission propagation in save_pretrained." + }, + { + "left": "issue:41720", + "right": "issue:45161", + "accept": false, + "reason": "Both mention parallelism/device setup, but they concern different models and distinct failures." + }, + { + "left": "issue:28530", + "right": "issue:33666", + "accept": false, + "reason": "Early stopping metric resolution and Qwen2-VL multi-GPU training are not the same bug." + }, + { + "left": "issue:43531", + "right": "issue:43572", + "accept": false, + "reason": "Qwen3-MoE sliding_window behavior and missing pad_token_idx in StableLmConfig are separate config/model issues." + }, + { + "left": "issue:44671", + "right": "issue:44743", + "accept": false, + "reason": "CamemBERT masked-LM predictions and Qwen3.5 recurrent-state reset are unrelated model behaviors." + }, + { + "left": "issue:43526", + "right": "issue:43881", + "accept": false, + "reason": "BeitImageProcessorFast label reduction bug is unrelated to glm-4v-9b loading 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": 21, + "review_unit_index": 15, + "review_unit_count": 56, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-473", + "nodes": [ + "issue:28530", + "issue:29159", + "issue:30064", + "issue:33357", + "issue:34567", + "issue:34634", + "issue:35707", + "issue:36032", + "issue:41720", + "issue:42490", + "issue:42915", + "issue:43316", + "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:44077", + "issue:44387", + "issue:44451", + "issue:44492", + "issue:44509", + "issue:44534", + "issue:44589", + "issue:44625", + "issue:44898", + "issue:44913", + "issue:44991", + "issue:45059", + "issue:45072", + "issue:45357", + "issue:45406", + "issue:45464", + "issue:45479", + "issue:45701", + "issue:45704", + "issue:45759", + "issue:45859", + "issue:46402", + "issue:46665" + ], + "soft_pairs": [ + "issue:43316|issue:44913", + "issue:42915|issue:43322", + "issue:28530|issue:30064", + "issue:43540|issue:43782", + "issue:33357|issue:35707", + "issue:43550|issue:43653", + "issue:43873|issue:45704", + "issue:43606|issue:43720", + "issue:29159|issue:36032", + "issue:30064|issue:34634", + "issue:44387|issue:45701", + "issue:44534|issue:46665", + "issue:44492|issue:44509", + "issue:43388|issue:45059", + "issue:41720|issue:43746", + "issue:43531|issue:45759", + "issue:42490|issue:43329", + "issue:43606|issue:43653", + "issue:45406|issue:45464", + "issue:34567|issue:35707", + "issue:34634|issue:41720", + "issue:45357|issue:46402", + "issue:44451|issue:44991", + "issue:45759|issue:45859", + "issue:43531|issue:44589", + "issue:44625|issue:45479", + "issue:44077|issue:44625", + "issue:43644|issue:46665", + "issue:44898|issue:45072" + ], + "prepared_review_unit_hash": "e98a17af1a0fd3e7c0b296ee7616b5578fa91d6a22389e6a2b3cc9ed80269505", + "component_budget": { + "node_count": 473, + "item_count": 473, + "soft_pair_count": 1738, + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 29, + "serialized_chars": 36440, + "estimated_input_tokens": 9110, + "estimated_eval_tokens": 18476 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "These issues span unrelated bugs across configs, tokenizers, multimodal loading, quantization/offload, and docs. None of the soft pairs look like safe duplicates or the same underlying change.", + "confidence": 0.91, + "canonical_issue_reason": "\"Transformers v5 fills non-persistent buffers with junk\" is a concrete, regression-style bug with a clear runtime symptom and broad impact, making it a strong anchor issue.", + "canonical_pr_reason": null, + "best_issue_reason": "This is the most representative standalone bug report in the set: specific, actionable, and broad enough to serve as a good canonical issue.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43316", + "right": "issue:44913", + "accept": false, + "reason": "Both are config/serialization-related, but they affect different models and different fields (`Gemma3TextConfig` API vs `rotary_pct` reload behavior)." + }, + { + "left": "issue:42915", + "right": "issue:43322", + "accept": false, + "reason": "Different failure classes and code paths: Qwen3Moe FP8 config handling versus Llava Next loading segfault." + }, + { + "left": "issue:28530", + "right": "issue:30064", + "accept": false, + "reason": "Early-stopping metric selection and void segmentation-map processing are unrelated subsystems and symptoms." + }, + { + "left": "issue:43540", + "right": "issue:43782", + "accept": false, + "reason": "Different Qwen multimodal bugs: video input processing error versus `weight_only=True` loading error." + }, + { + "left": "issue:33357", + "right": "issue:35707", + "accept": false, + "reason": "MacOS bus error while loading a CLIP model is unrelated to progressive generation with `inputs_embeds` and `past_key_values`." + }, + { + "left": "issue:43550", + "right": "issue:43653", + "accept": false, + "reason": "Torch.compile/SDPA failure in Bamba and BigBird tokenizer special-token registration are different bugs." + }, + { + "left": "issue:43873", + "right": "issue:45704", + "accept": false, + "reason": "Quantization/offload behavior and a T5 RMSNorm memory leak are distinct problems with different fixes." + }, + { + "left": "issue:43606", + "right": "issue:43720", + "accept": false, + "reason": "Both involve loading/offload, but one is a CPU offload device mismatch and the other is packed-weight unpacking during accelerate loading; not the same defect." + }, + { + "left": "issue:29159", + "right": "issue:36032", + "accept": false, + "reason": "Slow/fast tokenizer inconsistency and a T5 tokenizer loading name conflict are separate tokenizer issues." + }, + { + "left": "issue:30064", + "right": "issue:34634", + "accept": false, + "reason": "Image processor segmentation-map handling is unrelated to BarkProcessor `voice_preset` behavior." + }, + { + "left": "issue:44387", + "right": "issue:45701", + "accept": false, + "reason": "CUDA memory growth under int4 quantization is not the same as a version-dependent tokenization change." + }, + { + "left": "issue:44534", + "right": "issue:46665", + "accept": false, + "reason": "Non-persistent buffer initialization junk and inconsistent `from_pretrained` behavior inside Python classes are different bugs." + }, + { + "left": "issue:44492", + "right": "issue:44509", + "accept": false, + "reason": "A cache-strategy typo and outdated docs for removed pipeline tasks are documentation issues of different scope." + }, + { + "left": "issue:43388", + "right": "issue:45059", + "accept": false, + "reason": "Metric gathering drops tuple labels in the last batch; SAM3 text/box behavior is a different multimodal-input bug." + }, + { + "left": "issue:41720", + "right": "issue:43746", + "accept": false, + "reason": "Qwen3 auto device mapping CUDA assert and GraniteSpeech PEFT local-checkpoint loading are unrelated loading/runtime failures." + }, + { + "left": "issue:43531", + "right": "issue:45759", + "accept": false, + "reason": "Sliding-window behavior in Qwen3-MoE and `from_config` not unwrapping `text_config` are different configuration bugs." + }, + { + "left": "issue:42490", + "right": "issue:43329", + "accept": false, + "reason": "Local-path import warning and an undefined-variable bug in multimodal token counting are unrelated." + }, + { + "left": "issue:43606", + "right": "issue:43653", + "accept": false, + "reason": "Device-mismatch offload bug versus tokenizer special-token registration bug; no shared root cause." + }, + { + "left": "issue:45406", + "right": "issue:45464", + "accept": false, + "reason": "Gemma4Processor missing `_tokenizer` in serve and Qwen3.5 streaming inference failure are different API/runtime issues." + }, + { + "left": "issue:34567", + "right": "issue:35707", + "accept": false, + "reason": "Trainer state token accounting and progressive generation are unrelated trainer/generation paths." + }, + { + "left": "issue:34634", + "right": "issue:41720", + "accept": false, + "reason": "BarkProcessor voice-preset handling and Qwen3 device-mapping CUDA asserts do not share a code path." + }, + { + "left": "issue:45357", + "right": "issue:46402", + "accept": false, + "reason": "Incorrect save_pretrained visual-encoder keys and SAM3 text-encoder weights not loading are different multimodal serialization/load bugs." + }, + { + "left": "issue:44451", + "right": "issue:44991", + "accept": false, + "reason": "Both are tokenizer-load failures, but they affect different models and likely different tokenizer regressions." + }, + { + "left": "issue:45759", + "right": "issue:45859", + "accept": false, + "reason": "Composite Qwen config unwrapping and CPU-offload rotary embedding compatibility are unrelated fixes." + }, + { + "left": "issue:43531", + "right": "issue:44589", + "accept": false, + "reason": "Sliding-window handling and missing Float8 storage lookup are completely different bug classes." + }, + { + "left": "issue:44625", + "right": "issue:45479", + "accept": false, + "reason": "`num_labels` propagation and zero-loss single-label classification are both classification-related but stem from different logic errors." + }, + { + "left": "issue:44077", + "right": "issue:44625", + "accept": false, + "reason": "PatchTSMixer `post_init` policy and Qwen3.5 `num_labels` propagation are unrelated API/config issues." + }, + { + "left": "issue:43644", + "right": "issue:46665", + "accept": false, + "reason": "Non-persistent buffer junk and inconsistent `from_pretrained` in Python classes are different failure modes." + }, + { + "left": "issue:44898", + "right": "issue:45072", + "accept": false, + "reason": "Perceiver positional encoding interpolation and bf16 dtype mismatch in SwitchTransformers/TimmWrapperModel 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": 21, + "review_unit_index": 16, + "review_unit_count": 56, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-473", + "nodes": [ + "issue:30064", + "issue:36010", + "issue:39692", + "issue:41720", + "issue:42886", + "issue:43295", + "issue:43381", + "issue:43421", + "issue:43531", + "issue:43540", + "issue:43550", + "issue:43645", + "issue:43653", + "issue:43720", + "issue:43746", + "issue:43782", + "issue:43819", + "issue:43844", + "issue:43873", + "issue:43927", + "issue:43994", + "issue:44062", + "issue:44186", + "issue:44368", + "issue:44451", + "issue:44514", + "issue:44545", + "issue:44610", + "issue:44671", + "issue:44843", + "issue:44918", + "issue:44936", + "issue:44987", + "issue:45042", + "issue:45072", + "issue:45081", + "issue:45356", + "issue:45357", + "issue:45381", + "issue:45704", + "issue:45759", + "issue:45834", + "issue:45907", + "issue:45923", + "issue:46032", + "issue:46050", + "issue:46294", + "issue:46463" + ], + "soft_pairs": [ + "issue:44062|issue:44451", + "issue:44368|issue:45357", + "issue:43381|issue:43844", + "issue:44987|issue:45042", + "issue:39692|issue:44610", + "issue:44514|issue:44545", + "issue:44918|issue:44936", + "issue:43421|issue:43927", + "issue:45704|issue:45923", + "issue:43746|issue:46050", + "issue:41720|issue:45704", + "issue:43381|issue:45704", + "issue:43531|issue:43645", + "issue:42886|issue:44843", + "issue:46032|issue:46463", + "issue:45759|issue:45923", + "issue:43782|issue:45381", + "issue:43421|issue:43653", + "issue:43994|issue:44671", + "issue:39692|issue:43873", + "issue:43746|issue:43873", + "issue:43421|issue:44186", + "issue:45081|issue:45356", + "issue:30064|issue:36010", + "issue:43720|issue:43819", + "issue:45356|issue:45907", + "issue:43653|issue:43720", + "issue:43540|issue:46294", + "issue:43295|issue:45834", + "issue:43550|issue:45072" + ], + "prepared_review_unit_hash": "392866393daa0b82ec856dd429d39a25d6f3f1cc5267f4d6b0349c3845f3f7b2", + "component_budget": { + "node_count": 473, + "item_count": 473, + "soft_pair_count": 1738, + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 30, + "serialized_chars": 37057, + "estimated_input_tokens": 9265, + "estimated_eval_tokens": 18786 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly noisy and heterogeneous: most items are unrelated Transformers bugs spanning tokenizers, multimodal processors, quantization, and training/runtime regressions. Only one soft edge looks like a true duplicate pair (`Qwen2_5_VLProcessor.apply_chat_template` with `padding=False`), while the rest should stay separate.", + "confidence": 0.87, + "canonical_issue_reason": "Issue 43421 is the strongest representative of the cluster\u2019s main tokenizer/runtime theme: it describes a concrete core-library bug around special-token changes not propagating to the post-processor, and it sits near several other tokenizer-related reports.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 43421 is the best overall issue to keep because it is the most general and reusable bug report in the set, with the clearest cross-cutting impact compared with the many narrower model-specific regressions.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44062", + "right": "issue:44451", + "accept": false, + "reason": "Different problems: `AddedToken(..., special=...)` constructor conflict vs a model-loading failure for `vesteinn/ScandiBERT`." + }, + { + "left": "issue:44368", + "right": "issue:45357", + "accept": false, + "reason": "Both are warnings/regressions in multimodal setups, but one is about `tie_word_embeddings` and the other about incorrect visual encoder keys on save/load." + }, + { + "left": "issue:43381", + "right": "issue:43844", + "accept": false, + "reason": "Eval-mode gradient checkpointing behavior and ZeRO-3 gradient growth are distinct training-path issues." + }, + { + "left": "issue:44987", + "right": "issue:45042", + "accept": false, + "reason": "Unrelated failures: loading `physical-intelligence/fast` vs PIL image processors incorrectly requiring torchvision." + }, + { + "left": "issue:39692", + "right": "issue:44610", + "accept": false, + "reason": "Different model/documentation bugs: SigLIP2 example errors vs OmDet-Turbo image-size mismatch." + }, + { + "left": "issue:44514", + "right": "issue:44545", + "accept": true, + "reason": "Same concrete bug: `Qwen2_5_VLProcessor.apply_chat_template` crashes on batched input when `padding=False`." + }, + { + "left": "issue:44918", + "right": "issue:44936", + "accept": false, + "reason": "Different code paths: Qwen3.5 embedding unpacking in TRL SFT vs `trainer.evaluate()` failing after `trainer.train()`." + }, + { + "left": "issue:43421", + "right": "issue:43927", + "accept": false, + "reason": "Related tokenizer state, but one is runtime special-token/post-processor updates and the other is config persistence for custom token IDs." + }, + { + "left": "issue:45704", + "right": "issue:45923", + "accept": false, + "reason": "Different subsystems: T5/apex RMSNorm memory leak vs missing `supports_gradient_checkpointing` on a trust-remote-code model." + }, + { + "left": "issue:43746", + "right": "issue:46050", + "accept": false, + "reason": "Adapter-loading failure for local checkpoints vs quantized-model save/load breakage are not the same bug." + }, + { + "left": "issue:41720", + "right": "issue:45704", + "accept": false, + "reason": "Qwen3 auto device-mapping CUDA assert is unrelated to the T5/apex RMSNorm leak." + }, + { + "left": "issue:43381", + "right": "issue:45704", + "accept": false, + "reason": "Gradient checkpointing in eval mode and the T5 apex memory leak are separate issues." + }, + { + "left": "issue:43531", + "right": "issue:43645", + "accept": false, + "reason": "Sliding-window handling for Qwen3-MoE is unrelated to notebook-defined custom model initialization in Transformers 5.0.0." + }, + { + "left": "issue:42886", + "right": "issue:44843", + "accept": false, + "reason": "Both concern offline/tokenizer loading, but one is cache resolution for tokenizers and the other is an unconditional `model_info()` call in `_patch_mistral_regex`." + }, + { + "left": "issue:46032", + "right": "issue:46463", + "accept": false, + "reason": "Mamba2Mixer cache correctness with `seq_len > 1` is unrelated to T5 decoder-input validation messaging." + }, + { + "left": "issue:45759", + "right": "issue:45923", + "accept": false, + "reason": "`AutoModelForCausalLM.from_config` unwrapping `text_config` is a different composite-config bug from a missing gradient-checkpointing flag." + }, + { + "left": "issue:43782", + "right": "issue:45381", + "accept": false, + "reason": "Qwen3VL weight-only loading and Qwen2.5-VL video `vision_position_ids` are different multimodal failures." + }, + { + "left": "issue:43994", + "right": "issue:44671", + "accept": false, + "reason": "Different symptoms and models: incorrect SigLIP2 results vs CamemBERT masked-LM predictions." + }, + { + "left": "issue:39692", + "right": "issue:43873", + "accept": false, + "reason": "Documentation/example issues for SigLIP2 are not the same as quantization/offloading behavior." + }, + { + "left": "issue:43746", + "right": "issue:43873", + "accept": false, + "reason": "PEFT adapter checkpoint loading and quantization offloading are separate model-loading paths." + }, + { + "left": "issue:43421", + "right": "issue:43653", + "accept": false, + "reason": "Both touch tokenizer special-token handling, but one is post-processor refresh on runtime edits and the other is BigBird mask-token registration causing empty decode output." + }, + { + "left": "issue:45081", + "right": "issue:45356", + "accept": false, + "reason": "Both involve Mistral/Kimi tokenizer regex handling, but the reported failures are not the same concrete bug." + }, + { + "left": "issue:30064", + "right": "issue:36010", + "accept": false, + "reason": "Image processor void-segmentation failure is unrelated to the `GenerationMixin` import regression." + }, + { + "left": "issue:43720", + "right": "issue:43819", + "accept": false, + "reason": "BitNet packed-weight unpacking during accelerate loading is unrelated to DAC `from_latents` vs forward-pass mismatch." + }, + { + "left": "issue:45356", + "right": "issue:45907", + "accept": false, + "reason": "Tokenizer codec/regex regression is unrelated to `list_repo_files` failing to forward `revision`." + }, + { + "left": "issue:43653", + "right": "issue:43720", + "accept": false, + "reason": "Both are CI-tagged issues, but BigBird special-token decoding and BitNet packed-weight loading are different bugs." + }, + { + "left": "issue:43540", + "right": "issue:46294", + "accept": false, + "reason": "Both are Qwen multimodal video issues, but one is a processing `ValueError` in Qwen3OmniMoe and the other is a video-feature/token mismatch in Qwen3.5-9B." + }, + { + "left": "issue:43295", + "right": "issue:45834", + "accept": false, + "reason": "Custom model processor/tokenizer access regression is unrelated to Kosmos2.5 long-OCR index errors." + }, + { + "left": "issue:43550", + "right": "issue:45072", + "accept": false, + "reason": "Different runtime paths: Bamba-9B-v2 torch.compile with SDPA vs dtype mismatches in SwitchTransformers/TimmWrapperModel." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet, and the soft-edge decisions are conservative overall. The only accepted pair, issue:44514 / issue:44545, is an evident duplicate with the same Qwen2_5_VLProcessor.apply_chat_template padding=False crash. The remaining pairs are plausibly related by theme but not the same concrete bug, so rejecting them is safe." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 21, + "review_unit_index": 17, + "review_unit_count": 56, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-473", + "nodes": [ + "issue:30064", + "issue:36331", + "issue:39692", + "issue:41720", + "issue:43322", + "issue:43335", + "issue:43388", + "issue:43531", + "issue:43540", + "issue:43550", + "issue:43596", + "issue:43611", + "issue:43618", + "issue:43643", + "issue:43644", + "issue:43653", + "issue:43697", + "issue:43716", + "issue:43749", + "issue:43756", + "issue:43819", + "issue:43927", + "issue:44008", + "issue:44062", + "issue:44186", + "issue:44220", + "issue:44442", + "issue:44448", + "issue:44485", + "issue:44488", + "issue:44509", + "issue:44661", + "issue:44671", + "issue:44821", + "issue:44843", + "issue:44877", + "issue:44898", + "issue:44991", + "issue:45081", + "issue:45084", + "issue:45183", + "issue:45479", + "issue:45584", + "issue:45685", + "issue:45834", + "issue:46032", + "issue:46665" + ], + "soft_pairs": [ + "issue:43596|issue:44220", + "issue:43550|issue:43927", + "issue:44062|issue:44488", + "issue:43618|issue:43697", + "issue:43611|issue:46665", + "issue:44186|issue:44898", + "issue:44220|issue:44661", + "issue:43653|issue:43819", + "issue:43531|issue:43643", + "issue:43335|issue:43756", + "issue:43550|issue:43819", + "issue:43540|issue:44008", + "issue:44843|issue:45081", + "issue:41720|issue:44991", + "issue:43643|issue:44877", + "issue:30064|issue:36331", + "issue:44821|issue:45084", + "issue:39692|issue:45834", + "issue:44186|issue:44442", + "issue:43322|issue:45685", + "issue:44448|issue:44671", + "issue:45584|issue:45834", + "issue:45183|issue:45479", + "issue:43388|issue:46032", + "issue:43716|issue:44008", + "issue:44485|issue:44509", + "issue:43644|issue:43749" + ], + "prepared_review_unit_hash": "ccca756ad017e74c71b72a71f8b12adc5d727fb9fae4a6585076010ba5a0d397", + "component_budget": { + "node_count": 473, + "item_count": 473, + "soft_pair_count": 1738, + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 27, + "serialized_chars": 35388, + "estimated_input_tokens": 8847, + "estimated_eval_tokens": 17950 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is not a duplicate set; it mixes many unrelated Transformers bugs across tokenizers, multimodal models, loading/config paths, and v5 regressions. None of the soft edges look like the same concrete underlying defect.", + "confidence": 0.94, + "canonical_issue_reason": "No clear canonical issue: the items are heterogeneous and do not collapse to one shared bug family.", + "canonical_pr_reason": null, + "best_issue_reason": "No best issue candidate stands out as representing the rest; the cluster lacks a single common code path or symptom.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43596", + "right": "issue:44220", + "accept": false, + "reason": "Different subsystems and failures: Deepspeed/BERT init index error vs audio fbank extraction." + }, + { + "left": "issue:43550", + "right": "issue:43927", + "accept": false, + "reason": "Unrelated bugs: torch.compile/SDPA failure vs DiaConfig losing custom token IDs on save/load." + }, + { + "left": "issue:44062", + "right": "issue:44488", + "accept": false, + "reason": "Both tokenizer-related, but one is an AddedToken kwarg conflict and the other is a model loading failure for a specific tokenizer." + }, + { + "left": "issue:43618", + "right": "issue:43697", + "accept": false, + "reason": "Both are v5 regressions, but they affect different models and code paths, so they are not the same bug." + }, + { + "left": "issue:43611", + "right": "issue:46665", + "accept": false, + "reason": "Both concern loading, but one is base_model_prefix behavior and the other is from_pretrained inside Python classes; distinct issues." + }, + { + "left": "issue:44186", + "right": "issue:44898", + "accept": false, + "reason": "Different domains and symptoms: LayoutLMv2 tokenizer/NER padding crash vs Perceiver image classification interpolation failure." + }, + { + "left": "issue:44220", + "right": "issue:44661", + "accept": false, + "reason": "Feature extraction bug vs add-new-model-like failure in TOKENIZER_MAPPING_NAMES; no shared concrete defect." + }, + { + "left": "issue:43653", + "right": "issue:43819", + "accept": false, + "reason": "Tokenizer special-token registration issue vs DAC from_latents mismatch; unrelated code paths." + }, + { + "left": "issue:43531", + "right": "issue:43643", + "accept": false, + "reason": "Sliding-window behavior in Qwen3-MoE is unrelated to trust_remote_code config field loss." + }, + { + "left": "issue:43335", + "right": "issue:43756", + "accept": false, + "reason": "Config creates sparse layers incorrectly vs RoPE layer drop in Smollm3; different models and bugs." + }, + { + "left": "issue:43550", + "right": "issue:43819", + "accept": false, + "reason": "torch.compile/SDPA failure is unrelated to DAC latent decoding mismatch." + }, + { + "left": "issue:43540", + "right": "issue:44008", + "accept": false, + "reason": "Different multimodal failures: video input processing in Qwen3OmniMoe vs Gemma3n variable collision." + }, + { + "left": "issue:44843", + "right": "issue:45081", + "accept": false, + "reason": "Same helper function area, but different failures: offline model_info call vs backend_tokenizer attribute error." + }, + { + "left": "issue:41720", + "right": "issue:44991", + "accept": false, + "reason": "Qwen3 auto device mapping cuda assert is unrelated to tokenizer loading for EMBEDDIA/est-roberta." + }, + { + "left": "issue:43643", + "right": "issue:44877", + "accept": false, + "reason": "Both are config-loading problems, but the root causes and affected objects differ." + }, + { + "left": "issue:30064", + "right": "issue:36331", + "accept": false, + "reason": "Image segmentation map processing bug vs CustomTrainer compute_loss signature change; unrelated." + }, + { + "left": "issue:44821", + "right": "issue:45084", + "accept": false, + "reason": "AutoImageProcessor URL loading failure vs template-node compilation error; no shared bug." + }, + { + "left": "issue:39692", + "right": "issue:45834", + "accept": false, + "reason": "Doc example/model mismatch and quantization errors are unrelated to Kosmos2.5 long-ocr index errors." + }, + { + "left": "issue:44186", + "right": "issue:44442", + "accept": false, + "reason": "Tokenizer crash on NER/padding is not the same as AutoTokenizer failing to load a FastSpeech2Conformer tokenizer." + }, + { + "left": "issue:44448", + "right": "issue:44671", + "accept": false, + "reason": "Both are v5 output regressions, but they affect different models and likely different causes." + }, + { + "left": "issue:45584", + "right": "issue:45834", + "accept": false, + "reason": "Whisper empty transcription failure after token alignment is unrelated to Kosmos2.5 long-input indexing." + }, + { + "left": "issue:45183", + "right": "issue:45479", + "accept": false, + "reason": "Input validation/XOR messaging bug vs degenerate zero-loss classification behavior; different underlying defects." + }, + { + "left": "issue:43388", + "right": "issue:46032", + "accept": false, + "reason": "Metrics gathering last-batch bug is unrelated to Mamba2Mixer use_cache incorrect results." + }, + { + "left": "issue:43716", + "right": "issue:44008", + "accept": false, + "reason": "Different multimodal model families and failures: dtype mismatch in Mistral-3 vs Gemma3n attribute collision." + }, + { + "left": "issue:44485", + "right": "issue:44509", + "accept": false, + "reason": "RoPE implementation discussion is unrelated to docs still mentioning removed pipeline tasks." + }, + { + "left": "issue:43644", + "right": "issue:43749", + "accept": false, + "reason": "Both are v5/loading related, but one is non-persistent buffers filled with junk and the other is broken FSDP CPU RAM efficient loading." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 21, + "review_unit_index": 18, + "review_unit_count": 56, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-473", + "nodes": [ + "issue:28218", + "issue:29159", + "issue:34634", + "issue:36010", + "issue:36331", + "issue:42175", + "issue:43262", + "issue:43421", + "issue:43540", + "issue:43577", + "issue:43582", + "issue:43606", + "issue:43645", + "issue:43720", + "issue:43749", + "issue:43782", + "issue:43818", + "issue:43819", + "issue:43825", + "issue:43827", + "issue:43950", + "issue:43994", + "issue:44291", + "issue:44479", + "issue:44561", + "issue:44610", + "issue:44671", + "issue:44805", + "issue:44811", + "issue:44857", + "issue:44936", + "issue:44977", + "issue:44991", + "issue:45003", + "issue:45020", + "issue:45072", + "issue:45081", + "issue:45381", + "issue:45701", + "issue:45834", + "issue:45987", + "issue:46192", + "issue:46285", + "issue:46291", + "issue:46294", + "issue:46463", + "issue:46512", + "issue:46665" + ], + "soft_pairs": [ + "issue:43606|issue:45072", + "issue:43582|issue:46512", + "issue:43950|issue:45987", + "issue:34634|issue:36010", + "issue:44479|issue:45834", + "issue:43825|issue:43994", + "issue:43577|issue:43818", + "issue:43421|issue:44811", + "issue:43782|issue:44977", + "issue:46294|issue:46463", + "issue:43540|issue:45381", + "issue:44805|issue:44936", + "issue:44991|issue:46291", + "issue:44561|issue:45020", + "issue:44610|issue:45072", + "issue:43720|issue:45072", + "issue:28218|issue:29159", + "issue:43825|issue:43827", + "issue:43577|issue:43582", + "issue:43749|issue:44291", + "issue:45987|issue:46192", + "issue:43645|issue:46665", + "issue:44671|issue:45701", + "issue:42175|issue:45081", + "issue:34634|issue:36331", + "issue:43582|issue:44857", + "issue:45003|issue:46285", + "issue:43262|issue:43819" + ], + "prepared_review_unit_hash": "4fb6955d9525ba4762d8687fb97df70f8b6cd246dd25cf63fefaca75df43b34f", + "component_budget": { + "node_count": 473, + "item_count": 473, + "soft_pair_count": 1738, + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 28, + "serialized_chars": 36429, + "estimated_input_tokens": 9108, + "estimated_eval_tokens": 18472 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a grab bag of unrelated regression reports. The only likely duplicate pair is the tokenizer spacing/inconsistency reports; the rest are too model- or feature-specific to safely merge.", + "confidence": 0.38, + "canonical_issue_reason": "issue:45701 is the broadest, most umbrella-like issue in the set, and it best represents the general tokenizer/regression theme among the items.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:45701 has the widest scope and strongest chance of serving as the representative issue for the cluster's tokenizer-regression subset.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43606", + "right": "issue:45072", + "accept": false, + "reason": "Both are inference mismatches, but they describe different models and different symptoms (device offload vs bfloat16 dtype mismatch)." + }, + { + "left": "issue:43582", + "right": "issue:46512", + "accept": false, + "reason": "Unrelated: AppleSilicon warmup TypeError vs stale config dtype after from_config." + }, + { + "left": "issue:43950", + "right": "issue:45987", + "accept": false, + "reason": "Different regressions: non-persistent buffer corruption vs StaticCache length return type." + }, + { + "left": "issue:34634", + "right": "issue:36010", + "accept": false, + "reason": "BarkProcessor voice preset failure and GenerationMixin import failure are unrelated." + }, + { + "left": "issue:44479", + "right": "issue:45834", + "accept": false, + "reason": "Different vision/video failures in different models; no clear shared bug." + }, + { + "left": "issue:43825", + "right": "issue:43994", + "accept": false, + "reason": "One is an error-message/doc issue around pipeline deprecation; the other is a model output regression." + }, + { + "left": "issue:43577", + "right": "issue:43818", + "accept": false, + "reason": "Different BLIP2 dtype propagation issue vs Video-LLaVA architecture/weight issue." + }, + { + "left": "issue:43421", + "right": "issue:44811", + "accept": false, + "reason": "Both involve tokenizers, but one is runtime special-token post-processing and the other is batch_decode skip_special_tokens behavior." + }, + { + "left": "issue:43782", + "right": "issue:44977", + "accept": false, + "reason": "Different Qwen generation/loading failures with different triggers and code paths." + }, + { + "left": "issue:46294", + "right": "issue:46463", + "accept": false, + "reason": "Qwen video feature/token mismatch vs T5 decoder-input validation messaging are unrelated." + }, + { + "left": "issue:43540", + "right": "issue:45381", + "accept": false, + "reason": "Both are video-input bugs, but they affect different Qwen families and likely different preprocessing paths." + }, + { + "left": "issue:44805", + "right": "issue:44936", + "accept": false, + "reason": "Mask shape IndexError and trainer.evaluate() failing after train are not the same underlying bug." + }, + { + "left": "issue:44991", + "right": "issue:46291", + "accept": false, + "reason": "Tokenizer load failure for a specific model and kernels import/version constraint are unrelated." + }, + { + "left": "issue:44561", + "right": "issue:45020", + "accept": false, + "reason": "Both touch remote_code compatibility, but the titles point to different failure modes and 45020 is a broad umbrella rather than the same concrete bug." + }, + { + "left": "issue:44610", + "right": "issue:45072", + "accept": false, + "reason": "OmDet input-size mismatch and dtype mismatch in inference are different issues." + }, + { + "left": "issue:43720", + "right": "issue:45072", + "accept": false, + "reason": "Packed-weight loading failure and bfloat16 dtype mismatch are separate accelerate/loading regressions." + }, + { + "left": "issue:28218", + "right": "issue:29159", + "accept": true, + "reason": "Both describe the same tokenizer inconsistency around added tokens/spacing; 29159 is the broader version of the same symptom." + }, + { + "left": "issue:43825", + "right": "issue:43827", + "accept": false, + "reason": "One is an error-message regression, the other is stale docs; related topic, but not the same underlying bug/change." + }, + { + "left": "issue:43577", + "right": "issue:43582", + "accept": false, + "reason": "Different dtype handling bug vs AppleSilicon warmup TypeError." + }, + { + "left": "issue:43749", + "right": "issue:44291", + "accept": false, + "reason": "Both are loading-related, but they describe different breakages in different code paths." + }, + { + "left": "issue:45987", + "right": "issue:46192", + "accept": false, + "reason": "Both are processor/return-shape issues, but they affect different parameters and code paths." + }, + { + "left": "issue:43645", + "right": "issue:46665", + "accept": false, + "reason": "Both mention from_pretrained in class-like contexts, but the titles are too broad and point to different scenarios to treat as the same bug." + }, + { + "left": "issue:44671", + "right": "issue:45701", + "accept": false, + "reason": "CamemBERT masked-LM regression is model-specific; 45701 is a broad tokenization-change report, not the same concrete bug." + }, + { + "left": "issue:42175", + "right": "issue:45081", + "accept": false, + "reason": "TensorFlow backend packaging and Mistral tokenizer patch crash are unrelated." + }, + { + "left": "issue:34634", + "right": "issue:36331", + "accept": false, + "reason": "BarkProcessor voice preset and CustomTrainer.compute_loss signature mismatch are unrelated." + }, + { + "left": "issue:43582", + "right": "issue:44857", + "accept": false, + "reason": "Different numerical/device issues: AppleSilicon warmup TypeError vs CUDA AMP crash." + }, + { + "left": "issue:45003", + "right": "issue:46285", + "accept": false, + "reason": "sys.modules access bug and CLIP text-model structure inconsistency are unrelated." + }, + { + "left": "issue:43262", + "right": "issue:43819", + "accept": false, + "reason": "Sampling-rate defaulting in audio chat templates and DAC latent/forward mismatch are different bugs." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The analysis is grounded in the packet and stays conservative. The umbrella choice of issue:45701 is reasonable given its broad tokenizer-change title, and the only accepted soft-edge pair (28218/29159) is well supported by the titles. The rejected pairs mostly look safely non-mergeable as single bugfixes based on the information provided." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 21, + "review_unit_index": 19, + "review_unit_count": 56, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-473", + "nodes": [ + "issue:36010", + "issue:36331", + "issue:38175", + "issue:39692", + "issue:42175", + "issue:42490", + "issue:43257", + "issue:43329", + "issue:43452", + "issue:43526", + "issue:43531", + "issue:43550", + "issue:43638", + "issue:43645", + "issue:43653", + "issue:43697", + "issue:43720", + "issue:43742", + "issue:43746", + "issue:43782", + "issue:43819", + "issue:43854", + "issue:43906", + "issue:43927", + "issue:44060", + "issue:44368", + "issue:44387", + "issue:44554", + "issue:44589", + "issue:44610", + "issue:44617", + "issue:44811", + "issue:44912", + "issue:44936", + "issue:44977", + "issue:44991", + "issue:45020", + "issue:45042", + "issue:45081", + "issue:45084", + "issue:45325", + "issue:45372", + "issue:45406", + "issue:45538", + "issue:45600", + "issue:46209", + "issue:46291", + "issue:46326" + ], + "soft_pairs": [ + "issue:43653|issue:44912", + "issue:43782|issue:45325", + "issue:45081|issue:45406", + "issue:43526|issue:43697", + "issue:45372|issue:46326", + "issue:44060|issue:44368", + "issue:43329|issue:45600", + "issue:44589|issue:45042", + "issue:43742|issue:45020", + "issue:45084|issue:45538", + "issue:43526|issue:44554", + "issue:43550|issue:44912", + "issue:43906|issue:46209", + "issue:43452|issue:44991", + "issue:43638|issue:44936", + "issue:43257|issue:43329", + "issue:43819|issue:43927", + "issue:36010|issue:43645", + "issue:36010|issue:36331", + "issue:43720|issue:44610", + "issue:43531|issue:45042", + "issue:42175|issue:46291", + "issue:38175|issue:44610", + "issue:39692|issue:44912", + "issue:44387|issue:44617", + "issue:43746|issue:44811", + "issue:42490|issue:43257", + "issue:43854|issue:44977" + ], + "prepared_review_unit_hash": "df735a76fb0a470a220a1f8afdfc12e6080a8cf4d4ad56916003b7416dec58c3", + "component_budget": { + "node_count": 473, + "item_count": 473, + "soft_pair_count": 1738, + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 28, + "serialized_chars": 36420, + "estimated_input_tokens": 9105, + "estimated_eval_tokens": 18466 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This is not a true duplicate cluster: the items cover unrelated Transformers regressions across tokenizers, multimodal processors, quantization, training, packaging, and backend-specific behavior. Issue 45020 is the broadest umbrella, but it is still too general to absorb the rest as duplicates.", + "confidence": 0.93, + "canonical_issue_reason": "issue:45020 is the broadest, highest-level regression report in the set (recent Transformers versions breaking remote_code models), so it is the closest thing to an umbrella issue, though not a true duplicate of the others.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:45020 is the best representative of the cluster because it captures the general theme of version-induced loading breakages; the remaining issues are too specific and materially different to serve as one canonical duplicate target.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43653", + "right": "issue:44912", + "accept": false, + "reason": "Both involve model/loading behavior, but one is a tokenizer special-token decode bug and the other is MXFP4 loading fallback; different code paths and symptoms." + }, + { + "left": "issue:43782", + "right": "issue:45325", + "accept": false, + "reason": "Qwen3VL weight_only loading failure is unrelated to Qwen2.5-VL rope index scaling; different models and bugs." + }, + { + "left": "issue:45081", + "right": "issue:45406", + "accept": false, + "reason": "Tokenizer regex patch crash in Mistral loading is not the same as Gemma4Processor missing _tokenizer in transformers serve." + }, + { + "left": "issue:43526", + "right": "issue:43697", + "accept": false, + "reason": "BEiT reduce_labels returning one label and RTDetrV2 differing outputs are distinct model bugs with no shared failure mechanism." + }, + { + "left": "issue:45372", + "right": "issue:46326", + "accept": false, + "reason": "Both touch mistral_common, but one is a missing import breaking Gemma 4 processor loading and the other is continue_final_message behavior; not the same defect." + }, + { + "left": "issue:44060", + "right": "issue:44368", + "accept": false, + "reason": "Both mention tied-weights warnings in Qwen3 variants, but one is an internal incorrect tying bug and the other is a user-facing LoRA warning; not one concrete issue." + }, + { + "left": "issue:43329", + "right": "issue:45600", + "accept": false, + "reason": "Undefined video-token helper variables and removed Sam3LiteText mappings are unrelated regressions despite both affecting multimodal internals." + }, + { + "left": "issue:44589", + "right": "issue:45042", + "accept": false, + "reason": "Missing Float8 storage support and PIL backend requiring torchvision are unrelated backend/environment failures." + }, + { + "left": "issue:43742", + "right": "issue:45020", + "accept": false, + "reason": "MobileLLM key error and remote_code breakage are different loading problems affecting different code paths." + }, + { + "left": "issue:45084", + "right": "issue:45538", + "accept": false, + "reason": "Template compilation failure and CLIPTokenizer model_max_length are unrelated tokenizer/template issues." + }, + { + "left": "issue:43526", + "right": "issue:44554", + "accept": false, + "reason": "A label reduction bug in BeitImageProcessorFast is unrelated to the MPS attention correctness issue." + }, + { + "left": "issue:43550", + "right": "issue:44912", + "accept": false, + "reason": "torch.compile SDPA failure in Bamba and MXFP4 quantization fallback in git-oss-20b are different model/runtime bugs." + }, + { + "left": "issue:43906", + "right": "issue:46209", + "accept": false, + "reason": "An isolated reproduction of issue 38071 is not the same as FP8 inference failing on B200 GPUs." + }, + { + "left": "issue:43452", + "right": "issue:44991", + "accept": false, + "reason": "GGUF file handling breakage and tokenizer loading failure for a specific model are separate loading regressions." + }, + { + "left": "issue:43638", + "right": "issue:44936", + "accept": false, + "reason": "DeepSpeed ZeRO-3 index error during training and trainer.evaluate() failing after train are distinct trainer/runtime issues." + }, + { + "left": "issue:43257", + "right": "issue:43329", + "accept": false, + "reason": "Qwen3 MOE weight conversion under accelerate+deepspeed is unrelated to the multimodal token counting video-branch bug." + }, + { + "left": "issue:43819", + "right": "issue:43927", + "accept": false, + "reason": "DAC from_latents mismatch and DiaConfig custom token IDs lost after save/load are different serialization/inference bugs." + }, + { + "left": "issue:36010", + "right": "issue:43645", + "accept": false, + "reason": "GenerationMixin import failure and Jupyter custom model initialization breakage do not share the same failure mode." + }, + { + "left": "issue:36010", + "right": "issue:36331", + "accept": false, + "reason": "GenerationMixin import failure is unrelated to CustomTrainer.compute_loss receiving an unexpected keyword argument." + }, + { + "left": "issue:43720", + "right": "issue:44610", + "accept": false, + "reason": "BitNet packed-weight unpacking during accelerate loading and OmDet-Turbo processor size mismatch are unrelated defects." + }, + { + "left": "issue:43531", + "right": "issue:45042", + "accept": false, + "reason": "Qwen3-MoE sliding_window behavior and PIL backend torchvision dependency are different issues in different subsystems." + }, + { + "left": "issue:42175", + "right": "issue:46291", + "accept": false, + "reason": "Missing tensorflow in the torch extra and kernels v0.15 import-time version/revision validation are unrelated packaging/import failures." + }, + { + "left": "issue:38175", + "right": "issue:44610", + "accept": false, + "reason": "Unexpected zero probabilities in SigLIP2 and OmDet-Turbo input-size mismatch are unrelated model/processor bugs." + }, + { + "left": "issue:39692", + "right": "issue:44912", + "accept": false, + "reason": "Broken SigLIP2 docs example and MXFP4 quantization fallback are different problems; one is documentation/example correctness, the other runtime loading." + }, + { + "left": "issue:44387", + "right": "issue:44617", + "accept": false, + "reason": "Both mention memory, but int4 quantization reserved-memory growth and Sam3Video CUDA OOM are not the same underlying bug." + }, + { + "left": "issue:43746", + "right": "issue:44811", + "accept": false, + "reason": "PEFT local-checkpoint loading for GraniteSpeech and Whisper batch_decode ignoring skip_special_tokens are unrelated issues." + }, + { + "left": "issue:42490", + "right": "issue:43257", + "accept": false, + "reason": "Local-path trailing-slash import warning and Qwen3 MOE weight conversion are unrelated." + }, + { + "left": "issue:43854", + "right": "issue:44977", + "accept": false, + "reason": "GLM-4.7-Flash loading in tests and Qwen3.5 flash-attention generation are separate model-specific 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": 21, + "review_unit_index": 20, + "review_unit_count": 56, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-473", + "nodes": [ + "issue:28282", + "issue:29159", + "issue:33357", + "issue:34567", + "issue:34689", + "issue:40990", + "issue:42915", + "issue:43257", + "issue:43262", + "issue:43329", + "issue:43404", + "issue:43425", + "issue:43450", + "issue:43454", + "issue:43525", + "issue:43577", + "issue:43606", + "issue:43643", + "issue:43650", + "issue:43696", + "issue:43697", + "issue:43742", + "issue:43749", + "issue:43906", + "issue:43927", + "issue:43931", + "issue:44291", + "issue:44479", + "issue:44568", + "issue:44756", + "issue:44811", + "issue:44821", + "issue:44912", + "issue:44987", + "issue:45042", + "issue:45127", + "issue:45137", + "issue:45356", + "issue:45357", + "issue:45405", + "issue:45459", + "issue:45538", + "issue:45759", + "issue:45803", + "issue:45859", + "issue:46291", + "issue:46421", + "issue:46424" + ], + "soft_pairs": [ + "issue:43262|issue:44811", + "issue:43643|issue:45759", + "issue:43257|issue:43696", + "issue:46421|issue:46424", + "issue:45459|issue:45803", + "issue:43329|issue:44479", + "issue:43650|issue:43906", + "issue:43257|issue:43931", + "issue:43329|issue:43450", + "issue:42915|issue:43742", + "issue:43606|issue:45859", + "issue:43749|issue:44756", + "issue:34567|issue:40990", + "issue:44291|issue:45042", + "issue:43454|issue:43577", + "issue:42915|issue:44821", + "issue:29159|issue:34567", + "issue:44568|issue:45356", + "issue:43927|issue:44912", + "issue:43525|issue:46291", + "issue:44987|issue:45538", + "issue:43697|issue:45357", + "issue:43404|issue:45127", + "issue:28282|issue:34689", + "issue:42915|issue:45538", + "issue:43425|issue:46291", + "issue:45137|issue:45405", + "issue:33357|issue:34567" + ], + "prepared_review_unit_hash": "1dbb12db71b1f3ec8707869dd24b76d4484c11218e6c2ef5ba5f0fc9d47ca03f", + "component_budget": { + "node_count": 473, + "item_count": 473, + "soft_pair_count": 1738, + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 28, + "serialized_chars": 36276, + "estimated_input_tokens": 9069, + "estimated_eval_tokens": 18394 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is highly heterogeneous: tokenizer, multimodal, cache, loading, training, and offload issues are all mixed together. No soft pair appears to describe the same underlying bug or change.", + "confidence": 0.87, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": "issue:46424 is the most suitable representative here: it is open, concrete, and already has multiple inbound references, so it is the clearest actionable issue in an otherwise mixed set.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43262", + "right": "issue:44811", + "accept": false, + "reason": "Both are audio-related, but they target different APIs and bugs: chat template sampling-rate defaulting vs batch_decode skip_special_tokens handling." + }, + { + "left": "issue:43643", + "right": "issue:45759", + "accept": false, + "reason": "Both involve config/model loading, but one is about missing fields with trust_remote_code and the other about not unwrapping text_config in from_config." + }, + { + "left": "issue:43257", + "right": "issue:43696", + "accept": false, + "reason": "Qwen3 MoE weight conversion with accelerate/deepspeed is unrelated to a GPT-oss CUDA OOM failure." + }, + { + "left": "issue:46421", + "right": "issue:46424", + "accept": false, + "reason": "Both mention cache, but one is about spurious recompilation from lazy init and the other about max_cache_length only applying to Quantized Cache." + }, + { + "left": "issue:45459", + "right": "issue:45803", + "accept": false, + "reason": "Both are exception-handling complaints, but they affect different tokenizer/processor code paths and different failure modes." + }, + { + "left": "issue:43329", + "right": "issue:44479", + "accept": false, + "reason": "Both mention video regressions, but one is a specific undefined-variable bug in token counting and the other is a broader multi-model v5.3 regression." + }, + { + "left": "issue:43650", + "right": "issue:43906", + "accept": false, + "reason": "No shared bug is evident; one title is a placeholder and the other is an isolated reproduction of a different issue." + }, + { + "left": "issue:43257", + "right": "issue:43931", + "accept": false, + "reason": "Both are Qwen3-related load problems, but one is about MoE weight conversion and the other is a shape mismatch for Qwen3-VL." + }, + { + "left": "issue:43329", + "right": "issue:43450", + "accept": false, + "reason": "Different video bugs: undefined variables in multimodal token counting versus incorrect batched video processor output shapes." + }, + { + "left": "issue:42915", + "right": "issue:43742", + "accept": false, + "reason": "Different loading failures with different models and causes: FineGrainedFP8Config vs a KeyError on MobileLLM-125M." + }, + { + "left": "issue:43606", + "right": "issue:45859", + "accept": false, + "reason": "Both involve CPU offload, but they are distinct model-specific incompatibilities with different underlying code paths." + }, + { + "left": "issue:43749", + "right": "issue:44756", + "accept": false, + "reason": "One is FSDP CPU RAM-efficient loading being broken; the other is a Strix Halo mmap/OOM workaround. Not the same bug." + }, + { + "left": "issue:34567", + "right": "issue:40990", + "accept": false, + "reason": "Training-state token counting is unrelated to perplexity being high on gpt-oss-20b." + }, + { + "left": "issue:44291", + "right": "issue:45042", + "accept": false, + "reason": "Different problems entirely: init_empty_weights argument handling vs image processor torchvision dependency requirements." + }, + { + "left": "issue:43454", + "right": "issue:43577", + "accept": false, + "reason": "Different model-loading bugs: untied lm_head weights in AyaVision versus dtype propagation in Blip2." + }, + { + "left": "issue:42915", + "right": "issue:44821", + "accept": false, + "reason": "No underlying bug match: Qwen3 MoE FP8 config failure is unrelated to loading an AutoImageProcessor from URL." + }, + { + "left": "issue:29159", + "right": "issue:34567", + "accept": false, + "reason": "Tokenizer slow/fast inconsistency is unrelated to TrainerState token-count updating." + }, + { + "left": "issue:44568", + "right": "issue:45356", + "accept": false, + "reason": "Both are tokenizer regressions, but they affect different models and behaviors; no evidence they share one fix." + }, + { + "left": "issue:43927", + "right": "issue:44912", + "accept": false, + "reason": "DiaConfig save/load token IDs and MXFP4 quantization fallback are unrelated load-path bugs." + }, + { + "left": "issue:43525", + "right": "issue:46291", + "accept": false, + "reason": "A missing pad_token_id on Llama4Config is unrelated to the kernels v0.15 import failure in transformers.pipelines." + }, + { + "left": "issue:44987", + "right": "issue:45538", + "accept": false, + "reason": "Different tokenizer issues: loading physical-intelligence/fast under v5.1.0 versus CLIPTokenizer's absurd max length." + }, + { + "left": "issue:43697", + "right": "issue:45357", + "accept": false, + "reason": "Different regressions: RTDetrV2 output changes in v5 versus Qwen3.5 save_pretrained key mapping." + }, + { + "left": "issue:43404", + "right": "issue:45127", + "accept": false, + "reason": "Both involve weight tying in a broad sense, but they are different models and different bugs: tie failure vs LoRA merge collapse." + }, + { + "left": "issue:28282", + "right": "issue:34689", + "accept": false, + "reason": "ImportError from missing PyTorch is unrelated to a model-loading regression in Transformers 4.46.2." + }, + { + "left": "issue:42915", + "right": "issue:45538", + "accept": false, + "reason": "No shared underlying issue: Qwen3 FP8 config failure versus CLIPTokenizer max-length behavior." + }, + { + "left": "issue:43425", + "right": "issue:46291", + "accept": false, + "reason": "Torch version incompatibility is not the same as the kernels v0.15 package requirement failure." + }, + { + "left": "issue:45137", + "right": "issue:45405", + "accept": false, + "reason": "DeepSpeed ZeRO3 deque error is unrelated to the PEFT version bump/release issue." + }, + { + "left": "issue:33357", + "right": "issue:34567", + "accept": false, + "reason": "MacOS bus error with CLIP is unrelated to TrainerState token-count not updating." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 21, + "review_unit_index": 21, + "review_unit_count": 56, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-473", + "nodes": [ + "issue:24643", + "issue:27957", + "issue:28282", + "issue:29127", + "issue:29159", + "issue:34567", + "issue:36331", + "issue:38175", + "issue:41720", + "issue:42175", + "issue:42915", + "issue:43257", + "issue:43317", + "issue:43452", + "issue:43493", + "issue:43540", + "issue:43577", + "issue:43582", + "issue:43742", + "issue:43749", + "issue:43756", + "issue:43824", + "issue:43994", + "issue:44062", + "issue:44291", + "issue:44479", + "issue:44488", + "issue:44560", + "issue:44568", + "issue:44589", + "issue:44737", + "issue:44792", + "issue:44821", + "issue:44849", + "issue:44898", + "issue:44964", + "issue:44987", + "issue:44991", + "issue:45072", + "issue:45081", + "issue:45381", + "issue:45488", + "issue:45538", + "issue:45701", + "issue:45907", + "issue:46399", + "issue:46463", + "issue:46512" + ], + "soft_pairs": [ + "issue:42175|issue:43824", + "issue:43577|issue:46512", + "issue:43452|issue:43582", + "issue:28282|issue:29127", + "issue:43742|issue:44291", + "issue:44062|issue:44991", + "issue:43317|issue:45538", + "issue:44849|issue:45701", + "issue:41720|issue:43540", + "issue:42915|issue:43749", + "issue:43540|issue:44560", + "issue:44479|issue:45381", + "issue:43540|issue:44479", + "issue:24643|issue:29159", + "issue:44568|issue:45488", + "issue:38175|issue:44898", + "issue:44987|issue:45907", + "issue:27957|issue:28282", + "issue:42175|issue:44488", + "issue:43493|issue:43994", + "issue:43257|issue:43756", + "issue:44291|issue:44821", + "issue:34567|issue:36331", + "issue:44792|issue:44964", + "issue:44737|issue:46463", + "issue:44062|issue:45081", + "issue:43577|issue:45072", + "issue:43756|issue:46399", + "issue:43742|issue:44589" + ], + "prepared_review_unit_hash": "486907808cbbc0762b0b718fc7b3b3fe3e9cbcb34e251cd78d2606654bdc8f16", + "component_budget": { + "node_count": 473, + "item_count": 473, + "soft_pair_count": 1738, + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 29, + "serialized_chars": 36522, + "estimated_input_tokens": 9131, + "estimated_eval_tokens": 18518 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Mostly unrelated transformer issues, with two clear duplicate-like pairs: one around Qwen video-input regressions and one around SigLIP2 discrepancy. The rest are too different to merge.", + "confidence": 0.91, + "canonical_issue_reason": "issue:45488 is the strongest standalone report in this set: high discussion, inbound references, and a clear, user-facing regression in core tokenizer behavior affecting a widely used model family.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:45488 has the best mix of activity, impact, and diagnostic value, making it the best anchor issue for triage even though the cluster is not a single duplicate group.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:42175", + "right": "issue:43824", + "accept": false, + "reason": "Different problems: backend extras packaging vs a missing model class import." + }, + { + "left": "issue:43577", + "right": "issue:46512", + "accept": false, + "reason": "Both involve dtype handling, but one is a BLIP2 loading bug and the other is stale config metadata from from_config; not the same code-path bug." + }, + { + "left": "issue:43452", + "right": "issue:43582", + "accept": false, + "reason": "Unrelated: GGUF loading breakage vs Apple Silicon allocator warmup TypeError." + }, + { + "left": "issue:28282", + "right": "issue:29127", + "accept": false, + "reason": "Completely different: missing PyTorch import vs LayoutLMv3 box-validation error messaging." + }, + { + "left": "issue:43742", + "right": "issue:44291", + "accept": false, + "reason": "Different loader failures with different causes: model config/key loading vs init_empty_weights argument handling." + }, + { + "left": "issue:44062", + "right": "issue:44991", + "accept": false, + "reason": "Both are tokenizer-load failures, but they point to different regressions and different model-specific paths." + }, + { + "left": "issue:43317", + "right": "issue:45538", + "accept": false, + "reason": "Device-map/offload loading bug vs CLIP tokenizer max_length misconfiguration." + }, + { + "left": "issue:44849", + "right": "issue:45701", + "accept": false, + "reason": "Different issues: Qwen3.5 hidden-state handling vs a broader tokenization-change complaint." + }, + { + "left": "issue:41720", + "right": "issue:43540", + "accept": false, + "reason": "Both mention Qwen models, but one is auto device mapping cuda assert and the other is video input processing failure." + }, + { + "left": "issue:42915", + "right": "issue:43749", + "accept": false, + "reason": "Distinct training/runtime paths: Qwen3 MoE FP8 config vs FSDP CPU RAM efficient loading." + }, + { + "left": "issue:43540", + "right": "issue:44560", + "accept": false, + "reason": "Both are video-input failures, but they affect different Qwen model families and appear to be separate regressions." + }, + { + "left": "issue:44479", + "right": "issue:45381", + "accept": true, + "reason": "These describe the same Qwen/VL video-input regression; 45381 is a specific manifestation of the broader regression described in 44479." + }, + { + "left": "issue:24643", + "right": "issue:29159", + "accept": false, + "reason": "Training DeepSpeed weight-shape error is unrelated to tokenizer inconsistency." + }, + { + "left": "issue:44568", + "right": "issue:45488", + "accept": false, + "reason": "Both are tokenizer regressions, but they affect different tokenizers and different behavior; not one concrete bug." + }, + { + "left": "issue:38175", + "right": "issue:44898", + "accept": false, + "reason": "Different vision-model failures with different symptoms and model families." + }, + { + "left": "issue:44987", + "right": "issue:45907", + "accept": false, + "reason": "One is a model-loading failure, the other is a repo-files API forwarding bug." + }, + { + "left": "issue:27957", + "right": "issue:28282", + "accept": false, + "reason": "Unrelated: Flash Attention 2 on XLM-Roberta vs missing PyTorch import." + }, + { + "left": "issue:42175", + "right": "issue:44488", + "accept": false, + "reason": "Packaging/backend dependency issue vs a specific model-loading problem." + }, + { + "left": "issue:43493", + "right": "issue:43994", + "accept": true, + "reason": "Same underlying SigLIP2 implementation discrepancy: one reports mismatch with JAX, the other shows incorrect outputs in HF AutoModel/pipeline." + }, + { + "left": "issue:43257", + "right": "issue:43756", + "accept": false, + "reason": "Different model/config bugs: Qwen3 MoE weight conversion vs SmolLM3 RoPE layer drop." + }, + { + "left": "issue:44291", + "right": "issue:44821", + "accept": false, + "reason": "Init-empty-weights argument handling is unrelated to loading AutoImageProcessor from URL." + }, + { + "left": "issue:34567", + "right": "issue:36331", + "accept": false, + "reason": "TrainerState token counting and CustomTrainer compute_loss signature are separate trainer API issues." + }, + { + "left": "issue:44792", + "right": "issue:44964", + "accept": false, + "reason": "Both involve multimodal models, but the failures and code paths are different." + }, + { + "left": "issue:44737", + "right": "issue:46463", + "accept": false, + "reason": "XLNet CPU device placement bug is unrelated to T5 decoder-input validation messaging." + }, + { + "left": "issue:44062", + "right": "issue:45081", + "accept": false, + "reason": "Different tokenizer-regression surfaces and different failing code paths." + }, + { + "left": "issue:43577", + "right": "issue:45072", + "accept": false, + "reason": "Different dtype bugs in different model wrappers; not the same concrete defect." + }, + { + "left": "issue:43756", + "right": "issue:46399", + "accept": false, + "reason": "SmolLM3 RoPE-layer issue is unrelated to Qwen3.5 rotary-pos-embedding registration failure." + }, + { + "left": "issue:43742", + "right": "issue:44589", + "accept": false, + "reason": "Different loader/runtime failures with no clear shared root cause." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays mostly conservative. The two accepted pairs are reasonably supported by the titles and context, and the rejected pairs are generally distinct enough to avoid over-merging. The chosen anchor issue (45488) is defensible from the activity/reference signals. " + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 21, + "review_unit_index": 22, + "review_unit_count": 56, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-473", + "nodes": [ + "issue:24643", + "issue:28282", + "issue:29127", + "issue:32090", + "issue:35141", + "issue:41762", + "issue:43404", + "issue:43452", + "issue:43526", + "issue:43531", + "issue:43643", + "issue:43688", + "issue:43697", + "issue:43720", + "issue:43749", + "issue:43756", + "issue:43906", + "issue:44466", + "issue:44488", + "issue:44568", + "issue:44625", + "issue:44737", + "issue:44821", + "issue:44843", + "issue:44877", + "issue:44898", + "issue:44913", + "issue:44936", + "issue:45005", + "issue:45042", + "issue:45081", + "issue:45200", + "issue:45216", + "issue:45325", + "issue:45357", + "issue:45375", + "issue:45381", + "issue:45405", + "issue:45479", + "issue:45588", + "issue:45600", + "issue:45823", + "issue:45920", + "issue:45987", + "issue:46121", + "issue:46463", + "issue:46468", + "issue:46491" + ], + "soft_pairs": [ + "issue:44568|issue:46491", + "issue:45042|issue:45081", + "issue:43756|issue:45325", + "issue:43906|issue:46468", + "issue:45823|issue:46463", + "issue:44737|issue:45588", + "issue:45405|issue:45600", + "issue:43531|issue:44821", + "issue:43404|issue:45987", + "issue:43452|issue:44488", + "issue:43526|issue:45479", + "issue:45357|issue:45920", + "issue:43906|issue:45381", + "issue:44625|issue:44877", + "issue:44913|issue:46121", + "issue:43720|issue:44898", + "issue:43697|issue:45216", + "issue:24643|issue:29127", + "issue:28282|issue:32090", + "issue:43643|issue:45375", + "issue:45200|issue:46463", + "issue:28282|issue:35141", + "issue:41762|issue:44936", + "issue:44466|issue:45005", + "issue:43688|issue:43749", + "issue:43452|issue:44843" + ], + "prepared_review_unit_hash": "195cf426f9986be599b98f0653aaacf0c5a651125b7e611a0a5d9828af07df3d", + "component_budget": { + "node_count": 473, + "item_count": 473, + "soft_pair_count": 1738, + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 26, + "serialized_chars": 36107, + "estimated_input_tokens": 9027, + "estimated_eval_tokens": 18310 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Most pairs are clearly unrelated despite superficial keyword overlap. The only strong duplicate-like pair is the tied-weights v5 serialization reports (#44466 and #45005).", + "confidence": 0.81, + "canonical_issue_reason": "#45005 is the better canonical issue because it is broader and more representative of the tied-weights regression, while #44466 reads like a specific manifestation of the same problem (device-dependent lm_head.weight serialization).", + "canonical_pr_reason": null, + "best_issue_reason": "#45005 is the best issue to keep as the primary tracker for the tied-weights bug: it describes the underlying regression at a higher level and can absorb narrower repros like #44466.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44568", + "right": "issue:46491", + "accept": false, + "reason": "Different models and different failure modes: tokenizer special-token insertion vs CodeLlama encode/decode spacing regression." + }, + { + "left": "issue:45042", + "right": "issue:45081", + "accept": false, + "reason": "Unrelated bugs: PIL/torchvision dependency handling vs Mistral regex/tokenizer crash." + }, + { + "left": "issue:43756", + "right": "issue:45325", + "accept": false, + "reason": "Both mention RoPE, but they affect different models and different logic paths; not the same bug." + }, + { + "left": "issue:43906", + "right": "issue:46468", + "accept": false, + "reason": "An isolated reproduction of one issue is not evidence of the same bug as the Nemotron example failure; the titles indicate different problems." + }, + { + "left": "issue:45823", + "right": "issue:46463", + "accept": false, + "reason": "Gemma4 device-map/PLE mismatch is unrelated to T5 decoder input validation." + }, + { + "left": "issue:44737", + "right": "issue:45588", + "accept": false, + "reason": "XLNet CPU arange bug and flash-attention s_aux=None crash are different code paths and symptoms." + }, + { + "left": "issue:45405", + "right": "issue:45600", + "accept": false, + "reason": "PEFT version bump is packaging metadata; auto_mappings.py CI breakage is a removed-config reference bug." + }, + { + "left": "issue:43531", + "right": "issue:44821", + "accept": false, + "reason": "Qwen3-MoE sliding_window behavior and AutoImageProcessor URL loading are unrelated." + }, + { + "left": "issue:43404", + "right": "issue:45987", + "accept": false, + "reason": "lm_head tying in Mistral3 and StaticCache.get_seq_length() returning a tensor are separate defects." + }, + { + "left": "issue:43452", + "right": "issue:44488", + "accept": false, + "reason": "Both concern loading, but one is gguf_file path handling and the other is a model/tokenizer load failure for a specific repo; no shared concrete bug." + }, + { + "left": "issue:43526", + "right": "issue:45479", + "accept": false, + "reason": "BEiT reduce_labels returning one label is unrelated to zero loss with num_labels=1." + }, + { + "left": "issue:45357", + "right": "issue:45920", + "accept": false, + "reason": "Qwen3.5 save_pretrained key serialization and AutoTokenizer wrong token IDs are different regressions." + }, + { + "left": "issue:44625", + "right": "issue:44877", + "accept": false, + "reason": "Both involve config loading, but one is num_labels propagation and the other is strict config rejection for granite_speech; not the same bug." + }, + { + "left": "issue:44913", + "right": "issue:46121", + "accept": false, + "reason": "GPTNeoX rotary_pct resetting on reload and rope-validation TypeError for JSON-loaded lists are distinct configuration issues." + }, + { + "left": "issue:43720", + "right": "issue:44898", + "accept": false, + "reason": "BitNet packed-weight unpacking during accelerate loading is unrelated to Perceiver interpolation with non-default resolution." + }, + { + "left": "issue:43697", + "right": "issue:45216", + "accept": false, + "reason": "RTDetrV2 output regression and Qwen3.5 save_pretrained checkpoint corruption are different problems." + }, + { + "left": "issue:24643", + "right": "issue:29127", + "accept": false, + "reason": "DeepSpeed weight shape runtime error and LayoutLMv3 error-message clarity are unrelated." + }, + { + "left": "issue:28282", + "right": "issue:32090", + "accept": false, + "reason": "PyTorch-missing ImportError and Trainer _gpu_broadcast_one NoneType crash are unrelated." + }, + { + "left": "issue:43643", + "right": "issue:45375", + "accept": false, + "reason": "Missing fields with trust_remote_code and strict-dropping deepstack_visual_indexes are both config-related, but they are not the same concrete bug." + }, + { + "left": "issue:45200", + "right": "issue:46463", + "accept": false, + "reason": "Gemma 4 multimodal token defaults and T5 decoder input validation are unrelated." + }, + { + "left": "issue:41762", + "right": "issue:44936", + "accept": false, + "reason": "Gemma3 ZeRO-3 loading failure and trainer.evaluate() after trainer.train() are separate bugs." + }, + { + "left": "issue:44466", + "right": "issue:45005", + "accept": true, + "reason": "Both describe the same underlying tied-weights serialization regression in v5: inconsistent lm_head.weight/save-pretrained behavior for translation models, with #44466 being a specific manifestation." + }, + { + "left": "issue:43688", + "right": "issue:43749", + "accept": false, + "reason": "Aux-loss normalization in MoE models and broken FSDP CPU RAM efficient loading are unrelated." + }, + { + "left": "issue:43452", + "right": "issue:44843", + "accept": false, + "reason": "gguf_file loading failure and the offline mistral_regex/model_info bug are different tokenizer/loading issues." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The primary/canonical issue choice is grounded in the packet: #45005 is broader and #44466 is a narrower tied-weights serialization regression. The soft-edge rejections are conservative and stay within what the titles support, with no obvious overstatement of evidence." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 21, + "review_unit_index": 23, + "review_unit_count": 56, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-473", + "nodes": [ + "issue:29159", + "issue:33357", + "issue:38175", + "issue:42898", + "issue:43295", + "issue:43452", + "issue:43531", + "issue:43550", + "issue:43577", + "issue:43646", + "issue:43653", + "issue:43688", + "issue:43697", + "issue:43742", + "issue:43749", + "issue:43756", + "issue:43819", + "issue:43927", + "issue:44361", + "issue:44442", + "issue:44464", + "issue:44479", + "issue:44484", + "issue:44493", + "issue:44514", + "issue:44568", + "issue:44625", + "issue:44704", + "issue:44743", + "issue:44811", + "issue:44843", + "issue:44936", + "issue:44991", + "issue:45356", + "issue:45372", + "issue:45381", + "issue:45440", + "issue:45491", + "issue:45542", + "issue:45701", + "issue:45704", + "issue:45823", + "issue:46002", + "issue:46050", + "issue:46170", + "issue:46178", + "issue:46402", + "issue:46491" + ], + "soft_pairs": [ + "issue:43819|issue:44811", + "issue:44493|issue:46050", + "issue:44361|issue:44442", + "issue:43653|issue:44361", + "issue:38175|issue:43295", + "issue:45372|issue:46170", + "issue:44936|issue:45823", + "issue:42898|issue:46491", + "issue:45542|issue:46178", + "issue:43452|issue:43577", + "issue:43295|issue:45356", + "issue:44514|issue:45381", + "issue:43531|issue:43742", + "issue:43653|issue:44568", + "issue:43577|issue:43697", + "issue:43646|issue:43749", + "issue:44843|issue:44991", + "issue:29159|issue:33357", + "issue:43688|issue:45704", + "issue:43756|issue:46402", + "issue:44464|issue:44743", + "issue:43653|issue:43927", + "issue:44484|issue:44704", + "issue:43688|issue:45440", + "issue:43688|issue:45701", + "issue:44479|issue:44625", + "issue:43550|issue:44361", + "issue:45491|issue:46002" + ], + "prepared_review_unit_hash": "6fbdc79224d1e33431e1146c7031e1fc79e6f3b0bb6abfb8a3de1b6c1b49c28a", + "component_budget": { + "node_count": 473, + "item_count": 473, + "soft_pair_count": 1738, + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 28, + "serialized_chars": 36190, + "estimated_input_tokens": 9048, + "estimated_eval_tokens": 18352 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This set is mostly a grab bag of distinct tokenizer, model-loading, and generation regressions. I don\u2019t see any soft-edge pair that clearly describes the same underlying bug/change, so I would keep them separate.", + "confidence": 0.89, + "canonical_issue_reason": "issue:29159 is the broadest and most central representative of the tokenizer-behavior theme: it explicitly covers inconsistency between slow and fast tokenizer paths, which is the closest umbrella problem among the set.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:29159 best represents the cluster because it is generic, cross-model, and directly about tokenizer parity, unlike the more model-specific regressions.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43819", + "right": "issue:44811", + "accept": false, + "reason": "Different subsystems: DAC latent decoding vs Whisper batch_decode skip_special_tokens behavior." + }, + { + "left": "issue:44493", + "right": "issue:46050", + "accept": false, + "reason": "Unrelated bugs: position-id key warnings in configs vs quantized model save/load corruption." + }, + { + "left": "issue:44361", + "right": "issue:44442", + "accept": false, + "reason": "Different tokenizer classes and failures: MLuke task AttributeError vs FastSpeech2ConformerTokenizer loading." + }, + { + "left": "issue:43653", + "right": "issue:44361", + "accept": false, + "reason": "Both are tokenizer bugs, but one is BigBird special-token registration/decode and the other is MLuke task AttributeError; not the same concrete bug." + }, + { + "left": "issue:38175", + "right": "issue:43295", + "accept": false, + "reason": "Different problems: zero probabilities in a model vs processor/tokenizer regression for custom image inputs." + }, + { + "left": "issue:45372", + "right": "issue:46170", + "accept": false, + "reason": "Both are import/load failures, but the causes are different: mistral_common ReasoningEffort import vs Windows ROCm torch.distributed dependency issues." + }, + { + "left": "issue:44936", + "right": "issue:45823", + "accept": false, + "reason": "Training loop evaluate-after-train failure is unrelated to Gemma4 device-map mismatches in forward." + }, + { + "left": "issue:42898", + "right": "issue:46491", + "accept": false, + "reason": "Both involve tokenization behavior, but 42898 is a general v5 cleanup-space policy change while 46491 is a CodeLlama round-trip spacing regression." + }, + { + "left": "issue:45542", + "right": "issue:46178", + "accept": false, + "reason": "Different backend-registration issues: missing TensorFlow backend handling vs tensorflow_text backend mapping." + }, + { + "left": "issue:43452", + "right": "issue:43577", + "accept": false, + "reason": "Unrelated: gguf loading breaks from_pretrained paths, while the other is dtype propagation for Blip2 submodules." + }, + { + "left": "issue:43295", + "right": "issue:45356", + "accept": false, + "reason": "Processor/image-tokenizer regression is not the same bug as Kimi-K2.5 codec handling and warning behavior." + }, + { + "left": "issue:44514", + "right": "issue:45381", + "accept": false, + "reason": "Both are Qwen VL issues, but one is batched chat-template padding and the other is video vision_position_ids; different code paths." + }, + { + "left": "issue:43531", + "right": "issue:43742", + "accept": false, + "reason": "Different domains: Qwen3-MoE sliding_window logic vs MobileLLM-125M loading key errors." + }, + { + "left": "issue:43653", + "right": "issue:44568", + "accept": false, + "reason": "Both concern special tokens, but BigBird mask-token registration and mDeBERTa BOS/EOS insertion are distinct tokenizer bugs." + }, + { + "left": "issue:43577", + "right": "issue:43697", + "accept": false, + "reason": "Different failures: Blip2 dtype loading vs RTDetrV2 output divergence in v5." + }, + { + "left": "issue:43646", + "right": "issue:43749", + "accept": false, + "reason": "Custom model initialization breakage is unrelated to FSDP CPU/RAM-efficient loading." + }, + { + "left": "issue:44843", + "right": "issue:44991", + "accept": false, + "reason": "Both are tokenizer loading issues, but one is offline-mode model_info misuse and the other is a model-specific tokenizer load failure." + }, + { + "left": "issue:29159", + "right": "issue:33357", + "accept": false, + "reason": "Generic tokenizer parity issue vs Mac bus error on a CLIP community model; not the same bug." + }, + { + "left": "issue:43688", + "right": "issue:45704", + "accept": false, + "reason": "Loss normalization in MoE models is unrelated to a T5/apex RMSNorm memory leak." + }, + { + "left": "issue:43756", + "right": "issue:46402", + "accept": false, + "reason": "Different model-specific problems: RoPE layer dropping vs missing Sam3 text encoder weight loading." + }, + { + "left": "issue:44464", + "right": "issue:44743", + "accept": false, + "reason": "Both touch generation state, but compiled-forward chunked generation inconsistency is not the same as Qwen3.5 recurrent-state reset logic." + }, + { + "left": "issue:43653", + "right": "issue:43927", + "accept": false, + "reason": "Different special-token issues: BigBird mask-token registration vs DiaConfig custom token IDs lost on save/load." + }, + { + "left": "issue:44484", + "right": "issue:44704", + "accept": false, + "reason": "save_pretrained max_shard_size question is unrelated to AutoProcessor kwargs not reaching cached_file." + }, + { + "left": "issue:43688", + "right": "issue:45440", + "accept": false, + "reason": "Aux-loss normalization in MoE models is unrelated to DeepseekV3MoE implementation divergence." + }, + { + "left": "issue:43688", + "right": "issue:45701", + "accept": false, + "reason": "Generic MoE loss bug and broad tokenizer-version regression are not the same underlying issue." + }, + { + "left": "issue:44479", + "right": "issue:44625", + "accept": false, + "reason": "Video-input regression in Qwen variants is different from Qwen3.5 num_labels propagation." + }, + { + "left": "issue:43550", + "right": "issue:44361", + "accept": false, + "reason": "torch.compile/SDPA failure in Bamba is unrelated to MLukeTokenizer task AttributeError." + }, + { + "left": "issue:45491", + "right": "issue:46002", + "accept": false, + "reason": "Both mention NaNs, but the triggers differ: sliding-window all-padding embeddings vs classification-head init/checkout." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 21, + "review_unit_index": 24, + "review_unit_count": 56, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-473", + "nodes": [ + "issue:28218", + "issue:32090", + "issue:33357", + "issue:33666", + "issue:36010", + "issue:36331", + "issue:39692", + "issue:41720", + "issue:43262", + "issue:43278", + "issue:43295", + "issue:43452", + "issue:43493", + "issue:43504", + "issue:43525", + "issue:43550", + "issue:43572", + "issue:43577", + "issue:43606", + "issue:43643", + "issue:43688", + "issue:43720", + "issue:43742", + "issue:43825", + "issue:43927", + "issue:43994", + "issue:44361", + "issue:44451", + "issue:44464", + "issue:44466", + "issue:44514", + "issue:44554", + "issue:44561", + "issue:44610", + "issue:44625", + "issue:44734", + "issue:44821", + "issue:44964", + "issue:44977", + "issue:45290", + "issue:45405", + "issue:45584", + "issue:45823", + "issue:46050", + "issue:46192", + "issue:46209", + "issue:46365", + "issue:46612" + ], + "soft_pairs": [ + "issue:44734|issue:46612", + "issue:33666|issue:36010", + "issue:32090|issue:33666", + "issue:44464|issue:44977", + "issue:41720|issue:44464", + "issue:43742|issue:44821", + "issue:44466|issue:46050", + "issue:43504|issue:43720", + "issue:43504|issue:43550", + "issue:43643|issue:44561", + "issue:43577|issue:44554", + "issue:43504|issue:44610", + "issue:44514|issue:45290", + "issue:43720|issue:43927", + "issue:43606|issue:43927", + "issue:33666|issue:36331", + "issue:43262|issue:43295", + "issue:44554|issue:46365", + "issue:43278|issue:43688", + "issue:43720|issue:44361", + "issue:44977|issue:46209", + "issue:39692|issue:43493", + "issue:44964|issue:45405", + "issue:43493|issue:43825", + "issue:43452|issue:44451", + "issue:43525|issue:43572", + "issue:44625|issue:45584", + "issue:45823|issue:46050", + "issue:43994|issue:46192", + "issue:28218|issue:33357" + ], + "prepared_review_unit_hash": "3fdcd97b1e243ebc1a6e2f89b0624b8799670155c59d57c6d2e24beb5f5c1072", + "component_budget": { + "node_count": 473, + "item_count": 473, + "soft_pair_count": 1738, + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 30, + "serialized_chars": 36779, + "estimated_input_tokens": 9195, + "estimated_eval_tokens": 18646 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is highly heterogeneous: the soft pairs mostly share broad Transformer/model-loading/generation themes but not the same underlying bug. I would reject all soft edges. If forced to pick an issue anchor, issue 43504 is the most central by linkage density, but it is still not a strong true-duplicate hub.", + "confidence": 0.92, + "canonical_issue_reason": "issue:43504 has the densest soft-similarity neighborhood in this set (linked to 43720, 43550, and 44610) and is a concrete load-time regression, making it the best anchor if one must choose.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:43504 is the most representative issue in the cluster, but only weakly so; the cluster overall is too mixed to have a strong canonical issue.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44734", + "right": "issue:46612", + "accept": false, + "reason": "Both are generation bugs, but they hit different code paths: KV-cache continuation indexing vs beam-search cache reorder for several model families." + }, + { + "left": "issue:33666", + "right": "issue:36010", + "accept": false, + "reason": "Multi-GPU training trouble is unrelated to a GenerationMixin import failure; different symptoms and subsystems." + }, + { + "left": "issue:32090", + "right": "issue:33666", + "accept": false, + "reason": "Trainer/_gpu_broadcast_one TypeError is a distributed training error, not the same as Qwen2-VL multi-GPU training support/failure." + }, + { + "left": "issue:44464", + "right": "issue:44977", + "accept": false, + "reason": "Both concern generation quality, but one is chunked generation with compiled forward and the other is Qwen3.5 flash-attention behavior." + }, + { + "left": "issue:41720", + "right": "issue:44464", + "accept": false, + "reason": "A device-mapping cudaErrorAssert on A800 is a different failure mode from inconsistent outputs in compiled chunked generation." + }, + { + "left": "issue:43742", + "right": "issue:44821", + "accept": false, + "reason": "Loading a causal LM for MobileLLM is unrelated to loading AutoImageProcessor from URL; different object types and failure modes." + }, + { + "left": "issue:44466", + "right": "issue:46050", + "accept": false, + "reason": "Both touch save/load semantics, but one is tied-weight serialization by device and the other is sinq quantization persistence." + }, + { + "left": "issue:43504", + "right": "issue:43720", + "accept": false, + "reason": "Beit legacy-field loading and BitNet packed-weight unpacking are distinct model-specific loading bugs." + }, + { + "left": "issue:43504", + "right": "issue:43550", + "accept": false, + "reason": "A Beit pretrained preset load issue is unrelated to Bamba-9B-v2 torch.compile/SDPA failures." + }, + { + "left": "issue:43643", + "right": "issue:44561", + "accept": false, + "reason": "AutoConfig returning missing fields under trust_remote_code and removal of is_torch_fx_available are different compatibility regressions." + }, + { + "left": "issue:43577", + "right": "issue:44554", + "accept": false, + "reason": "Blip2 dtype casting inconsistency and an MPS attention correctness bug are unrelated." + }, + { + "left": "issue:43504", + "right": "issue:44610", + "accept": false, + "reason": "Legacy-field loading for Beit is not the same as OmDet-Turbo's processor/model input-size mismatch." + }, + { + "left": "issue:44514", + "right": "issue:45290", + "accept": false, + "reason": "Both involve apply_chat_template, but one crashes on batched padding=False and the other on assistant tool-call messages with no content." + }, + { + "left": "issue:43720", + "right": "issue:43927", + "accept": false, + "reason": "Packed-weight accelerate loading and DiaConfig custom token IDs after save/load are different persistence bugs." + }, + { + "left": "issue:43606", + "right": "issue:43927", + "accept": false, + "reason": "CPU offload device mismatch for bark-small is unrelated to DiaConfig losing custom token IDs." + }, + { + "left": "issue:33666", + "right": "issue:36331", + "accept": false, + "reason": "Qwen2-VL multi-GPU training and CustomTrainer.compute_loss signature mismatch are different trainer/distributed issues." + }, + { + "left": "issue:43262", + "right": "issue:43295", + "accept": false, + "reason": "Audio chat-template sample-rate defaults and processor.tokenizer regression are distinct processor API bugs." + }, + { + "left": "issue:44554", + "right": "issue:46365", + "accept": false, + "reason": "An MPS attention correctness issue is unrelated to importing Trainer failing when datasets is installed." + }, + { + "left": "issue:43278", + "right": "issue:43688", + "accept": false, + "reason": "Model dtype drift in eval and auxiliary-loss normalization are different training-time bugs." + }, + { + "left": "issue:43720", + "right": "issue:44361", + "accept": false, + "reason": "BitNet accelerate loading and MLukeTokenizer task AttributeError are unrelated model/tokenizer failures." + }, + { + "left": "issue:44977", + "right": "issue:46209", + "accept": false, + "reason": "Qwen3.5 flash-attention generation problems and FP8 inference failure on B200 are different hardware/kernel issues." + }, + { + "left": "issue:39692", + "right": "issue:43493", + "accept": false, + "reason": "SigLIP2 docs example errors and HF-vs-JAX implementation discrepancy are related to the same model family but not the same underlying bug." + }, + { + "left": "issue:44964", + "right": "issue:45405", + "accept": false, + "reason": "A model-loading regression for Phi-4 multimodal is not the same as a too-new MIN_PEFT_VERSION requirement." + }, + { + "left": "issue:43493", + "right": "issue:43825", + "accept": false, + "reason": "SigLIP2 implementation mismatch and an incorrect pipeline error message about translation support are unrelated." + }, + { + "left": "issue:43452", + "right": "issue:44451", + "accept": false, + "reason": "gguf_file breaking from_pretrained paths and ScandiBERT load failure are different loading issues." + }, + { + "left": "issue:43525", + "right": "issue:43572", + "accept": false, + "reason": "Both mention missing pad-token-related fields, but they affect different configs, fields, and models." + }, + { + "left": "issue:44625", + "right": "issue:45584", + "accept": false, + "reason": "Qwen3.5 num_labels propagation and Whisper empty-transcription failure are unrelated model behaviors." + }, + { + "left": "issue:45823", + "right": "issue:46050", + "accept": false, + "reason": "Gemma4 PLE device mismatch with auto device map is not the same as sinq quantization save/load breakage." + }, + { + "left": "issue:43994", + "right": "issue:46192", + "accept": false, + "reason": "SigLIP2 nonsense outputs and return_tensors being ignored are different API bugs." + }, + { + "left": "issue:28218", + "right": "issue:33357", + "accept": false, + "reason": "Tokenizer extra-spacing after added token and MacOS bus error with a CLIP model 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": 21, + "review_unit_index": 25, + "review_unit_count": 56, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-473", + "nodes": [ + "issue:33357", + "issue:33666", + "issue:39692", + "issue:43257", + "issue:43278", + "issue:43295", + "issue:43299", + "issue:43334", + "issue:43493", + "issue:43504", + "issue:43540", + "issue:43572", + "issue:43575", + "issue:43606", + "issue:43646", + "issue:43653", + "issue:43701", + "issue:43824", + "issue:43844", + "issue:43927", + "issue:43994", + "issue:44112", + "issue:44291", + "issue:44361", + "issue:44464", + "issue:44466", + "issue:44488", + "issue:44493", + "issue:44568", + "issue:44589", + "issue:44792", + "issue:44843", + "issue:44849", + "issue:44964", + "issue:44977", + "issue:45200", + "issue:45229", + "issue:45357", + "issue:45405", + "issue:45520", + "issue:45584", + "issue:45759", + "issue:45812", + "issue:45823", + "issue:46209", + "issue:46285", + "issue:46424", + "issue:46489" + ], + "soft_pairs": [ + "issue:43278|issue:43844", + "issue:43493|issue:43844", + "issue:39692|issue:43295", + "issue:44849|issue:46209", + "issue:33357|issue:33666", + "issue:44488|issue:44843", + "issue:43646|issue:44466", + "issue:44568|issue:46489", + "issue:43278|issue:43701", + "issue:43927|issue:44361", + "issue:44568|issue:45812", + "issue:43540|issue:44977", + "issue:44792|issue:45405", + "issue:43257|issue:43299", + "issue:43575|issue:45405", + "issue:44589|issue:45520", + "issue:43504|issue:43606", + "issue:44849|issue:45357", + "issue:43334|issue:43572", + "issue:43606|issue:44112", + "issue:43824|issue:44589", + "issue:43334|issue:45759", + "issue:44464|issue:46424", + "issue:44493|issue:46285", + "issue:43504|issue:43653", + "issue:45200|issue:45229", + "issue:43572|issue:44291", + "issue:44464|issue:45584", + "issue:45229|issue:45823", + "issue:43994|issue:44493", + "issue:43844|issue:44964" + ], + "prepared_review_unit_hash": "bcfb3589755502632e33adcf63cc6a7d4a70eff678f6636b6f0d08c7f1a80a3f", + "component_budget": { + "node_count": 473, + "item_count": 473, + "soft_pair_count": 1738, + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 31, + "serialized_chars": 37130, + "estimated_input_tokens": 9283, + "estimated_eval_tokens": 18822 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a loose collection of unrelated Transformers regressions, with many tokenizer/config/model-loading issues across different model families. None of the soft pairs look like the same underlying bug or change, so I rejected them all. If one issue has to represent the cluster, the broadest and most user-impactful is the Granite tokenizer regression.", + "confidence": 0.43, + "canonical_issue_reason": "issue:45812 is the best umbrella representative: it describes a broad, silent v4\u2192v5 tokenizer regression affecting all Granite models, which is more general and higher-impact than the more model-specific reports here.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:45812 has the clearest cluster-representative value because it is broad, regression-framed, and likely to matter to multiple downstream users; the rest are mostly narrower, unrelated edge cases.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43278", + "right": "issue:43844", + "accept": false, + "reason": "Different bugs: dtype change in evaluation vs. exploding gradients with HfDeepSpeedConfig/ZeRO-3." + }, + { + "left": "issue:43493", + "right": "issue:43844", + "accept": false, + "reason": "SigLIP2 implementation discrepancy is unrelated to the DeepSpeed gradient problem." + }, + { + "left": "issue:39692", + "right": "issue:43295", + "accept": false, + "reason": "Documentation/example errors for SigLIP2 are not the same as the v4.57.5 processor/tokenizer regression." + }, + { + "left": "issue:44849", + "right": "issue:46209", + "accept": false, + "reason": "Different model families and failure modes: Qwen3.5 hidden-states behavior vs FP8 inference on B200." + }, + { + "left": "issue:33357", + "right": "issue:33666", + "accept": false, + "reason": "CLIP bus error on macOS is unrelated to Qwen2-VL multi-GPU training." + }, + { + "left": "issue:44488", + "right": "issue:44843", + "accept": false, + "reason": "Both involve loading/tokenizer code, but one is a specific model load failure and the other is an offline-mode regression from unconditional hub calls." + }, + { + "left": "issue:43646", + "right": "issue:44466", + "accept": false, + "reason": "Custom model initialization breakage is not the same as device-dependent serialization of lm_head.weight." + }, + { + "left": "issue:44568", + "right": "issue:46489", + "accept": false, + "reason": "Both are tokenizer regressions, but they concern different models and different concrete failures." + }, + { + "left": "issue:43278", + "right": "issue:43701", + "accept": false, + "reason": "Embedding dtype mismatch is unrelated to resume_from_checkpoint key mismatch." + }, + { + "left": "issue:43927", + "right": "issue:44361", + "accept": false, + "reason": "DiaConfig save/load token ID loss is a different issue from MLukeTokenizer task-time AttributeError." + }, + { + "left": "issue:44568", + "right": "issue:45812", + "accept": false, + "reason": "Granite token ID regression and mdeberta add_special_tokens regression are separate tokenizer bugs." + }, + { + "left": "issue:43540", + "right": "issue:44977", + "accept": false, + "reason": "Video input processing error in Qwen3OmniMoe is not the same as Qwen3.5 flash-attention generation issues." + }, + { + "left": "issue:44792", + "right": "issue:45405", + "accept": false, + "reason": "A janus model test failure is unrelated to the PEFT minimum-version bump." + }, + { + "left": "issue:43257", + "right": "issue:43299", + "accept": false, + "reason": "Both are Qwen3 MOE-related, but one is accelerate/deepspeed weight conversion and the other is Qwen3VL loading; not the same bug." + }, + { + "left": "issue:43575", + "right": "issue:45405", + "accept": false, + "reason": "Tensor-parallel OOM is unrelated to a dependency version bump." + }, + { + "left": "issue:44589", + "right": "issue:45520", + "accept": false, + "reason": "Float8 storage lookup failure and flash_attn import KeyError are different platform/runtime issues." + }, + { + "left": "issue:43504", + "right": "issue:43606", + "accept": false, + "reason": "Legacy-field load failure for Beit is not the same as Bark CPU-offload device mismatch." + }, + { + "left": "issue:44849", + "right": "issue:45357", + "accept": false, + "reason": "Qwen3.5 hidden-states bug and incorrect visual encoder save_pretrained keys are distinct regression symptoms." + }, + { + "left": "issue:43334", + "right": "issue:43572", + "accept": false, + "reason": "Missing pad_token_id in Qwen3-VL text config is unrelated to StableLm pad_token_idx." + }, + { + "left": "issue:43606", + "right": "issue:44112", + "accept": false, + "reason": "A real CPU-offload device mismatch is not the same as a stale GraniteSpeech CI test." + }, + { + "left": "issue:43824", + "right": "issue:44589", + "accept": false, + "reason": "Qwen2.5-VL import error and Float8 storage lookup failure are unrelated." + }, + { + "left": "issue:43334", + "right": "issue:45759", + "accept": false, + "reason": "Both touch Qwen multimodal config loading, but the concrete failure modes differ: missing pad_token_id vs text_config unwrapping." + }, + { + "left": "issue:44464", + "right": "issue:46424", + "accept": false, + "reason": "Chunked generation inconsistency with compiled forward is not the same as max_cache_length handling in cache config." + }, + { + "left": "issue:44493", + "right": "issue:46285", + "accept": false, + "reason": "Unexpected position_id keys across many models is a different structural regression than CLIP text-model wrapper inconsistency." + }, + { + "left": "issue:43504", + "right": "issue:43653", + "accept": false, + "reason": "Beit legacy-field loading and BigBirdTokenizer special-token registration are unrelated." + }, + { + "left": "issue:45200", + "right": "issue:45229", + "accept": false, + "reason": "Text-only mm_token_type_ids defaulting is unrelated to Gemma4 multi-GPU CUDA OOM." + }, + { + "left": "issue:43572", + "right": "issue:44291", + "accept": false, + "reason": "StableLm pad_token_idx regression and init_empty_weights unexpected argument are different initialization problems." + }, + { + "left": "issue:44464", + "right": "issue:45584", + "accept": false, + "reason": "Compiled-forward chunked generation inconsistency is unrelated to Whisper empty-transcription failure after align_special_tokens." + }, + { + "left": "issue:45229", + "right": "issue:45823", + "accept": false, + "reason": "Both are Gemma4 issues, but one is inference OOM and the other is a device mismatch during forward." + }, + { + "left": "issue:43994", + "right": "issue:44493", + "accept": false, + "reason": "SigLIP2 nonsensical outputs and widespread position_id key warnings are not the same bug." + }, + { + "left": "issue:43844", + "right": "issue:44964", + "accept": false, + "reason": "DeepSpeed gradient anomaly is unrelated to Phi-4-multimodal model loading 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": 21, + "review_unit_index": 26, + "review_unit_count": 56, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-473", + "nodes": [ + "issue:15354", + "issue:28218", + "issue:29159", + "issue:29942", + "issue:30333", + "issue:32090", + "issue:34567", + "issue:37428", + "issue:38175", + "issue:39692", + "issue:42490", + "issue:43065", + "issue:43232", + "issue:43295", + "issue:43450", + "issue:43493", + "issue:43540", + "issue:43550", + "issue:43572", + "issue:43575", + "issue:43696", + "issue:43701", + "issue:43844", + "issue:43873", + "issue:43957", + "issue:44186", + "issue:44423", + "issue:44554", + "issue:44610", + "issue:44734", + "issue:44849", + "issue:44987", + "issue:45072", + "issue:45084", + "issue:45198", + "issue:45200", + "issue:45356", + "issue:45406", + "issue:45446", + "issue:45520", + "issue:45588", + "issue:45759", + "issue:45823", + "issue:46050", + "issue:46121", + "issue:46178", + "issue:46192", + "issue:46369" + ], + "soft_pairs": [ + "issue:43873|issue:46050", + "issue:45588|issue:46050", + "issue:29942|issue:30333", + "issue:42490|issue:43696", + "issue:29159|issue:32090", + "issue:43065|issue:46369", + "issue:45084|issue:45520", + "issue:28218|issue:34567", + "issue:44423|issue:44734", + "issue:44554|issue:45446", + "issue:15354|issue:38175", + "issue:45406|issue:46121", + "issue:43493|issue:43701", + "issue:45072|issue:45198", + "issue:45588|issue:45759", + "issue:43540|issue:45200", + "issue:43295|issue:43450", + "issue:37428|issue:43957", + "issue:44186|issue:46192", + "issue:45200|issue:45823", + "issue:45823|issue:46192", + "issue:43295|issue:44610", + "issue:43295|issue:43873", + "issue:45520|issue:46178", + "issue:43450|issue:43575", + "issue:43550|issue:43844", + "issue:43232|issue:44849", + "issue:43572|issue:44987", + "issue:44554|issue:45356", + "issue:39692|issue:46192" + ], + "prepared_review_unit_hash": "9146da04106aa5e94d1713322633bf0b6211a9a9965cab368a50d9014e45bcd0", + "component_budget": { + "node_count": 473, + "item_count": 473, + "soft_pair_count": 1738, + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 30, + "serialized_chars": 36779, + "estimated_input_tokens": 9195, + "estimated_eval_tokens": 18646 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a loose grab-bag of transformer bug reports spanning tokenizer regressions, multimodal processor issues, generation/serve crashes, flash-attention, quantization, and loading/import edge cases. The soft pairs mostly share only broad subsystem overlap, not the same underlying defect.", + "confidence": 0.31, + "canonical_issue_reason": "issue:43295 is the broadest user-facing regression here around processor/tokenizer behavior in a recent release, and it sits closest to the multimodal/tokenization theme that recurs across several nearby issues.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:43295 is the most representative anchor because it describes a concrete regression with clear user impact and a general processor/tokenizer path, making it the best single issue to summarize this otherwise heterogeneous set.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43873", + "right": "issue:46050", + "accept": false, + "reason": "Quantization offloading behavior and save/load corruption are different failure modes; same broad area, not the same bug." + }, + { + "left": "issue:45588", + "right": "issue:46050", + "accept": false, + "reason": "Flash-attention sink handling crash is unrelated to sinq save_pretrained/from_pretrained breakage." + }, + { + "left": "issue:29942", + "right": "issue:30333", + "accept": false, + "reason": "FlashAttention2 CI failures and MLflow reporting on job failure are unrelated subsystems." + }, + { + "left": "issue:42490", + "right": "issue:43696", + "accept": false, + "reason": "Trailing-slash import warning is a path-parsing issue; the other is a model OOM on load." + }, + { + "left": "issue:29159", + "right": "issue:32090", + "accept": false, + "reason": "Tokenizer slow/fast inconsistency is unrelated to the trainer NoneType GPU broadcast error." + }, + { + "left": "issue:43065", + "right": "issue:46369", + "accept": false, + "reason": "A dummy Conv2d placeholder in a decoder is not the same as memory overflow when using the SAM3 lite text model." + }, + { + "left": "issue:45084", + "right": "issue:45520", + "accept": false, + "reason": "Compilation of non-template nodes and flash_attn backend import mapping are different code paths." + }, + { + "left": "issue:28218", + "right": "issue:34567", + "accept": false, + "reason": "Added-token spacing and num_input_tokens_seen bookkeeping are unrelated tokenizer vs trainer-state issues." + }, + { + "left": "issue:44423", + "right": "issue:44734", + "accept": false, + "reason": "Both involve serve, but one is a multimodal continuous-batching crash and the other is KV-cache continuation indexing." + }, + { + "left": "issue:44554", + "right": "issue:45446", + "accept": false, + "reason": "MPS attention correctness and a PyTorch version check for AuxRequest import are unrelated defects." + }, + { + "left": "issue:15354", + "right": "issue:38175", + "accept": false, + "reason": "TorchScript generator-expression export failure is unrelated to SigLIP2 zero probabilities." + }, + { + "left": "issue:45406", + "right": "issue:46121", + "accept": false, + "reason": "Gemma4Processor missing _tokenizer and rope param list conversion are different bugs in different layers." + }, + { + "left": "issue:43493", + "right": "issue:43701", + "accept": false, + "reason": "SigLIP2 implementation parity and checkpoint resume key mismatch do not describe the same underlying problem." + }, + { + "left": "issue:45072", + "right": "issue:45198", + "accept": false, + "reason": "Dtype mismatches in bfloat16 inference are unrelated to Wav2Vec2 save_pretrained/tokenization failure." + }, + { + "left": "issue:45588", + "right": "issue:45759", + "accept": false, + "reason": "Flash-attention s_aux=None crash and composite Qwen config unwrapping are different model-integration bugs." + }, + { + "left": "issue:43540", + "right": "issue:45200", + "accept": false, + "reason": "Video input processing error in Qwen3OmniMoe is not the same as Gemma4 mm_token_type_ids defaults." + }, + { + "left": "issue:43295", + "right": "issue:43450", + "accept": false, + "reason": "Processor/tokenizer regression in one issue and batched video output shape in the other are distinct multimodal bugs." + }, + { + "left": "issue:37428", + "right": "issue:43957", + "accept": false, + "reason": "Missing flash-support import and meta-device loading breakage are different import/loading problems." + }, + { + "left": "issue:44186", + "right": "issue:46192", + "accept": false, + "reason": "LayoutLMv2 tokenizer crashes on NER/batching differ from return_tensors being ignored with text_kwargs." + }, + { + "left": "issue:45200", + "right": "issue:45823", + "accept": false, + "reason": "Defaulting mm_token_type_ids and device-map-induced device mismatch are not the same defect." + }, + { + "left": "issue:45823", + "right": "issue:46192", + "accept": false, + "reason": "Device mismatch during forward is unrelated to return_tensors being ignored." + }, + { + "left": "issue:43295", + "right": "issue:44610", + "accept": false, + "reason": "A processor/tokenizer regression is not the same as OmDet-Turbo producing the wrong image size." + }, + { + "left": "issue:45520", + "right": "issue:46178", + "accept": false, + "reason": "flash_attn KeyError on Python 3.13 and tensorflow_text backend mapping validation are separate import-registry issues." + }, + { + "left": "issue:43450", + "right": "issue:43575", + "accept": false, + "reason": "Batched video shape bug and tensor-parallel OOM for Qwen2-57B are unrelated." + }, + { + "left": "issue:43550", + "right": "issue:43844", + "accept": false, + "reason": "torch.compile SDPA failure and ZeRO-3 gradient explosion are different training/runtime bugs." + }, + { + "left": "issue:43232", + "right": "issue:44849", + "accept": false, + "reason": "Generation kwargs update after sync_gpus and output_hidden_states behavior in Qwen3.5 are not the same issue." + }, + { + "left": "issue:43572", + "right": "issue:44987", + "accept": false, + "reason": "Missing pad_token_idx in StableLmConfig and a transformers>=5.1 loading failure are separate regression reports." + }, + { + "left": "issue:44554", + "right": "issue:45356", + "accept": false, + "reason": "MPS attention correctness and Kimi tokenizer codec regression are unrelated." + }, + { + "left": "issue:39692", + "right": "issue:46192", + "accept": false, + "reason": "Documentation example errors for SigLIP2 are not the same as return_tensors being ignored when text_kwargs is passed." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 21, + "review_unit_index": 27, + "review_unit_count": 56, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-473", + "nodes": [ + "issue:28218", + "issue:35141", + "issue:37428", + "issue:38175", + "issue:41720", + "issue:42490", + "issue:42915", + "issue:43262", + "issue:43278", + "issue:43334", + "issue:43388", + "issue:43450", + "issue:43452", + "issue:43493", + "issue:43525", + "issue:43526", + "issue:43572", + "issue:43653", + "issue:43819", + "issue:43824", + "issue:43827", + "issue:43856", + "issue:43901", + "issue:43950", + "issue:43957", + "issue:43975", + "issue:44112", + "issue:44442", + "issue:44466", + "issue:44479", + "issue:44521", + "issue:44610", + "issue:44792", + "issue:44857", + "issue:44898", + "issue:45127", + "issue:45198", + "issue:45200", + "issue:45290", + "issue:45325", + "issue:45381", + "issue:45412", + "issue:45563", + "issue:46192", + "issue:46209", + "issue:46285", + "issue:46512" + ], + "soft_pairs": [ + "issue:44521|issue:45290", + "issue:43572|issue:43957", + "issue:37428|issue:43824", + "issue:45381|issue:46209", + "issue:42490|issue:43493", + "issue:44466|issue:45127", + "issue:38175|issue:43493", + "issue:43653|issue:44112", + "issue:43278|issue:43856", + "issue:28218|issue:35141", + "issue:43452|issue:46512", + "issue:43950|issue:46285", + "issue:44898|issue:45412", + "issue:42915|issue:43824", + "issue:43334|issue:43525", + "issue:43262|issue:43450", + "issue:43824|issue:43975", + "issue:45200|issue:45325", + "issue:44442|issue:45198", + "issue:44479|issue:46192", + "issue:44792|issue:45563", + "issue:43388|issue:43526", + "issue:43819|issue:44112", + "issue:43827|issue:43901", + "issue:41720|issue:46209", + "issue:44610|issue:44857" + ], + "prepared_review_unit_hash": "e7e9decde8a025d02ca0d1e9d9f344f1dff01a249304f5e114ee3b0b0f1f1e84", + "component_budget": { + "node_count": 473, + "item_count": 473, + "soft_pair_count": 1738, + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 26, + "serialized_chars": 35354, + "estimated_input_tokens": 8839, + "estimated_eval_tokens": 17934 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "All soft pairs look like distinct issues rather than duplicates; they mostly share broad subsystems or API surfaces but not the same concrete bug. I therefore rejected every soft edge and selected one strong open issue as the best representative issue overall.", + "confidence": 0.94, + "canonical_issue_reason": "issue:44521 is the strongest open, user-facing report here: it has multiple inbound references, clear reproducible behavior, and a focused multimodal `apply_chat_template` regression.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:44521 is the most actionable overall because it is open, well-scoped, and appears to impact a common multimodal code path with active discussion.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44521", + "right": "issue:45290", + "accept": false, + "reason": "Both involve `apply_chat_template`, but one is about zeroed assistant masks for multimodal inputs and the other is a crash with tool-call messages; different failure modes and likely different fixes." + }, + { + "left": "issue:43572", + "right": "issue:43957", + "accept": false, + "reason": "These are separate loading/config bugs: missing `pad_token_idx` in one model config versus failures when loading models on `torch.device(\"meta\")`." + }, + { + "left": "issue:37428", + "right": "issue:43824", + "accept": false, + "reason": "Both are import errors, but for different symbols and likely unrelated code paths (`_flash_supports_window_size` vs `Qwen2_5_VLForConditionalGeneration`)." + }, + { + "left": "issue:45381", + "right": "issue:46209", + "accept": false, + "reason": "One is a Qwen2.5-VL video position-id regression; the other is an FP8 inference failure on B200. Different hardware/model issues." + }, + { + "left": "issue:42490", + "right": "issue:43493", + "accept": false, + "reason": "Local-path slash import warning and SigLIP2 implementation mismatch are unrelated bugs." + }, + { + "left": "issue:44466", + "right": "issue:45127", + "accept": false, + "reason": "One concerns inconsistent `lm_head.weight` serialization by device; the other concerns LoRA merge behavior with extended vocab and tied embeddings. Not the same bug." + }, + { + "left": "issue:38175", + "right": "issue:43493", + "accept": false, + "reason": "Both mention SigLIP2, but one is a specific zero-probability symptom and the other is a broader HF-vs-JAX discrepancy; insufficient evidence they are the same concrete issue." + }, + { + "left": "issue:43653", + "right": "issue:44112", + "accept": false, + "reason": "A tokenizer special-token registration bug and a GraniteSpeech CI device-override test failure are different problems." + }, + { + "left": "issue:43278", + "right": "issue:43856", + "accept": false, + "reason": "Different topics: dtype drift in evaluation versus memory inefficiency during Qwen3 MoE training." + }, + { + "left": "issue:28218", + "right": "issue:35141", + "accept": false, + "reason": "Tokenizer spacing after added tokens is unrelated to output embedding reinitialization after resizing embeddings." + }, + { + "left": "issue:43452", + "right": "issue:46512", + "accept": false, + "reason": "GGUF loading breakage and stale `config.dtype` after `from_config(dtype=...)` are separate model-loading regressions." + }, + { + "left": "issue:43950", + "right": "issue:46285", + "accept": false, + "reason": "Non-persistent buffer corruption on load is not the same as CLIP text model structural inconsistency." + }, + { + "left": "issue:44898", + "right": "issue:45412", + "accept": false, + "reason": "Perceiver interpolation failure and RT-DETR memory not being released are unrelated." + }, + { + "left": "issue:42915", + "right": "issue:43824", + "accept": false, + "reason": "Qwen3Moe FP8 failure and a missing Qwen2.5-VL import are different issues." + }, + { + "left": "issue:43334", + "right": "issue:43525", + "accept": false, + "reason": "Both are missing `pad_token_id` on configs, but they affect different model classes and likely require separate fixes." + }, + { + "left": "issue:43262", + "right": "issue:43450", + "accept": false, + "reason": "Audio chat-template sampling-rate defaulting and video batch-shape handling are different processor bugs." + }, + { + "left": "issue:43824", + "right": "issue:43975", + "accept": false, + "reason": "Importing a missing class and incorrect detokenization for DeepSeek are unrelated." + }, + { + "left": "issue:45200", + "right": "issue:45325", + "accept": false, + "reason": "Gemma 4 multimodal token-type defaults and Qwen2.5-VL temporal position-id scaling are different multimodal regressions." + }, + { + "left": "issue:44442", + "right": "issue:45198", + "accept": false, + "reason": "FastSpeech2Conformer tokenizer loading and Wav2Vec2 save/tokenization failures are distinct audio-model problems." + }, + { + "left": "issue:44479", + "right": "issue:46192", + "accept": false, + "reason": "A video-input regression in Qwen models is unrelated to `return_tensors` being ignored when `text_kwargs` is passed." + }, + { + "left": "issue:44792", + "right": "issue:45563", + "accept": false, + "reason": "A failed Janus image-generation test and a stale warning for `num_return_sequences` are not the same bug." + }, + { + "left": "issue:43388", + "right": "issue:43526", + "accept": false, + "reason": "Metric gathering drops tuple labels in the last batch, whereas BeitImageProcessorFast reduces labels to a single item; different code paths." + }, + { + "left": "issue:43819", + "right": "issue:44112", + "accept": false, + "reason": "DAC latent reconstruction mismatch and a GraniteSpeech CI test failure are unrelated." + }, + { + "left": "issue:43827", + "right": "issue:43901", + "accept": false, + "reason": "Both are pipeline docs issues, but they concern different pipelines and different deprecated/changed behaviors." + }, + { + "left": "issue:41720", + "right": "issue:46209", + "accept": false, + "reason": "Different hardware/model failures: auto device mapping on A800 versus FP8 inference on B200." + }, + { + "left": "issue:44610", + "right": "issue:44857", + "accept": false, + "reason": "OmDet-Turbo preprocessing shape mismatch and LwDetrImageLoss AMP/CUDA crash are separate detection-model issues." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 21, + "review_unit_index": 28, + "review_unit_count": 56, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-473", + "nodes": [ + "issue:28282", + "issue:29127", + "issue:29942", + "issue:32090", + "issue:33357", + "issue:37428", + "issue:38175", + "issue:43278", + "issue:43335", + "issue:43425", + "issue:43504", + "issue:43550", + "issue:43606", + "issue:43611", + "issue:43653", + "issue:43720", + "issue:43819", + "issue:43825", + "issue:43931", + "issue:44112", + "issue:44206", + "issue:44351", + "issue:44448", + "issue:44514", + "issue:44521", + "issue:44561", + "issue:44743", + "issue:44855", + "issue:44857", + "issue:45083", + "issue:45127", + "issue:45200", + "issue:45245", + "issue:45290", + "issue:45412", + "issue:45520", + "issue:45584", + "issue:45588", + "issue:45593", + "issue:45759", + "issue:45859", + "issue:46170", + "issue:46234", + "issue:46285", + "issue:46326", + "issue:46431", + "issue:46512", + "issue:46566" + ], + "soft_pairs": [ + "issue:44561|issue:46234", + "issue:38175|issue:45412", + "issue:43335|issue:43550", + "issue:44206|issue:45584", + "issue:45759|issue:46285", + "issue:43819|issue:45083", + "issue:44857|issue:45593", + "issue:45520|issue:46170", + "issue:43606|issue:45083", + "issue:43335|issue:43653", + "issue:29942|issue:32090", + "issue:37428|issue:43425", + "issue:43550|issue:44112", + "issue:28282|issue:33357", + "issue:44448|issue:44743", + "issue:43611|issue:43931", + "issue:45759|issue:46512", + "issue:44514|issue:44521", + "issue:43335|issue:43504", + "issue:44855|issue:45083", + "issue:43278|issue:43825", + "issue:45290|issue:46326", + "issue:46431|issue:46566", + "issue:43720|issue:44112", + "issue:29127|issue:32090", + "issue:37428|issue:44351", + "issue:43335|issue:45083", + "issue:43335|issue:44112", + "issue:45127|issue:45245", + "issue:43335|issue:45593", + "issue:45588|issue:45859", + "issue:45200|issue:45588" + ], + "prepared_review_unit_hash": "6c79e578bb7ba8ef8afa85f564e4cb96a1cb26cd8ef4ba6c7f7413a0dd440cd1", + "component_budget": { + "node_count": 473, + "item_count": 473, + "soft_pair_count": 1738, + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 32, + "serialized_chars": 37603, + "estimated_input_tokens": 9401, + "estimated_eval_tokens": 19058 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is heterogeneous: most items are separate bugs across unrelated subsystems, so none of the soft pairs should be merged as duplicates. The only loosely coherent area is chat-template/multimodal handling, but even there the issues describe different failure modes.", + "confidence": 0.6, + "canonical_issue_reason": "issue:44521 is the best representative issue: it is open, has the most discussion/inbound references, and sits in the most coherent subgroup in this set (chat_template / multimodal input handling).", + "canonical_pr_reason": null, + "best_issue_reason": "issue:44521 is the strongest global issue candidate because it is the most active and broadly scoped within the only clearly related theme in the cluster.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44561", + "right": "issue:46234", + "accept": false, + "reason": "Both are import-utils-related, but they are different missing symbols and different breakages: `is_torch_fx_available` removal vs missing `is_falcon_mamba_ssm_available`." + }, + { + "left": "issue:38175", + "right": "issue:45412", + "accept": false, + "reason": "Different symptoms and code paths: zero probabilities in SigLIP2 vs RT-DETR memory not released on delete." + }, + { + "left": "issue:43335", + "right": "issue:43550", + "accept": false, + "reason": "One is a config bug in SwitchTransformers layer creation; the other is a Bamba SDPA/torch.compile failure. Not the same bug." + }, + { + "left": "issue:44206", + "right": "issue:45584", + "accept": false, + "reason": "Different feature areas: `LasrFeatureExtractor` center-arg regression vs Whisper empty transcription after `align_special_tokens`." + }, + { + "left": "issue:45759", + "right": "issue:46285", + "accept": false, + "reason": "Both involve model construction, but one is Qwen multimodal `text_config` unwrapping and the other is CLIP text-model structure inconsistency." + }, + { + "left": "issue:43819", + "right": "issue:45083", + "accept": false, + "reason": "Different model-specific helper bugs: DAC `from_latents`/STE mismatch vs qwen3_omni_moe feature-length helper behavior." + }, + { + "left": "issue:44857", + "right": "issue:45593", + "accept": false, + "reason": "Unrelated training-time failures: float16 AMP crash in LwDetrImageLoss vs D-FINE auxiliary-loss handling when denoising is off." + }, + { + "left": "issue:45520", + "right": "issue:46170", + "accept": false, + "reason": "Both are import/dependency issues, but they are on different platforms and dependencies: flash_attn on Python 3.13 vs torch.distributed on Windows ROCm." + }, + { + "left": "issue:43606", + "right": "issue:45083", + "accept": false, + "reason": "Different problems: CPU offload device mismatch in bark-small vs a qwen3_omni_moe helper-function issue." + }, + { + "left": "issue:43335", + "right": "issue:43653", + "accept": false, + "reason": "SwitchTransformers sparse-layer creation and BigBirdTokenizer special-token registration are unrelated bugs." + }, + { + "left": "issue:29942", + "right": "issue:32090", + "accept": false, + "reason": "Flash Attention 2 test failures and Trainer `_gpu_broadcast_one` NoneType broadcast errors are different code paths." + }, + { + "left": "issue:37428", + "right": "issue:43425", + "accept": false, + "reason": "One is a missing import from flash-attention utils; the other is broad Torch 2.10 incompatibility. Not the same issue." + }, + { + "left": "issue:43550", + "right": "issue:44112", + "accept": false, + "reason": "Bamba SDPA/compile failure and GraniteSpeech CI stale device override are distinct failures." + }, + { + "left": "issue:28282", + "right": "issue:33357", + "accept": false, + "reason": "Generic torch-missing ImportError vs MacOS bus error with a community CLIP model are not the same underlying bug." + }, + { + "left": "issue:44448", + "right": "issue:44743", + "accept": false, + "reason": "Pegasus output regression across v4/v5 and Qwen3 recurrent-state reset are separate model behavior bugs." + }, + { + "left": "issue:43611", + "right": "issue:43931", + "accept": false, + "reason": "Base-model-prefix loading breakage and Qwen3-VL weight-shape mismatch are different loading failures." + }, + { + "left": "issue:43278", + "right": "issue:43825", + "accept": false, + "reason": "A dtype-staleness bug in `from_config` is unrelated to a pipeline error-message regression." + }, + { + "left": "issue:45290", + "right": "issue:46326", + "accept": false, + "reason": "Both touch chat template logic, but one crashes on assistant tool-call messages and the other concerns `continue_final_message` for mistral_common tokenizers." + }, + { + "left": "issue:46431", + "right": "issue:46566", + "accept": false, + "reason": "Processor chat-template absence and DiffusionGemma meta-device loading are unrelated." + }, + { + "left": "issue:43720", + "right": "issue:44112", + "accept": false, + "reason": "BitNet packed-weight unpacking during accelerate loading is not the same as a GraniteSpeech CI device override test failure." + }, + { + "left": "issue:29127", + "right": "issue:32090", + "accept": false, + "reason": "LayoutLMV3 error-message clarity issue vs Trainer broadcast TypeError; different functionality and severity." + }, + { + "left": "issue:37428", + "right": "issue:44351", + "accept": false, + "reason": "Different missing imports: `_flash_supports_window_size` vs `HybridCache`." + }, + { + "left": "issue:43335", + "right": "issue:45083", + "accept": false, + "reason": "SwitchTransformers sparse-layer construction and qwen3_omni_moe feature-length helper behavior are unrelated." + }, + { + "left": "issue:43335", + "right": "issue:44112", + "accept": false, + "reason": "Different subsystems and failure modes: model config bug vs CI stale device override." + }, + { + "left": "issue:45127", + "right": "issue:45245", + "accept": false, + "reason": "LoRA merge/model collapse with tied embeddings is unrelated to category cardinality limits." + }, + { + "left": "issue:43335", + "right": "issue:45593", + "accept": false, + "reason": "SwitchTransformers sparse-layer config bug and D-FINE auxiliary-loss omission are different issues." + }, + { + "left": "issue:45588", + "right": "issue:45859", + "accept": false, + "reason": "Both mention attention-related paths, but one is an `s_aux=None` crash in flash_attention integration and the other is a CPU-offload incompatibility in Qwen3_5MoeTextRotaryEmbedding." + }, + { + "left": "issue:45200", + "right": "issue:45588", + "accept": false, + "reason": "Gemma 4 `mm_token_type_ids` defaulting and a flash_attention `s_aux=None` crash are separate 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": 21, + "review_unit_index": 29, + "review_unit_count": 56, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-473", + "nodes": [ + "issue:10000", + "issue:15354", + "issue:26413", + "issue:27957", + "issue:28218", + "issue:29942", + "issue:31501", + "issue:32090", + "issue:33357", + "issue:34689", + "issue:35141", + "issue:38175", + "issue:39692", + "issue:40444", + "issue:42490", + "issue:43278", + "issue:43329", + "issue:43381", + "issue:43526", + "issue:43550", + "issue:43575", + "issue:43606", + "issue:43701", + "issue:43819", + "issue:43854", + "issue:44206", + "issue:44360", + "issue:44448", + "issue:44493", + "issue:44561", + "issue:44779", + "issue:44855", + "issue:44857", + "issue:44977", + "issue:45072", + "issue:45083", + "issue:45229", + "issue:45290", + "issue:45310", + "issue:45381", + "issue:45459", + "issue:45478", + "issue:45823", + "issue:45920", + "issue:46192", + "issue:46234", + "issue:46399", + "issue:46655" + ], + "soft_pairs": [ + "issue:39692|issue:43526", + "issue:34689|issue:35141", + "issue:26413|issue:31501", + "issue:42490|issue:43854", + "issue:43819|issue:44493", + "issue:45459|issue:46234", + "issue:45310|issue:45478", + "issue:43606|issue:44857", + "issue:44448|issue:44779", + "issue:39692|issue:44493", + "issue:44206|issue:45823", + "issue:28218|issue:32090", + "issue:44360|issue:44561", + "issue:44857|issue:45072", + "issue:43701|issue:45083", + "issue:38175|issue:44493", + "issue:32090|issue:33357", + "issue:42490|issue:43575", + "issue:45290|issue:45381", + "issue:43550|issue:45083", + "issue:29942|issue:35141", + "issue:40444|issue:43526", + "issue:45290|issue:46655", + "issue:45920|issue:46192", + "issue:44448|issue:44977", + "issue:27957|issue:34689", + "issue:43278|issue:43329", + "issue:44977|issue:45229", + "issue:44855|issue:46399", + "issue:10000|issue:15354", + "issue:43278|issue:43381" + ], + "prepared_review_unit_hash": "efe77f0dc5ce3529c0e120cd5027a3acd33548e460256645020ebad854c99f93", + "component_budget": { + "node_count": 473, + "item_count": 473, + "soft_pair_count": 1738, + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 31, + "serialized_chars": 36984, + "estimated_input_tokens": 9246, + "estimated_eval_tokens": 18748 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly heterogeneous: nearly all pairs are unrelated bug reports that only share broad Transformer/regression language. The only clear duplicate is the repeated Qwen3.5 MoE `from_pretrained` issue (45310/45478).", + "confidence": 0.9, + "canonical_issue_reason": "issue:45310 is the earlier, fuller report of the same Qwen3.5 MoE `from_pretrained` regression as issue:45478, so it is the better canonical issue for that duplicate pair.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:45920 is the broadest and most representative standalone issue in the set: it describes a multi-model tokenizer regression, is open, and already has inbound references.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:39692", + "right": "issue:43526", + "accept": false, + "reason": "SigLIP2 doc-example errors and BEiT label reduction are different components and failure modes." + }, + { + "left": "issue:34689", + "right": "issue:35141", + "accept": false, + "reason": "Llama 3.2 model-loading failure is unrelated to token embedding resizing/post_init reinitialization." + }, + { + "left": "issue:26413", + "right": "issue:31501", + "accept": false, + "reason": "A checkpoint-resume epoch_iterator bug is not the same as FineWeb SLM startup failure." + }, + { + "left": "issue:42490", + "right": "issue:43854", + "accept": false, + "reason": "Local-path trailing-slash warning and GLM-4.7-Flash loading correctness are different loading issues." + }, + { + "left": "issue:43819", + "right": "issue:44493", + "accept": false, + "reason": "DAC `from_latents` mismatch and position-id key warnings concern different model paths." + }, + { + "left": "issue:45459", + "right": "issue:46234", + "accept": false, + "reason": "Tokenizer decode-error masking with missing protobuf is unrelated to a missing import-utils availability symbol." + }, + { + "left": "issue:45310", + "right": "issue:45478", + "accept": true, + "reason": "Same Qwen3.5 MoE `from_pretrained` regression, with matching title and symptom; 45478 is a later duplicate." + }, + { + "left": "issue:43606", + "right": "issue:44857", + "accept": false, + "reason": "CPU offload device mismatch for `suno/bark-small` is not the same bug as AMP/CUDA float16 loss failure." + }, + { + "left": "issue:44448", + "right": "issue:44779", + "accept": false, + "reason": "Pegasus v4/v5 output drift and DeepSeek tokenizer regression are both regressions, but not the same code-path bug." + }, + { + "left": "issue:39692", + "right": "issue:44493", + "accept": false, + "reason": "SigLIP2 documentation errors do not match the generic position-id key issue affecting many models." + }, + { + "left": "issue:44206", + "right": "issue:45823", + "accept": false, + "reason": "Unsupported `center` arg in feature extraction is unrelated to Gemma4 PLE device placement mismatch." + }, + { + "left": "issue:28218", + "right": "issue:32090", + "accept": false, + "reason": "Added-token spacing behavior and `_gpu_broadcast_one` NoneType errors are unrelated tokenizer vs trainer bugs." + }, + { + "left": "issue:44360", + "right": "issue:44561", + "accept": false, + "reason": "A missing ReLU in DSA indexer is unrelated to removal of `is_torch_fx_available` breaking `trust_remote_code` models." + }, + { + "left": "issue:44857", + "right": "issue:45072", + "accept": false, + "reason": "Both mention dtype/precision, but they affect different models and different execution paths." + }, + { + "left": "issue:43701", + "right": "issue:45083", + "accept": false, + "reason": "Checkpoint key mismatch and qwen3_omni_moe feature-length behavior are distinct issues." + }, + { + "left": "issue:38175", + "right": "issue:44493", + "accept": false, + "reason": "Zero probabilities in SigLIP2 are not the same as unexpected position-id key handling." + }, + { + "left": "issue:32090", + "right": "issue:33357", + "accept": false, + "reason": "Trainer broadcast NoneType failure and MacOS bus error are different runtime failures." + }, + { + "left": "issue:42490", + "right": "issue:43575", + "accept": false, + "reason": "Import warning for a trailing slash is unrelated to TP load OOM for Qwen2-57B." + }, + { + "left": "issue:45290", + "right": "issue:45381", + "accept": false, + "reason": "Chat-template tool-call crash and qwen2.5-vl video position-id bug are different multimodal code paths." + }, + { + "left": "issue:43550", + "right": "issue:45083", + "accept": false, + "reason": "Bamba SDPA compile failure and qwen3_omni_moe feature-length behavior are unrelated." + }, + { + "left": "issue:29942", + "right": "issue:35141", + "accept": false, + "reason": "Flash Attention 2 test failures are not the same bug as embedding reinitialization after resize." + }, + { + "left": "issue:40444", + "right": "issue:43526", + "accept": false, + "reason": "Qwen2.5-VL iterable-dataset finetuning failure and BEiT label reduction are unrelated." + }, + { + "left": "issue:45290", + "right": "issue:46655", + "accept": false, + "reason": "Chat template tool-call handling and SmolVLM video padding are different bugs." + }, + { + "left": "issue:45920", + "right": "issue:46192", + "accept": false, + "reason": "A v5 tokenizer wrong-token-ID regression is broader and different from `return_tensors` being ignored when `text_kwargs` is passed." + }, + { + "left": "issue:44448", + "right": "issue:44977", + "accept": false, + "reason": "Pegasus generation drift and Qwen3.5 flash-attention generation failure are distinct model-specific problems." + }, + { + "left": "issue:27957", + "right": "issue:34689", + "accept": false, + "reason": "XLMRoberta flash-attention support and Llama 3.2 model loading are unrelated." + }, + { + "left": "issue:43278", + "right": "issue:43329", + "accept": false, + "reason": "BF16-to-FP32 dtype change in evaluation is unrelated to the missing video helper functions in multimodal token counting." + }, + { + "left": "issue:44977", + "right": "issue:45229", + "accept": false, + "reason": "Qwen3.5 flash-attention generation failure is unrelated to Gemma4 multi-GPU CUDA OOM." + }, + { + "left": "issue:44855", + "right": "issue:46399", + "accept": false, + "reason": "Python 3.13 `IndentationError` in a scripted model import is not the same as a missing decorator for rotary-pos-emb registration." + }, + { + "left": "issue:10000", + "right": "issue:15354", + "accept": false, + "reason": "German DistilBertModel issue and GeneratorExp/JIT export failure are unrelated." + }, + { + "left": "issue:43278", + "right": "issue:43381", + "accept": false, + "reason": "Embedding dtype drift in eval and gradient checkpointing in eval mode are different concerns." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is well grounded in the packet: the cluster is indeed mostly heterogeneous, and the duplicate Qwen3.5 MoE from_pretrained pair (45310/45478) is the only clearly supported duplicate. The soft-edge calls are conservative overall and do not overstate equivalence between different bugs." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 21, + "review_unit_index": 30, + "review_unit_count": 56, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-473", + "nodes": [ + "issue:25251", + "issue:29127", + "issue:32090", + "issue:33290", + "issue:33357", + "issue:34567", + "issue:37428", + "issue:39692", + "issue:41720", + "issue:42175", + "issue:42490", + "issue:42915", + "issue:42994", + "issue:43278", + "issue:43381", + "issue:43412", + "issue:43575", + "issue:43611", + "issue:43653", + "issue:43701", + "issue:43746", + "issue:43873", + "issue:43927", + "issue:43976", + "issue:44186", + "issue:44361", + "issue:44448", + "issue:44466", + "issue:44561", + "issue:44568", + "issue:44743", + "issue:44779", + "issue:44857", + "issue:44871", + "issue:44977", + "issue:45198", + "issue:45229", + "issue:45372", + "issue:45406", + "issue:45440", + "issue:45491", + "issue:45563", + "issue:45588", + "issue:45698", + "issue:45803", + "issue:45823", + "issue:46121", + "issue:46209" + ], + "soft_pairs": [ + "issue:44857|issue:45588", + "issue:43927|issue:45698", + "issue:43746|issue:46209", + "issue:44561|issue:44779", + "issue:37428|issue:42175", + "issue:43873|issue:45229", + "issue:32090|issue:33290", + "issue:25251|issue:32090", + "issue:43976|issue:45440", + "issue:44448|issue:44466", + "issue:41720|issue:45229", + "issue:42490|issue:43873", + "issue:44568|issue:45491", + "issue:45588|issue:45823", + "issue:44977|issue:45563", + "issue:45372|issue:45588", + "issue:43278|issue:43575", + "issue:33357|issue:39692", + "issue:45198|issue:45803", + "issue:29127|issue:34567", + "issue:43412|issue:45588", + "issue:44871|issue:44977", + "issue:43381|issue:43701", + "issue:43653|issue:44743", + "issue:44361|issue:45198", + "issue:44186|issue:45198", + "issue:43611|issue:46121", + "issue:45372|issue:45406", + "issue:42915|issue:42994" + ], + "prepared_review_unit_hash": "f28ff716a012f25e1ee77cc95c817c29d9e0f29a322327198b801aed4782b135", + "component_budget": { + "node_count": 473, + "item_count": 473, + "soft_pair_count": 1738, + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 29, + "serialized_chars": 36517, + "estimated_input_tokens": 9130, + "estimated_eval_tokens": 18516 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The soft pairs are thematic false positives rather than true duplicates; each pair targets a different bug surface, so there is no safe canonical issue or PR for this cluster.", + "confidence": 0.95, + "canonical_issue_reason": "No single issue is a defensible canonical: the cluster spans unrelated regressions across tokenizers, training, inference, packaging, and docs, and the soft pairs do not share the same concrete bug.", + "canonical_pr_reason": null, + "best_issue_reason": "No issue is a strong global representative because the items are not one underlying problem; any pick would be arbitrary and misleading for dedup triage.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44857", + "right": "issue:45588", + "accept": false, + "reason": "Different failures: one is an AMP/CUDA crash in LwDetrImageLoss, the other is a flash-attention AttributeError on `s_aux=None`." + }, + { + "left": "issue:43927", + "right": "issue:45698", + "accept": false, + "reason": "Both involve save/load behavior, but one is DiaConfig custom token IDs and the other is wrong custom module resolution after `save_pretrained`; different root causes." + }, + { + "left": "issue:43746", + "right": "issue:46209", + "accept": false, + "reason": "Unrelated problems: PEFT adapter loading from local checkpoints vs FP8 inference failure on B200." + }, + { + "left": "issue:44561", + "right": "issue:44779", + "accept": false, + "reason": "Both are v5 regressions, but one is a removed utility breaking remote-code models and the other is a tokenizer output bug for DeepSeek; not the same code path." + }, + { + "left": "issue:37428", + "right": "issue:42175", + "accept": false, + "reason": "Different areas entirely: flash-attention import failure vs missing TensorFlow in a torch-only install." + }, + { + "left": "issue:43873", + "right": "issue:45229", + "accept": false, + "reason": "Both mention memory/offloading, but one is quantized offloading behavior and the other is multi-GPU CUDA OOM for Gemma4; not the same fix." + }, + { + "left": "issue:32090", + "right": "issue:33290", + "accept": false, + "reason": "Trainer broadcast TypeError and DeepSpeed Adafactor OOM are distinct training failures with different mechanisms." + }, + { + "left": "issue:25251", + "right": "issue:32090", + "accept": false, + "reason": "Pipeline top_k nesting behavior is unrelated to the Trainer `_gpu_broadcast_one` TypeError." + }, + { + "left": "issue:43976", + "right": "issue:45440", + "accept": false, + "reason": "Python version compatibility issue vs DeepSeekV3MoE divergence from remote implementation; no shared underlying bug." + }, + { + "left": "issue:44448", + "right": "issue:44466", + "accept": false, + "reason": "Both are v5 behavior changes, but one is generation/output mismatch and the other is serialization of tied weights; different concrete bugs." + }, + { + "left": "issue:41720", + "right": "issue:45229", + "accept": false, + "reason": "Both are large-model inference failures, but one is a cudaErrorAssert on Qwen3 auto device mapping and the other is CUDA OOM on Gemma4 multi-GPU." + }, + { + "left": "issue:42490", + "right": "issue:43873", + "accept": false, + "reason": "Import warning for trailing slash paths is unrelated to quantized offloading behavior." + }, + { + "left": "issue:44568", + "right": "issue:45491", + "accept": false, + "reason": "Tokenizer special-token handling and Gemma3 NaN embeddings with mixed-length batching are different issues." + }, + { + "left": "issue:45588", + "right": "issue:45823", + "accept": false, + "reason": "Both touch Gemma4/attention, but one is a flash-attention crash and the other is a device-mapping mismatch in PLE forward; not mergeable as one fix." + }, + { + "left": "issue:44977", + "right": "issue:45563", + "accept": false, + "reason": "Flash-attention generation bug and a stale warning in paged generate() are unrelated." + }, + { + "left": "issue:45372", + "right": "issue:45406", + "accept": false, + "reason": "Gemma4 processor loading fails for different reasons: missing `ReasoningEffort` import vs missing `_tokenizer` on serve; same family, different bugs." + }, + { + "left": "issue:43278", + "right": "issue:43575", + "accept": false, + "reason": "Embedding dtype drift during eval is unrelated to tensor-parallel OOM on Qwen2-57B." + }, + { + "left": "issue:33357", + "right": "issue:39692", + "accept": false, + "reason": "CLIP bus error on macOS and SigLIP2 doc example errors are different issues, with no shared code-path evidence." + }, + { + "left": "issue:45198", + "right": "issue:45803", + "accept": false, + "reason": "Wav2Vec2 save/tokenization failure and Fuyu batch tensor conversion swallowing errors are distinct tokenizer/data-processing bugs." + }, + { + "left": "issue:29127", + "right": "issue:34567", + "accept": false, + "reason": "LayoutLMv3 error-message clarity and TrainerState token-count tracking are unrelated." + }, + { + "left": "issue:43412", + "right": "issue:45588", + "accept": false, + "reason": "Executorch export failures for Gemma3n and a flash-attention crash on sink-less models do not share the same underlying bug." + }, + { + "left": "issue:44871", + "right": "issue:44977", + "accept": false, + "reason": "Gemma3 EOS token config mismatch and Qwen3.5 flash-attention generation failure are separate model-specific issues." + }, + { + "left": "issue:43381", + "right": "issue:43701", + "accept": false, + "reason": "Eval-mode checkpointing restriction and resume checkpoint key mismatch are different training-state problems." + }, + { + "left": "issue:43653", + "right": "issue:44743", + "accept": false, + "reason": "BigBirdTokenizer special-token registration and Qwen3.5 recurrent-state reset with cache are unrelated." + }, + { + "left": "issue:44361", + "right": "issue:45198", + "accept": false, + "reason": "Both affect tokenizers, but MLuke task AttributeError and Wav2Vec2 save/tokenization failure are different bugs." + }, + { + "left": "issue:44186", + "right": "issue:45198", + "accept": false, + "reason": "LayoutLMv2 NER/batching crash and Wav2Vec2 save/tokenization failure are not the same code path." + }, + { + "left": "issue:43611", + "right": "issue:46121", + "accept": false, + "reason": "base_model_prefix loading regression and rope-validation TypeError are unrelated implementation bugs." + }, + { + "left": "issue:45372", + "right": "issue:45588", + "accept": false, + "reason": "Gemma4 processor import failure and flash-attention `s_aux=None` crash are separate failures even though both mention Gemma4/attention." + }, + { + "left": "issue:42915", + "right": "issue:42994", + "accept": false, + "reason": "Qwen3Moe FP8 config failure and quantized model saving failure are different quantization-related problems, 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": 21, + "review_unit_index": 31, + "review_unit_count": 56, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-473", + "nodes": [ + "issue:27957", + "issue:29127", + "issue:29942", + "issue:30064", + "issue:33357", + "issue:35141", + "issue:42490", + "issue:42915", + "issue:43065", + "issue:43066", + "issue:43334", + "issue:43381", + "issue:43421", + "issue:43526", + "issue:43531", + "issue:43575", + "issue:43653", + "issue:43701", + "issue:43756", + "issue:43824", + "issue:43931", + "issue:43994", + "issue:44186", + "issue:44206", + "issue:44265", + "issue:44279", + "issue:44387", + "issue:44442", + "issue:44462", + "issue:44479", + "issue:45072", + "issue:45198", + "issue:45245", + "issue:45250", + "issue:45372", + "issue:45412", + "issue:45440", + "issue:45459", + "issue:45491", + "issue:45563", + "issue:45593", + "issue:45712", + "issue:45834", + "issue:46121", + "issue:46369", + "issue:46399", + "issue:46640", + "issue:46655" + ], + "soft_pairs": [ + "issue:44206|issue:44479", + "issue:45459|issue:45712", + "issue:43065|issue:45412", + "issue:29942|issue:30064", + "issue:43065|issue:43994", + "issue:43756|issue:45440", + "issue:43653|issue:44442", + "issue:45372|issue:45459", + "issue:45412|issue:46369", + "issue:45245|issue:45563", + "issue:29127|issue:33357", + "issue:44265|issue:45198", + "issue:43066|issue:44462", + "issue:43994|issue:45412", + "issue:27957|issue:29942", + "issue:45440|issue:45593", + "issue:43575|issue:43701", + "issue:43421|issue:45491", + "issue:29127|issue:35141", + "issue:42490|issue:43381", + "issue:43526|issue:45834", + "issue:43575|issue:44387", + "issue:44442|issue:45072", + "issue:42915|issue:43931", + "issue:43994|issue:46369", + "issue:43526|issue:46655", + "issue:43526|issue:44186", + "issue:43824|issue:44279", + "issue:46121|issue:46399", + "issue:45250|issue:46640", + "issue:27957|issue:29127", + "issue:43334|issue:43531" + ], + "prepared_review_unit_hash": "68c19df0be18907903005f8681cc4b99d7c3be2e012a7b2aef0b6d01fd8dec88", + "component_budget": { + "node_count": 473, + "item_count": 473, + "soft_pair_count": 1738, + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 32, + "serialized_chars": 37148, + "estimated_input_tokens": 9287, + "estimated_eval_tokens": 18830 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Most soft matches are broad-topic similarities rather than true duplicates. The only noticeable mini-theme is Flash Attention 2, but even there the issues describe different concrete failures, so I would not merge any soft pair.", + "confidence": 0.88, + "canonical_issue_reason": "issue:29942 is the clearest actionable Flash Attention 2 bug report in this set and the best representative of the only loosely related subgroup.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:29942 is the most representative issue here because it is concrete and broad enough to serve as a reference for the FA2-related subset, even though the overall cluster is heterogeneous.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44206", + "right": "issue:44479", + "accept": false, + "reason": "Different failures: unsupported feature-processor arg vs Qwen video-input regression; same area, not the same bug." + }, + { + "left": "issue:45459", + "right": "issue:45712", + "accept": false, + "reason": "Tokenizer error masking from missing protobuf is unrelated to leftover dummy classes leaking into exports." + }, + { + "left": "issue:43065", + "right": "issue:45412", + "accept": false, + "reason": "Dummy conv in Sam3PixelDecoder and RT-DETR memory leak are unrelated code paths and symptoms." + }, + { + "left": "issue:29942", + "right": "issue:30064", + "accept": false, + "reason": "Flash Attention 2 test failures and void segmentation-map processing are unrelated issues." + }, + { + "left": "issue:43065", + "right": "issue:43994", + "accept": false, + "reason": "Sam3PixelDecoder dummy layer vs SigLIP2 bad outputs are different model bugs." + }, + { + "left": "issue:43756", + "right": "issue:45440", + "accept": false, + "reason": "Different model implementations and failure modes; no shared concrete bug." + }, + { + "left": "issue:43653", + "right": "issue:44442", + "accept": false, + "reason": "BigBirdTokenizer special-token registration and FastSpeech2ConformerTokenizer loading are separate tokenizer issues." + }, + { + "left": "issue:45372", + "right": "issue:45459", + "accept": false, + "reason": "Dependency import failure in Gemma 4 processor loading is unrelated to protobuf error masking in tokenizers." + }, + { + "left": "issue:45412", + "right": "issue:46369", + "accept": false, + "reason": "Both mention memory, but one is a deletion/GC leak and the other is an overflow when loading a model; not the same bug." + }, + { + "left": "issue:45245", + "right": "issue:45563", + "accept": false, + "reason": "Category-limit runtime error and stale paged-generate warning are unrelated." + }, + { + "left": "issue:29127", + "right": "issue:33357", + "accept": false, + "reason": "LayoutLMv3 box validation error and MacOS bus error on CLIP are unrelated." + }, + { + "left": "issue:44265", + "right": "issue:45198", + "accept": false, + "reason": "torch.export failure with torch_compilable_check is unrelated to Wav2Vec2 save/tokenization failure." + }, + { + "left": "issue:43066", + "right": "issue:44462", + "accept": false, + "reason": "Wrong tokenizer decoder type in v5 and AutoTokenizer ignoring tokenizer.json are different loading/configuration bugs." + }, + { + "left": "issue:43994", + "right": "issue:45412", + "accept": false, + "reason": "SigLIP2 nonsensical outputs and RT-DETR memory release issues are unrelated." + }, + { + "left": "issue:27957", + "right": "issue:29942", + "accept": false, + "reason": "Both touch Flash Attention 2, but one is XLMRoberta-specific support and the other is FA2 test failures; not the same concrete bug." + }, + { + "left": "issue:45440", + "right": "issue:45593", + "accept": false, + "reason": "DeepseekV3 divergence and D-FINE auxiliary-loss behavior are different issues." + }, + { + "left": "issue:43575", + "right": "issue:43701", + "accept": false, + "reason": "Tensor-parallel OOM when loading a model and resume-from-checkpoint key mismatch are unrelated." + }, + { + "left": "issue:43421", + "right": "issue:45491", + "accept": false, + "reason": "Tokenizer post-processor update behavior and Gemma3 NaN embeddings come from different layers of the stack." + }, + { + "left": "issue:29127", + "right": "issue:35141", + "accept": false, + "reason": "Error-message clarity for LayoutLMv3 boxes and embedding reinitialization after resize are unrelated." + }, + { + "left": "issue:42490", + "right": "issue:43381", + "accept": false, + "reason": "Import warning for trailing-slash paths and gradient-checkpointing eval-mode restriction are unrelated." + }, + { + "left": "issue:43526", + "right": "issue:45834", + "accept": false, + "reason": "BEiT image label reduction bug and Kosmos2.5 OCR index error are different processor problems." + }, + { + "left": "issue:43575", + "right": "issue:44387", + "accept": false, + "reason": "Both involve OOM, but the causes differ: tensor-parallel loading vs increased reserved memory under int4 quantization." + }, + { + "left": "issue:44442", + "right": "issue:45072", + "accept": false, + "reason": "Tokenizer loading failure and bfloat16 dtype mismatch are unrelated." + }, + { + "left": "issue:42915", + "right": "issue:43931", + "accept": false, + "reason": "FineGrainedFP8Config failure for Qwen3Moe and Qwen3-VL weight-shape mismatch are different issues." + }, + { + "left": "issue:43994", + "right": "issue:46369", + "accept": false, + "reason": "SigLIP2 output quality bug and Sam3-LiteText memory overflow are unrelated." + }, + { + "left": "issue:43526", + "right": "issue:46655", + "accept": false, + "reason": "BEiTImageProcessorFast label reduction and SmolVLM video padding are unrelated processor bugs." + }, + { + "left": "issue:43526", + "right": "issue:44186", + "accept": false, + "reason": "Different components and failure modes: BEiT image label reduction vs LayoutLMv2 tokenizer crashes." + }, + { + "left": "issue:43824", + "right": "issue:44279", + "accept": false, + "reason": "Missing Qwen2.5-VL import and a generic dependency issue are not the same concrete problem." + }, + { + "left": "issue:46121", + "right": "issue:46399", + "accept": false, + "reason": "RoPE-validation TypeError and Qwen3.5 kernel registration crash are unrelated." + }, + { + "left": "issue:45250", + "right": "issue:46640", + "accept": false, + "reason": "Both are vague Flash Attention 2 reports, but the titles do not identify the same underlying bug or code path." + }, + { + "left": "issue:27957", + "right": "issue:29127", + "accept": false, + "reason": "XLMRoberta Flash Attention 2 support and LayoutLMv3 box error are unrelated." + }, + { + "left": "issue:43334", + "right": "issue:43531", + "accept": false, + "reason": "Qwen3-VL missing pad_token_id and Qwen3-MoE sliding_window issue are different model/config 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": 21, + "review_unit_index": 32, + "review_unit_count": 56, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-473", + "nodes": [ + "issue:27957", + "issue:33290", + "issue:34567", + "issue:34689", + "issue:35141", + "issue:42175", + "issue:42490", + "issue:43232", + "issue:43262", + "issue:43316", + "issue:43381", + "issue:43388", + "issue:43493", + "issue:43525", + "issue:43611", + "issue:43646", + "issue:43720", + "issue:43723", + "issue:43749", + "issue:43761", + "issue:43827", + "issue:43873", + "issue:43931", + "issue:44062", + "issue:44077", + "issue:44079", + "issue:44188", + "issue:44206", + "issue:44442", + "issue:44464", + "issue:44479", + "issue:44743", + "issue:44811", + "issue:44871", + "issue:44898", + "issue:44977", + "issue:45003", + "issue:45216", + "issue:45245", + "issue:45440", + "issue:45479", + "issue:45491", + "issue:45588", + "issue:45600", + "issue:45859", + "issue:45907", + "issue:46234" + ], + "soft_pairs": [ + "issue:27957|issue:33290", + "issue:43761|issue:45859", + "issue:33290|issue:34689", + "issue:33290|issue:35141", + "issue:45588|issue:45600", + "issue:33290|issue:34567", + "issue:43262|issue:44206", + "issue:43381|issue:43873", + "issue:44977|issue:45245", + "issue:44464|issue:45491", + "issue:43646|issue:44077", + "issue:44442|issue:44898", + "issue:44743|issue:44811", + "issue:42490|issue:44743", + "issue:45003|issue:45440", + "issue:34689|issue:43611", + "issue:43720|issue:44442", + "issue:42175|issue:46234", + "issue:43723|issue:45479", + "issue:43316|issue:43525", + "issue:43493|issue:43827", + "issue:44077|issue:44479", + "issue:43761|issue:44079", + "issue:44062|issue:45907", + "issue:43873|issue:44871", + "issue:43761|issue:44188", + "issue:43232|issue:43388", + "issue:43749|issue:43931", + "issue:43931|issue:45216" + ], + "prepared_review_unit_hash": "df76f2e05279feb25d0e73e35b876f57a88c4fd019cf04a90d5b24076d24e387", + "component_budget": { + "node_count": 473, + "item_count": 473, + "soft_pair_count": 1738, + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 29, + "serialized_chars": 35916, + "estimated_input_tokens": 8979, + "estimated_eval_tokens": 18214 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a grab-bag of unrelated transformer bugs/regressions; none of the soft-linked pairs look like true duplicates. I would anchor the cluster on the most concrete regression report, but not merge any of the candidate pairs.", + "confidence": 0.23, + "canonical_issue_reason": "No single issue is a clear duplicate canonical for the whole set, but issue #43761 is the strongest anchor because it describes a specific, reproducible v5 regression in one code path with a clear symptom and model target.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue #43761 is the most triage-friendly representative: specific regression, explicit failure mode, and narrow scope. Many others are either broader, more speculative, or belong to different subsystems.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:27957", + "right": "issue:33290", + "accept": false, + "reason": "Different problems: Flash Attention 2 support for XLM-Roberta vs Deepspeed+Adafactor OOM." + }, + { + "left": "issue:43761", + "right": "issue:45859", + "accept": false, + "reason": "Both are regressions, but they hit different models and different paths: CLIPVisionModel output states vs Qwen3.5 CPU offload." + }, + { + "left": "issue:33290", + "right": "issue:34689", + "accept": false, + "reason": "OOM during optimizer/deepspeed use is unrelated to model loading breakage for Llama 3.2 Vision." + }, + { + "left": "issue:33290", + "right": "issue:35141", + "accept": false, + "reason": "Optimizer memory issue vs embedding resize/post_init reinitialization bug." + }, + { + "left": "issue:45588", + "right": "issue:45600", + "accept": false, + "reason": "One is a flash_attention sinkless-model crash; the other is stale auto_mappings references breaking CI." + }, + { + "left": "issue:33290", + "right": "issue:34567", + "accept": false, + "reason": "Training memory OOM is unrelated to TrainerState token-count tracking." + }, + { + "left": "issue:43262", + "right": "issue:44206", + "accept": false, + "reason": "Audio processor sampling-rate default mismatch is unrelated to a feature extractor center-arg crash." + }, + { + "left": "issue:43381", + "right": "issue:43873", + "accept": false, + "reason": "Gradient-checkpointing eval-mode incompatibility is not the same as quantization offloading failure." + }, + { + "left": "issue:44977", + "right": "issue:45245", + "accept": false, + "reason": "Qwen3.5 flash-attention generation bug is unrelated to the 2^24 category-count limit error." + }, + { + "left": "issue:44464", + "right": "issue:45491", + "accept": false, + "reason": "Compiled-forward chunked generation inconsistency is unrelated to Gemma3 mixed-length NaN embeddings." + }, + { + "left": "issue:43646", + "right": "issue:44077", + "accept": false, + "reason": "Custom model initialization breakage is not the same as patchtsmixer optional post_init validation." + }, + { + "left": "issue:44442", + "right": "issue:44898", + "accept": false, + "reason": "Tokenizer loading failure is unrelated to Perceiver non-default-resolution image classification." + }, + { + "left": "issue:44743", + "right": "issue:44811", + "accept": false, + "reason": "Qwen3.5 recurrent-state reset bug is unrelated to Whisper batch_decode skip_special_tokens handling." + }, + { + "left": "issue:42490", + "right": "issue:44743", + "accept": false, + "reason": "Local-path import warning does not match the Qwen3.5 cache/recurrent-state bug." + }, + { + "left": "issue:45003", + "right": "issue:45440", + "accept": false, + "reason": "Unsafe sys.modules access is unrelated to DeepseekV3MoE divergence from remote implementation." + }, + { + "left": "issue:34689", + "right": "issue:43611", + "accept": false, + "reason": "Llama 3.2 Vision model-loading regression is not the same as base_model_prefix loading support in v5." + }, + { + "left": "issue:43720", + "right": "issue:44442", + "accept": false, + "reason": "BitNet packed-weight loading bug is unrelated to FastSpeech2ConformerTokenizer auto-loading." + }, + { + "left": "issue:42175", + "right": "issue:46234", + "accept": false, + "reason": "Backend packaging/install issue is unrelated to missing import_utils symbol for falcon_mamba_ssm." + }, + { + "left": "issue:43723", + "right": "issue:45479", + "accept": false, + "reason": "Tokenizer loading in v5 is unrelated to single-label classification zero-loss behavior." + }, + { + "left": "issue:43316", + "right": "issue:43525", + "accept": false, + "reason": "Gemma3TextConfig API discrepancy is unrelated to Llama4Config missing pad_token_id." + }, + { + "left": "issue:43493", + "right": "issue:43827", + "accept": false, + "reason": "SigLIP2 implementation mismatch is unrelated to docs still referencing removed pipeline()." + }, + { + "left": "issue:44077", + "right": "issue:44479", + "accept": false, + "reason": "Patchtsmixer post_init restriction is unrelated to video-input regression in Qwen VL models." + }, + { + "left": "issue:43761", + "right": "issue:44079", + "accept": false, + "reason": "CLIPVisionModel hidden_states regression is unrelated to ModelOutput None-key assignment." + }, + { + "left": "issue:44062", + "right": "issue:45907", + "accept": false, + "reason": "AddedToken keyword conflict is unrelated to list_repo_files not forwarding revision." + }, + { + "left": "issue:43873", + "right": "issue:44871", + "accept": false, + "reason": "Quantization offloading issue is unrelated to Gemma-3 eos_token_id config mismatch." + }, + { + "left": "issue:43761", + "right": "issue:44188", + "accept": false, + "reason": "These are both runtime regressions, but one is hidden_states propagation and the other is attention-kernel divergence under torch.compile." + }, + { + "left": "issue:43232", + "right": "issue:43388", + "accept": false, + "reason": "Generation kwargs update after sync_gpus is unrelated to gather_for_metrics dropping tuple labels." + }, + { + "left": "issue:43749", + "right": "issue:43931", + "accept": false, + "reason": "FSDP CPU RAM efficient loading breakage is unrelated to Qwen3-VL weight-shape mismatch." + }, + { + "left": "issue:43931", + "right": "issue:45216", + "accept": false, + "reason": "Model load shape mismatch is unrelated to Qwen3.5 save_pretrained checkpoint corruption." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 21, + "review_unit_index": 33, + "review_unit_count": 56, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-473", + "nodes": [ + "issue:16998", + "issue:22355", + "issue:25251", + "issue:27957", + "issue:28282", + "issue:32090", + "issue:42175", + "issue:42617", + "issue:42907", + "issue:43295", + "issue:43316", + "issue:43452", + "issue:43550", + "issue:43575", + "issue:43606", + "issue:43643", + "issue:43646", + "issue:43701", + "issue:43716", + "issue:43742", + "issue:43825", + "issue:44077", + "issue:44162", + "issue:44190", + "issue:44442", + "issue:44464", + "issue:44496", + "issue:44534", + "issue:44610", + "issue:44792", + "issue:44855", + "issue:44871", + "issue:44913", + "issue:44987", + "issue:45137", + "issue:45584", + "issue:45698", + "issue:45759", + "issue:45803", + "issue:46050", + "issue:46178", + "issue:46234", + "issue:46291", + "issue:46365", + "issue:46369", + "issue:46421", + "issue:46506" + ], + "soft_pairs": [ + "issue:44987|issue:46178", + "issue:43316|issue:45759", + "issue:44855|issue:45137", + "issue:43643|issue:44913", + "issue:44442|issue:45803", + "issue:43701|issue:46506", + "issue:44190|issue:44496", + "issue:42907|issue:45698", + "issue:44464|issue:46421", + "issue:42617|issue:43575", + "issue:25251|issue:28282", + "issue:27957|issue:32090", + "issue:43550|issue:43701", + "issue:43575|issue:43742", + "issue:46234|issue:46291", + "issue:43606|issue:43701", + "issue:43295|issue:43825", + "issue:16998|issue:22355", + "issue:43295|issue:44871", + "issue:44871|issue:45584", + "issue:43316|issue:43716", + "issue:42175|issue:43316", + "issue:44077|issue:44534", + "issue:45803|issue:46365", + "issue:44871|issue:46050", + "issue:43646|issue:44913", + "issue:44610|issue:46369", + "issue:43452|issue:44496", + "issue:44162|issue:44792" + ], + "prepared_review_unit_hash": "83eeef5be26049c0ebfa198a669e92fc98ec893b1819a7c6810078ce14e3a340", + "component_budget": { + "node_count": 473, + "item_count": 473, + "soft_pair_count": 1738, + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 29, + "serialized_chars": 35635, + "estimated_input_tokens": 8909, + "estimated_eval_tokens": 18074 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "These issues are mostly unrelated Transformers bugs spanning config loading, generation, backend availability, and model-specific regressions. I do not see any soft pair that is the same underlying bug/change, so the cluster should not be merged as duplicates. If a single anchor is needed, issue 45698 is the broadest and most active open loading-related report, but it is not a true duplicate of the others.", + "confidence": 0.84, + "canonical_issue_reason": "No true duplicate anchor is apparent across this set; issue 45698 is the broadest and most active open report about save_pretrained/from_pretrained custom-module loading, so it is the least-bad cluster anchor if one must be chosen.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 45698 has the widest scope, is still open, and has the strongest discussion/reference signal among the listed issues, making it the best representative issue for this noisy loading-regression cluster.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44987", + "right": "issue:46178", + "accept": false, + "reason": "Different failures: one is model loading under transformers>=5.1.0, the other is an invalid backend name for tensorflow_text." + }, + { + "left": "issue:43316", + "right": "issue:45759", + "accept": false, + "reason": "Both touch multimodal config APIs, but the concrete problems differ: Gemma3TextConfig API mismatch vs Qwen composite text_config unwrapping." + }, + { + "left": "issue:44855", + "right": "issue:45137", + "accept": false, + "reason": "Completely different code paths: Python 3.13 TorchScript parsing vs DeepSpeed ZeRO3 deque handling." + }, + { + "left": "issue:43643", + "right": "issue:44913", + "accept": false, + "reason": "Both are config persistence issues, but one is trust_remote_code field loss and the other is GPTNeoX rotary_pct not reloading." + }, + { + "left": "issue:44442", + "right": "issue:45803", + "accept": false, + "reason": "Unrelated tokenizer loading failure vs a bare-except bug in Fuyu tensor conversion." + }, + { + "left": "issue:43701", + "right": "issue:46506", + "accept": false, + "reason": "Checkpoint resume key mismatch is unrelated to PPDocLayoutV3 CPU/CUDA result differences." + }, + { + "left": "issue:44190", + "right": "issue:44496", + "accept": false, + "reason": "Different loading problems: local dataset handling in a training script vs missing model_type in config.json." + }, + { + "left": "issue:42907", + "right": "issue:45698", + "accept": false, + "reason": "Both involve save/load, but one is dequantized model saving and the other is wrong custom-module resolution after save_pretrained." + }, + { + "left": "issue:44464", + "right": "issue:46421", + "accept": false, + "reason": "Both mention chunked generation/compiled execution, but one is output inconsistency and the other is lazy cache initialization causing recompilation." + }, + { + "left": "issue:42617", + "right": "issue:43575", + "accept": false, + "reason": "3d_parallel.py failure and Qwen2 tp OOM are different runtime/resource issues." + }, + { + "left": "issue:25251", + "right": "issue:28282", + "accept": false, + "reason": "Pipeline top_k nesting bug is unrelated to missing PyTorch import support." + }, + { + "left": "issue:27957", + "right": "issue:32090", + "accept": false, + "reason": "Flash Attention 2 model support and Trainer _gpu_broadcast_one NoneType error are different failures." + }, + { + "left": "issue:43550", + "right": "issue:43701", + "accept": false, + "reason": "torch.compile/SDPA model bug is unrelated to checkpoint resume key mismatch." + }, + { + "left": "issue:43575", + "right": "issue:43742", + "accept": false, + "reason": "OOM on a large Qwen model and a key error loading MobileLLM are different loading issues." + }, + { + "left": "issue:46234", + "right": "issue:46291", + "accept": false, + "reason": "Missing import_utils symbol and kernels version/revision import failure are distinct dependency/API problems." + }, + { + "left": "issue:43606", + "right": "issue:43701", + "accept": false, + "reason": "CPU offload device mismatch does not match the checkpoint resume key mismatch bug." + }, + { + "left": "issue:43295", + "right": "issue:43825", + "accept": false, + "reason": "Custom processor/tokenizer regression is unrelated to a v5 pipeline error-message bug." + }, + { + "left": "issue:16998", + "right": "issue:22355", + "accept": false, + "reason": "model_max_length on DeBERTa-V3 and missing transformers.onnx are unrelated concerns." + }, + { + "left": "issue:43295", + "right": "issue:44871", + "accept": false, + "reason": "One is processor.tokenizer access regression, the other is eos_token_id config inconsistency in Gemma-3." + }, + { + "left": "issue:44871", + "right": "issue:45584", + "accept": false, + "reason": "Both involve special-token behavior, but one is Gemma-3 eos_token_id mismatch and the other is Whisper empty-transcription generation after align_special_tokens." + }, + { + "left": "issue:43316", + "right": "issue:43716", + "accept": false, + "reason": "Gemma3TextConfig API mismatch and Mistral-3 image-preprocessor dtype mismatch are different multimodal issues." + }, + { + "left": "issue:42175", + "right": "issue:43316", + "accept": false, + "reason": "Backend packaging omission with pip install '.[torch]' is unrelated to Gemma3TextConfig API behavior." + }, + { + "left": "issue:44077", + "right": "issue:44534", + "accept": false, + "reason": "Optional post_init restriction for patchtsmixer is unrelated to non-persistent buffers being filled with junk in v5." + }, + { + "left": "issue:45803", + "right": "issue:46365", + "accept": false, + "reason": "Bare except swallowing errors in Fuyu is unrelated to importing Trainer conflicting with datasets." + }, + { + "left": "issue:44871", + "right": "issue:46050", + "accept": false, + "reason": "Gemma-3 eos_token_id mismatch is not the same as quantized model save/load breakage after sinq." + }, + { + "left": "issue:43646", + "right": "issue:44913", + "accept": false, + "reason": "Custom model initialization breakage in v5 is different from GPTNeoX rotary_pct not persisting on reload." + }, + { + "left": "issue:44610", + "right": "issue:46369", + "accept": false, + "reason": "OmDet-Turbo preprocessing size mismatch and sam3-litetext memory overflow are unrelated." + }, + { + "left": "issue:43452", + "right": "issue:44496", + "accept": false, + "reason": "gguf_file loading breakage and missing model_type in config.json are separate model-loading failures." + }, + { + "left": "issue:44162", + "right": "issue:44792", + "accept": false, + "reason": "ESM2 breakage and a janus image-generation test failure are unrelated model-specific 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": 21, + "review_unit_index": 34, + "review_unit_count": 56, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-473", + "nodes": [ + "issue:35141", + "issue:41950", + "issue:42490", + "issue:43065", + "issue:43352", + "issue:43388", + "issue:43404", + "issue:43441", + "issue:43475", + "issue:43504", + "issue:43525", + "issue:43531", + "issue:43540", + "issue:43572", + "issue:43650", + "issue:43673", + "issue:43688", + "issue:43761", + "issue:44190", + "issue:44242", + "issue:44479", + "issue:44514", + "issue:44534", + "issue:44617", + "issue:44734", + "issue:44743", + "issue:44792", + "issue:44855", + "issue:44913", + "issue:44964", + "issue:44977", + "issue:45127", + "issue:45200", + "issue:45406", + "issue:45542", + "issue:45584", + "issue:45588", + "issue:45600", + "issue:45632", + "issue:45698", + "issue:45803", + "issue:46170", + "issue:46178", + "issue:46202", + "issue:46234", + "issue:46424", + "issue:46617", + "issue:46640" + ], + "soft_pairs": [ + "issue:45632|issue:46424", + "issue:45803|issue:46170", + "issue:43650|issue:46202", + "issue:43404|issue:43441", + "issue:43065|issue:44617", + "issue:43540|issue:44514", + "issue:43761|issue:44479", + "issue:45127|issue:45200", + "issue:43572|issue:46178", + "issue:43388|issue:43688", + "issue:45542|issue:46234", + "issue:43352|issue:46640", + "issue:43504|issue:45698", + "issue:43673|issue:44743", + "issue:44792|issue:44977", + "issue:44792|issue:45600", + "issue:41950|issue:43475", + "issue:43525|issue:43531", + "issue:44734|issue:45406", + "issue:44190|issue:44964", + "issue:44534|issue:44913", + "issue:35141|issue:42490", + "issue:44617|issue:45584", + "issue:45632|issue:46617", + "issue:43761|issue:44242", + "issue:44855|issue:45588" + ], + "prepared_review_unit_hash": "46f4ac241fdffe7c689eb46c54b47449c4174f5a7c9d024a9532ec676bc53625", + "component_budget": { + "node_count": 473, + "item_count": 473, + "soft_pair_count": 1738, + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 26, + "serialized_chars": 35524, + "estimated_input_tokens": 8881, + "estimated_eval_tokens": 18018 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is highly heterogeneous: the soft edges mostly share only broad subsystem keywords (loading, cache, flash attention, video, etc.) but do not describe the same underlying bug or a mergeable combined fix. I would keep the issues separate; issue:45698 is the best anchor only because it is the broadest and most referenced loading regression in the set.", + "confidence": 0.94, + "canonical_issue_reason": "issue:45698 is the most central/general item here: it concerns a broad `from_pretrained`/`save_pretrained` loading regression, has the strongest discussion/inbound activity, and is the best representative anchor if one issue must be chosen.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:45698 is the best overall issue candidate because it is active, broadly scoped, and more likely to serve as a canonical reference than the narrower model-specific or unrelated bugs.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:45632", + "right": "issue:46424", + "accept": false, + "reason": "Both are cache-related, but one is a trust_remote_code path collision and the other is max_cache_length behavior for quantized cache; different bugs." + }, + { + "left": "issue:45803", + "right": "issue:46170", + "accept": false, + "reason": "Different failure modes: one is a bare-except in tensor conversion, the other is a Windows/ROCm torch.distributed import dependency issue." + }, + { + "left": "issue:43650", + "right": "issue:46202", + "accept": false, + "reason": "Completely unrelated: placeholder text body vs Gemma pipeline roundtrip/tokenization failure." + }, + { + "left": "issue:43404", + "right": "issue:43441", + "accept": false, + "reason": "One is weight tying in Mistral3; the other is FlashAttention 2 support for Ministral-3. Same family, different bug." + }, + { + "left": "issue:43065", + "right": "issue:44617", + "accept": false, + "reason": "Sam3PixelDecoder dummy Conv2d is an architecture/implementation issue, while Sam3Video CUDA OOM is a memory usage problem." + }, + { + "left": "issue:43540", + "right": "issue:44514", + "accept": false, + "reason": "Both touch Qwen3 models, but one is video-input processing and the other is chat template batching/padding." + }, + { + "left": "issue:43761", + "right": "issue:44479", + "accept": false, + "reason": "CLIPVisionModel hidden_states regression is unrelated to the Qwen video input regression." + }, + { + "left": "issue:45127", + "right": "issue:45200", + "accept": false, + "reason": "LoRA merge with extended vocabulary and Gemma 4 mm_token_type_ids defaulting to zeros are different model-behavior bugs." + }, + { + "left": "issue:43572", + "right": "issue:46178", + "accept": false, + "reason": "StableLmConfig pad_token_idx regression is unrelated to a missing tensorflow_text backend mapping." + }, + { + "left": "issue:43388", + "right": "issue:43688", + "accept": false, + "reason": "Both involve training utilities/losses, but one drops labels in gather_for_metrics and the other misnormalizes auxiliary loss." + }, + { + "left": "issue:45542", + "right": "issue:46234", + "accept": false, + "reason": "Different integration problems: missing TensorFlow backend handling vs missing import_utils symbol." + }, + { + "left": "issue:43352", + "right": "issue:46640", + "accept": false, + "reason": "One is a specific model not supporting Flash Attention 2 yet; the other is a vague generic flash-attention issue without the same concrete code path." + }, + { + "left": "issue:43504", + "right": "issue:45698", + "accept": false, + "reason": "Legacy field on Beit preset loading is unrelated to wrong custom-module selection after save_pretrained." + }, + { + "left": "issue:43673", + "right": "issue:44743", + "accept": false, + "reason": "Both mention cache, but one is GenerationMixin cache missing in chunked_prefill while the other resets recurrent states in modular_qwen3_5 with seq_len>1." + }, + { + "left": "issue:44792", + "right": "issue:44977", + "accept": false, + "reason": "Janus image-generation test failure is unrelated to Qwen3.5 flash-attention generation issues." + }, + { + "left": "issue:44792", + "right": "issue:45600", + "accept": false, + "reason": "A failed Janus test case is unrelated to auto_mappings referencing removed Sam3LiteText configs." + }, + { + "left": "issue:41950", + "right": "issue:43475", + "accept": false, + "reason": "Video-classification processor lookup and missing fpn_position_embeddings in Sam3Video are different SAM/video bugs." + }, + { + "left": "issue:43525", + "right": "issue:43531", + "accept": false, + "reason": "Missing pad_token_id in Llama4Config and Qwen3-MoE sliding_window behavior are unrelated config/model issues." + }, + { + "left": "issue:44734", + "right": "issue:45406", + "accept": false, + "reason": "KV-cache continuation indexing in serve is unrelated to Gemma4Processor missing `_tokenizer`." + }, + { + "left": "issue:44190", + "right": "issue:44964", + "accept": false, + "reason": "Local dataset loading in an example script is unrelated to loading a Phi-4 multimodal model." + }, + { + "left": "issue:44534", + "right": "issue:44913", + "accept": false, + "reason": "Non-persistent buffer initialization and GPTNeoXConfig rotary_pct reload persistence are different regression classes." + }, + { + "left": "issue:35141", + "right": "issue:42490", + "accept": false, + "reason": "Token-embedding resizing/tie_word_embedding bug is unrelated to the local-path trailing-slash import warning." + }, + { + "left": "issue:44617", + "right": "issue:45584", + "accept": false, + "reason": "Sam3Video CUDA OOM is a resource issue, while Whisper empty-transcription failure after align_special_tokens is a generation/tokenization bug." + }, + { + "left": "issue:45632", + "right": "issue:46617", + "accept": false, + "reason": "Both involve trust_remote_code and cache, but one is a leaf-directory collision and the other is a FileNotFoundError when loading from cache." + }, + { + "left": "issue:43761", + "right": "issue:44242", + "accept": false, + "reason": "Hidden_states regression in CLIPVisionModel is unrelated to missing load balancing loss when output_router_logits=False." + }, + { + "left": "issue:44855", + "right": "issue:45588", + "accept": false, + "reason": "Python 3.13 indentation/parsing failure in DebertaV2Model is unrelated to a flash_attention.py AttributeError on sink-less models." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 21, + "review_unit_index": 35, + "review_unit_count": 56, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-473", + "nodes": [ + "issue:25251", + "issue:27957", + "issue:39692", + "issue:41720", + "issue:43065", + "issue:43295", + "issue:43335", + "issue:43540", + "issue:43550", + "issue:43644", + "issue:43673", + "issue:43827", + "issue:43931", + "issue:43994", + "issue:44060", + "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:44855", + "issue:44877", + "issue:44913", + "issue:44933", + "issue:44938", + "issue:45127", + "issue:45245", + "issue:45276", + "issue:45357", + "issue:45362", + "issue:45381", + "issue:45542", + "issue:45563", + "issue:45584", + "issue:45600", + "issue:45803", + "issue:45907", + "issue:46369", + "issue:46421", + "issue:46489", + "issue:46655" + ], + "soft_pairs": [ + "issue:43931|issue:44060", + "issue:43827|issue:43994", + "issue:43673|issue:46421", + "issue:45563|issue:45584", + "issue:45381|issue:46369", + "issue:44479|issue:44734", + "issue:43644|issue:44913", + "issue:44933|issue:45362", + "issue:43065|issue:44610", + "issue:25251|issue:27957", + "issue:43335|issue:44855", + "issue:41720|issue:45245", + "issue:43550|issue:44190", + "issue:44077|issue:44877", + "issue:44464|issue:45563", + "issue:45381|issue:46655", + "issue:43335|issue:44190", + "issue:44617|issue:46369", + "issue:45127|issue:45357", + "issue:44779|issue:46489", + "issue:39692|issue:44190", + "issue:44112|issue:45600", + "issue:44521|issue:45381", + "issue:43540|issue:44734", + "issue:44737|issue:45803", + "issue:43295|issue:45276", + "issue:44938|issue:45542", + "issue:45276|issue:45584", + "issue:44462|issue:45907" + ], + "prepared_review_unit_hash": "28abaff7cd1effc8647d14b771b0eb2193dc7f77730622423ef5e8abe0c00923", + "component_budget": { + "node_count": 473, + "item_count": 473, + "soft_pair_count": 1738, + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 29, + "serialized_chars": 36331, + "estimated_input_tokens": 9083, + "estimated_eval_tokens": 18422 + }, + "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. The only plausible duplicate pair is the DeepSeek tokenizer regression reports (44779 and 46489); the rest differ by model, subsystem, or failure mode and should stay separate.", + "confidence": 0.87, + "canonical_issue_reason": "issue:46489 is the most specific and actionable report in the only real duplicate pair: it narrows the problem to DeepSeek-Coder v1 tokenizer output on v5+ and explicitly references a gap in PR #44801's fix. issue:44779 is the broader sibling report.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:46489 is the best representative issue overall because it has the clearest scope, an explicit link to the prior fix, and a concrete regression description.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43931", + "right": "issue:44060", + "accept": false, + "reason": "Different models and different bugs: Qwen3-VL weight shape mismatch vs Qwen3-Next tied-weights warning." + }, + { + "left": "issue:43827", + "right": "issue:43994", + "accept": false, + "reason": "Both mention pipeline/vision workflows, but one is docs about removed pipelines and the other is an AutoModel/pipeline output correctness problem for SigLIP2." + }, + { + "left": "issue:43673", + "right": "issue:46421", + "accept": false, + "reason": "Both involve chunked prefill/cache, but one is a missing GenerationMixin cache bug and the other is lazy cache init causing recompilation; not the same code path." + }, + { + "left": "issue:45563", + "right": "issue:45584", + "accept": false, + "reason": "Unrelated failures: a stale warning in paged generate vs Whisper empty-transcription handling after align_special_tokens." + }, + { + "left": "issue:45381", + "right": "issue:46369", + "accept": false, + "reason": "Different models and symptoms: Qwen2.5-VL video position ids vs Sam3-LiteText memory overflow." + }, + { + "left": "issue:44479", + "right": "issue:44734", + "accept": false, + "reason": "Both are generation-related, but one is a Qwen video-input regression and the other is a server-side KV-cache indexing crash." + }, + { + "left": "issue:43644", + "right": "issue:44913", + "accept": false, + "reason": "One is a framework-wide non-persistent buffer initialization bug; the other is a GPTNeoX config reload persistence issue." + }, + { + "left": "issue:44933", + "right": "issue:45362", + "accept": false, + "reason": "Completely different problems: missing image_utils import vs Qwen3.5 chat crash." + }, + { + "left": "issue:43065", + "right": "issue:44610", + "accept": false, + "reason": "Both touch vision models, but one is a dummy Conv2d in Sam3PixelDecoder and the other is a processor/model image-size mismatch in OmDet-Turbo." + }, + { + "left": "issue:25251", + "right": "issue:27957", + "accept": false, + "reason": "No underlying overlap: pipeline top_k nesting bug vs XLM-RoBERTa Flash Attention 2 support." + }, + { + "left": "issue:43335", + "right": "issue:44855", + "accept": false, + "reason": "Different failure modes and code paths: SwitchTransformers config creates sparse layers incorrectly vs a Python 3.13 IndentationError in DebertaV2Model." + }, + { + "left": "issue:41720", + "right": "issue:45245", + "accept": false, + "reason": "Unrelated runtime failures: Qwen3 auto device mapping cudaErrorAssert vs category-count limit in data processing." + }, + { + "left": "issue:43550", + "right": "issue:44190", + "accept": false, + "reason": "Different subsystems and bugs: torch.compile/SDPA issue in Bamba vs dataset loading in an image classification example." + }, + { + "left": "issue:44077", + "right": "issue:44877", + "accept": false, + "reason": "Both concern config handling, but they are different models and distinct validation issues." + }, + { + "left": "issue:44464", + "right": "issue:45563", + "accept": false, + "reason": "Chunked generation inconsistency under compilation is unrelated to a stale warning about num_return_sequences." + }, + { + "left": "issue:45381", + "right": "issue:46655", + "accept": false, + "reason": "Both are video-input bugs, but the models and defect types differ: wrong vision_position_ids vs incorrect padding calculation." + }, + { + "left": "issue:43335", + "right": "issue:44190", + "accept": false, + "reason": "No meaningful overlap: SwitchTransformers sparse-layer config bug vs dataset loading script failure." + }, + { + "left": "issue:44617", + "right": "issue:46369", + "accept": false, + "reason": "Both mention memory pressure, but they affect different Sam3 variants and likely different causes." + }, + { + "left": "issue:45127", + "right": "issue:45357", + "accept": false, + "reason": "Different issues in different workflows: LoRA merge with extended vocab vs save_pretrained visual encoder key handling." + }, + { + "left": "issue:44779", + "right": "issue:46489", + "accept": true, + "reason": "Both describe the same DeepSeek tokenizer regression on v5+: incorrect outputs compared with v4, with 46489 explicitly calling out the remaining gap in the prior fix." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet, and the soft-edge verdicts stay conservative. The DeepSeek tokenizer pair (44779/46489) is the only clearly duplicate-looking match from the provided titles, while the other pairs are reasonably rejected as different models, subsystems, or failure modes." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 21, + "review_unit_index": 36, + "review_unit_count": 56, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-473", + "nodes": [ + "issue:27957", + "issue:32090", + "issue:35141", + "issue:36246", + "issue:38617", + "issue:41628", + "issue:41950", + "issue:42175", + "issue:42796", + "issue:43295", + "issue:43404", + "issue:43479", + "issue:43526", + "issue:43531", + "issue:43643", + "issue:43650", + "issue:43673", + "issue:43854", + "issue:43901", + "issue:43976", + "issue:44016", + "issue:44112", + "issue:44186", + "issue:44220", + "issue:44246", + "issue:44279", + "issue:44295", + "issue:44496", + "issue:44521", + "issue:44683", + "issue:44734", + "issue:44964", + "issue:44987", + "issue:45003", + "issue:45083", + "issue:45216", + "issue:45245", + "issue:45335", + "issue:45375", + "issue:45446", + "issue:45584", + "issue:45600", + "issue:45632", + "issue:46082", + "issue:46121", + "issue:46234", + "issue:46424", + "issue:46506" + ], + "soft_pairs": [ + "issue:45632|issue:46234", + "issue:44246|issue:45003", + "issue:42796|issue:44521", + "issue:46082|issue:46121", + "issue:42175|issue:44279", + "issue:45245|issue:45584", + "issue:44964|issue:45600", + "issue:43854|issue:44964", + "issue:44112|issue:44964", + "issue:43643|issue:44496", + "issue:27957|issue:35141", + "issue:44186|issue:44734", + "issue:36246|issue:41628", + "issue:43404|issue:45216", + "issue:44295|issue:46506", + "issue:36246|issue:38617", + "issue:43650|issue:44220", + "issue:41950|issue:44016", + "issue:43479|issue:45375", + "issue:32090|issue:35141", + "issue:43673|issue:46424", + "issue:43531|issue:45600", + "issue:44016|issue:44220", + "issue:45083|issue:46506", + "issue:43295|issue:45335", + "issue:43526|issue:43901", + "issue:44683|issue:45446", + "issue:43976|issue:44987" + ], + "prepared_review_unit_hash": "c9cb3e6141a9065d9230acae04aa49374ff680c3558096a31fb275e1c50320eb", + "component_budget": { + "node_count": 473, + "item_count": 473, + "soft_pair_count": 1738, + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 28, + "serialized_chars": 36129, + "estimated_input_tokens": 9033, + "estimated_eval_tokens": 18322 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This set is mostly a grab bag of unrelated closed issues across imports, multimodal configs, tokenizers, cache behavior, CI, and docs. The soft links are only superficial keyword/subsystem overlaps, so none look like true duplicates or mergeable PR-equivalents.", + "confidence": 0.94, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": "Issue 44964 is the strongest standalone bug report here: it describes a concrete, reproducible model-loading failure with a clear target and user impact. It is not a duplicate hub for the rest of the set, but it is the best representative issue if one must be chosen.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:45632", + "right": "issue:46234", + "accept": false, + "reason": "Both touch trust_remote_code/import utilities, but one is a cache-path collision and the other is a missing availability helper; different bugs." + }, + { + "left": "issue:44246", + "right": "issue:45003", + "accept": false, + "reason": "Both mention import behavior, but one is slow imports and the other is unsafe sys.modules access; not the same failure mode." + }, + { + "left": "issue:42796", + "right": "issue:44521", + "accept": false, + "reason": "Tokenizer BPE-marker handling and multimodal assistant_masks are unrelated bugs in different code paths." + }, + { + "left": "issue:46082", + "right": "issue:46121", + "accept": false, + "reason": "One is LlamaConfig head_dim validation, the other is rope-parameter JSON list handling; distinct configuration bugs." + }, + { + "left": "issue:42175", + "right": "issue:44279", + "accept": false, + "reason": "Backend packaging dependency mismatch vs a vague dependency issue; no shared concrete code-path problem." + }, + { + "left": "issue:45245", + "right": "issue:45584", + "accept": false, + "reason": "Category-count limit and Whisper empty transcription are unrelated runtime issues." + }, + { + "left": "issue:44964", + "right": "issue:45600", + "accept": false, + "reason": "Model loading failure for Phi-4 multimodal vs stale auto_mappings references; different underlying causes." + }, + { + "left": "issue:43854", + "right": "issue:44964", + "accept": false, + "reason": "Both are model-loading failures, but they involve different models and likely different loader problems." + }, + { + "left": "issue:44112", + "right": "issue:44964", + "accept": false, + "reason": "A CI test failure in GraniteSpeech is not the same as Phi-4 multimodal load regression." + }, + { + "left": "issue:43643", + "right": "issue:44496", + "accept": false, + "reason": "Both are config/load errors, but one drops fields with trust_remote_code while the other is an unrecognized model_type; not duplicates." + }, + { + "left": "issue:27957", + "right": "issue:35141", + "accept": false, + "reason": "Flash Attention compatibility and embedding reinitialization after resize_token_embeddings are different bugs." + }, + { + "left": "issue:44186", + "right": "issue:44734", + "accept": false, + "reason": "Tokenizer padding/truncation crash vs serve KV-cache continuation indexing bug; different paths and symptoms." + }, + { + "left": "issue:36246", + "right": "issue:41628", + "accept": false, + "reason": "Missing Qwen2_5_VLImageProcessor export vs missing AutoImageProcessor import are related only at a high level, not the same bug." + }, + { + "left": "issue:43404", + "right": "issue:45216", + "accept": false, + "reason": "Weight tying in Mistral3ForConditionalGeneration and save_pretrained checkpoint correctness for Qwen3.5 are separate model-specific issues." + }, + { + "left": "issue:44295", + "right": "issue:46506", + "accept": false, + "reason": "Position_ids buffer access error and CPU/CUDA output mismatch in PPDocLayoutV3 are unrelated correctness problems." + }, + { + "left": "issue:36246", + "right": "issue:38617", + "accept": false, + "reason": "Missing model export and missing configuration_utils symbol are different import/export regressions." + }, + { + "left": "issue:43650", + "right": "issue:44220", + "accept": false, + "reason": "Placeholder text issue and fbank feature extraction bug are not the same underlying problem." + }, + { + "left": "issue:41950", + "right": "issue:44016", + "accept": false, + "reason": "Video pipeline resolving image processors and a notebook syntax error are entirely unrelated." + }, + { + "left": "issue:43479", + "right": "issue:45375", + "accept": false, + "reason": "Different multimodal config classes and different missing/default fields; no shared bug." + }, + { + "left": "issue:32090", + "right": "issue:35141", + "accept": false, + "reason": "Trainer broadcast TypeError and embedding reinitialization after resize are unrelated." + }, + { + "left": "issue:43673", + "right": "issue:46424", + "accept": false, + "reason": "Both involve cache, but one is a missing GenerationMixin cache in v5 and the other is max_cache_length handling for Quantized Cache only." + }, + { + "left": "issue:43531", + "right": "issue:45600", + "accept": false, + "reason": "Qwen3-MoE sliding_window behavior and stale auto_mappings references are different issues." + }, + { + "left": "issue:44016", + "right": "issue:44220", + "accept": false, + "reason": "Notebook syntax error and audio feature-extraction bug do not share a code-path." + }, + { + "left": "issue:45083", + "right": "issue:46506", + "accept": false, + "reason": "Helper length computation in qwen3_omni_moe vs CPU/CUDA divergence in PPDocLayoutV3 are unrelated." + }, + { + "left": "issue:43295", + "right": "issue:45335", + "accept": false, + "reason": "Processor/tokenizer regression and untied decoder.embed_tokens after resize are different problems." + }, + { + "left": "issue:43526", + "right": "issue:43901", + "accept": false, + "reason": "ImageProcessorFast reduce_labels bug and stale docs for return_all_scores are unrelated." + }, + { + "left": "issue:44683", + "right": "issue:45446", + "accept": false, + "reason": "Both are flex_attention-related, but one is a torch>=2.9 compilation failure and the other is a version-check/import guard bug; not the same fix." + }, + { + "left": "issue:43976", + "right": "issue:44987", + "accept": false, + "reason": "A Python-version compatibility issue and a specific model-loading failure on 5.1.0 are too different to merge." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 21, + "review_unit_index": 37, + "review_unit_count": 56, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-473", + "nodes": [ + "issue:38175", + "issue:39692", + "issue:41950", + "issue:42796", + "issue:42907", + "issue:43295", + "issue:43299", + "issue:43381", + "issue:43441", + "issue:43577", + "issue:43650", + "issue:43673", + "issue:43824", + "issue:43901", + "issue:43976", + "issue:44016", + "issue:44062", + "issue:44075", + "issue:44242", + "issue:44279", + "issue:44336", + "issue:44351", + "issue:44368", + "issue:44496", + "issue:44617", + "issue:44779", + "issue:44792", + "issue:44933", + "issue:45020", + "issue:45042", + "issue:45200", + "issue:45216", + "issue:45278", + "issue:45335", + "issue:45357", + "issue:45405", + "issue:45584", + "issue:45588", + "issue:45593", + "issue:45759", + "issue:45812", + "issue:46082", + "issue:46123", + "issue:46369", + "issue:46399", + "issue:46424", + "issue:46512", + "issue:46515" + ], + "soft_pairs": [ + "issue:44496|issue:45042", + "issue:42907|issue:43441", + "issue:43901|issue:44496", + "issue:43381|issue:45593", + "issue:43976|issue:44279", + "issue:43673|issue:44336", + "issue:43441|issue:45588", + "issue:46123|issue:46369", + "issue:44016|issue:44062", + "issue:43824|issue:44933", + "issue:44792|issue:46424", + "issue:44496|issue:45759", + "issue:43299|issue:45216", + "issue:44279|issue:46515", + "issue:45335|issue:45584", + "issue:46082|issue:46399", + "issue:44279|issue:45405", + "issue:44075|issue:44368", + "issue:42796|issue:43650", + "issue:43577|issue:45216", + "issue:39692|issue:43901", + "issue:43295|issue:44779", + "issue:43976|issue:45405", + "issue:44351|issue:45278", + "issue:46082|issue:46512", + "issue:43299|issue:45357", + "issue:44242|issue:45200", + "issue:44779|issue:45812", + "issue:41950|issue:45020", + "issue:38175|issue:44617" + ], + "prepared_review_unit_hash": "15f37f944dd2babb6f9e11ffacce1098a64f683e64760865951fd68e3489baba", + "component_budget": { + "node_count": 473, + "item_count": 473, + "soft_pair_count": 1738, + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 30, + "serialized_chars": 36568, + "estimated_input_tokens": 9142, + "estimated_eval_tokens": 18540 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This set is a loose cluster of Transformers regression/model-specific bugs, not a duplicate set. The closest thematic umbrella is recent-version breakage and import/loading regressions, but the individual reports mostly target different code paths and models.", + "confidence": 0.82, + "canonical_issue_reason": "issue:45278 is the broadest representative of the cluster: it captures multiple import failures after upgrading Transformers, which matches several other version-regression reports better than the more model-specific items.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:45020 is the best single issue for a cluster label because it is the most general statement of the shared theme: recent Transformers versions breaking model loading, especially with remote code.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44496", + "right": "issue:45042", + "accept": false, + "reason": "Both are model-loading failures, but one is an unrecognized-model/config issue and the other is PIL image processor requiring torchvision; different root causes and code paths." + }, + { + "left": "issue:42907", + "right": "issue:43441", + "accept": false, + "reason": "Both mention Ministral, but one is about saving dequantized models and the other is FlashAttention initialization; not the same underlying bug." + }, + { + "left": "issue:43901", + "right": "issue:44496", + "accept": false, + "reason": "One is a docs mismatch for TextClassificationPipeline, the other is an unrecognized-model config error; unrelated problems." + }, + { + "left": "issue:43381", + "right": "issue:45593", + "accept": false, + "reason": "Gradient checkpointing in eval mode and missing auxiliary losses when denoising is off are different training-path bugs." + }, + { + "left": "issue:43976", + "right": "issue:44279", + "accept": false, + "reason": "Python version compatibility and a generic dependency issue are too vague to be the same concrete bug." + }, + { + "left": "issue:43673", + "right": "issue:44336", + "accept": false, + "reason": "Chunked prefill cache handling and ANSI codes in loading_report are unrelated." + }, + { + "left": "issue:43441", + "right": "issue:45588", + "accept": false, + "reason": "FlashAttention failure for Ministral and a sink-less-model crash in a different flash_attention integration path are not the same bug." + }, + { + "left": "issue:46123", + "right": "issue:46369", + "accept": false, + "reason": "Both concern memory/performance symptoms in SAM3-related work, but one is a dropped-results batching bug and the other is a memory overflow report." + }, + { + "left": "issue:44016", + "right": "issue:44062", + "accept": false, + "reason": "A notebook syntax error and an AddedToken keyword collision are distinct issues." + }, + { + "left": "issue:43824", + "right": "issue:44933", + "accept": false, + "reason": "Both are import-related, but one is a missing Qwen2_5_VL symbol and the other is a nonexistent image_utils import; different symbols and fixes." + }, + { + "left": "issue:44792", + "right": "issue:46424", + "accept": false, + "reason": "A janus image-generation test failure and a cache config max length bug are unrelated." + }, + { + "left": "issue:44496", + "right": "issue:45759", + "accept": false, + "reason": "Both involve model/config loading, but one is an unrecognized OLMo model and the other is unwrapping text_config for composite Qwen multimodal configs; different underlying problems." + }, + { + "left": "issue:43299", + "right": "issue:45216", + "accept": false, + "reason": "Qwen3VL MoE loading failure and Qwen3.5 save_pretrained regression are different code paths." + }, + { + "left": "issue:44279", + "right": "issue:46515", + "accept": false, + "reason": "A generic dependency issue and a torch 2.4.1 incompatibility report are too nonspecific to merge as one bug." + }, + { + "left": "issue:45335", + "right": "issue:45584", + "accept": false, + "reason": "resize_token_embeddings not affecting decoder embeddings and Whisper empty-transcription failure after token alignment are unrelated." + }, + { + "left": "issue:46082", + "right": "issue:46399", + "accept": false, + "reason": "A LlamaConfig head_dim validation issue and a missing decorator for apply_rotary_pos_emb registration are different failures." + }, + { + "left": "issue:44279", + "right": "issue:45405", + "accept": false, + "reason": "A generic dependency issue and a MIN_PEFT_VERSION bump are not the same bug; one is about requirements, the other about release availability." + }, + { + "left": "issue:44075", + "right": "issue:44368", + "accept": false, + "reason": "SGD args not being used and a tie_word_embeddings warning in Qwen3.5 fine-tuning are unrelated." + }, + { + "left": "issue:42796", + "right": "issue:43650", + "accept": false, + "reason": "Ministral tokenizer BPE-marker handling and the placeholder 'ADD THE DATA' issue do not describe the same technical problem." + }, + { + "left": "issue:43577", + "right": "issue:45216", + "accept": false, + "reason": "Blip2 dtype propagation and Qwen3.5 incorrect save_pretrained checkpointing are different model-specific bugs." + }, + { + "left": "issue:39692", + "right": "issue:43901", + "accept": false, + "reason": "SigLIP2 doc example errors and TextClassificationPipeline docs mentioning return_all_scores are both documentation-related, but they are separate examples and behaviors." + }, + { + "left": "issue:43295", + "right": "issue:44779", + "accept": false, + "reason": "A regression around processor.tokenizer/images handling is not the same as Deepseek tokenizer output corruption." + }, + { + "left": "issue:43976", + "right": "issue:45405", + "accept": false, + "reason": "Python version support and unreleased PEFT minimum version are different compatibility problems." + }, + { + "left": "issue:44351", + "right": "issue:45278", + "accept": false, + "reason": "Missing HybridCache import and broad import errors after upgrade are both import-related, but not the same missing symbol or fix." + }, + { + "left": "issue:46082", + "right": "issue:46512", + "accept": false, + "reason": "Explicit head_dim validation and stale model.config.dtype after from_config(dtype=...) are unrelated configuration bugs." + }, + { + "left": "issue:43299", + "right": "issue:45357", + "accept": false, + "reason": "Qwen3VL loading failure and incorrect visual encoder keys on save_pretrained are separate multimodal model regressions." + }, + { + "left": "issue:44242", + "right": "issue:45200", + "accept": false, + "reason": "Load balancing loss gating and Gemma 4 mm_token_type_ids defaults are different training semantics bugs." + }, + { + "left": "issue:44779", + "right": "issue:45812", + "accept": false, + "reason": "Deepseek tokenizer regression and Granite AutoTokenizer wrong token IDs are both tokenizer issues, but they affect different models and likely different code paths." + }, + { + "left": "issue:41950", + "right": "issue:45020", + "accept": false, + "reason": "A video-classification pipeline processor lookup bug and broad remote_code model breakage are not the same bug." + }, + { + "left": "issue:38175", + "right": "issue:44617", + "accept": false, + "reason": "Unexpected zero probabilities in SigLIP2 and SAM3 video CUDA OOM are completely different model/runtime issues." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 21, + "review_unit_index": 38, + "review_unit_count": 56, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-473", + "nodes": [ + "issue:33290", + "issue:41950", + "issue:42175", + "issue:42994", + "issue:43299", + "issue:43317", + "issue:43441", + "issue:43479", + "issue:43525", + "issue:43650", + "issue:43653", + "issue:43673", + "issue:43723", + "issue:43761", + "issue:43854", + "issue:43901", + "issue:43976", + "issue:43994", + "issue:44016", + "issue:44162", + "issue:44261", + "issue:44263", + "issue:44291", + "issue:44361", + "issue:44462", + "issue:44466", + "issue:44521", + "issue:44623", + "issue:44671", + "issue:44683", + "issue:44749", + "issue:44779", + "issue:45081", + "issue:45103", + "issue:45276", + "issue:45325", + "issue:45372", + "issue:45397", + "issue:45542", + "issue:45593", + "issue:45600", + "issue:45656", + "issue:45712", + "issue:45800", + "issue:45823", + "issue:45834", + "issue:45907", + "issue:46209" + ], + "soft_pairs": [ + "issue:43299|issue:43854", + "issue:43479|issue:43525", + "issue:43650|issue:46209", + "issue:41950|issue:45823", + "issue:44466|issue:45276", + "issue:44683|issue:45542", + "issue:44521|issue:45325", + "issue:45103|issue:45372", + "issue:43653|issue:45593", + "issue:44462|issue:45081", + "issue:44016|issue:44291", + "issue:43441|issue:45834", + "issue:33290|issue:45656", + "issue:44361|issue:45593", + "issue:45081|issue:45907", + "issue:42994|issue:43317", + "issue:44261|issue:44263", + "issue:44623|issue:44749", + "issue:45600|issue:45712", + "issue:44623|issue:45397", + "issue:42175|issue:43976", + "issue:44671|issue:44779", + "issue:43901|issue:43994", + "issue:43441|issue:44683", + "issue:43673|issue:43761", + "issue:45542|issue:45800", + "issue:43723|issue:44162" + ], + "prepared_review_unit_hash": "298c8000ec09cdfdabaa3272bfdca135dad91c0dab865c83594cf781e98df003", + "component_budget": { + "node_count": 473, + "item_count": 473, + "soft_pair_count": 1738, + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 27, + "serialized_chars": 35825, + "estimated_input_tokens": 8957, + "estimated_eval_tokens": 18170 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "These issues are largely unrelated single-bug reports across different models, configs, tokenizer paths, and environment problems; I don\u2019t see any true duplicate pairs worth merging.", + "confidence": 0.69, + "canonical_issue_reason": "issue:44521 is the strongest standalone report here: it has a clear regression description, reproducer focus, and active discussion/inbound references, making it the best representative issue in this set.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:44521 is the most actionable and well-scoped issue among a mostly unrelated set, with enough detail to serve as a reference point if needed.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43299", + "right": "issue:43854", + "accept": false, + "reason": "Both are model-loading failures, but they concern different models and different failure modes (Qwen3VL MoE vs GLM-4.7-Flash unit-test loading)." + }, + { + "left": "issue:43479", + "right": "issue:43525", + "accept": false, + "reason": "Different config bugs: one misinitializes Phi4MultimodalConfig subconfigs, the other lacks pad_token_id on Llama4Config." + }, + { + "left": "issue:43650", + "right": "issue:46209", + "accept": false, + "reason": "Unrelated topics: a vague data request versus an FP8 inference failure on B200." + }, + { + "left": "issue:41950", + "right": "issue:45823", + "accept": false, + "reason": "Different subsystems and symptoms: pipeline processor selection vs Gemma4 device-map forward mismatch." + }, + { + "left": "issue:44466", + "right": "issue:45276", + "accept": false, + "reason": "Both involve Gemma/v5 model state handling, but one is serialization of tied lm_head weights and the other is resize_token_embeddings behavior." + }, + { + "left": "issue:44683", + "right": "issue:45542", + "accept": false, + "reason": "Completely different bugs: torch>=2.9 flex_attention compilation vs a TensorFlow backend dependency/install error." + }, + { + "left": "issue:44521", + "right": "issue:45325", + "accept": false, + "reason": "Both are multimodal/VL related, but one is chat-template assistant masks and the other is Qwen2.5-VL rope/position-id scaling." + }, + { + "left": "issue:45103", + "right": "issue:45372", + "accept": false, + "reason": "Different failure classes: auto_docstring annotation handling vs Gemma processor loading due to a missing mistral_common import." + }, + { + "left": "issue:43653", + "right": "issue:45593", + "accept": false, + "reason": "Tokenizer special-token registration bug versus a training-loss issue in D-FINE with denoising disabled." + }, + { + "left": "issue:44462", + "right": "issue:45081", + "accept": false, + "reason": "Both are tokenizer-related, but one is repository tokenizer discovery and the other is a Mistral regex patch crash on a different code path." + }, + { + "left": "issue:44016", + "right": "issue:44291", + "accept": false, + "reason": "Notebook syntax error and init_empty_weights argument mismatch are unrelated." + }, + { + "left": "issue:43441", + "right": "issue:45834", + "accept": false, + "reason": "Different models and bugs: FlashAttention failure for Ministral-3 versus OCR index error in Kosmos2.5." + }, + { + "left": "issue:33290", + "right": "issue:45656", + "accept": false, + "reason": "Both mention deepspeed/optimizer behavior, but one is Adafactor OOM and the other is optimizer.step being called twice." + }, + { + "left": "issue:44361", + "right": "issue:45593", + "accept": false, + "reason": "Tokenizer AttributeError on tasks is unrelated to D-FINE auxiliary-loss behavior." + }, + { + "left": "issue:45081", + "right": "issue:45907", + "accept": false, + "reason": "Different issues: a tokenizer regex patch crash versus repo file listing not forwarding revision." + }, + { + "left": "issue:42994", + "right": "issue:43317", + "accept": false, + "reason": "Both touch quantization/dequantization, but one is saving quantized models and the other is loading a dequantized model with device_map=auto." + }, + { + "left": "issue:44261", + "right": "issue:44263", + "accept": false, + "reason": "Distinct bugs in different components: MLA q_a_layernorm config precision vs GlmMoeDsaIndexer split return values." + }, + { + "left": "issue:44623", + "right": "issue:44749", + "accept": false, + "reason": "Processor save_pretrained file persistence is unrelated to the Chinese report about slower filtering after upgrade." + }, + { + "left": "issue:45600", + "right": "issue:45712", + "accept": false, + "reason": "Both are CI/repo-integrity related, but one is stale auto_mappings references and the other is leftover dummy_pt_objects classes." + }, + { + "left": "issue:44623", + "right": "issue:45397", + "accept": false, + "reason": "Different code paths: processor.save_pretrained missing files versus Gemma-4 zero3 from_pretrained." + }, + { + "left": "issue:42175", + "right": "issue:43976", + "accept": false, + "reason": "Different environment problems: missing TensorFlow in a torch install versus Python version incompatibility in Transformers 5.1.0." + }, + { + "left": "issue:44671", + "right": "issue:44779", + "accept": false, + "reason": "Both are v5 regressions in model outputs, but they affect different models and likely different tokenization/modeling paths." + }, + { + "left": "issue:43901", + "right": "issue:43994", + "accept": false, + "reason": "Docs mismatch for return_all_scores is not the same bug as nonsensical SigLIP2 model/pipeline outputs." + }, + { + "left": "issue:43441", + "right": "issue:44683", + "accept": false, + "reason": "Different failure modes: Ministral FlashAttention compatibility versus compiled flex_attention failing on newer torch." + }, + { + "left": "issue:43673", + "right": "issue:43761", + "accept": false, + "reason": "Unrelated regressions: chunked_prefill cache handling versus CLIPVisionModel hidden_states not returned." + }, + { + "left": "issue:45542", + "right": "issue:45800", + "accept": false, + "reason": "TensorBoard/TensorFlow backend install issue is unrelated to torch/Transformers version incompatibility." + }, + { + "left": "issue:43723", + "right": "issue:44162", + "accept": false, + "reason": "AutoTokenizer loading regression and ESM2 model breakage are different problems with no shared concrete code path." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 21, + "review_unit_index": 39, + "review_unit_count": 56, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-473", + "nodes": [ + "issue:36246", + "issue:41628", + "issue:41950", + "issue:42491", + "issue:42757", + "issue:42886", + "issue:43299", + "issue:43525", + "issue:43526", + "issue:43531", + "issue:43540", + "issue:43550", + "issue:43643", + "issue:43784", + "issue:43824", + "issue:43881", + "issue:43901", + "issue:43976", + "issue:44016", + "issue:44188", + "issue:44220", + "issue:44242", + "issue:44246", + "issue:44295", + "issue:44297", + "issue:44336", + "issue:44373", + "issue:44485", + "issue:44496", + "issue:44589", + "issue:44871", + "issue:45003", + "issue:45103", + "issue:45362", + "issue:45397", + "issue:45542", + "issue:45561", + "issue:45563", + "issue:45593", + "issue:45823", + "issue:46132", + "issue:46291", + "issue:46399", + "issue:46424", + "issue:46506", + "issue:46614", + "issue:46655", + "issue:46665" + ], + "soft_pairs": [ + "issue:43901|issue:44246", + "issue:46291|issue:46399", + "issue:43525|issue:43881", + "issue:44016|issue:45823", + "issue:43824|issue:43976", + "issue:44188|issue:44336", + "issue:43525|issue:43643", + "issue:44589|issue:45362", + "issue:44336|issue:44373", + "issue:44297|issue:45397", + "issue:43784|issue:45542", + "issue:41628|issue:43976", + "issue:44246|issue:44496", + "issue:45561|issue:45563", + "issue:46614|issue:46665", + "issue:44242|issue:45593", + "issue:45103|issue:46399", + "issue:42491|issue:43299", + "issue:43526|issue:45003", + "issue:44295|issue:45397", + "issue:43531|issue:45362", + "issue:44246|issue:44871", + "issue:44336|issue:44485", + "issue:42886|issue:46132", + "issue:43550|issue:46506", + "issue:44336|issue:46424", + "issue:41950|issue:44220", + "issue:43540|issue:46655", + "issue:36246|issue:42757" + ], + "prepared_review_unit_hash": "b09e42712fb5202ed33e9ebdf4975d9d17d03fa9e7fa1ddad7339ed113de4e26", + "component_budget": { + "node_count": 473, + "item_count": 473, + "soft_pair_count": 1738, + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 29, + "serialized_chars": 36300, + "estimated_input_tokens": 9075, + "estimated_eval_tokens": 18406 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is thematically related (mostly Transformers/HF bug reports), but the soft pairs are not true duplicates: they usually differ by model family, API surface, or failure mode. No PRs are present, so no PR canonical applies.", + "confidence": 0.93, + "canonical_issue_reason": "issue:41628 is the broadest, most generic user-facing failure here: a top-level Transformers import error. It best represents the cluster\u2019s import/API breakage theme without overfitting to a single model family.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:41628 is the strongest exemplar for triage because it is concise, high-signal, and broadly applicable across the import-related subset of reports.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43901", + "right": "issue:44246", + "accept": false, + "reason": "Docs mismatch for text classification versus intermittent slow import; unrelated symptoms and code paths." + }, + { + "left": "issue:46291", + "right": "issue:46399", + "accept": false, + "reason": "Different failures: pipeline import conflict with kernels versioning vs Qwen3.5 kernel registration error." + }, + { + "left": "issue:43525", + "right": "issue:43881", + "accept": false, + "reason": "Llama4 missing config field and GLM-4V loading failure are different model-specific bugs." + }, + { + "left": "issue:44016", + "right": "issue:45823", + "accept": false, + "reason": "Notebook syntax error and Gemma4 device-mapping bug are unrelated." + }, + { + "left": "issue:43824", + "right": "issue:43976", + "accept": false, + "reason": "Qwen2.5 VL import error versus Python-version compatibility issue; not the same bug." + }, + { + "left": "issue:44188", + "right": "issue:44336", + "accept": false, + "reason": "torch.compile attention-kernel divergence is unrelated to ANSI output in loading_report." + }, + { + "left": "issue:43525", + "right": "issue:43643", + "accept": false, + "reason": "Missing pad_token_id in Llama4Config is unrelated to trust_remote_code returning incomplete AutoConfig fields." + }, + { + "left": "issue:44589", + "right": "issue:45362", + "accept": false, + "reason": "Storage type lookup failure and Qwen3.5 chat crash are different runtime errors." + }, + { + "left": "issue:44336", + "right": "issue:44373", + "accept": false, + "reason": "Terminal ANSI formatting bug and wrong docstring are unrelated." + }, + { + "left": "issue:44297", + "right": "issue:45397", + "accept": false, + "reason": "Tokenizer config mismatch and Gemma-4 ZeRO3 loading failure are distinct issues." + }, + { + "left": "issue:43784", + "right": "issue:45542", + "accept": false, + "reason": "sentence-transformers import NameError and undefined tf backend from tensorboard-only install are different dependency bugs." + }, + { + "left": "issue:41628", + "right": "issue:43976", + "accept": false, + "reason": "Top-level import error versus Python support regression; not the same underlying defect." + }, + { + "left": "issue:44246", + "right": "issue:44871", + "accept": false, + "reason": "Slow import and Gemma eos_token_id inconsistency are unrelated." + }, + { + "left": "issue:45561", + "right": "issue:45563", + "accept": false, + "reason": "pytest-xdist race in testing utils versus stale warning in paged generate are different code paths." + }, + { + "left": "issue:46614", + "right": "issue:46665", + "accept": false, + "reason": "Stale phonemizer language state and from_pretrained behavior inside Python classes are unrelated." + }, + { + "left": "issue:44242", + "right": "issue:45593", + "accept": false, + "reason": "Load-balancing loss omission in MoE routing is not the same as D-FINE auxiliary-loss behavior." + }, + { + "left": "issue:45103", + "right": "issue:46399", + "accept": false, + "reason": "auto_docstring annotation crash and Qwen3.5 kernel registration bug are unrelated." + }, + { + "left": "issue:42491", + "right": "issue:43299", + "accept": false, + "reason": "Both involve Qwen/MoE, but one is LoRA compatibility across branches and the other is a model loading regression; different bugs." + }, + { + "left": "issue:43526", + "right": "issue:45003", + "accept": false, + "reason": "BeitImageProcessorFast reduce_labels bug and sys.modules access in modeling_utils are unrelated." + }, + { + "left": "issue:44295", + "right": "issue:45397", + "accept": false, + "reason": "position_ids buffer read error and Gemma-4 ZeRO3 loading failure are different failures." + }, + { + "left": "issue:43531", + "right": "issue:45362", + "accept": false, + "reason": "Qwen3-MoE sliding_window issue and Qwen3.5 chat crash are distinct model behaviors." + }, + { + "left": "issue:44246", + "right": "issue:44871", + "accept": false, + "reason": "Performance complaint about import time and eos_token_id config mismatch do not share the same underlying bug." + }, + { + "left": "issue:44336", + "right": "issue:44485", + "accept": false, + "reason": "ANSI code emission in loading_report and GLM-5 RoPE implementation are unrelated." + }, + { + "left": "issue:42886", + "right": "issue:46132", + "accept": false, + "reason": "Offline cache loading bug and AttentionInterface.register behavior change are different subsystems." + }, + { + "left": "issue:43550", + "right": "issue:46506", + "accept": false, + "reason": "torch.compile SDPA failure in Bamba and CPU/CUDA output divergence in PPDocLayoutV3 are not the same bug." + }, + { + "left": "issue:44336", + "right": "issue:46424", + "accept": false, + "reason": "Terminal ANSI formatting and cache max length semantics are unrelated." + }, + { + "left": "issue:41950", + "right": "issue:44220", + "accept": false, + "reason": "video-classification pipeline image-processor lookup and fbank feature extraction are different media-processing paths." + }, + { + "left": "issue:43540", + "right": "issue:46655", + "accept": false, + "reason": "Qwen3OmniMoe video-input processing error and SmolVLM padding bug are both video-related but clearly different defects." + }, + { + "left": "issue:36246", + "right": "issue:42757", + "accept": false, + "reason": "Qwen2.5 VL import failure and huggingface_hub is_offline_mode import failure are different package/API breaks." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 21, + "review_unit_index": 40, + "review_unit_count": 56, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-473", + "nodes": [ + "issue:36246", + "issue:43257", + "issue:43262", + "issue:43352", + "issue:43475", + "issue:43596", + "issue:43673", + "issue:43901", + "issue:44038", + "issue:44060", + "issue:44077", + "issue:44162", + "issue:44242", + "issue:44246", + "issue:44295", + "issue:44351", + "issue:44466", + "issue:44479", + "issue:44496", + "issue:44561", + "issue:44734", + "issue:44749", + "issue:44779", + "issue:44857", + "issue:44877", + "issue:44987", + "issue:45003", + "issue:45137", + "issue:45200", + "issue:45250", + "issue:45292", + "issue:45335", + "issue:45375", + "issue:45412", + "issue:45459", + "issue:45561", + "issue:45584", + "issue:45600", + "issue:45701", + "issue:45759", + "issue:45859", + "issue:46123", + "issue:46170", + "issue:46369", + "issue:46399", + "issue:46424", + "issue:46506", + "issue:46655" + ], + "soft_pairs": [ + "issue:36246|issue:44351", + "issue:44779|issue:45292", + "issue:45003|issue:45561", + "issue:45859|issue:46399", + "issue:44734|issue:46424", + "issue:44479|issue:46655", + "issue:43901|issue:45003", + "issue:43257|issue:45137", + "issue:43475|issue:46369", + "issue:44496|issue:45003", + "issue:44242|issue:44857", + "issue:43673|issue:44561", + "issue:44295|issue:44749", + "issue:44162|issue:45200", + "issue:43262|issue:44246", + "issue:45375|issue:45759", + "issue:36246|issue:43596", + "issue:45292|issue:45584", + "issue:44561|issue:46399", + "issue:45292|issue:45701", + "issue:45137|issue:45600", + "issue:44877|issue:45375", + "issue:44038|issue:46506", + "issue:44466|issue:45335", + "issue:44987|issue:45459", + "issue:45412|issue:46123", + "issue:43352|issue:45250", + "issue:44479|issue:46170", + "issue:44060|issue:44077" + ], + "prepared_review_unit_hash": "6daba197ba25058752b1833286e70c71657b3663b925274f2dfa2819e596d0e8", + "component_budget": { + "node_count": 473, + "item_count": 473, + "soft_pair_count": 1738, + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 29, + "serialized_chars": 36472, + "estimated_input_tokens": 9118, + "estimated_eval_tokens": 18492 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This set is heterogeneous: import/export regressions, model-specific bugs, cache issues, docs mismatches, and unrelated runtime failures. None of the soft pairs look like true duplicates of the same underlying bug or change.", + "confidence": 0.88, + "canonical_issue_reason": "No single issue is a defensible canonical for the whole set: the items span unrelated models and subsystems, and the apparent similarities are mostly broad keyword overlap (imports, Qwen, video, cache, Deepspeed, etc.).", + "canonical_pr_reason": null, + "best_issue_reason": "If a representative issue is needed, issue:44479 is the broadest user-facing regression report here, covering multiple Qwen VL variants and a concrete v5.3.0 video-input breakage.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:36246", + "right": "issue:44351", + "accept": false, + "reason": "Both are import errors, but they involve different missing symbols (`Qwen2_5_VLImageProcessor` vs `HybridCache`) and likely different export paths." + }, + { + "left": "issue:44779", + "right": "issue:45292", + "accept": false, + "reason": "Tokenizer output regression and `resize_token_embeddings` not updating output embeddings are different bugs in different code paths." + }, + { + "left": "issue:45003", + "right": "issue:45561", + "accept": false, + "reason": "`sys.modules` access in model loading and a pytest-xdist file race are unrelated failures." + }, + { + "left": "issue:45859", + "right": "issue:46399", + "accept": false, + "reason": "Both touch Qwen3.5, but one is CPU-offload incompatibility in rotary embedding and the other is a missing decorator causing kernel registration failure." + }, + { + "left": "issue:44734", + "right": "issue:46424", + "accept": false, + "reason": "KV-cache continuation crash and cache max-length handling are separate cache features with different symptoms." + }, + { + "left": "issue:44479", + "right": "issue:46655", + "accept": false, + "reason": "Both concern video processing, but 44479 is a broad Qwen VL input regression while 46655 is a SmolVLM padding-calculation bug." + }, + { + "left": "issue:43901", + "right": "issue:45003", + "accept": false, + "reason": "Docs inconsistency for `return_all_scores` is not the same bug as unsafe `sys.modules` access." + }, + { + "left": "issue:43257", + "right": "issue:45137", + "accept": false, + "reason": "Both involve DeepSpeed/zero3-related loading, but one is Qwen3 MOE weight conversion and the other is a deque underflow during init." + }, + { + "left": "issue:43475", + "right": "issue:46369", + "accept": false, + "reason": "Sam3VisionEncoderOutput missing an attribute is a structural output bug; memory overflow in sam3-litetext is a separate resource issue." + }, + { + "left": "issue:44496", + "right": "issue:45003", + "accept": false, + "reason": "Unrecognized model loading and unsafe `sys.modules` access are different failure modes." + }, + { + "left": "issue:44242", + "right": "issue:44857", + "accept": false, + "reason": "Missing load-balancing loss and an AMP/CUDA loss crash are not the same defect." + }, + { + "left": "issue:43673", + "right": "issue:44561", + "accept": false, + "reason": "Chunked prefill cache handling and removal of `is_torch_fx_available` affect different APIs and code paths." + }, + { + "left": "issue:44295", + "right": "issue:44749", + "accept": false, + "reason": "A buffer read error for `position_ids` and a v4-to-v5 slowdown report are unrelated." + }, + { + "left": "issue:44162", + "right": "issue:45200", + "accept": false, + "reason": "ESM2 breakage and Gemma 4 token-type defaults are unrelated model-specific issues." + }, + { + "left": "issue:43262", + "right": "issue:44246", + "accept": false, + "reason": "Audio chat-template sampling-rate defaulting and slow import times are different subsystems." + }, + { + "left": "issue:45375", + "right": "issue:45759", + "accept": false, + "reason": "Missing config field in Qwen3_5MoeVisionConfig and `from_config` not unwrapping `text_config` are distinct composite-config bugs." + }, + { + "left": "issue:36246", + "right": "issue:43596", + "accept": false, + "reason": "ImportError for a VL image processor and a zero3/BertModel index error are unrelated." + }, + { + "left": "issue:45292", + "right": "issue:45584", + "accept": false, + "reason": "Token embedding resize behavior and Whisper empty-transcription generation failure are different bugs." + }, + { + "left": "issue:44561", + "right": "issue:46399", + "accept": false, + "reason": "A v5 trust_remote_code regression from removing `is_torch_fx_available` is not the same as the Qwen3.5 kernel-registration error." + }, + { + "left": "issue:45292", + "right": "issue:45701", + "accept": false, + "reason": "Both mention tokenization/embedding effects, but one is embedding resizing and the other is changed tokenization behavior across versions." + }, + { + "left": "issue:45137", + "right": "issue:45600", + "accept": false, + "reason": "Zero3 deque underflow and stale auto_mappings references are unrelated." + }, + { + "left": "issue:44877", + "right": "issue:45375", + "accept": false, + "reason": "Strict config loading for granite_speech and a missing Qwen3_5MoeVisionConfig field are different config-validation problems." + }, + { + "left": "issue:44038", + "right": "issue:46506", + "accept": false, + "reason": "Qwen3-VL-Moe regression in v5.0 and CPU/CUDA output differences for PPDocLayoutV3 are not the same issue." + }, + { + "left": "issue:44466", + "right": "issue:45335", + "accept": false, + "reason": "Both concern embedding/weight updates, but one is tied-weight serialization and the other is `resize_token_embeddings` not reaching decoder embeddings." + }, + { + "left": "issue:44987", + "right": "issue:45459", + "accept": false, + "reason": "Model loading failure on v5.1+ and tokenizer error masking when protobuf is absent are separate problems." + }, + { + "left": "issue:45412", + "right": "issue:46123", + "accept": false, + "reason": "Memory not released for RT-DETR and final partial batch handling in MaskGenerationPipeline are unrelated." + }, + { + "left": "issue:43352", + "right": "issue:45250", + "accept": false, + "reason": "One is a model-specific Flash Attention 2 unsupported error; the other is just a generic Flash Attention 2 topic, not the same concrete bug." + }, + { + "left": "issue:44479", + "right": "issue:46170", + "accept": false, + "reason": "Video-input regression in Qwen VL models and Windows/ROCm torch.distributed import/type errors are unrelated." + }, + { + "left": "issue:44060", + "right": "issue:44077", + "accept": false, + "reason": "Incorrect tied-weights warning and `patchtsmixer` post_init policy are different issues." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 21, + "review_unit_index": 41, + "review_unit_count": 56, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-473", + "nodes": [ + "issue:42886", + "issue:42994", + "issue:43065", + "issue:43232", + "issue:43408", + "issue:43475", + "issue:43482", + "issue:43540", + "issue:43643", + "issue:43646", + "issue:43653", + "issue:43673", + "issue:43701", + "issue:43906", + "issue:43931", + "issue:44162", + "issue:44188", + "issue:44297", + "issue:44361", + "issue:44368", + "issue:44483", + "issue:44623", + "issue:44749", + "issue:44869", + "issue:44871", + "issue:44877", + "issue:44977", + "issue:44995", + "issue:45200", + "issue:45229", + "issue:45230", + "issue:45356", + "issue:45375", + "issue:45397", + "issue:45478", + "issue:45563", + "issue:45584", + "issue:45588", + "issue:45656", + "issue:45685", + "issue:45987", + "issue:46082", + "issue:46123", + "issue:46132", + "issue:46170", + "issue:46202", + "issue:46421", + "issue:46431" + ], + "soft_pairs": [ + "issue:44995|issue:45200", + "issue:44368|issue:44977", + "issue:43653|issue:44162", + "issue:43475|issue:43540", + "issue:45397|issue:45656", + "issue:42886|issue:45229", + "issue:44877|issue:46170", + "issue:44995|issue:45588", + "issue:44368|issue:45563", + "issue:43065|issue:43408", + "issue:43906|issue:46202", + "issue:43931|issue:44368", + "issue:45685|issue:46431", + "issue:45987|issue:46082", + "issue:44483|issue:46202", + "issue:44188|issue:46202", + "issue:44188|issue:46421", + "issue:45584|issue:46170", + "issue:43065|issue:46123", + "issue:44297|issue:44623", + "issue:44361|issue:44749", + "issue:44749|issue:45656", + "issue:43701|issue:44749", + "issue:44623|issue:45656", + "issue:43646|issue:44162", + "issue:44871|issue:46132", + "issue:43232|issue:43673", + "issue:42994|issue:43482", + "issue:44869|issue:45356", + "issue:45230|issue:45397", + "issue:43475|issue:43643", + "issue:45375|issue:45478" + ], + "prepared_review_unit_hash": "92684c07073cc0f7f8a79df623b195e412e5a9881b34883d1c3829b2c6da6586", + "component_budget": { + "node_count": 473, + "item_count": 473, + "soft_pair_count": 1738, + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 32, + "serialized_chars": 37186, + "estimated_input_tokens": 9297, + "estimated_eval_tokens": 18850 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The set is heterogeneous: tokenizer/save/load regressions, generation/cache issues, model-specific config bugs, and unrelated framework/runtime problems. None of the proposed soft edges looks like the same underlying bug or a mergeable duplicate PR.", + "confidence": 0.93, + "canonical_issue_reason": "No clear canonical issue: the items span unrelated regressions across tokenizers, generation, configs, save/load, and model-specific bugs.", + "canonical_pr_reason": null, + "best_issue_reason": "No single issue is a safe duplicate anchor; the cluster is too mixed to consolidate around one report.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44995", + "right": "issue:45200", + "accept": false, + "reason": "Different failures: stale indexer cache on second forward pass vs Gemma mm_token_type_ids defaulting for fine-tuning." + }, + { + "left": "issue:44368", + "right": "issue:44977", + "accept": false, + "reason": "Unrelated: a warning about tie_word_embeddings vs a flash-attention generation bug in Qwen3.5." + }, + { + "left": "issue:43653", + "right": "issue:44162", + "accept": false, + "reason": "Different models and symptoms: BigBirdTokenizer special-token decode bug vs ESM2 regression." + }, + { + "left": "issue:43475", + "right": "issue:43540", + "accept": false, + "reason": "Both are SAM3/Qwen video-related, but one is a missing attribute and the other is a video-input ValueError; not the same bug." + }, + { + "left": "issue:45397", + "right": "issue:45656", + "accept": false, + "reason": "Different code paths: Gemma zero3 from_pretrained failure vs deepspeed optimizer stepping twice." + }, + { + "left": "issue:42886", + "right": "issue:45229", + "accept": false, + "reason": "Tokenizer cache/offline loading bug is unrelated to Gemma4 multi-GPU OOM." + }, + { + "left": "issue:44877", + "right": "issue:46170", + "accept": false, + "reason": "Strict config loading for granite_speech is unrelated to Windows/ROCm torch.distributed import failures." + }, + { + "left": "issue:44995", + "right": "issue:45588", + "accept": false, + "reason": "Different bugs: stale cache on forward pass vs flash_attention.py crashing on s_aux=None." + }, + { + "left": "issue:44368", + "right": "issue:45563", + "accept": false, + "reason": "A tie_word_embeddings warning is not the same as a stale num_return_sequences warning in paged generate()." + }, + { + "left": "issue:43065", + "right": "issue:43408", + "accept": false, + "reason": "Related SAM3 area, but dummy Conv2d in the decoder is not the same as a model-type warning on instantiation." + }, + { + "left": "issue:43906", + "right": "issue:46202", + "accept": false, + "reason": "Unrelated: isolated reproduction of an old issue vs Gemma pipeline roundtrip failure on \u2581." + }, + { + "left": "issue:43931", + "right": "issue:44368", + "accept": false, + "reason": "Qwen3-VL weight-shape mismatch is unrelated to a tie_word_embeddings warning." + }, + { + "left": "issue:45685", + "right": "issue:46431", + "accept": false, + "reason": "MPS histogram/int error is unrelated to missing chat template in processor.apply_chat_template()." + }, + { + "left": "issue:45987", + "right": "issue:46082", + "accept": false, + "reason": "StaticCache return-type contract bug is unrelated to LlamaConfig head_dim validation." + }, + { + "left": "issue:44483", + "right": "issue:46202", + "accept": false, + "reason": "Chat/completions request rejection is unrelated to Gemma pipeline roundtrip tokenization behavior." + }, + { + "left": "issue:44188", + "right": "issue:46202", + "accept": false, + "reason": "torch.compile attention-kernel divergence is unrelated to Gemma roundtrip decoding on \u2581." + }, + { + "left": "issue:44188", + "right": "issue:46421", + "accept": false, + "reason": "Both mention cache/compile, but one is branching-induced kernel divergence and the other is lazy cache init causing recompilation; different bugs." + }, + { + "left": "issue:45584", + "right": "issue:46170", + "accept": false, + "reason": "Whisper empty-transcription failure is unrelated to Windows/ROCm import issues." + }, + { + "left": "issue:43065", + "right": "issue:46123", + "accept": false, + "reason": "Different SAM3 and pipeline bugs: dummy Conv2d vs is_last never true on the final batch." + }, + { + "left": "issue:44297", + "right": "issue:44623", + "accept": false, + "reason": "Both involve save_pretrained, but tokenizer_class mismatch and missing processor files are different defects." + }, + { + "left": "issue:44361", + "right": "issue:44749", + "accept": false, + "reason": "MLukeTokenizer task AttributeError is unrelated to a performance regression in filtering after upgrade." + }, + { + "left": "issue:44749", + "right": "issue:45656", + "accept": false, + "reason": "Performance slowdown after upgrade is unrelated to double optimizer.step() under deepspeed." + }, + { + "left": "issue:43701", + "right": "issue:44749", + "accept": false, + "reason": "resume_from_checkpoint key mismatch is unrelated to the filtering performance regression." + }, + { + "left": "issue:44623", + "right": "issue:45656", + "accept": false, + "reason": "Missing processor.save_pretrained files is unrelated to deepspeed optimizer double-step behavior." + }, + { + "left": "issue:43646", + "right": "issue:44162", + "accept": false, + "reason": "Custom model initialization breakage in v5 is unrelated to ESM2 being broken." + }, + { + "left": "issue:44871", + "right": "issue:46132", + "accept": false, + "reason": "Gemma EOS config mismatch is unrelated to AttentionInterface.register changing function behavior." + }, + { + "left": "issue:43232", + "right": "issue:43673", + "accept": false, + "reason": "Both generation-related, but one is sync_gpus/update_model_kwargs and the other is missing cache during chunked prefill; not the same bug." + }, + { + "left": "issue:42994", + "right": "issue:43482", + "accept": false, + "reason": "Quantized model saving failure is unrelated to Qwen2.5-GGUF loading failure." + }, + { + "left": "issue:44869", + "right": "issue:45356", + "accept": false, + "reason": "Whisper timestamp decode crash on trailing replacement char is unrelated to Kimi-K2.5 tokenizer codec handling." + }, + { + "left": "issue:45230", + "right": "issue:45397", + "accept": false, + "reason": "\"Bug report\" gives no concrete overlap with Gemma4 zero3 from_pretrained." + }, + { + "left": "issue:43475", + "right": "issue:43643", + "accept": false, + "reason": "Different problems: missing Sam3VisionEncoderOutput attribute vs AutoConfig missing fields under trust_remote_code." + }, + { + "left": "issue:45375", + "right": "issue:45478", + "accept": false, + "reason": "Related Qwen3.5 family, but missing deepstack_visual_indexes being dropped by strict is not the same as a from_pretrained error in transformers>=5.4.0." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 21, + "review_unit_index": 42, + "review_unit_count": 56, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-473", + "nodes": [ + "issue:28282", + "issue:30333", + "issue:30990", + "issue:33290", + "issue:36010", + "issue:41628", + "issue:42175", + "issue:42371", + "issue:42673", + "issue:42831", + "issue:42994", + "issue:43232", + "issue:43278", + "issue:43475", + "issue:43504", + "issue:43525", + "issue:43704", + "issue:43723", + "issue:43784", + "issue:43824", + "issue:43873", + "issue:43976", + "issue:44038", + "issue:44188", + "issue:44222", + "issue:44351", + "issue:44496", + "issue:44623", + "issue:44792", + "issue:44933", + "issue:44945", + "issue:44964", + "issue:45081", + "issue:45092", + "issue:45103", + "issue:45137", + "issue:45310", + "issue:45375", + "issue:45381", + "issue:45406", + "issue:45561", + "issue:45600", + "issue:45712", + "issue:45753", + "issue:46202", + "issue:46326", + "issue:46399" + ], + "soft_pairs": [ + "issue:28282|issue:30990", + "issue:42371|issue:45092", + "issue:44038|issue:44623", + "issue:41628|issue:44933", + "issue:45310|issue:45375", + "issue:43475|issue:45381", + "issue:42994|issue:43525", + "issue:44964|issue:45137", + "issue:43723|issue:46326", + "issue:43784|issue:46399", + "issue:45406|issue:46202", + "issue:43232|issue:44188", + "issue:43873|issue:44945", + "issue:44945|issue:45753", + "issue:41628|issue:44351", + "issue:45081|issue:45103", + "issue:42673|issue:43704", + "issue:44038|issue:45310", + "issue:42994|issue:43824", + "issue:44222|issue:45310", + "issue:36010|issue:41628", + "issue:42831|issue:44222", + "issue:43278|issue:45137", + "issue:43232|issue:44792", + "issue:42175|issue:44933", + "issue:44496|issue:45561", + "issue:43504|issue:45561", + "issue:45561|issue:45600", + "issue:43976|issue:44933", + "issue:30333|issue:33290", + "issue:42175|issue:45712" + ], + "prepared_review_unit_hash": "09aab3d309f2c9970b28d66888e6136301ed32391577735218848974ce0b5258", + "component_budget": { + "node_count": 473, + "item_count": 473, + "soft_pair_count": 1738, + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 31, + "serialized_chars": 36243, + "estimated_input_tokens": 9061, + "estimated_eval_tokens": 18378 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly heterogeneous, but it contains one clear duplicate pair about a Qwen3ForCausalLM VRAM leak. The rest of the soft edges look like broad thematic similarity rather than the same underlying bug.", + "confidence": 0.62, + "canonical_issue_reason": "issue:42673 is the best canonical issue because it has an exact duplicate twin (issue:43704) with the same title and underlying VRAM leak bug, making it the strongest representative of the only clear duplicate in the set.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:42673 is the most suitable issue to anchor triage: it is concrete, has a direct duplicate, and is cleaner than the other mostly unrelated or loosely similar import/model bugs.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:28282", + "right": "issue:30990", + "accept": false, + "reason": "PyTorch import failure vs Sentence Transformers loading hang; unrelated symptoms and code paths." + }, + { + "left": "issue:42371", + "right": "issue:45092", + "accept": false, + "reason": "TF32 API warning/settings issue vs old InternVL2 meta-init incompatibility; different bugs." + }, + { + "left": "issue:44038", + "right": "issue:44623", + "accept": false, + "reason": "Qwen3-VL-Moe load bug vs processor.save_pretrained missing files; not the same failure." + }, + { + "left": "issue:41628", + "right": "issue:44933", + "accept": false, + "reason": "Different import errors in different modules; similar surface area only, not same bug." + }, + { + "left": "issue:45310", + "right": "issue:45375", + "accept": false, + "reason": "Both are Qwen3.5/Moe-related, but one is from_pretrained failure and the other is a missing config field; not clearly the same bug." + }, + { + "left": "issue:43475", + "right": "issue:45381", + "accept": false, + "reason": "Different model families and different runtime failures; no evidence of one underlying defect." + }, + { + "left": "issue:42994", + "right": "issue:43525", + "accept": false, + "reason": "Quantized model save failure vs missing pad_token_id on Llama4Config; unrelated." + }, + { + "left": "issue:44964", + "right": "issue:45137", + "accept": false, + "reason": "Multimodal model loading error vs DeepSpeed ZeRO3 deque IndexError; unrelated." + }, + { + "left": "issue:43723", + "right": "issue:46326", + "accept": false, + "reason": "Tokenizer loading in v5 vs continue_final_message on mistral_common tokenizers; different code paths and bugs." + }, + { + "left": "issue:43784", + "right": "issue:46399", + "accept": false, + "reason": "Sentence-transformers import NameError vs Qwen3.5 kernel registration error; unrelated." + }, + { + "left": "issue:45406", + "right": "issue:46202", + "accept": false, + "reason": "Gemma processor attribute error vs Gemma pipeline roundtrip token issue; not the same bug." + }, + { + "left": "issue:43232", + "right": "issue:44188", + "accept": false, + "reason": "Generation kwargs update after sync_gpus vs torch.compile attention-kernel branching; different issues." + }, + { + "left": "issue:43873", + "right": "issue:44945", + "accept": false, + "reason": "Quantization/offloading behavior vs pipeline-parallel output correctness; unrelated." + }, + { + "left": "issue:44945", + "right": "issue:45753", + "accept": false, + "reason": "Pipeline parallelism output bug vs Qwen3.5 image-specific infinite loop; different failures." + }, + { + "left": "issue:41628", + "right": "issue:44351", + "accept": false, + "reason": "Both are import-name problems, but for different symbols and likely different missing-export causes." + }, + { + "left": "issue:45081", + "right": "issue:45103", + "accept": false, + "reason": "Mistral tokenizer patch crash vs auto_docstring annotations crash; unrelated." + }, + { + "left": "issue:42673", + "right": "issue:43704", + "accept": true, + "reason": "Exact duplicate titles and the same Qwen3ForCausalLM VRAM leak in multiple dataloader threads." + }, + { + "left": "issue:44038", + "right": "issue:45310", + "accept": false, + "reason": "Both mention Qwen/Moe, but they target different versions and different from_pretrained failures." + }, + { + "left": "issue:42994", + "right": "issue:43824", + "accept": false, + "reason": "Quantized model saving vs missing Qwen2_5_VLForConditionalGeneration import; unrelated." + }, + { + "left": "issue:44222", + "right": "issue:45310", + "accept": false, + "reason": "FP8 save_pretrained issue vs Qwen3.5 from_pretrained error; different code paths." + }, + { + "left": "issue:36010", + "right": "issue:41628", + "accept": false, + "reason": "Both are import errors, but for different APIs and different breakages; too broad to merge." + }, + { + "left": "issue:42831", + "right": "issue:44222", + "accept": false, + "reason": "Accuracy regression in FineGrainedFP8 vs FP8 save_pretrained bug; not the same defect." + }, + { + "left": "issue:43278", + "right": "issue:45137", + "accept": false, + "reason": "Embedding dtype mismatch during eval vs DeepSpeed deque IndexError; unrelated." + }, + { + "left": "issue:43232", + "right": "issue:44792", + "accept": false, + "reason": "Generation state bug vs janus image-generation test failure; no clear shared root cause." + }, + { + "left": "issue:42175", + "right": "issue:44933", + "accept": false, + "reason": "Packaging/backend dependency omission vs missing image_utils import; different problems." + }, + { + "left": "issue:44496", + "right": "issue:45561", + "accept": false, + "reason": "Unrecognized model/config issue vs xdist file race; unrelated." + }, + { + "left": "issue:43504", + "right": "issue:45561", + "accept": false, + "reason": "Legacy config field loading bug vs test-worker file race; unrelated." + }, + { + "left": "issue:45561", + "right": "issue:45600", + "accept": false, + "reason": "Captured-info race vs broken auto_mappings references; different CI failures." + }, + { + "left": "issue:43976", + "right": "issue:44933", + "accept": false, + "reason": "Python-version compatibility problem vs missing import from image_utils; unrelated." + }, + { + "left": "issue:30333", + "right": "issue:33290", + "accept": false, + "reason": "MLflow reporting on failed jobs vs DeepSpeed OOM; unrelated." + }, + { + "left": "issue:42175", + "right": "issue:45712", + "accept": false, + "reason": "Both touch torch packaging/dummy objects, but one is missing TF backend in torch install and the other is leftover dummy classes; different fixes." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is well grounded in the packet. The only clear canonical pair is issue:42673 and issue:43704, which share the exact same title and appear to describe the same Qwen3ForCausalLM VRAM leak. The other pairwise rejections are cautious and consistent with the issue titles and metadata, without overstating evidence." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 21, + "review_unit_index": 43, + "review_unit_count": 56, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-473", + "nodes": [ + "issue:22355", + "issue:27957", + "issue:28282", + "issue:30333", + "issue:30990", + "issue:34689", + "issue:35141", + "issue:36331", + "issue:42222", + "issue:42831", + "issue:42898", + "issue:43116", + "issue:43295", + "issue:43377", + "issue:43479", + "issue:43723", + "issue:43784", + "issue:43824", + "issue:43874", + "issue:44016", + "issue:44060", + "issue:44075", + "issue:44079", + "issue:44190", + "issue:44222", + "issue:44242", + "issue:44295", + "issue:44297", + "issue:44568", + "issue:44623", + "issue:44734", + "issue:44779", + "issue:44869", + "issue:44933", + "issue:45092", + "issue:45103", + "issue:45125", + "issue:45200", + "issue:45310", + "issue:45356", + "issue:45405", + "issue:45656", + "issue:45712", + "issue:45812", + "issue:46077", + "issue:46123", + "issue:46231" + ], + "soft_pairs": [ + "issue:46077|issue:46123", + "issue:30990|issue:35141", + "issue:42898|issue:44568", + "issue:43377|issue:44242", + "issue:44623|issue:45310", + "issue:43479|issue:44869", + "issue:42222|issue:45200", + "issue:44016|issue:46231", + "issue:42898|issue:45812", + "issue:43723|issue:44568", + "issue:43874|issue:45092", + "issue:43479|issue:44079", + "issue:42898|issue:44779", + "issue:43784|issue:43824", + "issue:44734|issue:45103", + "issue:42898|issue:45356", + "issue:44295|issue:45656", + "issue:42898|issue:43295", + "issue:27957|issue:30333", + "issue:45405|issue:45712", + "issue:30333|issue:34689", + "issue:44060|issue:45125", + "issue:44075|issue:46231", + "issue:43116|issue:44190", + "issue:44222|issue:44297", + "issue:44933|issue:45712", + "issue:30333|issue:36331", + "issue:30333|issue:35141", + "issue:22355|issue:28282", + "issue:44297|issue:45656", + "issue:42831|issue:44623" + ], + "prepared_review_unit_hash": "abfae1369dfe5e3527608f917cdf1d762b5845db3f858c0ad730189427dfa183", + "component_budget": { + "node_count": 473, + "item_count": 473, + "soft_pair_count": 1738, + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 31, + "serialized_chars": 36621, + "estimated_input_tokens": 9156, + "estimated_eval_tokens": 18568 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The candidate pairs are mostly loose topical similarities, not true duplicates; they span unrelated tokenizer, training, multimodal, serving, and docs bugs. No pair looks safe to merge as the same underlying issue/change.", + "confidence": 0.95, + "canonical_issue_reason": "No single true canonical issue stands out because the set is heterogeneous; issue:42898 is the closest umbrella candidate only because several pairs are tokenizer/regression themed, but it does not subsume the rest.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:42898 is the broadest and most representative regression-style issue in this set, but only as a weak umbrella candidate, not a duplicate target for the others.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:46077", + "right": "issue:46123", + "accept": false, + "reason": "Different bugs: Gemma4 attention-mask construction vs pipeline batch-finalization logic." + }, + { + "left": "issue:30990", + "right": "issue:35141", + "accept": false, + "reason": "Unrelated symptoms and code paths: loading hangs vs embedding resize/post-init reinitialization." + }, + { + "left": "issue:42898", + "right": "issue:44568", + "accept": false, + "reason": "Both tokenizer regressions, but different models and failures; not the same concrete bug." + }, + { + "left": "issue:43377", + "right": "issue:44242", + "accept": false, + "reason": "Different functionality: MIMI batching/padding-mask mismatch vs MoE load-balancing loss calculation." + }, + { + "left": "issue:44623", + "right": "issue:45310", + "accept": false, + "reason": "Processor save-pretrained file emission vs Qwen3.5 MoE from_pretrained loading error are separate issues." + }, + { + "left": "issue:43479", + "right": "issue:44869", + "accept": false, + "reason": "Multimodal config default initialization is unrelated to Whisper timestamp decoding crash." + }, + { + "left": "issue:42222", + "right": "issue:45200", + "accept": false, + "reason": "Different model families and bugs: vitpose breakage vs Gemma4 text-only mm_token_type_ids defaulting." + }, + { + "left": "issue:44016", + "right": "issue:46231", + "accept": false, + "reason": "Notebook syntax error is unrelated to a docs parameter-name mismatch." + }, + { + "left": "issue:42898", + "right": "issue:45812", + "accept": false, + "reason": "Both are tokenizer regressions, but they affect different model families and likely different code paths." + }, + { + "left": "issue:43723", + "right": "issue:44568", + "accept": false, + "reason": "General tokenizer-loading failure vs missing BOS/EOS insertion for one tokenizer." + }, + { + "left": "issue:43874", + "right": "issue:45092", + "accept": false, + "reason": "Different multimodal bugs: missing image-patch helper vs remote-code/meta-init incompatibility." + }, + { + "left": "issue:43479", + "right": "issue:44079", + "accept": false, + "reason": "Config default construction bug is unrelated to ModelOutput key assignment behavior." + }, + { + "left": "issue:42898", + "right": "issue:44779", + "accept": false, + "reason": "Same broad tokenizer-regression theme, but the affected models and failures differ; not one duplicate." + }, + { + "left": "issue:43784", + "right": "issue:43824", + "accept": false, + "reason": "Separate import failures in different integration points and different missing symbols." + }, + { + "left": "issue:44734", + "right": "issue:45103", + "accept": false, + "reason": "Serving KV-cache indexing crash is unrelated to auto-docstring annotation handling." + }, + { + "left": "issue:42898", + "right": "issue:45356", + "accept": false, + "reason": "Both are tokenizer regressions, but they concern different model-specific codec/regex behavior." + }, + { + "left": "issue:44295", + "right": "issue:45656", + "accept": false, + "reason": "Position-ids buffer access bug is unrelated to deepspeed double optimizer stepping." + }, + { + "left": "issue:42898", + "right": "issue:43295", + "accept": false, + "reason": "Tokenizer cleanup regression and processor.tokenizer access regression are different code paths." + }, + { + "left": "issue:27957", + "right": "issue:30333", + "accept": false, + "reason": "Flash Attention 2 support for XLMRoberta is unrelated to MLFlow reporting after training failure." + }, + { + "left": "issue:45405", + "right": "issue:45712", + "accept": false, + "reason": "Version bump for peft dependency is unrelated to leftover dummy-class cleanup." + }, + { + "left": "issue:30333", + "right": "issue:34689", + "accept": false, + "reason": "No shared underlying bug: MLFlow failure reporting vs model-loading regression for Llama 3.2 Vision." + }, + { + "left": "issue:44060", + "right": "issue:45125", + "accept": false, + "reason": "Different model-specific tensor-parallel / tied-weights issues." + }, + { + "left": "issue:44075", + "right": "issue:46231", + "accept": false, + "reason": "Optimizer argument handling bug is unrelated to documentation parameter naming." + }, + { + "left": "issue:43116", + "right": "issue:44190", + "accept": false, + "reason": "Different example scripts and failures: classification output bug vs local dataset loading." + }, + { + "left": "issue:44222", + "right": "issue:44297", + "accept": false, + "reason": "FP8 save_pretrained issue and tokenizer_config class mismatch are unrelated." + }, + { + "left": "issue:44933", + "right": "issue:45712", + "accept": false, + "reason": "Missing import from image_utils is unrelated to dummy-class cleanup." + }, + { + "left": "issue:30333", + "right": "issue:36331", + "accept": false, + "reason": "MLFlow reporting after failure and CustomTrainer signature breakage are different issues." + }, + { + "left": "issue:30333", + "right": "issue:35141", + "accept": false, + "reason": "Training-job reporting bug is unrelated to embedding reinitialization on resize." + }, + { + "left": "issue:22355", + "right": "issue:28282", + "accept": false, + "reason": "Both are import errors, but one is missing transformers.onnx and the other is missing PyTorch entirely." + }, + { + "left": "issue:44297", + "right": "issue:45656", + "accept": false, + "reason": "Tokenizer save-pretrained metadata mismatch is unrelated to deepspeed optimizer stepping." + }, + { + "left": "issue:42831", + "right": "issue:44623", + "accept": false, + "reason": "FineGrainedFP8 accuracy issue is unrelated to processor save-pretrained missing files." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 21, + "review_unit_index": 44, + "review_unit_count": 56, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-473", + "nodes": [ + "issue:30990", + "issue:38175", + "issue:41628", + "issue:42757", + "issue:42796", + "issue:42831", + "issue:43122", + "issue:43232", + "issue:43475", + "issue:43479", + "issue:43644", + "issue:43723", + "issue:44190", + "issue:44206", + "issue:44222", + "issue:44261", + "issue:44263", + "issue:44279", + "issue:44295", + "issue:44297", + "issue:44360", + "issue:44373", + "issue:44448", + "issue:44484", + "issue:44485", + "issue:44623", + "issue:44734", + "issue:44871", + "issue:44977", + "issue:45071", + "issue:45092", + "issue:45103", + "issue:45245", + "issue:45310", + "issue:45357", + "issue:45381", + "issue:45834", + "issue:46082", + "issue:46132", + "issue:46209", + "issue:46305", + "issue:46328", + "issue:46421", + "issue:46424", + "issue:46431", + "issue:46655", + "issue:46665" + ], + "soft_pairs": [ + "issue:44448|issue:46132", + "issue:44222|issue:44623", + "issue:42796|issue:43723", + "issue:43479|issue:43644", + "issue:45245|issue:46132", + "issue:43479|issue:46665", + "issue:41628|issue:42757", + "issue:44297|issue:45310", + "issue:45071|issue:46082", + "issue:44373|issue:44485", + "issue:44263|issue:45092", + "issue:43723|issue:44977", + "issue:44222|issue:44484", + "issue:43122|issue:46209", + "issue:43232|issue:46424", + "issue:44206|issue:45103", + "issue:44190|issue:46082", + "issue:46209|issue:46431", + "issue:43232|issue:45834", + "issue:44734|issue:46421", + "issue:43475|issue:46305", + "issue:44295|issue:44623", + "issue:46305|issue:46655", + "issue:43723|issue:44871", + "issue:44263|issue:44360", + "issue:45381|issue:46305", + "issue:44261|issue:45092", + "issue:30990|issue:38175", + "issue:45092|issue:45357", + "issue:44279|issue:46328", + "issue:43644|issue:46305", + "issue:44263|issue:44373", + "issue:42831|issue:45310" + ], + "prepared_review_unit_hash": "1dcb1e0ea53d216cea08b67edcd5a17fc25a033edbf9dadbde4f9d819650cff5", + "component_budget": { + "node_count": 473, + "item_count": 473, + "soft_pair_count": 1738, + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 33, + "serialized_chars": 36710, + "estimated_input_tokens": 9178, + "estimated_eval_tokens": 18612 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is heterogeneous: it mixes unrelated tokenizer, import, config, generation, save_pretrained, cache, and model-specific bugs, so there is no single duplicate center.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44448", + "right": "issue:46132", + "accept": false, + "reason": "Different problems: Pegasus v4/v5 output regression vs AttentionInterface.register behavior change." + }, + { + "left": "issue:44222", + "right": "issue:44623", + "accept": false, + "reason": "Different save_pretrained bugs: FP8 MoE saving vs processor file omission." + }, + { + "left": "issue:42796", + "right": "issue:43723", + "accept": false, + "reason": "Tokenizer BPE marker handling vs AutoTokenizer loading failure; different failure modes." + }, + { + "left": "issue:43479", + "right": "issue:43644", + "accept": false, + "reason": "Config initialization from None vs junk non-persistent buffers; unrelated code paths." + }, + { + "left": "issue:45245", + "right": "issue:46132", + "accept": false, + "reason": "Category-count limit error vs registered attention function behavior; unrelated." + }, + { + "left": "issue:43479", + "right": "issue:46665", + "accept": false, + "reason": "Default multimodal config init vs from_pretrained behavior inside Python classes." + }, + { + "left": "issue:41628", + "right": "issue:42757", + "accept": false, + "reason": "Missing AutoImageProcessor import vs missing is_offline_mode import; different symbols." + }, + { + "left": "issue:44297", + "right": "issue:45310", + "accept": false, + "reason": "Tokenizer config mismatch on save vs Qwen3.5 MoE load error; different stages and causes." + }, + { + "left": "issue:45071", + "right": "issue:46082", + "accept": false, + "reason": "PretrainedConfig type-checking regression vs head_dim validation in LlamaConfig." + }, + { + "left": "issue:44373", + "right": "issue:44485", + "accept": false, + "reason": "Docstring bug vs RoPE implementation issue; not the same underlying defect." + }, + { + "left": "issue:44263", + "right": "issue:45092", + "accept": false, + "reason": "DSA indexer torch.split handling vs remote-code meta-init incompatibility; unrelated." + }, + { + "left": "issue:43723", + "right": "issue:44977", + "accept": false, + "reason": "Tokenizer loading in v5 vs flash-attention generation failure; different code paths." + }, + { + "left": "issue:44222", + "right": "issue:44484", + "accept": false, + "reason": "FP8 save_pretrained bug vs max_shard_size default question; not duplicates." + }, + { + "left": "issue:43122", + "right": "issue:46209", + "accept": false, + "reason": "Tokenization version regression vs FP8 inference failure on B200; unrelated." + }, + { + "left": "issue:43232", + "right": "issue:46424", + "accept": false, + "reason": "Generation kwargs after sync_gpus vs cache max_length configuration; different bugs." + }, + { + "left": "issue:44206", + "right": "issue:45103", + "accept": false, + "reason": "Unsupported center arg in feature extractor vs auto_docstring AttributeError; unrelated." + }, + { + "left": "issue:44190", + "right": "issue:46082", + "accept": false, + "reason": "Local dataset loading script issue vs LlamaConfig head_dim validation; different areas." + }, + { + "left": "issue:46209", + "right": "issue:46431", + "accept": false, + "reason": "FP8 on B200 inference failure vs missing chat template error; unrelated." + }, + { + "left": "issue:43232", + "right": "issue:45834", + "accept": false, + "reason": "Generation bookkeeping bug vs long OCR index error in Kosmos2.5; different models/issues." + }, + { + "left": "issue:44734", + "right": "issue:46421", + "accept": false, + "reason": "Serve KV-cache continuation indexing bug vs lazy cache init recompilation; different cache problems." + }, + { + "left": "issue:43475", + "right": "issue:46305", + "accept": false, + "reason": "SAM 3 video attribute error vs MoonshineStreaming off-by-one masking; unrelated." + }, + { + "left": "issue:44295", + "right": "issue:44623", + "accept": false, + "reason": "position_ids buffer read error vs processor.save_pretrained missing files; different APIs." + }, + { + "left": "issue:46305", + "right": "issue:46655", + "accept": false, + "reason": "Streaming mask off-by-one vs SmolVLM video padding calculation; different video bugs." + }, + { + "left": "issue:43723", + "right": "issue:44871", + "accept": false, + "reason": "Tokenizer loading failure vs eos_token_id inconsistency between tokenizer and model config." + }, + { + "left": "issue:44263", + "right": "issue:44360", + "accept": false, + "reason": "DSA indexer torch.split issue vs missing ReLU in DSA indexer; different logic errors." + }, + { + "left": "issue:45381", + "right": "issue:46305", + "accept": false, + "reason": "Qwen2.5-VL vision_position_ids bug vs MoonshineStreaming window bug; unrelated models." + }, + { + "left": "issue:44261", + "right": "issue:45092", + "accept": false, + "reason": "MLA RMS norm epsilon precision issue vs meta-init incompatibility; unrelated." + }, + { + "left": "issue:30990", + "right": "issue:38175", + "accept": false, + "reason": "Sentence Transformers loading hang vs zero probabilities in SigLIP2; different symptoms and models." + }, + { + "left": "issue:45092", + "right": "issue:45357", + "accept": false, + "reason": "Old InternVL2 meta-init incompatibility vs incorrect visual encoder keys saved in Qwen3.5." + }, + { + "left": "issue:44279", + "right": "issue:46328", + "accept": false, + "reason": "General dependency issue vs SequenceFeatureExtractor.pad performance regression; unrelated." + }, + { + "left": "issue:43644", + "right": "issue:46305", + "accept": false, + "reason": "Non-persistent buffer corruption vs streaming mask off-by-one; different bug classes." + }, + { + "left": "issue:44263", + "right": "issue:44373", + "accept": false, + "reason": "Runtime indexer bug vs docstring correction; not the same issue." + }, + { + "left": "issue:42831", + "right": "issue:45310", + "accept": false, + "reason": "FineGrainedFP8 accuracy issue vs Qwen3.5 MoE load error; distinct FP8-related problems." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 21, + "review_unit_index": 45, + "review_unit_count": 56, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-473", + "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:43550", + "issue:43576", + "issue:43646", + "issue:43756", + "issue:43784", + "issue:43824", + "issue:43976", + "issue:44038", + "issue:44060", + "issue:44117", + "issue:44188", + "issue:44261", + "issue:44263", + "issue:44295", + "issue:44297", + "issue:44373", + "issue:44483", + "issue:44568", + "issue:44704", + "issue:44829", + "issue:44945", + "issue:45278", + "issue:45406", + "issue:45593", + "issue:45753", + "issue:46002", + "issue:46050", + "issue:46132", + "issue:46369", + "issue:46402" + ], + "soft_pairs": [ + "issue:42907|issue:46050", + "issue:44829|issue:46002", + "issue:42831|issue:44038", + "issue:44483|issue:45406", + "issue:22355|issue:27957", + "issue:41628|issue:43824", + "issue:43408|issue:46402", + "issue:22355|issue:28218", + "issue:42831|issue:45753", + "issue:39401|issue:44568", + "issue:42757|issue:45278", + "issue:27957|issue:30990", + "issue:42886|issue:43066", + "issue:42886|issue:42994", + "issue:43232|issue:45593", + "issue:43408|issue:46369", + "issue:44188|issue:44373", + "issue:42796|issue:43122", + "issue:43550|issue:46132", + "issue:44261|issue:44373", + "issue:43646|issue:46002", + "issue:42371|issue:44263", + "issue:43576|issue:43784", + "issue:44295|issue:44297", + "issue:42907|issue:44945", + "issue:30990|issue:33357", + "issue:43756|issue:44060", + "issue:44117|issue:44704", + "issue:43976|issue:44945" + ], + "prepared_review_unit_hash": "e059ffbb337cb2caa2e197ab74185c2087154fd29aeba39c9e8b64505293bb82", + "component_budget": { + "node_count": 473, + "item_count": 473, + "soft_pair_count": 1738, + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 29, + "serialized_chars": 35884, + "estimated_input_tokens": 8971, + "estimated_eval_tokens": 18198 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a mixed bag of unrelated Transformers issues across tokenization, imports, quantization, attention, and model-specific regressions. None of the soft pairs look like true duplicates of the same concrete bug; they are only loosely similar by subsystem or symptom.", + "confidence": 0.81, + "canonical_issue_reason": "Issue 45278 is the broadest representative of the import-regression/version-upgrade subset, which is the closest recurring theme in this set.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 45278 is the best overall representative because it captures the widest, most general failure mode here: post-upgrade import breakage.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:42907", + "right": "issue:46050", + "accept": false, + "reason": "Both concern quantized save/load, but one is dequantized Ministral/Devstral save_pretrained failure and the other is sinq round-trip breakage; different bug paths." + }, + { + "left": "issue:44829", + "right": "issue:46002", + "accept": false, + "reason": "Both mention training instability, but one is flash_attention_3 degenerate training and the other is NaNs after checkout/init; not the same defect." + }, + { + "left": "issue:42831", + "right": "issue:44038", + "accept": false, + "reason": "FineGrainedFP8 accuracy loss and Qwen3-VL-Moe model issues are unrelated concrete problems." + }, + { + "left": "issue:44483", + "right": "issue:45406", + "accept": false, + "reason": "One is an API request rejection issue, the other is a Gemma4Processor `_tokenizer` crash; unrelated." + }, + { + "left": "issue:22355", + "right": "issue:27957", + "accept": false, + "reason": "ONNX import failure and XLMRoberta Flash Attention 2 support are different issues." + }, + { + "left": "issue:41628", + "right": "issue:43824", + "accept": false, + "reason": "Different missing imports (`AutoImageProcessor` vs `Qwen2_5_VLForConditionalGeneration`), so these are separate bugs." + }, + { + "left": "issue:43408", + "right": "issue:46402", + "accept": false, + "reason": "Both involve sam3, but one is a model-type warning and the other is missing text encoder weights; not the same code-path problem." + }, + { + "left": "issue:22355", + "right": "issue:28218", + "accept": false, + "reason": "Transformers ONNX import error and tokenizer added-token spacing bug are unrelated." + }, + { + "left": "issue:42831", + "right": "issue:45753", + "accept": false, + "reason": "FP8 accuracy regression and a Qwen3_5 image-induced infinite loop are different failures." + }, + { + "left": "issue:39401", + "right": "issue:44568", + "accept": false, + "reason": "Both are tokenizer-related, but one is wrong offset mappings and the other is missing BOS/EOS insertion; different behavior." + }, + { + "left": "issue:42757", + "right": "issue:45278", + "accept": false, + "reason": "A single `is_offline_mode` import error is not the same concrete bug as the broader post-upgrade import failures." + }, + { + "left": "issue:27957", + "right": "issue:30990", + "accept": false, + "reason": "Flash Attention 2 support for XLMRoberta and Sentence Transformers loading hang are unrelated." + }, + { + "left": "issue:42886", + "right": "issue:43066", + "accept": false, + "reason": "Offline cache loading failure and wrong tokenizer decoder type in v5 are separate tokenizer regressions." + }, + { + "left": "issue:42886", + "right": "issue:42994", + "accept": false, + "reason": "Offline tokenizer cache loading and quantized model saving are different code paths." + }, + { + "left": "issue:43232", + "right": "issue:45593", + "accept": false, + "reason": "Generation kwargs/sync bug and D-FINE auxiliary-loss behavior are unrelated." + }, + { + "left": "issue:43408", + "right": "issue:46369", + "accept": false, + "reason": "SAM3 model-type warning versus SAM3 litetext memory overflow are different bugs." + }, + { + "left": "issue:44188", + "right": "issue:44373", + "accept": false, + "reason": "A torch.compile attention-kernel divergence and a docstring typo are clearly not duplicates." + }, + { + "left": "issue:42796", + "right": "issue:43122", + "accept": false, + "reason": "Both are tokenizer regressions, but one is BPE-marker handling for Ministral and the other is broader tokenization drift across versions." + }, + { + "left": "issue:43550", + "right": "issue:46132", + "accept": false, + "reason": "Bamba SDPA/torch.compile failure and AttentionInterface.register behavior changes are different attention-related issues." + }, + { + "left": "issue:44261", + "right": "issue:44373", + "accept": false, + "reason": "MLA layernorm precision/config bug and a wrong docstring are unrelated." + }, + { + "left": "issue:43646", + "right": "issue:46002", + "accept": false, + "reason": "Custom model initialization breakage and NaNs in classification heads are different failures with different fixes." + }, + { + "left": "issue:42371", + "right": "issue:44263", + "accept": false, + "reason": "TF32 configuration warning and a torch.split return-value issue are unrelated." + }, + { + "left": "issue:43576", + "right": "issue:43784", + "accept": false, + "reason": "Broken `transformers env` CLI and sentence-transformers import NameError are separate issues." + }, + { + "left": "issue:44295", + "right": "issue:44297", + "accept": false, + "reason": "Reading a registered position_ids buffer and tokenizer_class mismatch in saved config are different problems." + }, + { + "left": "issue:42907", + "right": "issue:44945", + "accept": false, + "reason": "Quantized save/pretrained failure and pipeline-parallelism output corruption are unrelated." + }, + { + "left": "issue:30990", + "right": "issue:33357", + "accept": false, + "reason": "A loading hang and a MacOS bus error on CLIP are different failure modes." + }, + { + "left": "issue:43756", + "right": "issue:44060", + "accept": false, + "reason": "RoPE-layer dropping and tied-weights warning behavior are not the same bug." + }, + { + "left": "issue:44117", + "right": "issue:44704", + "accept": false, + "reason": "Tokenizer mapping None handling and AutoProcessor kwargs passthrough are separate plumbing bugs." + }, + { + "left": "issue:43976", + "right": "issue:44945", + "accept": false, + "reason": "Python version compatibility and pipeline-parallelism output issues are unrelated." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 21, + "review_unit_index": 46, + "review_unit_count": 56, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-473", + "nodes": [ + "issue:22355", + "issue:29127", + "issue:29159", + "issue:36296", + "issue:38617", + "issue:39401", + "issue:41628", + "issue:42175", + "issue:42757", + "issue:42898", + "issue:42907", + "issue:43116", + "issue:43122", + "issue:43278", + "issue:43408", + "issue:43452", + "issue:43475", + "issue:43493", + "issue:43784", + "issue:43824", + "issue:43937", + "issue:44038", + "issue:44188", + "issue:44222", + "issue:44295", + "issue:44297", + "issue:44351", + "issue:44360", + "issue:44485", + "issue:44488", + "issue:44521", + "issue:44625", + "issue:44704", + "issue:44779", + "issue:44829", + "issue:44938", + "issue:44945", + "issue:44964", + "issue:45200", + "issue:45216", + "issue:45278", + "issue:45290", + "issue:45478", + "issue:45600", + "issue:45753", + "issue:46326", + "issue:46328", + "issue:46614" + ], + "soft_pairs": [ + "issue:39401|issue:44488", + "issue:41628|issue:45278", + "issue:42907|issue:44964", + "issue:43784|issue:45600", + "issue:42907|issue:43937", + "issue:44360|issue:44485", + "issue:44704|issue:45290", + "issue:44625|issue:44829", + "issue:39401|issue:44521", + "issue:43122|issue:46326", + "issue:44038|issue:44297", + "issue:43122|issue:44521", + "issue:43116|issue:45200", + "issue:43408|issue:43475", + "issue:39401|issue:46614", + "issue:44038|issue:44295", + "issue:42907|issue:43452", + "issue:44038|issue:44222", + "issue:22355|issue:29159", + "issue:45216|issue:45478", + "issue:43824|issue:44351", + "issue:43122|issue:44779", + "issue:36296|issue:43116", + "issue:42757|issue:44351", + "issue:43278|issue:44945", + "issue:42175|issue:44938", + "issue:22355|issue:29127", + "issue:43493|issue:45753", + "issue:44188|issue:46328", + "issue:38617|issue:45278", + "issue:42898|issue:46614", + "issue:43937|issue:44485" + ], + "prepared_review_unit_hash": "4ec113c6bd7e03c28d22c5eeef637e40d4447fa4c83d0276c6207cd276a0e622", + "component_budget": { + "node_count": 473, + "item_count": 473, + "soft_pair_count": 1738, + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 32, + "serialized_chars": 36999, + "estimated_input_tokens": 9250, + "estimated_eval_tokens": 18756 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is heterogeneous: it mixes unrelated import errors, tokenizer regressions, model-specific bugs, and training/runtime issues. The pairwise similarities look lexical rather than describing the same underlying defect, so I would not merge any of the soft pairs as duplicates.", + "confidence": 0.74, + "canonical_issue_reason": "issue:45278 is the broadest representative of the import-error/version-upgrade subgroup and the most generic issue in the set, but the cluster overall does not look like one true duplicate family.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:45278 is the best single representative because it is broad and clearly describes a common user-facing failure mode, even though it is not a safe duplicate of most others.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:39401", + "right": "issue:44488", + "accept": false, + "reason": "Different failures: tokenizer offset mapping bug vs a model-loading/import problem for sleng-bert." + }, + { + "left": "issue:41628", + "right": "issue:45278", + "accept": false, + "reason": "Both are import-related, but one is a specific missing symbol and the other is a broad upgrade-induced import-error report." + }, + { + "left": "issue:42907", + "right": "issue:44964", + "accept": false, + "reason": "Unrelated code paths: dequantized save_pretrained failure vs loading a multimodal model." + }, + { + "left": "issue:43784", + "right": "issue:45600", + "accept": false, + "reason": "Both involve import-time breakage, but the missing-name issue in sentence-transformers is distinct from auto_mappings referencing removed configs." + }, + { + "left": "issue:42907", + "right": "issue:43937", + "accept": false, + "reason": "Different model behaviors: saving dequantized weights vs GenerationConfig validation." + }, + { + "left": "issue:44360", + "right": "issue:44485", + "accept": false, + "reason": "Completely different subsystems and symptoms; no shared bug mechanism is evident." + }, + { + "left": "issue:44704", + "right": "issue:45290", + "accept": false, + "reason": "Different APIs and failures: cached_file kwargs propagation vs chat template crashes on tool-call messages." + }, + { + "left": "issue:44625", + "right": "issue:44829", + "accept": false, + "reason": "One is config propagation for Qwen3.5 labels; the other is a flash_attention_3 training regression." + }, + { + "left": "issue:39401", + "right": "issue:44521", + "accept": false, + "reason": "Tokenizer offset mapping bug is unrelated to multimodal assistant_mask generation." + }, + { + "left": "issue:43122", + "right": "issue:46326", + "accept": false, + "reason": "Both are tokenizer-related, but they concern different tokenizer stacks and different failure modes." + }, + { + "left": "issue:44038", + "right": "issue:44297", + "accept": false, + "reason": "Different problems: Qwen3-VL-Moe loading vs tokenizer_class metadata written by save_pretrained." + }, + { + "left": "issue:43122", + "right": "issue:44521", + "accept": false, + "reason": "Different functionality: text tokenization drift vs multimodal chat-template masking." + }, + { + "left": "issue:43116", + "right": "issue:45200", + "accept": false, + "reason": "Different features: classification example script output vs Gemma 4 mm_token_type_ids defaults." + }, + { + "left": "issue:43408", + "right": "issue:43475", + "accept": false, + "reason": "These are separate SAM 3 video issues with different object/model attribute problems." + }, + { + "left": "issue:39401", + "right": "issue:46614", + "accept": false, + "reason": "Tokenizer offset_mapping vs phonemizer backend state are unrelated." + }, + { + "left": "issue:44038", + "right": "issue:44295", + "accept": false, + "reason": "Different bugs: Qwen3-VL-Moe loading vs reading position_ids after buffer registration." + }, + { + "left": "issue:42907", + "right": "issue:43452", + "accept": false, + "reason": "Saving dequantized models and gguf_file loading failures are different code paths." + }, + { + "left": "issue:44038", + "right": "issue:44222", + "accept": false, + "reason": "Both mention model support, but one is a load-time bug and the other is FP8 save_pretrained handling." + }, + { + "left": "issue:22355", + "right": "issue:29159", + "accept": false, + "reason": "Missing transformers.onnx module is unrelated to slow/fast tokenizer inconsistency." + }, + { + "left": "issue:45216", + "right": "issue:45478", + "accept": false, + "reason": "Related version and model family, but save_pretrained checkpoint corruption and from_pretrained errors are not the same concrete bug." + }, + { + "left": "issue:43824", + "right": "issue:44351", + "accept": false, + "reason": "Both are import errors, but they concern different missing classes from different modules." + }, + { + "left": "issue:43122", + "right": "issue:44779", + "accept": false, + "reason": "Both are tokenizer regressions, but there is no evidence they share the same tokenizer implementation or root cause." + }, + { + "left": "issue:36296", + "right": "issue:43116", + "accept": false, + "reason": "Tensor-parallel training bug is unrelated to the multi-label classification example script issue." + }, + { + "left": "issue:42757", + "right": "issue:44351", + "accept": false, + "reason": "Different import failures in different packages/modules." + }, + { + "left": "issue:43278", + "right": "issue:44945", + "accept": false, + "reason": "Embedding dtype mismatch during eval is unrelated to pipeline-parallel output corruption." + }, + { + "left": "issue:42175", + "right": "issue:44938", + "accept": false, + "reason": "One is missing TensorFlow in an install extra; the other is a Python 3.14 load failure." + }, + { + "left": "issue:22355", + "right": "issue:29127", + "accept": false, + "reason": "No module named transformers.onnx is unrelated to the LayoutLMv3 error-message issue." + }, + { + "left": "issue:43493", + "right": "issue:45753", + "accept": false, + "reason": "Different multimodal/model-specific bugs: SigLIP2 implementation drift vs Qwen3_5 image loop." + }, + { + "left": "issue:44188", + "right": "issue:46328", + "accept": false, + "reason": "Diverging attention kernels under torch.compile is unrelated to SequenceFeatureExtractor.pad inefficiency." + }, + { + "left": "issue:38617", + "right": "issue:45278", + "accept": false, + "reason": "Both are import failures, but they are different missing symbols and do not look like the same bug." + }, + { + "left": "issue:42898", + "right": "issue:46614", + "accept": false, + "reason": "Tokenization-space cleanup behavior and phonemizer language state are different regressions." + }, + { + "left": "issue:43937", + "right": "issue:44485", + "accept": false, + "reason": "Both are GLM-5 related, but they target different implementation areas 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": 21, + "review_unit_index": 47, + "review_unit_count": 56, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-473", + "nodes": [ + "issue:28348", + "issue:33453", + "issue:39401", + "issue:42222", + "issue:42371", + "issue:42673", + "issue:42796", + "issue:43064", + "issue:43066", + "issue:43116", + "issue:43441", + "issue:43493", + "issue:43611", + "issue:43673", + "issue:43825", + "issue:43827", + "issue:43874", + "issue:43906", + "issue:43937", + "issue:43976", + "issue:43992", + "issue:44038", + "issue:44075", + "issue:44117", + "issue:44291", + "issue:44355", + "issue:44380", + "issue:44625", + "issue:44704", + "issue:44779", + "issue:44936", + "issue:45071", + "issue:45072", + "issue:45083", + "issue:45356", + "issue:45357", + "issue:45405", + "issue:45446", + "issue:45478", + "issue:45563", + "issue:45588", + "issue:46002", + "issue:46077", + "issue:46231", + "issue:46328", + "issue:46614", + "issue:46640", + "issue:46655" + ], + "soft_pairs": [ + "issue:43066|issue:44779", + "issue:42673|issue:45072", + "issue:43976|issue:45446", + "issue:43116|issue:43493", + "issue:43611|issue:46002", + "issue:44355|issue:46231", + "issue:42371|issue:43874", + "issue:44038|issue:45478", + "issue:44625|issue:46002", + "issue:45357|issue:45478", + "issue:45083|issue:45446", + "issue:44075|issue:44936", + "issue:42222|issue:43441", + "issue:43116|issue:43825", + "issue:43992|issue:44704", + "issue:28348|issue:46640", + "issue:43064|issue:43673", + "issue:42673|issue:42796", + "issue:42673|issue:46655", + "issue:43937|issue:45563", + "issue:43827|issue:45446", + "issue:33453|issue:39401", + "issue:45356|issue:46614", + "issue:44380|issue:46077", + "issue:43673|issue:46328", + "issue:44117|issue:45588", + "issue:45071|issue:46614", + "issue:44291|issue:45071", + "issue:45405|issue:45446", + "issue:42673|issue:43906", + "issue:39401|issue:44779" + ], + "prepared_review_unit_hash": "7d6b90dd6fe33f850d98488c1307984d15e6455092207d242fd9c607f8fa609a", + "component_budget": { + "node_count": 473, + "item_count": 473, + "soft_pair_count": 1738, + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 31, + "serialized_chars": 36802, + "estimated_input_tokens": 9201, + "estimated_eval_tokens": 18658 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a loose thematic grouping of Transformers v5 regressions, but the soft pairs are not true duplicates: they span different models, code paths, and failure modes. No PRs are present.", + "confidence": 0.37, + "canonical_issue_reason": "issue:33453 is the broadest umbrella issue in the set, covering a tokenizer-loading regression theme that overlaps several of the tokenizer-focused reports.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:44779 is the strongest concrete anchor for the cluster: a clear, user-visible tokenizer regression in v5 with a specific before/after behavior change.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43066", + "right": "issue:44779", + "accept": false, + "reason": "Both are tokenizer regressions, but one is about decoder type handling while the other is DeepSeek output correctness; different bugs." + }, + { + "left": "issue:42673", + "right": "issue:45072", + "accept": false, + "reason": "VRAM leak in multi-thread dataloader use is unrelated to the bfloat16 dtype mismatch in inference." + }, + { + "left": "issue:43976", + "right": "issue:45446", + "accept": false, + "reason": "Different version-compatibility failures: Python version gating vs PyTorch version check for a flex_attention import." + }, + { + "left": "issue:43116", + "right": "issue:43493", + "accept": false, + "reason": "Multi-label classification returning empty results is unrelated to a SigLIP2 HF-vs-JAX model discrepancy." + }, + { + "left": "issue:43611", + "right": "issue:46002", + "accept": false, + "reason": "Model loading with base_model_prefix is a loading-path regression, while the other is NaNs in classification heads after checkout/init." + }, + { + "left": "issue:44355", + "right": "issue:46231", + "accept": false, + "reason": "Compiled Python file errors are unrelated to a documentation parameter-name mismatch in set_training()." + }, + { + "left": "issue:42371", + "right": "issue:43874", + "accept": false, + "reason": "TF32 behavior control is unrelated to the GLM46V image processor missing method / token-counting AttributeError." + }, + { + "left": "issue:44038", + "right": "issue:45478", + "accept": false, + "reason": "Both involve Qwen3-family models, but they are different from_pretrained failures on different model variants and versions." + }, + { + "left": "issue:44625", + "right": "issue:46002", + "accept": false, + "reason": "num_labels propagation in Qwen3.5 is a config-plumbing bug, not the same as NaNs in classification heads." + }, + { + "left": "issue:45357", + "right": "issue:45478", + "accept": false, + "reason": "Saving wrong visual encoder keys and failing from_pretrained are different Qwen3.5 bugs in different code paths." + }, + { + "left": "issue:45083", + "right": "issue:45446", + "accept": false, + "reason": "Unexpected helper output lengths in qwen3_omni_moe is unrelated to a flex_attention import version check." + }, + { + "left": "issue:44075", + "right": "issue:44936", + "accept": false, + "reason": "Ignoring SGD args and failing evaluate() after train() are distinct training-loop problems." + }, + { + "left": "issue:42222", + "right": "issue:43441", + "accept": false, + "reason": "VitPose model breakage and Ministral FlashAttention failure affect different model families and mechanisms." + }, + { + "left": "issue:43116", + "right": "issue:43825", + "accept": false, + "reason": "An example script returning empty multilabel results is unrelated to pipeline()'s translation-task error message." + }, + { + "left": "issue:43992", + "right": "issue:44704", + "accept": false, + "reason": "Missing embed_tokens.weight in UMT5 loading and AutoProcessor not forwarding kwargs are separate loading/processor issues." + }, + { + "left": "issue:28348", + "right": "issue:46640", + "accept": false, + "reason": "GPT2 FlashAttention 2 support request is not the same bug as a generic flash-attention issue report." + }, + { + "left": "issue:43064", + "right": "issue:43673", + "accept": false, + "reason": "Wrong optimizer states under FSDP2/PEFT/cpu_ram_efficient_loading is unrelated to missing GenerationMixin cache in chunked prefill." + }, + { + "left": "issue:42673", + "right": "issue:42796", + "accept": false, + "reason": "A VRAM leak in multithreaded loading is unrelated to tokenizer BPE marker handling." + }, + { + "left": "issue:42673", + "right": "issue:46655", + "accept": false, + "reason": "The VRAM leak report and the video padding calculation bug involve different subsystems and symptoms." + }, + { + "left": "issue:43937", + "right": "issue:45563", + "accept": false, + "reason": "An invalid GenerationConfig error is different from a stale warning about num_return_sequences." + }, + { + "left": "issue:43827", + "right": "issue:45446", + "accept": false, + "reason": "Docs still referencing pipeline() after removal is unrelated to the flex_attention PyTorch version check." + }, + { + "left": "issue:33453", + "right": "issue:39401", + "accept": false, + "reason": "Both are tokenizer regressions, but one is a broad loading regression and the other is a specific offset_mapping bug in Qwen3." + }, + { + "left": "issue:45356", + "right": "issue:46614", + "accept": false, + "reason": "Kimi tokenizer codec/warning regression is unrelated to stale espeak backend language state in phonemization." + }, + { + "left": "issue:44380", + "right": "issue:46077", + "accept": false, + "reason": "GPT2 attention scaling being ignored under SDPA/FlashAttention is a different attention-path bug from Gemma4 building causal masks." + }, + { + "left": "issue:43673", + "right": "issue:46328", + "accept": false, + "reason": "Missing generation cache during chunked prefill is unrelated to SequenceFeatureExtractor.pad conversion overhead." + }, + { + "left": "issue:44117", + "right": "issue:45588", + "accept": false, + "reason": "Tokenizer mapping None handling and flash_attention.py crashing on s_aux=None are unrelated code paths." + }, + { + "left": "issue:45071", + "right": "issue:46614", + "accept": false, + "reason": "PretrainedConfig type checking regression is unrelated to stale phonemizer language backend state." + }, + { + "left": "issue:44291", + "right": "issue:45071", + "accept": false, + "reason": "Unexpected _is_hf_initialized handling during init_empty_weights is a different loading bug than PretrainedConfig type checking." + }, + { + "left": "issue:45405", + "right": "issue:45446", + "accept": false, + "reason": "An unreleased PEFT version bump is unrelated to the PyTorch version check for flex_attention imports." + }, + { + "left": "issue:39401", + "right": "issue:44779", + "accept": false, + "reason": "Both are tokenizer issues, but one is Qwen3 offset mapping and the other is DeepSeek tokenization correctness; different underlying 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": 21, + "review_unit_index": 48, + "review_unit_count": 56, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-473", + "nodes": [ + "issue:38617", + "issue:39401", + "issue:41628", + "issue:42222", + "issue:42371", + "issue:42617", + "issue:43066", + "issue:43097", + "issue:43122", + "issue:43408", + "issue:43450", + "issue:43742", + "issue:43937", + "issue:43986", + "issue:44075", + "issue:44117", + "issue:44220", + "issue:44261", + "issue:44263", + "issue:44410", + "issue:44466", + "issue:44484", + "issue:44485", + "issue:44493", + "issue:44496", + "issue:44829", + "issue:44877", + "issue:44933", + "issue:44977", + "issue:45071", + "issue:45081", + "issue:45216", + "issue:45278", + "issue:45310", + "issue:45341", + "issue:45362", + "issue:45478", + "issue:45542", + "issue:45561", + "issue:45656", + "issue:46002", + "issue:46178", + "issue:46209", + "issue:46231", + "issue:46298", + "issue:46328", + "issue:46512", + "issue:46617" + ], + "soft_pairs": [ + "issue:44466|issue:46002", + "issue:43937|issue:44977", + "issue:44410|issue:45310", + "issue:43450|issue:43986", + "issue:45341|issue:45561", + "issue:45341|issue:46328", + "issue:45071|issue:46512", + "issue:44829|issue:45216", + "issue:44117|issue:45081", + "issue:38617|issue:41628", + "issue:42617|issue:45341", + "issue:43122|issue:46298", + "issue:42222|issue:46178", + "issue:42371|issue:44261", + "issue:45362|issue:46328", + "issue:43408|issue:43742", + "issue:44877|issue:46231", + "issue:42371|issue:46617", + "issue:44484|issue:45310", + "issue:39401|issue:46209", + "issue:43937|issue:44496", + "issue:44933|issue:45278", + "issue:44410|issue:45478", + "issue:38617|issue:44220", + "issue:44075|issue:45656", + "issue:43066|issue:43122", + "issue:44263|issue:44485", + "issue:42222|issue:45542", + "issue:43097|issue:44493" + ], + "prepared_review_unit_hash": "5685fd38b8ec01572c212b0f0008c3aa3406d4612749c2bb79d0864cce5ce7da", + "component_budget": { + "node_count": 473, + "item_count": 473, + "soft_pair_count": 1738, + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 29, + "serialized_chars": 35985, + "estimated_input_tokens": 8997, + "estimated_eval_tokens": 18250 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Mostly unrelated issues were grouped together; the soft pairs are only title-level similarities and do not clear the strict duplicate threshold. I would not merge any of these pairs as duplicates.", + "confidence": 0.88, + "canonical_issue_reason": "No single canonical duplicate stands out: the items span unrelated regressions across imports, tokenization, model loading, testing, optimizer behavior, and config handling.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 45278 is the broadest umbrella candidate because it captures the only obvious repeated theme here: upgrade-related import regressions.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44466", + "right": "issue:46002", + "accept": false, + "reason": "Both mention model weight behavior, but one is a serialization inconsistency and the other is NaNs in classification heads; different bugs and code paths." + }, + { + "left": "issue:43937", + "right": "issue:44977", + "accept": false, + "reason": "GenerationConfig validation for GLM-5 is unrelated to Qwen3.5 flash-attention generation failures." + }, + { + "left": "issue:44410", + "right": "issue:45310", + "accept": false, + "reason": "One is a missing layer projection in qwen3next, the other is a Qwen3.5 MoE from_pretrained error; not the same defect." + }, + { + "left": "issue:43450", + "right": "issue:43986", + "accept": false, + "reason": "Incorrect batched video processor shapes and missing-torchvision AutoProcessor crashes are different problems." + }, + { + "left": "issue:45341", + "right": "issue:45561", + "accept": false, + "reason": "A testing_utils bug and a pytest-xdist file race are not the same underlying issue." + }, + { + "left": "issue:45341", + "right": "issue:46328", + "accept": false, + "reason": "A testing helper bug and a SequenceFeatureExtractor.pad performance issue are unrelated." + }, + { + "left": "issue:45071", + "right": "issue:46512", + "accept": false, + "reason": "Type-checking breakage in PretrainedConfig and stale config.dtype after from_config are distinct issues." + }, + { + "left": "issue:44829", + "right": "issue:45216", + "accept": false, + "reason": "Flash-attention-3 training degeneration and save_pretrained checkpoint regression are different code paths." + }, + { + "left": "issue:44117", + "right": "issue:45081", + "accept": false, + "reason": "TOKENIZER_MAPPING_NAMES returning None and the Mistral regex patch crash are unrelated tokenizer bugs." + }, + { + "left": "issue:38617", + "right": "issue:41628", + "accept": false, + "reason": "Both are import errors, but they involve different missing symbols from different modules and do not look like one concrete bug." + }, + { + "left": "issue:42617", + "right": "issue:45341", + "accept": false, + "reason": "Running 3d_parallel.py and a testing_utils bug are unrelated." + }, + { + "left": "issue:43122", + "right": "issue:46298", + "accept": false, + "reason": "Tokenizer behavior regression and noisy logging are unrelated." + }, + { + "left": "issue:42222", + "right": "issue:46178", + "accept": false, + "reason": "Broken vitpose models and a missing tensorflow_text backend entry are different issues." + }, + { + "left": "issue:42371", + "right": "issue:44261", + "accept": false, + "reason": "TF32 API guidance and MLA q_a_layernorm precision error are unrelated." + }, + { + "left": "issue:45362", + "right": "issue:46328", + "accept": false, + "reason": "A chat crash for Qwen3.5-35B and a padding performance issue are not the same bug." + }, + { + "left": "issue:43408", + "right": "issue:43742", + "accept": false, + "reason": "A model-type warning for sam3_video/sam3_tracker and a MobileLLM key error are different failures." + }, + { + "left": "issue:44877", + "right": "issue:46231", + "accept": false, + "reason": "Strict config loading for granite_speech and a set_training doc parameter mismatch are unrelated." + }, + { + "left": "issue:42371", + "right": "issue:46617", + "accept": false, + "reason": "TF32 settings guidance and trust_remote_code cache FileNotFoundError are unrelated." + }, + { + "left": "issue:44484", + "right": "issue:45310", + "accept": false, + "reason": "save_pretrained sharding size behavior and Qwen3.5 MoE from_pretrained failure do not share a concrete bug." + }, + { + "left": "issue:39401", + "right": "issue:46209", + "accept": false, + "reason": "Qwen3 tokenizer offset mapping and FP8 inference on B200 are unrelated." + }, + { + "left": "issue:43937", + "right": "issue:44496", + "accept": false, + "reason": "GenerationConfig validation for GLM-5 and an unrecognized model/config error are different problems." + }, + { + "left": "issue:44933", + "right": "issue:45278", + "accept": false, + "reason": "A specific missing import from image_utils and a broad report of many import errors after upgrade are related in theme but not clearly the same bug." + }, + { + "left": "issue:44410", + "right": "issue:45478", + "accept": false, + "reason": "Missing qwen3next projections and Qwen3.5 MoE from_pretrained errors are not the same defect." + }, + { + "left": "issue:38617", + "right": "issue:44220", + "accept": false, + "reason": "A missing import from configuration_utils and a feature-extraction helper issue are unrelated." + }, + { + "left": "issue:44075", + "right": "issue:45656", + "accept": false, + "reason": "SGD args being ignored and optimizer.step() being called twice with deepspeed are different optimizer bugs." + }, + { + "left": "issue:43066", + "right": "issue:43122", + "accept": false, + "reason": "Both concern tokenizer regressions, but one is about decoder type and the other about differing tokenization output; not enough to treat as the same bug." + }, + { + "left": "issue:44263", + "right": "issue:44485", + "accept": false, + "reason": "A torch.split return-value issue in GlmMoeDsaIndexer and GLM-5 RoPE implementation are unrelated." + }, + { + "left": "issue:42222", + "right": "issue:45542", + "accept": false, + "reason": "Broken vitpose models and a TensorBoard-without-TensorFlow backend error are unrelated." + }, + { + "left": "issue:43097", + "right": "issue:44493", + "accept": false, + "reason": "Removal of tie_embeddings_and_encoder_decoder and unexpected position-id keys are different 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": 21, + "review_unit_index": 49, + "review_unit_count": 56, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-473", + "nodes": [ + "issue:27957", + "issue:38617", + "issue:41628", + "issue:42175", + "issue:42222", + "issue:42757", + "issue:43012", + "issue:43064", + "issue:43097", + "issue:43352", + "issue:43441", + "issue:43452", + "issue:43531", + "issue:43577", + "issue:43673", + "issue:43874", + "issue:43931", + "issue:43986", + "issue:44038", + "issue:44117", + "issue:44261", + "issue:44263", + "issue:44315", + "issue:44360", + "issue:44410", + "issue:44483", + "issue:44484", + "issue:44485", + "issue:44829", + "issue:44933", + "issue:44987", + "issue:45003", + "issue:45071", + "issue:45341", + "issue:45372", + "issue:45375", + "issue:45440", + "issue:45464", + "issue:45478", + "issue:45542", + "issue:45588", + "issue:45753", + "issue:46123", + "issue:46172", + "issue:46231", + "issue:46369", + "issue:46655", + "issue:46664" + ], + "soft_pairs": [ + "issue:43673|issue:45341", + "issue:42175|issue:42222", + "issue:43874|issue:44261", + "issue:41628|issue:45071", + "issue:45071|issue:45542", + "issue:43012|issue:43577", + "issue:43931|issue:44410", + "issue:44263|issue:46172", + "issue:44360|issue:46172", + "issue:44483|issue:45464", + "issue:44484|issue:45478", + "issue:27957|issue:43352", + "issue:43441|issue:44829", + "issue:43931|issue:44038", + "issue:43097|issue:43452", + "issue:45003|issue:45341", + "issue:43986|issue:45341", + "issue:45753|issue:46369", + "issue:44933|issue:46231", + "issue:43874|issue:44263", + "issue:44261|issue:44485", + "issue:44117|issue:44987", + "issue:43673|issue:46123", + "issue:43531|issue:45341", + "issue:45440|issue:46172", + "issue:43064|issue:44315", + "issue:43986|issue:45372", + "issue:43097|issue:45588", + "issue:38617|issue:42757", + "issue:46655|issue:46664", + "issue:44410|issue:45375" + ], + "prepared_review_unit_hash": "233c266f51f099656c69033308378f35196e58444db584bade6299e237f744a2", + "component_budget": { + "node_count": 473, + "item_count": 473, + "soft_pair_count": 1738, + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 31, + "serialized_chars": 36792, + "estimated_input_tokens": 9198, + "estimated_eval_tokens": 18652 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a heterogeneous mix of unrelated Transformers issues (import regressions, model-specific architecture bugs, backend/package dependency problems, and API behavior changes). I do not see any soft pair that clearly describes the same underlying bug or change, so all candidate edges should be rejected.", + "confidence": 0.96, + "canonical_issue_reason": "Issue 45372 is the strongest single anchor: it is an active, concrete regression report with a precise import failure, version scope, and clear downstream impact on Gemma 4 processor loading.", + "canonical_pr_reason": null, + "best_issue_reason": "45372 is the most representative issue to keep: it has a specific reproducible failure, explicit affected versions, and active references, making it more actionable than the other mostly model-specific reports.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43673", + "right": "issue:45341", + "accept": false, + "reason": "Different subsystems and failure modes: generation cache/chunked prefill vs a testing utility bug." + }, + { + "left": "issue:42175", + "right": "issue:42222", + "accept": false, + "reason": "One is a packaging/backend dependency issue; the other is a VitPose model breakage." + }, + { + "left": "issue:43874", + "right": "issue:44261", + "accept": false, + "reason": "Different models and bugs: missing image-patch method vs missing rms_norm_eps precision issue." + }, + { + "left": "issue:41628", + "right": "issue:45071", + "accept": false, + "reason": "Unrelated regressions: missing AutoImageProcessor import vs PreTrainedConfig type-checking breakage." + }, + { + "left": "issue:45071", + "right": "issue:45542", + "accept": false, + "reason": "Both touch backend-related code, but one is type checking and the other is a TensorFlow/tensorboard dependency problem." + }, + { + "left": "issue:43012", + "right": "issue:43577", + "accept": false, + "reason": "Both involve dtype/precision, but one is a compilation warning and the other is incorrect loaded module dtypes." + }, + { + "left": "issue:43931", + "right": "issue:44410", + "accept": false, + "reason": "Different Qwen model bugs: weight-shape mismatch on load vs missing projections in layer 0." + }, + { + "left": "issue:44263", + "right": "issue:46172", + "accept": false, + "reason": "Different model families and code paths: a torch.split indexing issue vs DeepSeek-V4 dtype/transpose behavior." + }, + { + "left": "issue:44360", + "right": "issue:46172", + "accept": false, + "reason": "Different underlying defects: missing ReLU in DSA indexer vs DeepSeek-V4 tensor conversion/transpose behavior." + }, + { + "left": "issue:44483", + "right": "issue:45464", + "accept": false, + "reason": "Both mention chat/completions, but one is a general API acceptance bug and the other is a Qwen3.5 streaming inference issue." + }, + { + "left": "issue:44484", + "right": "issue:45478", + "accept": false, + "reason": "Completely different topics: save_pretrained shard-size question vs Qwen3.5-Moe loading error." + }, + { + "left": "issue:27957", + "right": "issue:43352", + "accept": false, + "reason": "Same Flash Attention feature area, but different models and different support gaps; not the same bug." + }, + { + "left": "issue:43441", + "right": "issue:44829", + "accept": false, + "reason": "Different Flash Attention versions and symptoms: model loading failure vs degenerate training behavior." + }, + { + "left": "issue:43931", + "right": "issue:44038", + "accept": false, + "reason": "Related model family, but one is a specific shape mismatch and the other is a broader Transformers 5/Qwen3-VL-Moe issue." + }, + { + "left": "issue:43097", + "right": "issue:43452", + "accept": false, + "reason": "Unrelated breakages: removed config helper vs gguf_file tokenizer/model loading failure." + }, + { + "left": "issue:45003", + "right": "issue:45341", + "accept": false, + "reason": "Different areas: sys.modules access in modeling_utils vs a testing_utils bug." + }, + { + "left": "issue:43986", + "right": "issue:45341", + "accept": false, + "reason": "Different failure domains: AutoProcessor/video-model torchvision dependency vs testing utilities." + }, + { + "left": "issue:45753", + "right": "issue:46369", + "accept": false, + "reason": "Different models and symptoms: infinite loop on a specific image vs memory overflow in sam3-litetext." + }, + { + "left": "issue:44933", + "right": "issue:46231", + "accept": false, + "reason": "Unrelated: missing image_utils import vs documentation mismatch in set_training()." + }, + { + "left": "issue:43874", + "right": "issue:44263", + "accept": false, + "reason": "Different model components and bugs; no shared code path beyond general multimodal/model internals." + }, + { + "left": "issue:44261", + "right": "issue:44485", + "accept": false, + "reason": "One is a missing config epsilon precision bug; the other is a RoPE implementation discussion." + }, + { + "left": "issue:44117", + "right": "issue:44987", + "accept": false, + "reason": "Different loading failures: tokenizer mapping None handling vs a specific model load failure." + }, + { + "left": "issue:43673", + "right": "issue:46123", + "accept": false, + "reason": "Different pipelines and bugs: generation cache missing vs mask pipeline batch flag logic." + }, + { + "left": "issue:43531", + "right": "issue:45341", + "accept": false, + "reason": "Qwen3-MoE sliding window behavior is unrelated to the testing_utils bug." + }, + { + "left": "issue:45440", + "right": "issue:46172", + "accept": false, + "reason": "Same DeepSeek naming, but different model versions and different underlying implementation bugs." + }, + { + "left": "issue:43064", + "right": "issue:44315", + "accept": false, + "reason": "Different training/runtime issues: FSDP2 optimizer-state corruption vs Liger Kernel not applied with model_init." + }, + { + "left": "issue:43986", + "right": "issue:45372", + "accept": false, + "reason": "Both are loading-time import/dependency problems, but they affect different code paths and different missing dependencies." + }, + { + "left": "issue:43097", + "right": "issue:45588", + "accept": false, + "reason": "Different bug classes: removed tie_embeddings API vs flash_attention.py crashing on s_aux=None." + }, + { + "left": "issue:38617", + "right": "issue:42757", + "accept": false, + "reason": "Both are import errors, but for different symbols from different packages." + }, + { + "left": "issue:46655", + "right": "issue:46664", + "accept": false, + "reason": "Same video processor area, but padding calculation and num_frames semantics are distinct bugs." + }, + { + "left": "issue:44410", + "right": "issue:45375", + "accept": false, + "reason": "Both are Qwen3 vision/MoE-related, but one is missing projections and the other is a missing config field being dropped." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 21, + "review_unit_index": 50, + "review_unit_count": 56, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-473", + "nodes": [ + "issue:36683", + "issue:38617", + "issue:42890", + "issue:42907", + "issue:43012", + "issue:43064", + "issue:43066", + "issue:43097", + "issue:43208", + "issue:43352", + "issue:43502", + "issue:43550", + "issue:43582", + "issue:43825", + "issue:43873", + "issue:43906", + "issue:43976", + "issue:43986", + "issue:43992", + "issue:44016", + "issue:44038", + "issue:44062", + "issue:44077", + "issue:44117", + "issue:44162", + "issue:44222", + "issue:44351", + "issue:44368", + "issue:44380", + "issue:44410", + "issue:44484", + "issue:44655", + "issue:44704", + "issue:44936", + "issue:44938", + "issue:45103", + "issue:45478", + "issue:45479", + "issue:45584", + "issue:45712", + "issue:45753", + "issue:46050", + "issue:46132", + "issue:46172", + "issue:46178", + "issue:46298", + "issue:46365", + "issue:46421" + ], + "soft_pairs": [ + "issue:43097|issue:46050", + "issue:43986|issue:44484", + "issue:44038|issue:44410", + "issue:38617|issue:44351", + "issue:42890|issue:46298", + "issue:44062|issue:44117", + "issue:43097|issue:43873", + "issue:44936|issue:46365", + "issue:43502|issue:43992", + "issue:44704|issue:45103", + "issue:44380|issue:46421", + "issue:43986|issue:44117", + "issue:44938|issue:46178", + "issue:43986|issue:44016", + "issue:43208|issue:43550", + "issue:43825|issue:44655", + "issue:42907|issue:44655", + "issue:42907|issue:45753", + "issue:43012|issue:43550", + "issue:43012|issue:44368", + "issue:45712|issue:46298", + "issue:43992|issue:44484", + "issue:43066|issue:43906", + "issue:43352|issue:44380", + "issue:43352|issue:43976", + "issue:43097|issue:46298", + "issue:44162|issue:46172", + "issue:45584|issue:46365", + "issue:44380|issue:46132", + "issue:44077|issue:46365", + "issue:44222|issue:45478", + "issue:44380|issue:45479", + "issue:36683|issue:43582", + "issue:43064|issue:44380" + ], + "prepared_review_unit_hash": "470efb5e9774c15b51c051d42e22aedc04b301e8a7945cce9cad8978460f38d4", + "component_budget": { + "node_count": 473, + "item_count": 473, + "soft_pair_count": 1738, + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 34, + "serialized_chars": 37691, + "estimated_input_tokens": 9423, + "estimated_eval_tokens": 19102 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "These are mostly unrelated issue reports; I wouldn\u2019t merge any of the soft pairs as duplicates. The strongest representative issue is the v5 breaking-change report in 43097 because it is concrete, broadly relevant, and already referenced elsewhere.", + "confidence": 0.63, + "canonical_issue_reason": "issue:43097 is the clearest high-signal representative: a specific Transformers v5 API removal with inbound references and broad downstream impact.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:43097 is the best standalone issue to anchor the set because it is specific, impactful, and more central than the other reports here.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43097", + "right": "issue:46050", + "accept": false, + "reason": "Different problems: a removed v5 embedding-tie API vs broken save/load for quantized sinq models." + }, + { + "left": "issue:43986", + "right": "issue:44484", + "accept": false, + "reason": "One is a torchvision dependency crash in AutoProcessor; the other is a save_pretrained shard-size question." + }, + { + "left": "issue:44038", + "right": "issue:44410", + "accept": false, + "reason": "Both are Qwen-related, but they describe different models and different missing-component bugs." + }, + { + "left": "issue:38617", + "right": "issue:44351", + "accept": false, + "reason": "Both are import errors, but for different missing symbols and likely unrelated code paths." + }, + { + "left": "issue:42890", + "right": "issue:46298", + "accept": false, + "reason": "Test flakiness from missing set_seed is unrelated to a noisy logging message." + }, + { + "left": "issue:44062", + "right": "issue:44117", + "accept": false, + "reason": "Tokenizer construction error vs mapping lookup returning None; different failures in the tokenizer stack." + }, + { + "left": "issue:43097", + "right": "issue:43873", + "accept": false, + "reason": "Removed v5 API vs quantization/offloading behavior; no shared concrete bug." + }, + { + "left": "issue:44936", + "right": "issue:46365", + "accept": false, + "reason": "Trainer.evaluate breakage is not the same as an import failure involving datasets." + }, + { + "left": "issue:43502", + "right": "issue:43992", + "accept": false, + "reason": "Offline file lookup bug vs missing encoder weights on load; different code paths." + }, + { + "left": "issue:44704", + "right": "issue:45103", + "accept": false, + "reason": "AutoProcessor kwargs propagation and auto-docstring annotation parsing are unrelated." + }, + { + "left": "issue:44380", + "right": "issue:46421", + "accept": false, + "reason": "Attention backend scaling bug vs cache lazy-init recompilation issue; different subsystems." + }, + { + "left": "issue:43986", + "right": "issue:44117", + "accept": false, + "reason": "Video-model loading without torchvision is unrelated to tokenizer mapping returning None." + }, + { + "left": "issue:44938", + "right": "issue:46178", + "accept": false, + "reason": "Python 3.14 import failure vs unsupported backend mapping; different root causes." + }, + { + "left": "issue:43986", + "right": "issue:44016", + "accept": false, + "reason": "A dependency crash is not the same as a notebook syntax error." + }, + { + "left": "issue:43208", + "right": "issue:43550", + "accept": false, + "reason": "xLSTM training blockers and a Bamba torch.compile SDPA failure are separate bugs." + }, + { + "left": "issue:43825", + "right": "issue:44655", + "accept": false, + "reason": "Pipeline task messaging and pipeline serialization are different issues." + }, + { + "left": "issue:42907", + "right": "issue:44655", + "accept": false, + "reason": "Dequantized model save failure is not the same as inability to save Pipeline objects." + }, + { + "left": "issue:42907", + "right": "issue:45753", + "accept": false, + "reason": "Saving dequantized models and an infinite-loop inference bug are unrelated." + }, + { + "left": "issue:43012", + "right": "issue:43550", + "accept": false, + "reason": "A precision warning during compile is not the same as a model failure under SDPA compile." + }, + { + "left": "issue:43012", + "right": "issue:44368", + "accept": false, + "reason": "Both mention warnings, but they concern different models and different tie-embedding behavior." + }, + { + "left": "issue:45712", + "right": "issue:46298", + "accept": false, + "reason": "Repository dummy-class cleanup is unrelated to logging noise." + }, + { + "left": "issue:43992", + "right": "issue:44484", + "accept": false, + "reason": "Missing embed_tokens weights on load is not the same as the 50GB shard-size default question." + }, + { + "left": "issue:43066", + "right": "issue:43906", + "accept": false, + "reason": "A tokenizer decoder-type regression is not enough to identify the reproduction issue as the same bug." + }, + { + "left": "issue:43352", + "right": "issue:44380", + "accept": false, + "reason": "FA2 unsupported for Nemotron and GPT2 attention scaling ignored are different backend issues." + }, + { + "left": "issue:43352", + "right": "issue:43976", + "accept": false, + "reason": "Model-specific FA2 support and a Python-version compatibility problem are unrelated." + }, + { + "left": "issue:43097", + "right": "issue:46298", + "accept": false, + "reason": "A v5 API removal is unrelated to noisy logging output." + }, + { + "left": "issue:44162", + "right": "issue:46172", + "accept": false, + "reason": "ESM2 breakage and a DeepSeek tensor operation issue do not share the same underlying bug." + }, + { + "left": "issue:45584", + "right": "issue:46365", + "accept": false, + "reason": "Whisper empty-transcription failure is unrelated to importing Trainer." + }, + { + "left": "issue:44077", + "right": "issue:46365", + "accept": false, + "reason": "patchtsmixer post_init validation and Trainer import failure are different bugs." + }, + { + "left": "issue:44222", + "right": "issue:45478", + "accept": false, + "reason": "Both involve MoE/FP8, but one is save_pretrained and the other is from_pretrained; not enough to treat as the same concrete bug." + }, + { + "left": "issue:44351", + "right": "issue:45478", + "accept": false, + "reason": "HybridCache import failure and Qwen3.5 MoE loading error are unrelated." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 21, + "review_unit_index": 51, + "review_unit_count": 56, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-473", + "nodes": [ + "issue:36010", + "issue:36683", + "issue:38617", + "issue:42175", + "issue:42490", + "issue:42757", + "issue:42898", + "issue:42994", + "issue:43012", + "issue:43066", + "issue:43208", + "issue:43299", + "issue:43352", + "issue:43475", + "issue:43502", + "issue:43525", + "issue:43526", + "issue:43611", + "issue:43618", + "issue:43756", + "issue:43761", + "issue:43824", + "issue:43867", + "issue:43901", + "issue:43937", + "issue:43976", + "issue:44038", + "issue:44077", + "issue:44230", + "issue:44380", + "issue:44483", + "issue:44485", + "issue:44514", + "issue:44625", + "issue:44655", + "issue:44683", + "issue:44938", + "issue:45216", + "issue:45276", + "issue:45290", + "issue:45600", + "issue:45698", + "issue:46077", + "issue:46082", + "issue:46172", + "issue:46298", + "issue:46365" + ], + "soft_pairs": [ + "issue:42175|issue:46365", + "issue:42490|issue:43012", + "issue:43208|issue:43526", + "issue:42757|issue:43824", + "issue:43012|issue:44683", + "issue:44483|issue:44514", + "issue:45600|issue:46298", + "issue:44038|issue:44230", + "issue:43618|issue:43761", + "issue:43618|issue:44380", + "issue:45698|issue:46172", + "issue:38617|issue:43502", + "issue:43756|issue:44485", + "issue:43299|issue:43867", + "issue:36683|issue:43475", + "issue:42994|issue:44655", + "issue:44483|issue:45290", + "issue:44625|issue:46365", + "issue:42898|issue:43066", + "issue:45276|issue:46077", + "issue:36010|issue:42757", + "issue:43352|issue:43937", + "issue:43901|issue:44655", + "issue:43611|issue:44655", + "issue:36683|issue:44077", + "issue:43976|issue:44938", + "issue:43867|issue:45216", + "issue:44625|issue:46077", + "issue:46077|issue:46082", + "issue:36683|issue:43525", + "issue:43618|issue:45216", + "issue:36683|issue:44625" + ], + "prepared_review_unit_hash": "92ece040d5a43de40264a436b5abe0d81e4f4552c067f7473da2a8ae9b314212", + "component_budget": { + "node_count": 473, + "item_count": 473, + "soft_pair_count": 1738, + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 32, + "serialized_chars": 36268, + "estimated_input_tokens": 9067, + "estimated_eval_tokens": 18390 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is highly heterogeneous: most pairs only share superficial keywords like import, load, save, or v5, but they point to different models, configs, and code paths. I would not merge any of the soft edges as duplicates.", + "confidence": 0.92, + "canonical_issue_reason": "No single canonical issue stands out because the items span unrelated bugs across imports, model configs, serialization, attention behavior, and packaging/backend issues.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 45698 is the closest representative only because it has the strongest discussion/inbound activity and is a broad save/load regression, but it still does not unify the rest of the cluster.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:42175", + "right": "issue:46365", + "accept": false, + "reason": "Packaging/backend install behavior vs Trainer import blocked by datasets; different failure modes and root causes." + }, + { + "left": "issue:42490", + "right": "issue:43012", + "accept": false, + "reason": "Local-path import warning vs PyTorch bfloat16 compile warning; unrelated code paths." + }, + { + "left": "issue:43208", + "right": "issue:43526", + "accept": false, + "reason": "xLSTM training bugs and BeitImageProcessorFast reduce_labels are different models and bugs." + }, + { + "left": "issue:42757", + "right": "issue:43824", + "accept": false, + "reason": "Both are import errors, but they concern different missing symbols in different modules." + }, + { + "left": "issue:43012", + "right": "issue:44683", + "accept": false, + "reason": "A warning during bf16 compilation is not the same as compiled flex_attention failing on newer torch." + }, + { + "left": "issue:44483", + "right": "issue:44514", + "accept": false, + "reason": "Chat-completions request validation and apply_chat_template batch crash are different layers and symptoms." + }, + { + "left": "issue:45600", + "right": "issue:46298", + "accept": false, + "reason": "Stale auto_mappings references vs noisy logging; not the same underlying bug." + }, + { + "left": "issue:44038", + "right": "issue:44230", + "accept": false, + "reason": "Qwen3-VL-Moe loading regression vs fp8 inference support request; related family, different change." + }, + { + "left": "issue:43618", + "right": "issue:43761", + "accept": false, + "reason": "Both involve CLIP regression, but one drops attentions and the other drops hidden_states; distinct bugs." + }, + { + "left": "issue:43618", + "right": "issue:44380", + "accept": false, + "reason": "CLIP output regression and GPT2 attention-scaling behavior are unrelated code paths." + }, + { + "left": "issue:45698", + "right": "issue:46172", + "accept": false, + "reason": "Save/load custom-module resolution bug vs DeepSeek dtype/transpose bug; unrelated." + }, + { + "left": "issue:38617", + "right": "issue:43502", + "accept": false, + "reason": "Missing import from configuration_utils vs local_files_only still triggering network access; different issues." + }, + { + "left": "issue:43756", + "right": "issue:44485", + "accept": false, + "reason": "Smollm3 RoPE layer count bug vs GLM-5 RoPE implementation discussion; different models and defects." + }, + { + "left": "issue:43299", + "right": "issue:43867", + "accept": false, + "reason": "Qwen3VL-Moe loading breakage vs state_dict-sorted load failure; both loading-related but not the same bug." + }, + { + "left": "issue:36683", + "right": "issue:43475", + "accept": false, + "reason": "Missing vocab_size on Gemma3Config vs missing fpn_position_embeddings on SAM3 output; different objects and models." + }, + { + "left": "issue:42994", + "right": "issue:44655", + "accept": false, + "reason": "Quantized model saving failure vs pipeline save_pretrained failure; both saving-related but distinct root causes." + }, + { + "left": "issue:44483", + "right": "issue:45290", + "accept": false, + "reason": "Request rejection at /v1/chat/completions vs chat_template crash on tool-call messages; different problems." + }, + { + "left": "issue:44625", + "right": "issue:46365", + "accept": false, + "reason": "Qwen3.5 num_labels propagation vs Trainer import conflict with datasets; unrelated." + }, + { + "left": "issue:42898", + "right": "issue:43066", + "accept": false, + "reason": "Tokenization-space behavior change vs wrong tokenizer decoder type; both v5/tokenizer-related but not the same bug." + }, + { + "left": "issue:45276", + "right": "issue:46077", + "accept": false, + "reason": "Gemma4 embedding resizing bug vs Gemma4 bidirectional-attention masking bug; same model family, different fixes." + }, + { + "left": "issue:36010", + "right": "issue:42757", + "accept": false, + "reason": "GenerationMixin import path and is_offline_mode import path are different missing-symbol regressions." + }, + { + "left": "issue:43352", + "right": "issue:43937", + "accept": false, + "reason": "FlashAttention-2 unsupported for Nemotron vs invalid GenerationConfig for GLM-5; unrelated." + }, + { + "left": "issue:43901", + "right": "issue:44655", + "accept": false, + "reason": "Docs mismatch for return_all_scores vs pipeline save_pretrained failure; not a duplicate." + }, + { + "left": "issue:43976", + "right": "issue:44938", + "accept": false, + "reason": "Python version support regression vs Python 3.14 load failure; different environment problems." + }, + { + "left": "issue:43867", + "right": "issue:45216", + "accept": false, + "reason": "Sorted state_dict loading error vs Qwen3.5 save_pretrained regression; both serialization-related but different failure modes." + }, + { + "left": "issue:44625", + "right": "issue:46077", + "accept": false, + "reason": "Qwen3.5 label propagation vs Gemma4 attention-mask construction; unrelated." + }, + { + "left": "issue:36683", + "right": "issue:43525", + "accept": false, + "reason": "Missing config field on Gemma3 vs missing config field on Llama4; same pattern, but different model-specific bugs." + }, + { + "left": "issue:43618", + "right": "issue:45216", + "accept": false, + "reason": "CLIP attentions regression vs Qwen3.5 save_pretrained regression; unrelated." + }, + { + "left": "issue:36683", + "right": "issue:44625", + "accept": false, + "reason": "Different model families and different config propagation bugs." + }, + { + "left": "issue:36683", + "right": "issue:44077", + "accept": false, + "reason": "Gemma3 missing config attribute vs patchtsmixer post_init policy change; 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": 21, + "review_unit_index": 52, + "review_unit_count": 56, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-473", + "nodes": [ + "issue:28282", + "issue:33453", + "issue:36296", + "issue:36683", + "issue:41669", + "issue:42890", + "issue:42898", + "issue:43010", + "issue:43064", + "issue:43502", + "issue:43575", + "issue:43618", + "issue:43723", + "issue:43867", + "issue:43874", + "issue:43976", + "issue:43992", + "issue:44062", + "issue:44075", + "issue:44077", + "issue:44112", + "issue:44222", + "issue:44246", + "issue:44355", + "issue:44360", + "issue:44492", + "issue:44556", + "issue:44734", + "issue:44855", + "issue:44857", + "issue:44938", + "issue:44964", + "issue:45071", + "issue:45125", + "issue:45278", + "issue:45310", + "issue:45357", + "issue:45375", + "issue:45440", + "issue:45478", + "issue:45561", + "issue:45600", + "issue:45850", + "issue:46202", + "issue:46285", + "issue:46365", + "issue:46617", + "issue:46664" + ], + "soft_pairs": [ + "issue:43992|issue:45310", + "issue:43064|issue:44492", + "issue:44075|issue:46365", + "issue:43010|issue:45278", + "issue:44355|issue:44855", + "issue:44938|issue:45071", + "issue:43867|issue:45375", + "issue:43867|issue:46202", + "issue:36683|issue:43064", + "issue:42890|issue:45561", + "issue:41669|issue:44246", + "issue:43010|issue:45850", + "issue:36296|issue:43575", + "issue:43010|issue:43976", + "issue:43618|issue:44077", + "issue:43867|issue:45357", + "issue:43992|issue:45478", + "issue:28282|issue:33453", + "issue:43618|issue:46285", + "issue:43618|issue:44857", + "issue:44360|issue:45440", + "issue:42890|issue:44964", + "issue:42890|issue:44112", + "issue:42890|issue:45600", + "issue:42898|issue:46664", + "issue:43874|issue:46617", + "issue:43502|issue:44062", + "issue:43723|issue:44556", + "issue:44222|issue:45125", + "issue:44734|issue:46617" + ], + "prepared_review_unit_hash": "9cf530c9aaed79e8dc90656ad39f865144c135458743a50da4dfde3ff19fc7c4", + "component_budget": { + "node_count": 473, + "item_count": 473, + "soft_pair_count": 1738, + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 30, + "serialized_chars": 36255, + "estimated_input_tokens": 9064, + "estimated_eval_tokens": 18384 + }, + "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 v5 regression reports spanning model loading, tokenizers, tensor parallelism, tests, and import-time behavior. Only one soft edge looks like a plausible duplicate pair (import slowdown / import * issue).", + "confidence": 0.76, + "canonical_issue_reason": "issue:45278 is the best umbrella issue here: it is the broadest and most representative post-upgrade regression report, with enough detail to route multiple import-related breakages.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:45278 is the strongest single issue to anchor this cluster because it summarizes a wider class of v5 incompatibilities rather than a narrow model-specific bug.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43992", + "right": "issue:45310", + "accept": false, + "reason": "Different model families and failure modes: missing UMT5 embed weights vs Qwen3.5 MoE from_pretrained error." + }, + { + "left": "issue:43064", + "right": "issue:44492", + "accept": false, + "reason": "Trainer/FSDP optimizer-state bug is unrelated to a cache-strategy typo." + }, + { + "left": "issue:44075", + "right": "issue:46365", + "accept": false, + "reason": "SGD argument handling is unrelated to importing Trainer and dataset dependency conflicts." + }, + { + "left": "issue:43010", + "right": "issue:45278", + "accept": false, + "reason": "Cache/layer no_grad semantics are not the same underlying bug as broad import failures after upgrade." + }, + { + "left": "issue:44355", + "right": "issue:44855", + "accept": false, + "reason": "Both involve Python/import errors, but one is compiled-file execution and the other is a specific JIT parsing/indentation issue." + }, + { + "left": "issue:44938", + "right": "issue:45071", + "accept": false, + "reason": "Python 3.14 import failure and PretrainedConfig type-checking regression are different problems." + }, + { + "left": "issue:43867", + "right": "issue:45375", + "accept": false, + "reason": "State-dict sorting/load failure is unrelated to a missing Qwen3.5 vision config field." + }, + { + "left": "issue:43867", + "right": "issue:46202", + "accept": false, + "reason": "Model load ordering issues are not the same as Gemma pipeline roundtrip tokenization behavior." + }, + { + "left": "issue:36683", + "right": "issue:43064", + "accept": false, + "reason": "Gemma3Config missing vocab_size is unrelated to FSDP2/PEFT optimizer state corruption." + }, + { + "left": "issue:42890", + "right": "issue:45561", + "accept": false, + "reason": "Both are test failures, but one is seed nondeterminism and the other is an xdist file race." + }, + { + "left": "issue:41669", + "right": "issue:44246", + "accept": true, + "reason": "These point to the same underlying import-time slowdown problem; one reports slow imports, the other identifies the import-* source and cost." + }, + { + "left": "issue:43010", + "right": "issue:45850", + "accept": false, + "reason": "Generic 'breaks too many models' does not match the specific cache update no_grad bug." + }, + { + "left": "issue:36296", + "right": "issue:43575", + "accept": false, + "reason": "Generic tensor-parallel training bug and a Qwen2 TP OOM are not clearly the same defect." + }, + { + "left": "issue:43010", + "right": "issue:43976", + "accept": false, + "reason": "Cache update no_grad semantics are unrelated to Python version compatibility." + }, + { + "left": "issue:43618", + "right": "issue:44077", + "accept": false, + "reason": "CLIP attentions regression and patchtsmixer post_init policy are unrelated model issues." + }, + { + "left": "issue:43867", + "right": "issue:45357", + "accept": false, + "reason": "Different Qwen3.5 problems: state-dict/load ordering versus incorrect visual encoder keys on save." + }, + { + "left": "issue:43992", + "right": "issue:45478", + "accept": false, + "reason": "Same broad phrase 'from_pretrained error', but the model and missing component differ; not the same bug." + }, + { + "left": "issue:28282", + "right": "issue:33453", + "accept": false, + "reason": "AutoModel missing PyTorch and tokenizer loading regression are unrelated." + }, + { + "left": "issue:43618", + "right": "issue:46285", + "accept": false, + "reason": "CLIPOutput attentions regression is unrelated to CLIP text model structure divergence." + }, + { + "left": "issue:43618", + "right": "issue:44857", + "accept": false, + "reason": "CLIP attentions handling and LwDetr AMP crash are different code paths." + }, + { + "left": "issue:44360", + "right": "issue:45440", + "accept": false, + "reason": "DSA indexer/ReLU discussion is unrelated to DeepSeekV3MoE implementation divergence." + }, + { + "left": "issue:42890", + "right": "issue:44964", + "accept": false, + "reason": "Seed-related test flakiness is unrelated to Phi-4 multimodal model loading failures." + }, + { + "left": "issue:42890", + "right": "issue:44112", + "accept": false, + "reason": "A flaky integration test and a GraniteSpeech CI stale override issue are different defects." + }, + { + "left": "issue:42890", + "right": "issue:45600", + "accept": false, + "reason": "Test seeding flakiness is unrelated to auto_mappings references breaking CI." + }, + { + "left": "issue:42898", + "right": "issue:46664", + "accept": false, + "reason": "Tokenizer cleanup behavior in v5 is unrelated to SmolVLM video processor semantics." + }, + { + "left": "issue:43874", + "right": "issue:46617", + "accept": false, + "reason": "Missing image-patch counting API and trust_remote_code cache FileNotFoundError are different loading paths." + }, + { + "left": "issue:43502", + "right": "issue:44062", + "accept": false, + "reason": "Offline API-request leakage is unrelated to AddedToken keyword duplication." + }, + { + "left": "issue:43723", + "right": "issue:44556", + "accept": false, + "reason": "Tokenizer loading regression and checkpoint reload compatibility regression are not clearly the same bug." + }, + { + "left": "issue:44222", + "right": "issue:45125", + "accept": false, + "reason": "FP8 save_pretrained behavior and missing tensor-parallel plan are distinct issues." + }, + { + "left": "issue:44734", + "right": "issue:46617", + "accept": false, + "reason": "KV-cache continuation tensor indexing and cached trust_remote_code file lookup are unrelated." + }, + { + "left": "issue:43867", + "right": "issue:46617", + "accept": false, + "reason": "State-dict ordering/load problems do not match the trust_remote_code cache FileNotFoundError." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The cluster summary is well grounded in the packet: the issues are indeed a mixed set of unrelated v5 regressions, and the one accepted soft edge (41669/44246) is a conservative, mergeable duplicate pair about import-time slowdown from import-* usage. No soft-edge verdict appears overstated." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 21, + "review_unit_index": 53, + "review_unit_count": 56, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-473", + "nodes": [ + "issue:27957", + "issue:30333", + "issue:30990", + "issue:31795", + "issue:33453", + "issue:34689", + "issue:36296", + "issue:38617", + "issue:41628", + "issue:42548", + "issue:42757", + "issue:43010", + "issue:43065", + "issue:43232", + "issue:43299", + "issue:43475", + "issue:43502", + "issue:43526", + "issue:43576", + "issue:43824", + "issue:43827", + "issue:43901", + "issue:44222", + "issue:44230", + "issue:44315", + "issue:44351", + "issue:44355", + "issue:44360", + "issue:44387", + "issue:44485", + "issue:44556", + "issue:44655", + "issue:44749", + "issue:44908", + "issue:44938", + "issue:44961", + "issue:45071", + "issue:45125", + "issue:45375", + "issue:45468", + "issue:45478", + "issue:46132", + "issue:46143", + "issue:46144", + "issue:46209", + "issue:46617", + "issue:46664" + ], + "soft_pairs": [ + "issue:44387|issue:44556", + "issue:43010|issue:43065", + "issue:43502|issue:44485", + "issue:43824|issue:46617", + "issue:43576|issue:46132", + "issue:45125|issue:45478", + "issue:30333|issue:33453", + "issue:44355|issue:44749", + "issue:45125|issue:45375", + "issue:42548|issue:44351", + "issue:43010|issue:43232", + "issue:30990|issue:31795", + "issue:43502|issue:44222", + "issue:43502|issue:44961", + "issue:43827|issue:44961", + "issue:30333|issue:36296", + "issue:43299|issue:44556", + "issue:44355|issue:45071", + "issue:44355|issue:44655", + "issue:33453|issue:34689", + "issue:27957|issue:33453", + "issue:42757|issue:43576", + "issue:43576|issue:45071", + "issue:44315|issue:44360", + "issue:43576|issue:43824", + "issue:31795|issue:38617", + "issue:44961|issue:45468", + "issue:44351|issue:46617", + "issue:44485|issue:44961", + "issue:43576|issue:44938", + "issue:44908|issue:44961", + "issue:36296|issue:43901", + "issue:44222|issue:44230", + "issue:46143|issue:46144", + "issue:36296|issue:43526", + "issue:43475|issue:46617", + "issue:41628|issue:42548", + "issue:44315|issue:46209", + "issue:44485|issue:46664" + ], + "prepared_review_unit_hash": "bf82ddf7059586eb0b090f8663715dee38c7e302dd19957277d12ff5fd6e9487", + "component_budget": { + "node_count": 473, + "item_count": 473, + "soft_pair_count": 1738, + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 39, + "serialized_chars": 38108, + "estimated_input_tokens": 9527, + "estimated_eval_tokens": 19310 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "These items are mostly unrelated, one-off bugs and documentation issues rather than a single duplicate cluster. The only near-match is the two RoFormer issues, which both point to missing RoFormer/attention-interface plumbing; the rest should stay separate.", + "confidence": 0.89, + "canonical_issue_reason": "If a single issue must serve as the anchor, issue 46617 is the most concrete and generally actionable runtime bug: it describes a reproducible cache-loading failure with a clear code path. That said, the overall set is too heterogeneous for a true canonical duplicate target.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 46617 is the strongest standalone candidate because it reports a specific failure mode, affects a common load path, and is easier to validate and triage than the more model-specific or documentation-only reports.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44387", + "right": "issue:44556", + "accept": false, + "reason": "Both are version-regression reports, but one is int4 memory/OOM and the other is checkpoint reload failure; different symptoms and likely different fixes." + }, + { + "left": "issue:43010", + "right": "issue:43065", + "accept": false, + "reason": "One is about no_grad on cache/layer updates; the other is a Sam3PixelDecoder Conv2d placeholder issue. No shared code-path bug." + }, + { + "left": "issue:43502", + "right": "issue:44485", + "accept": false, + "reason": "local_files_only network leakage is unrelated to GLM-5 RoPE implementation." + }, + { + "left": "issue:43824", + "right": "issue:46617", + "accept": false, + "reason": "Both involve loading/imports, but one is a missing export for a specific model class and the other is a trust_remote_code cache FileNotFoundError; not the same bug." + }, + { + "left": "issue:43576", + "right": "issue:46132", + "accept": false, + "reason": "Broken `transformers env` CLI output and AttentionInterface.register behavior are separate issues in different areas." + }, + { + "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 from_pretrained regression; too different to merge." + }, + { + "left": "issue:30333", + "right": "issue:33453", + "accept": false, + "reason": "MLflow reporting on failed training jobs is unrelated to tokenizer-loading regression." + }, + { + "left": "issue:44355", + "right": "issue:44749", + "accept": false, + "reason": "Compiled-Python-file errors and a data-filtering slowdown after upgrade are unrelated symptoms." + }, + { + "left": "issue:45125", + "right": "issue:45375", + "accept": false, + "reason": "Missing `_tp_plan` for tensor parallelism and a missing vision-config field are different model-config bugs." + }, + { + "left": "issue:42548", + "right": "issue:44351", + "accept": false, + "reason": "Both are import errors, but for different symbols and likely different packaging/export causes." + }, + { + "left": "issue:43010", + "right": "issue:43232", + "accept": false, + "reason": "Cache/layer update no_grad and generation kwargs update after sync_gpus are separate generation/caching problems." + }, + { + "left": "issue:30990", + "right": "issue:31795", + "accept": false, + "reason": "Sentence Transformers loading hangs and docs confusion about forward arguments are unrelated." + }, + { + "left": "issue:43502", + "right": "issue:44222", + "accept": false, + "reason": "local_files_only network leakage is unrelated to FP8 save_pretrained behavior." + }, + { + "left": "issue:43502", + "right": "issue:44961", + "accept": false, + "reason": "A network-policy bug and a placeholder issue are not the same underlying change." + }, + { + "left": "issue:43827", + "right": "issue:44961", + "accept": false, + "reason": "Doc references to removed pipelines and an unrelated placeholder issue should not be merged." + }, + { + "left": "issue:30333", + "right": "issue:36296", + "accept": false, + "reason": "MLflow reporting and tensor-parallel training are unrelated." + }, + { + "left": "issue:43299", + "right": "issue:44556", + "accept": false, + "reason": "Both are upgrade regressions, but they target different model-loading failures and different version ranges." + }, + { + "left": "issue:44355", + "right": "issue:45071", + "accept": false, + "reason": "Compiled-file errors and `PretrainedConfig` type-checking breakage are unrelated." + }, + { + "left": "issue:44355", + "right": "issue:44655", + "accept": false, + "reason": "Running compiled Python files and saving Pipeline objects are different failures." + }, + { + "left": "issue:33453", + "right": "issue:34689", + "accept": false, + "reason": "Tokenizer-loading regression and Llama 3.2 model-loading regression are distinct bugs." + }, + { + "left": "issue:27957", + "right": "issue:33453", + "accept": false, + "reason": "Flash Attention 2 support for XLM-Roberta and tokenizer-loading regression are unrelated." + }, + { + "left": "issue:42757", + "right": "issue:43576", + "accept": false, + "reason": "Missing `is_offline_mode` import from huggingface_hub and broken env CLI are separate issues." + }, + { + "left": "issue:43576", + "right": "issue:45071", + "accept": false, + "reason": "CLI/env breakage and `PretrainedConfig` type checking are unrelated." + }, + { + "left": "issue:44315", + "right": "issue:44360", + "accept": false, + "reason": "Liger Kernel application and a DSA indexer ReLU discussion are unrelated." + }, + { + "left": "issue:43576", + "right": "issue:43824", + "accept": false, + "reason": "Broken env command and missing Qwen2.5-VL export are different code paths." + }, + { + "left": "issue:31795", + "right": "issue:38617", + "accept": false, + "reason": "Documentation confusion and an import error for `layer_type_validation` are unrelated." + }, + { + "left": "issue:44961", + "right": "issue:45468", + "accept": false, + "reason": "A placeholder issue and Gemma-4 audio positional encoding bug are not duplicates." + }, + { + "left": "issue:44351", + "right": "issue:46617", + "accept": false, + "reason": "Missing `HybridCache` export and cache FileNotFoundError are both loading-related but not the same defect." + }, + { + "left": "issue:44485", + "right": "issue:44961", + "accept": false, + "reason": "GLM-5 RoPE implementation discussion is unrelated to the placeholder issue." + }, + { + "left": "issue:43576", + "right": "issue:44938", + "accept": false, + "reason": "Broken env CLI and Python 3.14 load failure are separate regressions." + }, + { + "left": "issue:44908", + "right": "issue:44961", + "accept": false, + "reason": "Scheduler kwargs handling and the placeholder issue are unrelated." + }, + { + "left": "issue:36296", + "right": "issue:43901", + "accept": false, + "reason": "Tensor-parallel training and pipeline docs mention are different problems." + }, + { + "left": "issue:44222", + "right": "issue:44230", + "accept": false, + "reason": "FP8 save_pretrained and FP8 inference support for Qwen3-VL/Qwen3.5 MoE are related only at a broad feature level, not the same bug." + }, + { + "left": "issue:46143", + "right": "issue:46144", + "accept": true, + "reason": "Both describe the same RoFormer plumbing gap: kwargs are not propagated and the attention implementation does not use the attention interface, which points to one underlying integration defect." + }, + { + "left": "issue:36296", + "right": "issue:43526", + "accept": false, + "reason": "Tensor parallel training and BeitImageProcessorFast label reduction are unrelated." + }, + { + "left": "issue:43475", + "right": "issue:46617", + "accept": false, + "reason": "SAM3 vision encoder attribute bug and trust_remote_code cache loading failure are unrelated." + }, + { + "left": "issue:41628", + "right": "issue:42548", + "accept": false, + "reason": "Different import errors for different symbols; no shared underlying fix is evident." + }, + { + "left": "issue:44315", + "right": "issue:46209", + "accept": false, + "reason": "Liger Kernel application and FP8 inference on B200 are different feature/regression areas." + }, + { + "left": "issue:44485", + "right": "issue:46664", + "accept": false, + "reason": "GLM-5 RoPE implementation and SmolVLM video-processor semantics are unrelated." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet: it correctly treats the set as mostly heterogeneous and only flags the RoFormer pair as a plausible near-match. The individual soft-edge rejections are conservative overall, and the one accepted pair is reasonably supported by the issue titles." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 21, + "review_unit_index": 54, + "review_unit_count": 56, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-473", + "nodes": [ + "issue:31795", + "issue:33453", + "issue:41669", + "issue:42371", + "issue:42890", + "issue:43352", + "issue:43408", + "issue:43525", + "issue:43572", + "issue:43576", + "issue:43673", + "issue:43827", + "issue:43976", + "issue:44038", + "issue:44077", + "issue:44230", + "issue:44246", + "issue:44273", + "issue:44295", + "issue:44297", + "issue:44315", + "issue:44351", + "issue:44360", + "issue:44393", + "issue:44492", + "issue:44556", + "issue:44623", + "issue:44704", + "issue:44829", + "issue:44861", + "issue:44908", + "issue:45003", + "issue:45071", + "issue:45092", + "issue:45125", + "issue:45230", + "issue:45278", + "issue:45341", + "issue:45362", + "issue:45381", + "issue:45478", + "issue:45542", + "issue:45753", + "issue:45850", + "issue:46095", + "issue:46218", + "issue:46328", + "issue:46664" + ], + "soft_pairs": [ + "issue:45362|issue:45850", + "issue:43673|issue:44492", + "issue:44623|issue:45230", + "issue:44360|issue:46664", + "issue:44273|issue:45542", + "issue:44038|issue:45125", + "issue:44273|issue:45003", + "issue:44393|issue:45753", + "issue:44393|issue:45071", + "issue:45381|issue:46664", + "issue:43576|issue:44556", + "issue:41669|issue:43673", + "issue:43525|issue:45850", + "issue:43572|issue:45850", + "issue:44230|issue:46095", + "issue:44556|issue:45092", + "issue:45278|issue:45850", + "issue:44492|issue:45850", + "issue:42371|issue:44492", + "issue:41669|issue:46218", + "issue:43408|issue:44315", + "issue:42890|issue:44908", + "issue:44246|issue:44273", + "issue:43352|issue:44393", + "issue:43576|issue:44908", + "issue:43976|issue:44393", + "issue:44704|issue:44908", + "issue:44908|issue:45071", + "issue:44295|issue:45230", + "issue:44297|issue:45230", + "issue:41669|issue:44351", + "issue:44492|issue:46328", + "issue:41669|issue:45341", + "issue:43976|issue:45850", + "issue:44393|issue:45478", + "issue:43976|issue:46218", + "issue:44077|issue:44861", + "issue:43827|issue:44908", + "issue:31795|issue:33453", + "issue:44315|issue:44829", + "issue:43576|issue:44273" + ], + "prepared_review_unit_hash": "fbc70cc6aacc7423d1fd46b044c96a9fec0b76e3eced96a0a38607a177313e54", + "component_budget": { + "node_count": 473, + "item_count": 473, + "soft_pair_count": 1738, + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 41, + "serialized_chars": 39442, + "estimated_input_tokens": 9861, + "estimated_eval_tokens": 19978 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The soft pairs are mostly loose thematic matches, not true duplicates. They span unrelated regressions across model configs, import/lazy-loading, docs, checkpoints, and scheduler/API changes. No pull requests are present.", + "confidence": 0.42, + "canonical_issue_reason": "issue:45478 is the most concrete, actionable report in the set: it names the model family, version threshold, and failure mode clearly, while most other issues are broader or more ambiguous.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:45478 is the strongest standalone issue to anchor triage because it is specific and reproducible; it is more useful than the broad upgrade-breakage complaints.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:45362", + "right": "issue:45850", + "accept": false, + "reason": "A specific Qwen3.5 crash and a broad \"many models break\" complaint are only thematically similar, not the same bug." + }, + { + "left": "issue:43673", + "right": "issue:44492", + "accept": false, + "reason": "Chunked-prefill cache regression vs a typo in cache strategy text/labels; different code paths." + }, + { + "left": "issue:44623", + "right": "issue:45230", + "accept": false, + "reason": "Missing files in processor.save_pretrained is unrelated to a generic bug report." + }, + { + "left": "issue:44360", + "right": "issue:46664", + "accept": false, + "reason": "DSA/ReLU discussion and SmolVLM video num_frames semantics are unrelated." + }, + { + "left": "issue:44273", + "right": "issue:45542", + "accept": false, + "reason": "Lazy loading/import behavior is not the same as the TensorFlow backend error from tensorboard-only installs." + }, + { + "left": "issue:44038", + "right": "issue:45125", + "accept": false, + "reason": "Both mention Qwen3.5 MoE, but one is a general bug/support report and the other is a tensor-parallel _tp_plan issue." + }, + { + "left": "issue:44273", + "right": "issue:45003", + "accept": false, + "reason": "Lazy loading malfunction and unsafe sys.modules access are different root causes." + }, + { + "left": "issue:44393", + "right": "issue:45753", + "accept": false, + "reason": "Qwen3-VL 2D bbox hallucination is not the same as an infinite loop on a specific image." + }, + { + "left": "issue:44393", + "right": "issue:45071", + "accept": false, + "reason": "A model-specific vision bug and a PretrainedConfig type-checking regression are unrelated." + }, + { + "left": "issue:45381", + "right": "issue:46664", + "accept": false, + "reason": "Both are video-related, but qwen2.5-vl vision_position_ids and SmolVLM num_frames semantics are separate issues." + }, + { + "left": "issue:43576", + "right": "issue:44556", + "accept": false, + "reason": "Broken env command and checkpoint reload incompatibility are unrelated." + }, + { + "left": "issue:41669", + "right": "issue:43673", + "accept": false, + "reason": "Import-star slowdown is a different problem from cache missing during chunked prefill." + }, + { + "left": "issue:43525", + "right": "issue:45850", + "accept": false, + "reason": "Missing pad_token_id in Llama4Config is not the same as a broad v5 breakage complaint." + }, + { + "left": "issue:43572", + "right": "issue:45850", + "accept": false, + "reason": "StableLmConfig pad_token_idx regression is distinct from generic version breakage." + }, + { + "left": "issue:44230", + "right": "issue:46095", + "accept": false, + "reason": "FP8 support request for Qwen3-VL/Qwen3.5 MoE is unrelated to a DeepSeekV4 weight-conversion question." + }, + { + "left": "issue:44556", + "right": "issue:45092", + "accept": false, + "reason": "Both concern loading older checkpoints after v5 upgrades, but the concrete failure modes are different." + }, + { + "left": "issue:45278", + "right": "issue:45850", + "accept": false, + "reason": "Many import errors after upgrade is a broad complaint, not the same bug as a generic \"too many models break\" issue." + }, + { + "left": "issue:44492", + "right": "issue:45850", + "accept": false, + "reason": "A cache-strategy typo is not the same as a general version-breakage report." + }, + { + "left": "issue:42371", + "right": "issue:44492", + "accept": false, + "reason": "TF32 API settings guidance and cache-strategy typo are unrelated." + }, + { + "left": "issue:41669", + "right": "issue:46218", + "accept": false, + "reason": "Import-star slowdown and Qwen3_5 module import failure are different failures." + }, + { + "left": "issue:43408", + "right": "issue:44315", + "accept": false, + "reason": "sam3_video/sam3_tracker warning and Liger Kernel application are unrelated." + }, + { + "left": "issue:42890", + "right": "issue:44908", + "accept": false, + "reason": "Missing set_seed in tests is unrelated to inverse_sqrt scheduler kwargs handling." + }, + { + "left": "issue:44246", + "right": "issue:44273", + "accept": false, + "reason": "Both touch imports/lazy loading, but one is intermittent import slowness and the other is a general lazy-loading bug; too broad to merge." + }, + { + "left": "issue:43352", + "right": "issue:44393", + "accept": false, + "reason": "Nemotron Flash Attention 2 support and Qwen3-VL bbox hallucination are unrelated." + }, + { + "left": "issue:43576", + "right": "issue:44908", + "accept": false, + "reason": "transformers env command and scheduler kwargs are unrelated." + }, + { + "left": "issue:43976", + "right": "issue:44393", + "accept": false, + "reason": "Python version compatibility and Qwen3-VL bbox output are unrelated." + }, + { + "left": "issue:44704", + "right": "issue:44908", + "accept": false, + "reason": "AutoProcessor kwargs passing and inverse_sqrt scheduler kwargs are unrelated." + }, + { + "left": "issue:44908", + "right": "issue:45071", + "accept": false, + "reason": "Scheduler kwargs handling and PretrainedConfig type checking are different bugs." + }, + { + "left": "issue:44295", + "right": "issue:45230", + "accept": false, + "reason": "position_ids buffer access error is unrelated to a generic bug report." + }, + { + "left": "issue:44297", + "right": "issue:45230", + "accept": false, + "reason": "tokenizer_class mismatch on save is unrelated to a generic bug report." + }, + { + "left": "issue:41669", + "right": "issue:44351", + "accept": false, + "reason": "Import-star slowdown and missing HybridCache import are different import-path problems." + }, + { + "left": "issue:44492", + "right": "issue:46328", + "accept": false, + "reason": "Cache-strategy typo and SequenceFeatureExtractor.pad performance bug are unrelated." + }, + { + "left": "issue:41669", + "right": "issue:45341", + "accept": false, + "reason": "Model import-star slowdown and a small testing_utils bug are unrelated." + }, + { + "left": "issue:43976", + "right": "issue:45850", + "accept": false, + "reason": "Python-version support regression and generic v5 breakage are not the same issue." + }, + { + "left": "issue:44393", + "right": "issue:45478", + "accept": false, + "reason": "Qwen3-VL bbox hallucination and Qwen3.5 MoE from_pretrained failure involve different models and code paths." + }, + { + "left": "issue:43976", + "right": "issue:46218", + "accept": false, + "reason": "Python support regression and Qwen3_5 import failure are unrelated." + }, + { + "left": "issue:44077", + "right": "issue:44861", + "accept": false, + "reason": "Optional post_init policy and tied-weights-key AttributeError are different code paths." + }, + { + "left": "issue:43827", + "right": "issue:44908", + "accept": false, + "reason": "Docs still referencing pipeline() and scheduler kwargs handling are unrelated." + }, + { + "left": "issue:31795", + "right": "issue:33453", + "accept": false, + "reason": "Documentation confusion and tokenizer loading regression are unrelated." + }, + { + "left": "issue:44315", + "right": "issue:44829", + "accept": false, + "reason": "Liger Kernel not applied with model_init and flash_attention_3 degenerate training are unrelated." + }, + { + "left": "issue:43576", + "right": "issue:44273", + "accept": false, + "reason": "Broken env command and lazy-loading malfunction are only broadly related to tooling, not duplicate 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": 21, + "review_unit_index": 55, + "review_unit_count": 56, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-473", + "nodes": [ + "issue:28348", + "issue:30333", + "issue:31515", + "issue:31795", + "issue:33453", + "issue:34689", + "issue:36296", + "issue:38617", + "issue:41669", + "issue:42371", + "issue:42913", + "issue:43066", + "issue:43298", + "issue:43352", + "issue:43502", + "issue:43519", + "issue:43525", + "issue:43572", + "issue:43576", + "issue:43606", + "issue:43618", + "issue:43704", + "issue:44230", + "issue:44261", + "issue:44263", + "issue:44273", + "issue:44351", + "issue:44360", + "issue:44485", + "issue:44492", + "issue:44623", + "issue:44655", + "issue:44861", + "issue:44908", + "issue:44938", + "issue:44998", + "issue:45230", + "issue:45250", + "issue:45290", + "issue:45310", + "issue:45468", + "issue:45850", + "issue:46018", + "issue:46095", + "issue:46172", + "issue:46218", + "issue:46655" + ], + "soft_pairs": [ + "issue:43519|issue:46655", + "issue:44360|issue:45468", + "issue:44655|issue:44861", + "issue:44273|issue:44908", + "issue:45230|issue:45310", + "issue:43576|issue:46218", + "issue:43519|issue:44485", + "issue:43519|issue:44908", + "issue:41669|issue:43704", + "issue:46018|issue:46172", + "issue:31795|issue:34689", + "issue:31515|issue:31795", + "issue:43704|issue:44485", + "issue:44485|issue:45468", + "issue:45850|issue:46218", + "issue:43502|issue:43519", + "issue:42371|issue:43704", + "issue:43519|issue:44623", + "issue:44263|issue:45468", + "issue:43606|issue:45468", + "issue:43298|issue:46095", + "issue:44261|issue:45468", + "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:43525|issue:44861", + "issue:31515|issue:38617", + "issue:43572|issue:44861", + "issue:44230|issue:44623", + "issue:28348|issue:43352", + "issue:30333|issue:31515", + "issue:31515|issue:36296", + "issue:44230|issue:45310", + "issue:31515|issue:33453", + "issue:43298|issue:44230" + ], + "prepared_review_unit_hash": "acf7e1d6e2e52ec0a26e21eed2e16a5ae5ea6bd5a9a4f8754f7266293a6a377e", + "component_budget": { + "node_count": 473, + "item_count": 473, + "soft_pair_count": 1738, + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 43, + "serialized_chars": 39014, + "estimated_input_tokens": 9754, + "estimated_eval_tokens": 19764 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This set is not a true duplicate cluster; it mixes many unrelated Transformers issues across model loading, tokenizers, processors, configs, scheduling, and feature requests. No soft pair looks like the same underlying bug/change, so all are rejected.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43519", + "right": "issue:46655", + "accept": false, + "reason": "Both are processor calculation bugs, but they affect different models and different logic (timestamp vs padding)." + }, + { + "left": "issue:44360", + "right": "issue:45468", + "accept": false, + "reason": "Different models and different defects; no shared code-path bug is evident." + }, + { + "left": "issue:44655", + "right": "issue:44861", + "accept": false, + "reason": "Both touch save/load internals, but one is pipeline save_pretrained and the other is tied-weight key handling." + }, + { + "left": "issue:44273", + "right": "issue:44908", + "accept": false, + "reason": "Lazy loading and scheduler kwargs are unrelated subsystems." + }, + { + "left": "issue:45230", + "right": "issue:45310", + "accept": false, + "reason": "Generic bug report versus a specific Qwen3.5 MOE from_pretrained regression; not the same issue." + }, + { + "left": "issue:43576", + "right": "issue:46218", + "accept": false, + "reason": "Broken env command and Qwen import failure are unrelated." + }, + { + "left": "issue:43519", + "right": "issue:44485", + "accept": false, + "reason": "Different model processors; one is timestamp math, the other is RoPE implementation." + }, + { + "left": "issue:43519", + "right": "issue:44908", + "accept": false, + "reason": "Processor timestamp bug and scheduler kwargs bug are unrelated." + }, + { + "left": "issue:41669", + "right": "issue:43704", + "accept": false, + "reason": "Import-star cleanup and VRAM leak are different code changes/problems." + }, + { + "left": "issue:46018", + "right": "issue:46172", + "accept": false, + "reason": "Both are DeepSeek-V4 related, but they describe different concrete bugs and code paths." + }, + { + "left": "issue:31795", + "right": "issue:34689", + "accept": false, + "reason": "Documentation confusion and model loading regression are not the same underlying bug." + }, + { + "left": "issue:31515", + "right": "issue:31795", + "accept": false, + "reason": "Slow checkpoint loading and forward-doc confusion are unrelated." + }, + { + "left": "issue:43704", + "right": "issue:44485", + "accept": false, + "reason": "VRAM leak in Qwen3 differs from GLM-5 RoPE implementation bug." + }, + { + "left": "issue:44485", + "right": "issue:45468", + "accept": false, + "reason": "Different model families and different failure modes." + }, + { + "left": "issue:45850", + "right": "issue:46218", + "accept": false, + "reason": "Broad v5 breakage report is not the same as a specific Qwen import error." + }, + { + "left": "issue:43502", + "right": "issue:43519", + "accept": false, + "reason": "Local-files-only network leak and timestamp calculation are unrelated." + }, + { + "left": "issue:42371", + "right": "issue:43704", + "accept": false, + "reason": "TF32 settings guidance and VRAM leak are different issues." + }, + { + "left": "issue:43519", + "right": "issue:44623", + "accept": false, + "reason": "Processor timestamp math and processor.save_pretrained missing files are different bugs." + }, + { + "left": "issue:44263", + "right": "issue:45468", + "accept": false, + "reason": "Torch.split indexing bug and Gemma audio positional encoding bug are unrelated." + }, + { + "left": "issue:43606", + "right": "issue:45468", + "accept": false, + "reason": "CPU offload device mismatch is unrelated to Gemma audio encoding." + }, + { + "left": "issue:43298", + "right": "issue:46095", + "accept": false, + "reason": "Continuous batching audio support and DeepSeek-V4 weight conversion are different topics." + }, + { + "left": "issue:44261", + "right": "issue:45468", + "accept": false, + "reason": "RMS norm epsilon precision issue and Gemma audio positional encoding are unrelated." + }, + { + "left": "issue:43519", + "right": "issue:43704", + "accept": false, + "reason": "Processor timestamp bug and Qwen VRAM leak do not share the same code-path." + }, + { + "left": "issue:44351", + "right": "issue:46218", + "accept": false, + "reason": "Both are import failures, but for different modules and causes." + }, + { + "left": "issue:43352", + "right": "issue:46218", + "accept": false, + "reason": "Flash-attention support gap and Qwen import failure are unrelated." + }, + { + "left": "issue:42913", + "right": "issue:43066", + "accept": false, + "reason": "Both concern tokenizer v5 behavior, but the titles point to different concrete failures." + }, + { + "left": "issue:28348", + "right": "issue:45250", + "accept": false, + "reason": "Specific GPT2 support request versus a generic Flash Attention 2.0 issue; not the same change." + }, + { + "left": "issue:44908", + "right": "issue:44998", + "accept": false, + "reason": "Scheduler kwarg bug and unrelated placeholder issue." + }, + { + "left": "issue:44492", + "right": "issue:44998", + "accept": false, + "reason": "Cache strategy typo and unrelated placeholder issue." + }, + { + "left": "issue:43618", + "right": "issue:44998", + "accept": false, + "reason": "CLIPOutput attentions regression is unrelated to the placeholder issue." + }, + { + "left": "issue:44998", + "right": "issue:45850", + "accept": false, + "reason": "Placeholder issue versus broad v5 breakage report; no duplicate relation." + }, + { + "left": "issue:44938", + "right": "issue:44998", + "accept": false, + "reason": "Python 3.14 load failure and unrelated placeholder issue." + }, + { + "left": "issue:44861", + "right": "issue:45290", + "accept": false, + "reason": "Two different crash sites: tied-weights handling versus chat template/tool-call handling." + }, + { + "left": "issue:43525", + "right": "issue:44861", + "accept": false, + "reason": "Missing pad_token_id on Llama4Config and tied-weights list crash are distinct bugs." + }, + { + "left": "issue:31515", + "right": "issue:38617", + "accept": false, + "reason": "Slow from_pretrained loading and layer_type_validation import error are different failures." + }, + { + "left": "issue:43572", + "right": "issue:44861", + "accept": false, + "reason": "StableLm pad token config regression and tied-weights crash are unrelated." + }, + { + "left": "issue:44230", + "right": "issue:44623", + "accept": false, + "reason": "FP8 support request and missing processor save files are not the same issue." + }, + { + "left": "issue:28348", + "right": "issue:43352", + "accept": false, + "reason": "Both mention Flash Attention 2.0, but they target different models and separate support requests." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 21, + "review_unit_index": 56, + "review_unit_count": 56, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-473", + "nodes": [ + "issue:37934", + "issue:42936", + "issue:43240", + "issue:43668", + "issue:43717", + "issue:43813", + "issue:44031", + "issue:46018", + "issue:46144" + ], + "soft_pairs": [ + "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": "0abe26ce51b0176b9f4df243a9bbef12bc99d966f3469b311486a37cb88b6fda", + "component_budget": { + "node_count": 473, + "item_count": 473, + "soft_pair_count": 1738, + "serialized_chars": 901434, + "estimated_input_tokens": 225359, + "estimated_eval_tokens": 450974 + }, + "budget": { + "node_count": 9, + "item_count": 9, + "soft_pair_count": 12, + "serialized_chars": 8613, + "estimated_input_tokens": 2154, + "estimated_eval_tokens": 4564 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "These issues are a loose grab bag of unrelated bug reports across different models, configs, tokenizers, and utilities. None of the soft-similarity pairs looks like the same underlying bug or change, so there is no real duplicate cluster here.", + "confidence": 0.94, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": "If a representative issue must be chosen, issue 44031 is the broadest user-facing regression report in the set; the others are narrower, unrelated correctness bugs.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43717", + "right": "issue:46144", + "accept": false, + "reason": "Different components and bug classes: Mamba init_weights behavior vs RoFormer attention interface usage." + }, + { + "left": "issue:42936", + "right": "issue:46018", + "accept": false, + "reason": "Mask2Former ignore_value handling is unrelated to DeepSeek-V4 shared expert gating." + }, + { + "left": "issue:43668", + "right": "issue:44031", + "accept": false, + "reason": "A config type-hint problem is not the same as a tokenizer regex warning regression." + }, + { + "left": "issue:42936", + "right": "issue:43240", + "accept": false, + "reason": "One is about an unused model field; the other is about kwargs not reaching a loss function." + }, + { + "left": "issue:43240", + "right": "issue:46144", + "accept": false, + "reason": "Loss-function kwargs propagation and RoFormer attention interface are distinct code paths." + }, + { + "left": "issue:42936", + "right": "issue:44031", + "accept": false, + "reason": "Mask2Former ignore_value misuse does not match a tokenizer warning regression." + }, + { + "left": "issue:42936", + "right": "issue:46144", + "accept": false, + "reason": "Different model families and different defects; no shared underlying change." + }, + { + "left": "issue:42936", + "right": "issue:43813", + "accept": false, + "reason": "Unused ignore_value field and a quantization-typo string are unrelated." + }, + { + "left": "issue:43240", + "right": "issue:43668", + "accept": false, + "reason": "Passing kwargs to loss calculation is unrelated to an incorrect type annotation." + }, + { + "left": "issue:43668", + "right": "issue:43813", + "accept": false, + "reason": "Type hint correction vs typo in quantization operation name are not the same bug." + }, + { + "left": "issue:37934", + "right": "issue:43668", + "accept": false, + "reason": "RMS-norm semantics question for Llama4TextL2Norm is unrelated to ModernBERTConfig norm_eps typing." + }, + { + "left": "issue:42936", + "right": "issue:43668", + "accept": false, + "reason": "Mask2Former ignore_value usage and ModernBERT norm_eps typing concern different subsystems and fixes." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 2, + "component_count": 21, + "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": "49f022f3accb9978b93e355050abc8a4c23288524c26015e82980cef2c718b1e", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 2876, + "estimated_input_tokens": 719, + "estimated_eval_tokens": 1694 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 2876, + "estimated_input_tokens": 719, + "estimated_eval_tokens": 1694 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These three issues are not duplicates: they cover unrelated topics (tokenizer truncation behavior, a ReduceLROnPlateau-like training feature, and Wav2Vec2 word time offsets).", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": "No issue is a good cluster representative because the issues are about different features/problems in different parts of the library.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:10519", + "right": "issue:11307", + "accept": false, + "reason": "Different domains: truncation strategy for tokenization vs. Wav2Vec2 word timestamp extraction." + }, + { + "left": "issue:10519", + "right": "issue:11005", + "accept": false, + "reason": "No overlap in underlying bug or feature request; one is about input truncation, the other about learning-rate scheduling behavior." + }, + { + "left": "issue:11005", + "right": "issue:11307", + "accept": false, + "reason": "Unrelated requests: ReduceLROnPlateau-like functionality vs. audio model time offsets." + } + ] + }, + "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": 21, + "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": "92ad73f507ae42e1ab92a3660aef5c824e898d022e8cfe8a0fb61c0b544eaf59", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 2478, + "estimated_input_tokens": 620, + "estimated_eval_tokens": 1496 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 2478, + "estimated_input_tokens": 620, + "estimated_eval_tokens": 1496 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "The cluster appears to contain three distinct issues: one about Tapas tokenization mismatch, and two about model-loading/download failures that are not clearly the same bug. The model-loading issues share a broad theme, but the available titles point to different scenarios (private-model auth vs. a public model load failure), so I would not merge them as duplicates.", + "confidence": 0.46, + "canonical_issue_reason": "Issue 16618 is the broadest user-facing model-loading report in the set, so it is the best representative if one issue must be chosen.", + "canonical_pr_reason": null, + "best_issue_reason": "16618 is the most general of the issues and the closest to a core runtime failure, while 13244 is clearly unrelated and 15980 is a narrower auth/error-message case.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:13244", + "right": "issue:16618", + "accept": false, + "reason": "Tapas tokenization mismatch is unrelated to failing to load bert-base-uncased; they describe different subsystems and different failure modes." + }, + { + "left": "issue:15980", + "right": "issue:16618", + "accept": false, + "reason": "15980 is specifically about the error message when accessing a private model without login, while 16618 is a public-model load failure; the underlying bug is not clearly the same." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 4, + "component_count": 21, + "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": "2fdf89b6782c445044a49c54c955a694160d42ae2a79d9c9ff9dd0f67312693b", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1590, + "estimated_input_tokens": 398, + "estimated_eval_tokens": 1052 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1590, + "estimated_input_tokens": 398, + "estimated_eval_tokens": 1052 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "The two issues request unrelated community contributions: Flash Attention 2 support for more architectures vs GGUF support for more architectures. They share a broad theme (architecture support) but not the same underlying feature or 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:26350", + "right": "issue:33260", + "accept": false, + "reason": "Different requested capabilities for different backends/code paths; this is not the same underlying issue and should not be merged as duplicates." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 5, + "component_count": 21, + "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": "8949938030c2c6a074259a1bcff9ec2da2cdb35774889e731348bd9451eccf41", + "component_budget": { + "node_count": 8, + "item_count": 8, + "soft_pair_count": 24, + "serialized_chars": 13526, + "estimated_input_tokens": 3382, + "estimated_eval_tokens": 7020 + }, + "budget": { + "node_count": 8, + "item_count": 8, + "soft_pair_count": 24, + "serialized_chars": 13526, + "estimated_input_tokens": 3382, + "estimated_eval_tokens": 7020 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Most similarities are boilerplate from the same translation-request template, not true duplicates. I\u2019d only treat the exact same-title twins as duplicate-worthy; the different language requests are separate issues.", + "confidence": 0.82, + "canonical_issue_reason": "issue:43834 is the best anchor for the repeated placeholder-titled translation requests: it is the earliest of the identical generic items and sits in the densest duplicate-like subgroup.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:43834 is the strongest overall issue representative for this cluster because it is central to the repeated template-style items and is the cleanest canonical anchor among the issue set.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:27088", + "right": "issue:44370", + "accept": false, + "reason": "Different language targets (Turkish vs a separate translation request); the overlap is just the shared translation template." + }, + { + "left": "issue:27088", + "right": "issue:44365", + "accept": false, + "reason": "Same boilerplate theme, but not the same concrete translation request." + }, + { + "left": "issue:27088", + "right": "issue:43834", + "accept": false, + "reason": "Turkish translation request is distinct from the generic placeholder-titled request." + }, + { + "left": "issue:27088", + "right": "issue:38435", + "accept": false, + "reason": "Turkish and Romanian are separate translation tasks, not one underlying issue." + }, + { + "left": "issue:27088", + "right": "issue:28936", + "accept": false, + "reason": "Different target languages, so these are not duplicates." + }, + { + "left": "issue:28936", + "right": "issue:43834", + "accept": false, + "reason": "Spanish request vs generic placeholder title; too broad to be the same issue." + }, + { + "left": "issue:28936", + "right": "issue:44365", + "accept": false, + "reason": "Shared template language only; the translation target is different or unspecified." + }, + { + "left": "issue:28936", + "right": "issue:38435", + "accept": false, + "reason": "Spanish and Romanian translation requests are distinct." + }, + { + "left": "issue:28936", + "right": "issue:44370", + "accept": false, + "reason": "Different concrete translation request, despite similar boilerplate." + }, + { + "left": "issue:38435", + "right": "issue:44370", + "accept": false, + "reason": "Romanian request is separate from the generic placeholder issue." + }, + { + "left": "issue:38435", + "right": "issue:44365", + "accept": false, + "reason": "They look templated, but not clearly the same underlying translation task." + }, + { + "left": "issue:38435", + "right": "issue:43834", + "accept": false, + "reason": "Different language target; similarity is only from shared issue template." + }, + { + "left": "issue:38435", + "right": "issue:46222", + "accept": false, + "reason": "Romanian translation request and the Thai-titled issue are not the same concrete issue." + }, + { + "left": "issue:38435", + "right": "issue:46226", + "accept": false, + "reason": "Same boilerplate family, but no evidence they are the same request." + }, + { + "left": "issue:43834", + "right": "issue:44370", + "accept": true, + "reason": "Same exact generic title and near-identical body shape; this looks like a duplicate translation-request issue." + }, + { + "left": "issue:43834", + "right": "issue:44365", + "accept": true, + "reason": "Same exact generic title and highly similar content; likely duplicate reports of the same issue." + }, + { + "left": "issue:44365", + "right": "issue:44370", + "accept": true, + "reason": "Same exact placeholder title and near-identical body; strong duplicate signal." + }, + { + "left": "issue:43834", + "right": "issue:46222", + "accept": false, + "reason": "The titles do not match and one is a generic placeholder while the other is a concrete Thai-titled request." + }, + { + "left": "issue:43834", + "right": "issue:46226", + "accept": false, + "reason": "Different titles and no clear evidence of the same underlying translation request." + }, + { + "left": "issue:44365", + "right": "issue:46222", + "accept": false, + "reason": "Not the same concrete issue; the match is only superficial template similarity." + }, + { + "left": "issue:44365", + "right": "issue:46226", + "accept": false, + "reason": "Different request titles and no explicit common target." + }, + { + "left": "issue:44370", + "right": "issue:46222", + "accept": false, + "reason": "Shared boilerplate, but not the same underlying issue." + }, + { + "left": "issue:44370", + "right": "issue:46226", + "accept": false, + "reason": "Different titles and no concrete duplicate evidence." + }, + { + "left": "issue:46222", + "right": "issue:46226", + "accept": true, + "reason": "Same exact Thai title and matching body shape; these look like duplicate copies of the same issue." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays conservative: it treats the identical-title pairs as the only duplicate-worthy matches, while rejecting cross-language/template similarities as mere boilerplate. The canonical anchor choice for issue:43834 is supported by the shared title pattern and creation order among the generic placeholder items." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 6, + "component_count": 21, + "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-36817-18", + "nodes": [ + "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: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:43671|issue:43909", + "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": "db4056249a32a337fbb3fae9a034c405a9cf3f05fceeec4d20ad204617e3eb16", + "component_budget": { + "node_count": 18, + "item_count": 18, + "soft_pair_count": 40, + "serialized_chars": 24144, + "estimated_input_tokens": 6036, + "estimated_eval_tokens": 12328 + }, + "budget": { + "node_count": 18, + "item_count": 18, + "soft_pair_count": 40, + "serialized_chars": 24144, + "estimated_input_tokens": 6036, + "estimated_eval_tokens": 12328 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "This cluster is dominated by unrelated model-add/support requests that only share generic wording. They span different model families and modalities (text, vision, audio, ASR/TTS, OCR, architecture tweaks), so none of the soft edges look like true duplicates.", + "confidence": 0.35, + "canonical_issue_reason": "Issue 42738 is the broadest and most umbrella-like request in the set ('BERT-like models with RoPE'), so it is the best single representative of this mostly model-request cluster, though it should not subsume the other specific requests as duplicates.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 42738 is the best representative of the cluster because it is the most general feature request among otherwise model-specific asks.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:36817", + "right": "issue:41211", + "accept": false, + "reason": "EuroBert (text encoder) vs DEIMv2 (vision detector); same request pattern only." + }, + { + "left": "issue:36817", + "right": "issue:43671", + "accept": false, + "reason": "EuroBert vs Qwen3-TTS; different model families and modality." + }, + { + "left": "issue:36817", + "right": "issue:36879", + "accept": false, + "reason": "EuroBert vs RF-DETR; unrelated model domains." + }, + { + "left": "issue:36879", + "right": "issue:41211", + "accept": false, + "reason": "RF-DETR vs DEIMv2; both are detectors, but they are different concrete models and changes." + }, + { + "left": "issue:36879", + "right": "issue:43909", + "accept": false, + "reason": "RF-DETR vs LFM2.5 Audio; vision detector vs audio model." + }, + { + "left": "issue:36879", + "right": "issue:44121", + "accept": false, + "reason": "RF-DETR vs sparse transformer LM; different code paths and tasks." + }, + { + "left": "issue:36879", + "right": "issue:43837", + "accept": false, + "reason": "RF-DETR vs ASR model; different task and implementation." + }, + { + "left": "issue:36879", + "right": "issue:43671", + "accept": false, + "reason": "RF-DETR vs TTS model; different task and implementation." + }, + { + "left": "issue:37015", + "right": "issue:43837", + "accept": false, + "reason": "NeoBERT vs Qwen3-ASR; text encoder vs speech model." + }, + { + "left": "issue:37015", + "right": "issue:44121", + "accept": false, + "reason": "NeoBERT vs weight-sparse transformer; distinct model families." + }, + { + "left": "issue:37015", + "right": "issue:43909", + "accept": false, + "reason": "NeoBERT vs audio model; different modality." + }, + { + "left": "issue:36879", + "right": "issue:37015", + "accept": false, + "reason": "RF-DETR vs NeoBERT; unrelated families despite similar 'add model' wording." + }, + { + "left": "issue:37015", + "right": "issue:42503", + "accept": false, + "reason": "NeoBERT vs ModernVBERT; both BERT-like names, but separate model additions." + }, + { + "left": "issue:41115", + "right": "issue:43837", + "accept": false, + "reason": "MiniCPM3 vs Qwen3-ASR; different model families and task." + }, + { + "left": "issue:41115", + "right": "issue:44121", + "accept": false, + "reason": "MiniCPM3 vs weight-sparse transformer; unrelated requests." + }, + { + "left": "issue:41115", + "right": "issue:42503", + "accept": false, + "reason": "MiniCPM3 vs ModernVBERT; separate model-specific requests." + }, + { + "left": "issue:41115", + "right": "issue:45027", + "accept": false, + "reason": "MiniCPM3 vs Voxtral TTS; multimodal LLM vs speech synthesis support." + }, + { + "left": "issue:41211", + "right": "issue:43671", + "accept": false, + "reason": "DEIMv2 vs Qwen3-TTS; detector vs TTS." + }, + { + "left": "issue:42503", + "right": "issue:43837", + "accept": false, + "reason": "ModernVBERT vs Qwen3-ASR; different modalities and code paths." + }, + { + "left": "issue:42503", + "right": "issue:43671", + "accept": false, + "reason": "ModernVBERT vs Qwen3-TTS; different tasks and model families." + }, + { + "left": "issue:42503", + "right": "issue:42738", + "accept": false, + "reason": "ModernVBERT vs generic BERT-with-RoPE request; related wording, but not the same concrete change." + }, + { + "left": "issue:42738", + "right": "issue:45027", + "accept": false, + "reason": "BERT/RoPE architectural request vs Voxtral TTS; unrelated." + }, + { + "left": "issue:42738", + "right": "issue:46653", + "accept": false, + "reason": "BERT/RoPE request vs unrelated person-name issue; not the same kind of artifact." + }, + { + "left": "issue:43671", + "right": "issue:43837", + "accept": false, + "reason": "Qwen3-TTS vs Qwen3-ASR; same family name, but different concrete support and code paths." + }, + { + "left": "issue:43671", + "right": "issue:45027", + "accept": false, + "reason": "Qwen3-TTS vs Voxtral TTS; both TTS, but different models and likely different integrations." + }, + { + "left": "issue:43671", + "right": "issue:45177", + "accept": false, + "reason": "TTS vs OCR; unrelated modalities." + }, + { + "left": "issue:43671", + "right": "issue:44121", + "accept": false, + "reason": "TTS model vs sparse transformer LM; unrelated." + }, + { + "left": "issue:43671", + "right": "issue:43909", + "accept": false, + "reason": "Qwen3-TTS vs LFM2.5 Audio; related audio domain, but distinct model additions." + }, + { + "left": "issue:43837", + "right": "issue:43909", + "accept": false, + "reason": "ASR vs audio model; related domain, but different concrete requests." + }, + { + "left": "issue:43837", + "right": "issue:45322", + "accept": false, + "reason": "ASR vs vision-language encoder; different task and architecture." + }, + { + "left": "issue:43837", + "right": "issue:44121", + "accept": false, + "reason": "ASR vs sparse transformer; unrelated." + }, + { + "left": "issue:43909", + "right": "issue:44205", + "accept": false, + "reason": "Audio model vs SAM3-LiteText; different model families and use cases." + }, + { + "left": "issue:43909", + "right": "issue:46645", + "accept": false, + "reason": "Audio model request vs Granite SWA+sink option; one adds a model, the other changes a model option." + }, + { + "left": "issue:44121", + "right": "issue:45322", + "accept": false, + "reason": "Sparse transformer vs EUPE encoder; distinct architectures." + }, + { + "left": "issue:43837", + "right": "issue:45027", + "accept": false, + "reason": "ASR vs TTS; same broad area, but not the same underlying change." + }, + { + "left": "issue:37015", + "right": "issue:45027", + "accept": false, + "reason": "NeoBERT vs Voxtral TTS; unrelated model families." + }, + { + "left": "issue:43837", + "right": "issue:45177", + "accept": false, + "reason": "ASR vs OCR; different modalities and implementations." + }, + { + "left": "issue:45322", + "right": "issue:46133", + "accept": false, + "reason": "EUPE vs TIPSv2; both vision-language related, but different model requests." + }, + { + "left": "issue:42503", + "right": "issue:46653", + "accept": false, + "reason": "ModernVBERT vs unrelated person-name issue; no shared underlying change." + }, + { + "left": "issue:45027", + "right": "issue:46653", + "accept": false, + "reason": "Voxtral TTS vs unrelated person-name issue; no shared underlying change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 7, + "component_count": 21, + "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": "b6e543f1c4fe0896dfc058515a8be783ae9a1e2445148ca2871ce58e44d6e3f5", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1505, + "estimated_input_tokens": 377, + "estimated_eval_tokens": 1010 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1505, + "estimated_input_tokens": 377, + "estimated_eval_tokens": 1010 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These issues are unrelated: one is a docs request to add Pruna, the other is a Vision RoPE refactor.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": "No single issue is a good canonical representative because the topics do not overlap.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:38740", + "right": "issue:46443", + "accept": false, + "reason": "Different work items and no shared code-path or explicit link; a docs/framework addition is not the same as a Vision RoPE refactor." + } + ] + }, + "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": 21, + "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": "e2d80f8083cbcfca53a6bdf34fb9f0f2935a62ce50af90e86411b81b9fe6cc40", + "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": "The two issues are not duplicates: one requests block decoding support in general, while the other asks for DeepSeek v3.2 support. They refer to different feature areas and no concrete overlap is shown.", + "confidence": 0.94, + "canonical_issue_reason": "Issue #43037 is the clearer representative because it names a specific model-support request, making it more actionable and easier to track than the broader \"Set Block Decoding\" title.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue #43037 is the best issue to keep as the cluster representative since it is specific, self-contained, and less ambiguous than #41084.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:41084", + "right": "issue:43037", + "accept": false, + "reason": "The titles indicate different underlying requests: block decoding functionality versus support for a specific DeepSeek model version. They do not appear to describe the same bug or change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 9, + "component_count": 21, + "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": "1ed683f09599ef818f29dbfdd63ae7ce4e6336f6f57a896b328f63394f87d05f", + "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 two separate enhancement issues in Trainer: one about making `remove_unused_column` more customizable, and another about handling custom trainers when estimating FLOPS. They share only a broad Trainer context, not the same underlying bug or change.", + "confidence": 0.96, + "canonical_issue_reason": "Issue #43598 is the more general Trainer customizability request and is slightly broader/more central than #43600, but the two are not duplicates.", + "canonical_pr_reason": null, + "best_issue_reason": "#43598 is the better representative issue because it frames a reusable Trainer behavior change, whereas #43600 is a distinct FLOPS-estimation concern.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43598", + "right": "issue:43600", + "accept": false, + "reason": "Different concrete problems: one targets `remove_unused_column` behavior, the other targets FLOPS estimation for custom trainers. Same subsystem, but not the same fix." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 10, + "component_count": 21, + "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": "536fa9775f8cdc376504d2ec45e511a8dd2b8f7c310e2d12f0f57530d8c46b54", + "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": "These two issues are both about Hugging Face model-class registration, but they target different model families and different auto-classes: Qwen3-Omni registration for AutoModel/AutoModelForConditionalGeneration vs. Qwen3.5 sequence-classification support for AutoModelForSequenceClassification. They are not the same underlying bug/request.", + "confidence": 0.95, + "canonical_issue_reason": "Issue #43684 is the broader registration request and is more representative of the cluster topic, while #44405 is a separate, narrower model-class support request.", + "canonical_pr_reason": null, + "best_issue_reason": "#43684 is the better representative issue because it is the more general model-registration request and not tied to a specific downstream task class.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43684", + "right": "issue:44405", + "accept": false, + "reason": "Reject: different model families (Qwen3-Omni vs Qwen3.5/Qwen3_5Config) and different target auto-classes, so they do not describe the same code-path change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 11, + "component_count": 21, + "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": "4ee97d3a0666028b1bd04b37cb4389a241103091f927e865f9cdd972d989ad54", + "component_budget": { + "node_count": 4, + "item_count": 4, + "soft_pair_count": 6, + "serialized_chars": 4260, + "estimated_input_tokens": 1065, + "estimated_eval_tokens": 2386 + }, + "budget": { + "node_count": 4, + "item_count": 4, + "soft_pair_count": 6, + "serialized_chars": 4260, + "estimated_input_tokens": 1065, + "estimated_eval_tokens": 2386 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "The cluster is too thinly evidenced to confidently call the three \u201chuggingface\u201d issues duplicates. They share a title, but the packet only shows soft similarity and timing/body-length metadata, so the safer call is to leave the pairwise links unresolved. The \u201cFett\u201d issue does not appear related.", + "confidence": 0.44, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": "43845 is the best provisional anchor only because it is one of the repeated \u201chuggingface\u201d reports and is earliest among them, but the evidence is too weak to treat it as a confirmed canonical duplicate.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43835", + "right": "issue:43845", + "accept": false, + "reason": "Different titles and no concrete shared issue content; soft similarity alone is not enough." + }, + { + "left": "issue:43835", + "right": "issue:43846", + "accept": false, + "reason": "\u201cFett\u201d does not have enough evidence to match the huggingface reports." + }, + { + "left": "issue:43835", + "right": "issue:43859", + "accept": false, + "reason": "No concrete indication that the short \u201cFett\u201d report is the same as the huggingface reports." + }, + { + "left": "issue:43845", + "right": "issue:43846", + "accept": false, + "reason": "Same title and close timestamps are not enough on their own to confirm a duplicate bug report." + }, + { + "left": "issue:43845", + "right": "issue:43859", + "accept": false, + "reason": "The evidence is limited to title similarity and metadata, which is too weak to accept as the same underlying issue." + }, + { + "left": "issue:43846", + "right": "issue:43859", + "accept": false, + "reason": "These look similar, but there is no concrete content-level evidence that they describe the same bug." + } + ] + }, + "evaluator_result": { + "accept": false, + "feedback": "The cluster summary is a bit too strong for the packet. The three \u201chuggingface\u201d issues do share a title, but the evidence shown is limited to soft similarity plus timestamps/body lengths; the packet does not ground a confident claim that they are duplicates. In particular, the accepted soft-edge verdicts overstate the evidence by saying the reports \u201cstrongly suggest\u201d the same underlying issue. I would keep those pairwise verdicts conservative or unresolved unless there is more concrete content-level evidence." + }, + "evaluator_used": true, + "retried": true, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 12, + "component_count": 21, + "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": "4e579894b7a209a4e6fe42e78ac5daac04f5c5a40c26b5f7847506c5f7c47862", + "component_budget": { + "node_count": 5, + "item_count": 5, + "soft_pair_count": 9, + "serialized_chars": 5791, + "estimated_input_tokens": 1448, + "estimated_eval_tokens": 3152 + }, + "budget": { + "node_count": 5, + "item_count": 5, + "soft_pair_count": 9, + "serialized_chars": 5791, + "estimated_input_tokens": 1448, + "estimated_eval_tokens": 3152 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are all low-signal issue reports with spammy or generic titles and no bodies, links, or activity to establish a shared underlying bug. None of the pairs are strong enough to merge as duplicates.", + "confidence": 0.18, + "canonical_issue_reason": "issue:44418 is the least spam-like and most plausibly a real report, but it is still very generic and lacks any concrete problem description.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:44418 is the best available issue candidate only because it has a non-spam title; there is still no substantive evidence that it represents the cluster\u2019s true topic.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44371", + "right": "issue:44572", + "accept": false, + "reason": "Both are spam-like with no body detail; identical tokenless titles do not demonstrate the same underlying issue." + }, + { + "left": "issue:44371", + "right": "issue:44573", + "accept": false, + "reason": "Both are spam-like with no body detail; no concrete bug or request is visible to justify duplicate status." + }, + { + "left": "issue:44371", + "right": "issue:44418", + "accept": false, + "reason": "A spam title and a generic documentation suggestion are not enough to show the same issue." + }, + { + "left": "issue:44371", + "right": "issue:44937", + "accept": false, + "reason": "The titles are unrelated on their face; no evidence they refer to the same report." + }, + { + "left": "issue:44418", + "right": "issue:44572", + "accept": false, + "reason": "Generic documentation wording and spam are not the same underlying bug/change." + }, + { + "left": "issue:44418", + "right": "issue:44573", + "accept": false, + "reason": "Generic documentation wording and spam are not the same underlying bug/change." + }, + { + "left": "issue:44572", + "right": "issue:44573", + "accept": false, + "reason": "Both are spam-like, but there is still no concrete description showing a shared problem." + }, + { + "left": "issue:44572", + "right": "issue:44937", + "accept": false, + "reason": "Spam-like title versus a specific quoted-name title; no shared bug or change is evident." + }, + { + "left": "issue:44573", + "right": "issue:44937", + "accept": false, + "reason": "Spam-like title versus a specific quoted-name title; no shared bug or change is evident." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 13, + "component_count": 21, + "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": "6d372d41016017783d2b8a9158187f5c3fdc1fe875533ab5178d236cf3352954", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1646, + "estimated_input_tokens": 412, + "estimated_eval_tokens": 1080 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1646, + "estimated_input_tokens": 412, + "estimated_eval_tokens": 1080 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "The two closed issues appear to be duplicates of the same proposal/report: identical titles, highly similar bodies, and no distinguishing links or files. Issue #45145 is the better canonical choice because it was created earlier.", + "confidence": 0.99, + "canonical_issue_reason": "Issue #45145 was created earlier and appears to be the same report as #45151, with an identical title and highly similar content.", + "canonical_pr_reason": null, + "best_issue_reason": "#45145 is the best issue representative because it is the earlier of the two near-identical issues and there is no evidence of a distinct scope between them.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:45145", + "right": "issue:45151", + "accept": true, + "reason": "Accept: the issues have the same title and substantially overlapping content, with no evidence of different underlying problems; they look like duplicates of the same issue." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The judgment is grounded in the packet: the two issues share an identical title, similar body sizes, and a soft-similarity signal, with no explicit links or filenames suggesting separate scopes. The canonical choice of the earlier issue is reasonable, and the soft-edge accept is conservative enough here." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 14, + "component_count": 21, + "review_unit_index": 1, + "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": 15, + "component_count": 21, + "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-45672-2", + "nodes": [ + "issue:45672", + "issue:45674" + ], + "soft_pairs": [ + "issue:45672|issue:45674" + ], + "prepared_review_unit_hash": "abddce1de611a6aa333499c6ca10379d70e5a77cc8a2de3f2d6a3539b0c325ad", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1672, + "estimated_input_tokens": 418, + "estimated_eval_tokens": 1092 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1672, + "estimated_input_tokens": 418, + "estimated_eval_tokens": 1092 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are two separate 4-bit inference issues: one is Gemma4/audio-module specific and triggered by `torch.finfo()` on `uint8` weights, while the other is a `BitsAndBytesConfig` regression where `llm_int8_skip_modules` drops the default `lm_head` exclusion and causes an assertion failure. They share the broad 4-bit inference context but not the same bug or code path.", + "confidence": 0.93, + "canonical_issue_reason": "Issue 45674 is the better canonical issue because it describes a clearer, more generally applicable 4-bit inference regression in `BitsAndBytesConfig`, whereas 45672 is a narrower Gemma4/audio-module-specific failure.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 45674 is the strongest representative for the cluster since it is broader, more actionable, and tied to a concrete config behavior change rather than a model-specific symptom.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:45672", + "right": "issue:45674", + "accept": false, + "reason": "Both concern 4-bit inference, but the failure modes differ: `torch.finfo()` on `uint8` weights in Gemma4 audio modules versus `llm_int8_skip_modules` removing the default `lm_head` exclusion. Different triggering code paths and different underlying bugs." + } + ] + }, + "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": 21, + "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": "b03dab2f223092cd93760f27ff442cf0bb03b76aa7cd607cd1da611d4f15c15c", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1477, + "estimated_input_tokens": 370, + "estimated_eval_tokens": 996 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1477, + "estimated_input_tokens": 370, + "estimated_eval_tokens": 996 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These two closed issues do not look like duplicates of the same underlying problem; they have different names and dates and no body/links to connect them. The similarity is only superficial.", + "confidence": 0.98, + "canonical_issue_reason": "issue:45762 is the better placeholder canonical issue only because it is the earlier report; there is no evidence these two issues describe the same thing.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:45762 is marginally the best representative for triage since it is older, but neither issue appears to be a true duplicate candidate.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:45762", + "right": "issue:45763", + "accept": false, + "reason": "Different people and dates in the titles, with no shared body text, links, or supporting evidence of the same bug/request." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 17, + "component_count": 21, + "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": "806e20052bcb6ccfdbf43539fee2eec4b972598836afaf3712a2816bd54eac0c", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1481, + "estimated_input_tokens": 371, + "estimated_eval_tokens": 998 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1481, + "estimated_input_tokens": 371, + "estimated_eval_tokens": 998 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are two closed, content-less issue stubs with unrelated personal-name/date titles. They do not describe the same bug or request, so they should not be clustered as duplicates beyond weak textual similarity.", + "confidence": 0.98, + "canonical_issue_reason": "Issue 45764 is marginally earlier, but neither issue contains substantive problem details; there is no meaningful canonical issue here.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 45764 is the best placeholder only because it was created first; both issues are effectively empty and not good representative reports.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:45764", + "right": "issue:45765", + "accept": false, + "reason": "The titles are different personal names with different dates and no body content or bug description; this is 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": 18, + "component_count": 21, + "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": "18927839abe21d5d41bdb41331aa755587c55eaac150995cd97d173b0006fc93", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1542, + "estimated_input_tokens": 386, + "estimated_eval_tokens": 1028 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1542, + "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: one requests adding a Qwen3_5 token-classification model, while the other asks for FP8 training support. They should not be clustered as duplicates.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:45810", + "right": "issue:46295", + "accept": false, + "reason": "Different feature requests for different code paths: model-class support for token classification vs training precision support (FP8). 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": 21, + "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": "c7f33cf8dd2037965622ec8c310b8a495799561512db95fb0941eefc3d5a86a9", + "component_budget": { + "node_count": 5, + "item_count": 5, + "soft_pair_count": 8, + "serialized_chars": 5728, + "estimated_input_tokens": 1432, + "estimated_eval_tokens": 3120 + }, + "budget": { + "node_count": 5, + "item_count": 5, + "soft_pair_count": 8, + "serialized_chars": 5728, + "estimated_input_tokens": 1432, + "estimated_eval_tokens": 3120 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "There are two strong near-duplicate issue pairs: 45870/45871 and 45872/45873. Issue 45869 is related only by topic and should remain separate.", + "confidence": 0.89, + "canonical_issue_reason": "45870 is a clean representative for the first near-duplicate pair: it has the same title and body length as 45871, and it is the earlier of the two. With only metadata available, treat this as a likely duplicate rather than a proven exact match.", + "canonical_pr_reason": null, + "best_issue_reason": "45870 is the best global issue representative because it anchors one of the clearest near-duplicate pairs and is the earliest instance. 45869 is a separate dataset request, while 45872/45873 appear to be a different, separate pair.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:45869", + "right": "issue:45870", + "accept": false, + "reason": "Related financial domain, but this is a dataset bundle request versus a prediction example; not the same underlying issue." + }, + { + "left": "issue:45869", + "right": "issue:45871", + "accept": false, + "reason": "Related financial domain, but this is a dataset bundle request versus a prediction example; not the same underlying issue." + }, + { + "left": "issue:45870", + "right": "issue:45871", + "accept": true, + "reason": "Same title, same body length, and very close timestamps; this is a strong near-duplicate pair, though the packet only supports a likely-duplicate judgment." + }, + { + "left": "issue:45870", + "right": "issue:45872", + "accept": false, + "reason": "Both mention financial time series, but the titles indicate different requests: an example prediction issue versus a scraping-service request." + }, + { + "left": "issue:45870", + "right": "issue:45873", + "accept": false, + "reason": "Both mention financial time series, but the titles indicate different requests: an example prediction issue versus a scraping-service request." + }, + { + "left": "issue:45871", + "right": "issue:45872", + "accept": false, + "reason": "Both mention financial time series, but the titles indicate different requests: an example prediction issue versus a scraping-service request." + }, + { + "left": "issue:45871", + "right": "issue:45873", + "accept": false, + "reason": "Both mention financial time series, but the titles indicate different requests: an example prediction issue versus a scraping-service request." + }, + { + "left": "issue:45872", + "right": "issue:45873", + "accept": true, + "reason": "Same title, same body length, and very close timestamps; this is a strong near-duplicate pair, though the packet only supports a likely-duplicate judgment." + } + ] + }, + "evaluator_result": { + "accept": false, + "feedback": "The clustering is a bit too strong for the packet. The items only provide titles, lengths, timestamps, and soft-similarity signals; there is no body text to justify claims like \u201cfully duplicated,\u201d \u201ceffectively identical content,\u201d or \u201cdirect duplicates.\u201d The 45870/45871 and 45872/45873 pairs may be strong matches, but the verdicts should stay conservative (e.g. likely duplicates / near-identical) unless the packet shows exact content equivalence. The 45869 separation is fine." + }, + "evaluator_used": true, + "retried": true, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 20, + "component_count": 21, + "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": "3d6842affd6c26258b98074fb04761a37ee3430bb06b6c88cc06e4a211f17538", + "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": "The two issues are in the same static-cache / linear-attention area, but the packet only shows title-level similarity and one soft similarity edge. 46439 is the broader root-cause report, while 46441 is a model-specific crash that could be related but is not proven to be the same bug.", + "confidence": 0.58, + "canonical_issue_reason": "Issue 46439 is the best canonical issue because it describes the underlying mechanism more directly: static-cache `early_initialization` corrupting linear-attention shapes and causing runtime failure.", + "canonical_pr_reason": null, + "best_issue_reason": "46439 is the strongest anchor for the cluster since it captures the broader bug mechanism rather than one specific symptom.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:46439", + "right": "issue:46441", + "accept": false, + "reason": "Do not merge as duplicates on the current evidence: there are no shared targets/files or body-level links, and the packet only supports a possible relation in the same subsystem, not the same underlying defect." + } + ] + }, + "evaluator_result": { + "accept": false, + "feedback": "The cluster summary is mostly grounded, but the soft-edge verdict is too strong for the evidence available. The packet only gives title-level similarity and one `soft_similarity` edge with no shared targets/files; that supports a possible relation, not a confident accept. Issue 46441 could be related to the same static-cache/linear-attention area, but the packet does not establish that its `KeyError` is the same underlying bug as 46439\u2019s shape corruption. Keep the relationship tentative unless there is stronger issue-body evidence." + }, + "evaluator_used": true, + "retried": true, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 21, + "component_count": 21, + "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": "7e8761665fef41f111d8f662998438e7ba60455dcd7170e1b2056c5b40050a62", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1706, + "estimated_input_tokens": 427, + "estimated_eval_tokens": 1110 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1706, + "estimated_input_tokens": 427, + "estimated_eval_tokens": 1110 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These two issues are effectively exact duplicates: same title, same body length, and identical bug description about Wav2Vec2CTCTokenizer v5 leaking the word delimiter as a special token. The open issue is the better representative for ongoing tracking.", + "confidence": 0.98, + "canonical_issue_reason": "Issue #46552 is the better canonical issue because it is open and already has inbound references, so it is the more active tracker for the same bug.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue #46552 is the strongest issue candidate overall: same problem statement, open state, and existing references make it the most suitable single issue to keep.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:46551", + "right": "issue:46552", + "accept": true, + "reason": "Exact same title and identical described bug; both refer to the same tokenizer special-token leak, so this is a true duplicate pair." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "Grounded in the packet: the two issues have identical titles, identical body lengths, and the same soft-similarity evidence, so treating them as duplicates is conservative. The canonical choice of the open issue with inbound references is also supported." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 36, + "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-39895-14", + "nodes": [ + "pull_request:39895", + "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" + ], + "soft_pairs": [ + "pull_request:39895|pull_request:43424", + "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: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:45170", + "pull_request:44827|pull_request:46267", + "pull_request:44827|pull_request:46048", + "pull_request:44827|pull_request:45655", + "pull_request:44827|pull_request:45213", + "pull_request:43823|pull_request:45213", + "pull_request:43251|pull_request:43254" + ], + "prepared_review_unit_hash": "861040f44251a5a02dc79c15e9ec16927d6d5035fe2654cc74d6d8a1b11e9e03", + "component_budget": { + "node_count": 14, + "item_count": 14, + "soft_pair_count": 19, + "serialized_chars": 38084, + "estimated_input_tokens": 9521, + "estimated_eval_tokens": 19298 + }, + "budget": { + "node_count": 14, + "item_count": 14, + "soft_pair_count": 19, + "serialized_chars": 38084, + "estimated_input_tokens": 9521, + "estimated_eval_tokens": 19298 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Only one pair looks like a true duplicate: the two `fixed_cross_entropy` kwargs PRs (#43251 and #43254). The rest are unrelated model additions, refactors, or test/documentation changes that do not fix the same concrete bug/change.", + "confidence": 0.95, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #43251 is the better canonical choice: it explicitly targets issue #43240 and implements the broader `cross_entropy` kwargs passthrough in `fixed_cross_entropy`, while #43254 is just an alternate phrasing of the same fix.", + "best_issue_reason": null, + "best_pr_reason": "PR #43251 is the strongest representative for the cluster because it has the explicit issue link, a clear bug title, and the same underlying code-path change as #43254.", + "soft_edge_verdicts": [ + { + "left": "pull_request:39895", + "right": "pull_request:43424", + "accept": false, + "reason": "Both are PRs, but they are unrelated feature/test additions in different areas (VideoPrism model support vs ExecuTorch dynamic-shape export tests)." + }, + { + "left": "pull_request:41855", + "right": "pull_request:43823", + "accept": false, + "reason": "Mistral tokenizer helper methods and adding MobileLLM are unrelated changes; they do not fix the same code path." + }, + { + "left": "pull_request:41855", + "right": "pull_request:43424", + "accept": false, + "reason": "Tokenizer method additions and ExecuTorch export test coverage are different changes with no shared underlying bug." + }, + { + "left": "pull_request:42668", + "right": "pull_request:43823", + "accept": false, + "reason": "These are different processor/model robustness additions for different models; they are not the same bug or change." + }, + { + "left": "pull_request:43251", + "right": "pull_request:43823", + "accept": false, + "reason": "A cross-entropy kwargs fix and adding MobileLLM are unrelated." + }, + { + "left": "pull_request:43251", + "right": "pull_request:44827", + "accept": false, + "reason": "The kwargs passthrough fix in loss utils is unrelated to Mistral4 test/model updates." + }, + { + "left": "pull_request:43424", + "right": "pull_request:44827", + "accept": false, + "reason": "Dynamic-shape export tests and Mistral4 test/model changes are not the same concrete issue." + }, + { + "left": "pull_request:43424", + "right": "pull_request:43823", + "accept": false, + "reason": "Different model features/tests; no evidence they address the same underlying problem." + }, + { + "left": "pull_request:43823", + "right": "pull_request:45655", + "accept": false, + "reason": "Adding MobileLLM and fixing `cls.config` resolution are unrelated code changes." + }, + { + "left": "pull_request:43823", + "right": "pull_request:43995", + "accept": false, + "reason": "MobileLLM model addition and Falcon output-collection refactor are different work items." + }, + { + "left": "pull_request:43995", + "right": "pull_request:45213", + "accept": false, + "reason": "Falcon refactoring and the model-creation skill PR are unrelated." + }, + { + "left": "pull_request:43995", + "right": "pull_request:44827", + "accept": false, + "reason": "Falcon interface refactoring and Mistral4 test fixes are not the same underlying bug/change." + }, + { + "left": "pull_request:44827", + "right": "pull_request:45170", + "accept": false, + "reason": "Mistral4 test/model changes and `layrnorm -> layernorm` renaming across CLIP-like models are unrelated." + }, + { + "left": "pull_request:44827", + "right": "pull_request:46267", + "accept": false, + "reason": "Mistral4 fixes and chat response parsing updates target different subsystems and problems." + }, + { + "left": "pull_request:44827", + "right": "pull_request:46048", + "accept": false, + "reason": "Mistral4 test/model updates and adding the TDT loss kernel are unrelated." + }, + { + "left": "pull_request:44827", + "right": "pull_request:45655", + "accept": false, + "reason": "These PRs address different issues: Mistral4 test/model changes vs config resolution order in `modeling_utils`." + }, + { + "left": "pull_request:44827", + "right": "pull_request:45213", + "accept": false, + "reason": "A model/test fix and a model-creation skill/documentation PR are not the same change." + }, + { + "left": "pull_request:43823", + "right": "pull_request:45213", + "accept": false, + "reason": "Adding MobileLLM and the model-creation skill are not duplicate fixes; one adds a model, the other adds workflow tooling." + }, + { + "left": "pull_request:43251", + "right": "pull_request:43254", + "accept": true, + "reason": "Both PRs implement the same underlying fix in `fixed_cross_entropy`: forwarding supported kwargs such as `weight` and `label_smoothing` to `nn.functional.cross_entropy`. They are plausible mergeable duplicates." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "Grounded overall. The only supported duplicate/mergeable pair in the packet is #43251/#43254: both touch `src/transformers/loss/loss_utils.py`, share issue target #43240, and implement the same `fixed_cross_entropy` kwargs passthrough fix. The other soft-edge pairs are conservatively rejected as unrelated model additions, refactors, or test/docs changes." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 2, + "component_count": 36, + "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": "e655b8ef7072f36c856a1f76bbb1afd37d88060fc8a99af76d047e897bbb6fda", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 7226, + "estimated_input_tokens": 1807, + "estimated_eval_tokens": 3870 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 7226, + "estimated_input_tokens": 1807, + "estimated_eval_tokens": 3870 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both PRs are about the VibeVoice family, but they are not the same change: #40546 is the core VibeVoice implementation, while #42781 adds VibeVoice Realtime plus the acoustic tokenizer and generation-related follow-ups. They overlap by model family and shared auto-registration/docs, not by the same concrete fix.", + "confidence": 0.89, + "canonical_issue_reason": null, + "canonical_pr_reason": "#40546 is the better canonical PR for the cluster because it is the primary, more focused VibeVoice implementation and appears more mature/reviewed; #42781 is a later, broader realtime follow-up with draft status.", + "best_issue_reason": null, + "best_pr_reason": "#40546 is the best single representative of the cluster topic: it directly matches the base VibeVoice implementation and is less ambiguous than the broader realtime extension in #42781.", + "soft_edge_verdicts": [ + { + "left": "pull_request:40546", + "right": "pull_request:42781", + "accept": false, + "reason": "Reject: these are related VibeVoice workstreams, but not the same underlying change. #40546 implements core VibeVoice, while #42781 extends the family with realtime support and the acoustic tokenizer; they are not plausibly one merged PR for the same code-path bug/change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 3, + "component_count": 36, + "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": "2b88ca252ede4e1b2f42554c4aa0195d81a99a30fd65e01a1b60e973600fec09", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4675, + "estimated_input_tokens": 1169, + "estimated_eval_tokens": 2594 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4675, + "estimated_input_tokens": 1169, + "estimated_eval_tokens": 2594 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These PRs are unrelated: one adds TimesFM covariate forecasting, the other adds DINOv3 image classification support. They do not fix the same bug or implement the same change.", + "confidence": 0.99, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": "pull_request:41224 is the more self-contained and specific change set, with a clear model-classification feature addition and supporting code/tests/docs. It is not a duplicate of the TimesFM PR, just the stronger standalone PR in this cluster.", + "soft_edge_verdicts": [ + { + "left": "pull_request:40755", + "right": "pull_request:41224", + "accept": false, + "reason": "Different models, different tasks, and different code paths: TimesFM forecasting with covariates versus DINOv3 image classification. They are not the same underlying change and should not be merged as duplicates." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 4, + "component_count": 36, + "review_unit_index": 1, + "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": 5, + "component_count": 36, + "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": "e0358ba1d19df304f352eb5529a44a95de7ae716cdcf1b6c565e9fec499a5c43", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5934, + "estimated_input_tokens": 1484, + "estimated_eval_tokens": 3224 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5934, + "estimated_input_tokens": 1484, + "estimated_eval_tokens": 3224 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both PRs are broad GitHub Actions dependency bumps, but they target different action families and different workflow subsets. They are related maintenance, not the same underlying fix/change.", + "confidence": 0.94, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 42975 is the cleaner canonical representative: it has a clearer, single purpose\u2014updating workflow actions for Node 24 compatibility across many files.", + "best_issue_reason": null, + "best_pr_reason": "PR 42975 is the best standalone PR in the cluster because its intent is more explicit and it covers the broader workflow-action compatibility update.", + "soft_edge_verdicts": [ + { + "left": "pull_request:42975", + "right": "pull_request:42976", + "accept": false, + "reason": "Do not merge: 42975 upgrades core workflow actions like checkout/cache/upload-artifact for Node 24 compatibility, while 42976 bumps docker/build-push-action and a few other workflow actions. They touch overlapping workflow files but fix different dependency upgrades and are not the same concrete change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 6, + "component_count": 36, + "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": "b51af974b91ce1dbdfa2a6b0329c972665671169b8c14a9ac3541cbe5378a61e", + "component_budget": { + "node_count": 6, + "item_count": 6, + "soft_pair_count": 8, + "serialized_chars": 20308, + "estimated_input_tokens": 5077, + "estimated_eval_tokens": 10410 + }, + "budget": { + "node_count": 6, + "item_count": 6, + "soft_pair_count": 8, + "serialized_chars": 20308, + "estimated_input_tokens": 5077, + "estimated_eval_tokens": 10410 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "This cluster is not a duplicate set: it contains several unrelated model-addition PRs that only share generic boilerplate (docs/auto mappings). The only near-match is the Step3p7 Flash pair, but the WIP mini-model/script PR is support material, not the same change as the model integration PR.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": "pull_request:46562 is the most central substantive change in the only near-overlap: it is the actual Step3p7 Flash model implementation with docs/tests, whereas the other PRs are separate model additions and pull_request:46658 is just a mini-model/debug/script bundle.", + "best_issue_reason": null, + "best_pr_reason": "pull_request:46562 is the best representative PR for this cluster because it contains a complete concrete code-path change; the rest are different models or ancillary assets and should not be merged together as duplicates.", + "soft_edge_verdicts": [ + { + "left": "pull_request:42978", + "right": "pull_request:43448", + "accept": false, + "reason": "Different models (ViT NEPA vs Molmo). Shared docs and auto-registration files are generic model-onboarding boilerplate, not the same underlying change." + }, + { + "left": "pull_request:42978", + "right": "pull_request:43451", + "accept": false, + "reason": "Different models (ViT NEPA vs Molmo2). Overlap is only in shared framework registration/docs, which is too broad to treat as a duplicate PR." + }, + { + "left": "pull_request:42978", + "right": "pull_request:45490", + "accept": false, + "reason": "Different models (ViT NEPA vs CTSM). They touch common onboarding files, but they implement unrelated model additions." + }, + { + "left": "pull_request:43448", + "right": "pull_request:43451", + "accept": false, + "reason": "Molmo and Molmo2 are distinct model integrations with different files, configs, and tests; same naming family is not enough to merge them." + }, + { + "left": "pull_request:43448", + "right": "pull_request:45490", + "accept": false, + "reason": "Molmo vs CTSM are unrelated model additions; shared docs/auto files are standard repo plumbing, not duplicate evidence." + }, + { + "left": "pull_request:43451", + "right": "pull_request:45490", + "accept": false, + "reason": "Molmo2 and CTSM are different model families and code paths; the overlap is only in generic registration/configuration files." + }, + { + "left": "pull_request:45490", + "right": "pull_request:46562", + "accept": false, + "reason": "CTSM and Step3p7 Flash are different models; similar broad model-addition structure does not indicate the same bug or change." + }, + { + "left": "pull_request:46562", + "right": "pull_request:46658", + "accept": false, + "reason": "Both mention Step 3.7 Flash, but pull_request:46562 is the actual model integration while pull_request:46658 adds a mini model plus scripts/debug assets; they are not the same concrete change and would not plausibly be one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 7, + "component_count": 36, + "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:43498", + "pull_request:43665", + "pull_request:44815", + "pull_request:45114", + "pull_request:46380", + "pull_request:46539", + "pull_request:46579" + ], + "soft_pairs": [ + "pull_request:43291|pull_request:43665", + "pull_request:43291|pull_request:44815", + "pull_request:43291|pull_request:46380", + "pull_request:43498|pull_request:46579", + "pull_request:43665|pull_request:46539", + "pull_request:43665|pull_request:44815", + "pull_request:43665|pull_request:46380", + "pull_request:43665|pull_request:46579", + "pull_request:44815|pull_request:46539", + "pull_request:44815|pull_request:46380", + "pull_request:44815|pull_request:46579", + "pull_request:45114|pull_request:46579", + "pull_request:46380|pull_request:46539", + "pull_request:46380|pull_request:46579" + ], + "prepared_review_unit_hash": "76836826012d67fa074b38d3e13e125dd54c618b34be40d861123f894139890a", + "component_budget": { + "node_count": 8, + "item_count": 8, + "soft_pair_count": 14, + "serialized_chars": 23159, + "estimated_input_tokens": 5790, + "estimated_eval_tokens": 11836 + }, + "budget": { + "node_count": 8, + "item_count": 8, + "soft_pair_count": 14, + "serialized_chars": 23159, + "estimated_input_tokens": 5790, + "estimated_eval_tokens": 11836 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "The cluster is heterogeneous: it mixes unrelated PRs for Whisper tokenization/tests, tie_weights backward-compatibility, FP8 dequantization, doctest cleanup, generation watermarking validation, CI/runtime setup, and typo fixes. None of the soft edges look like the same underlying bug or change, so this is not a duplicate cluster.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": "No canonical PR: the items target different subsystems and code paths, with only superficial \"fix\"/cleanup similarity.", + "best_issue_reason": null, + "best_pr_reason": "No best PR for deduplication: none of these PRs is a plausible merged-into-one representative for the others.", + "soft_edge_verdicts": [ + { + "left": "pull_request:43291", + "right": "pull_request:43665", + "accept": false, + "reason": "Both touch tests, but they are for different models/code paths (Whisper tokenization vs CLIP/ViT test edits)." + }, + { + "left": "pull_request:43291", + "right": "pull_request:44815", + "accept": false, + "reason": "Whisper decoding/test changes are unrelated to the FP8 dequant/loading fix." + }, + { + "left": "pull_request:43291", + "right": "pull_request:46380", + "accept": false, + "reason": "Whisper tokenizer logic and generation watermarking validation are different bugs in different subsystems." + }, + { + "left": "pull_request:43498", + "right": "pull_request:46579", + "accept": false, + "reason": "Tie-weights backward compatibility is unrelated to generic typo cleanup across many files." + }, + { + "left": "pull_request:43665", + "right": "pull_request:46539", + "accept": false, + "reason": "Synthetic test assertion edits in model tests are unrelated to Phimoe CI memory-limit/setup changes." + }, + { + "left": "pull_request:43665", + "right": "pull_request:44815", + "accept": false, + "reason": "These are different fixes: test-file edits versus FP8/model-loading behavior." + }, + { + "left": "pull_request:43665", + "right": "pull_request:46380", + "accept": false, + "reason": "Only broad test-related similarity; one is test noise, the other is a generation config bug fix." + }, + { + "left": "pull_request:43665", + "right": "pull_request:46579", + "accept": false, + "reason": "Model test edits do not match a codebase-wide typo cleanup." + }, + { + "left": "pull_request:44815", + "right": "pull_request:46539", + "accept": false, + "reason": "FP8 dequantization logic and Phimoe integration CI changes do not share the same underlying bug." + }, + { + "left": "pull_request:44815", + "right": "pull_request:46380", + "accept": false, + "reason": "Different concrete problems: FP8 padding/dequant versus watermarking config validation." + }, + { + "left": "pull_request:44815", + "right": "pull_request:46579", + "accept": false, + "reason": "Shared filenames are incidental; the actual changes address unrelated code cleanup vs FP8/model-loading behavior." + }, + { + "left": "pull_request:45114", + "right": "pull_request:46579", + "accept": false, + "reason": "Doctest formatting fixes are unrelated to generic typo cleanup in code files." + }, + { + "left": "pull_request:46380", + "right": "pull_request:46539", + "accept": false, + "reason": "Generation watermarking validation is unrelated to Phimoe integration test CI/environment fixes." + }, + { + "left": "pull_request:46380", + "right": "pull_request:46579", + "accept": false, + "reason": "A validation fix in generation config is not the same change as broad typo corrections." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 8, + "component_count": 36, + "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-43488-2", + "nodes": [ + "pull_request:43488", + "pull_request:43532" + ], + "soft_pairs": [ + "pull_request:43488|pull_request:43532" + ], + "prepared_review_unit_hash": "f9c7c721b4328687472aacf50de261a9f55741a1aaa9c935edfbf2308bc7f62b", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4563, + "estimated_input_tokens": 1141, + "estimated_eval_tokens": 2538 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4563, + "estimated_input_tokens": 1141, + "estimated_eval_tokens": 2538 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "The two pull requests are not duplicates: one is a setup/version/dependency bot-check change, while the other is a CI/workflow and testing utility change. They only share a test file and broadly related repo context, not the same underlying fix.", + "confidence": 0.95, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #43532 is the better representative of the cluster because it is the broader, more substantive repository-change PR; #43488 is a narrow bot-format check focused on setup/version metadata.", + "best_issue_reason": null, + "best_pr_reason": "PR #43532 is the most suitable representative among the PRs, but only as a loose cluster anchor, not as a duplicate target; it changes CI/workflow behavior and related testing utilities rather than the same concrete code-path as #43488.", + "soft_edge_verdicts": [ + { + "left": "pull_request:43488", + "right": "pull_request:43532", + "accept": false, + "reason": "Different changes: #43488 alters setup/version/dependency metadata for a bot-format check, while #43532 rewires CI/workflow jobs and testing utilities. Shared test-file touchpoint is incidental, not the same underlying bug or 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": 9, + "component_count": 36, + "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": "c12590a45f886b39a2937557609572271bf846c9325b9520f5e09e7b7b239d8d", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 6082, + "estimated_input_tokens": 1521, + "estimated_eval_tokens": 3298 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 6082, + "estimated_input_tokens": 1521, + "estimated_eval_tokens": 3298 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Two open PRs add different segmentation pipelines: one for Promptable Concept Segmentation and one for Promptable Visual Segmentation. They share repository scaffolding and documentation patterns, but they target different features and model integrations, so they are not duplicates.", + "confidence": 0.95, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #43613 is the stronger standalone candidate: it adds the visual segmentation pipeline with broader model/test coverage and a more complete implementation footprint. PR #43612 is a separate concept-segmentation feature, not the same change.", + "best_issue_reason": null, + "best_pr_reason": "PR #43613 is the best representative of the cluster because it is the more fully scoped pipeline addition and appears more production-ready on its own.", + "soft_edge_verdicts": [ + { + "left": "pull_request:43612", + "right": "pull_request:43613", + "accept": false, + "reason": "Different underlying features. One adds Promptable Concept Segmentation, the other Promptable Visual Segmentation; they touch shared docs and pipeline registration, but the implementation targets and model/test files differ, so they should not be merged as duplicates." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 10, + "component_count": 36, + "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": "14d039ea58d99dae80e8813aac965ddebad09b562a7d35f8121de09338f0aa57", + "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": "Two open Trainer PRs touch the same file but fix different Trainer behaviors: gradient-accumulation loss scaling vs. input signature/column filtering. They are not duplicates.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": "No issue is present; between the two PRs, #43651 is the more self-contained behavior change and a slightly better representative of the cluster.", + "best_issue_reason": null, + "best_pr_reason": "#43651 is the stronger representative because it introduces a focused trainer behavior change in the training step, whereas #43663 is a separate refactor/override point for signature-column filtering.", + "soft_edge_verdicts": [ + { + "left": "pull_request:43651", + "right": "pull_request:43663", + "accept": false, + "reason": "Different underlying code paths and goals: #43651 changes loss normalization during training, while #43663 adds a helper for signature-based column filtering. They do not appear to fix the same bug or be 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": 11, + "component_count": 36, + "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": "8779d23c85ff12a4aa208a41b5d1ab751369c4cc89be9bfa956a5e2192be2878", + "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": "Both PRs touch the same CLI module and both deal with annotation-related NameErrors, but they fix different failure modes: one is about forward-reference typing in `serve.py`, the other is about `TypeAdapter` being unavailable when `pydantic` is missing. They are related, but not the same underlying bug.", + "confidence": 0.92, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 43842 is the more specific and concrete fix: it explicitly targets the `TypeAdapter` NameError when `pydantic` is not installed and makes the annotation safe at runtime. It has a clear issue link and a narrowly scoped change.", + "best_issue_reason": null, + "best_pr_reason": "PR 43842 is the best representative of the cluster because it addresses a clearly stated runtime bug with an explicit target issue and a focused patch in the affected code path.", + "soft_edge_verdicts": [ + { + "left": "pull_request:43656", + "right": "pull_request:43842", + "accept": false, + "reason": "Related but not duplicate: 43656 changes future-annotation handling and local model/processor type hints, while 43842 fixes `TypeAdapter` access when `pydantic` is absent. They address different NameError sources and would not naturally be merged as one bug fix." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 12, + "component_count": 36, + "review_unit_index": 1, + "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": 36, + "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: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:44066|pull_request:44071", + "pull_request:44018|pull_request:44068", + "pull_request:44066|pull_request:44068", + "pull_request:44018|pull_request:44066", + "pull_request:44019|pull_request:44722", + "pull_request:44019|pull_request:44085", + "pull_request:44086|pull_request:44722", + "pull_request:44018|pull_request:44019", + "pull_request:44019|pull_request:44071", + "pull_request:44068|pull_request:44116", + "pull_request:44068|pull_request:44722", + "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:44059|pull_request:44086", + "pull_request:44085|pull_request:44086", + "pull_request:44098|pull_request:44101", + "pull_request:44018|pull_request:44071", + "pull_request:44059|pull_request:44722", + "pull_request:44017|pull_request:44161", + "pull_request:44017|pull_request:44073", + "pull_request:44017|pull_request:44722", + "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:44056|pull_request:44161", + "pull_request:43997|pull_request:44018", + "pull_request:43997|pull_request:44161", + "pull_request:44074|pull_request:44161", + "pull_request:44018|pull_request:44161", + "pull_request:43997|pull_request:44071", + "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:43997|pull_request:44129", + "pull_request:44010|pull_request:44018", + "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" + ], + "prepared_review_unit_hash": "4356ea22d971a2dafecb4a0de8628b1638539cf48dda1db9fa8df759cf17a76d", + "component_budget": { + "node_count": 28, + "item_count": 28, + "soft_pair_count": 78, + "serialized_chars": 84893, + "estimated_input_tokens": 21224, + "estimated_eval_tokens": 42704 + }, + "budget": { + "node_count": 27, + "item_count": 27, + "soft_pair_count": 72, + "serialized_chars": 75090, + "estimated_input_tokens": 18773, + "estimated_eval_tokens": 37802 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a set of independent model-specific output-tracing refactors under the same umbrella issue (#43979), not one duplicate bug/change. The only strong duplicate-looking edges are within the GPT-Neo pair and the GPT-J/CodeGen pair(s).", + "confidence": 0.91, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 44066 is the clearest representative of the shared output-tracing refactor theme: it has an explicit issue link, touches the core GPT-J/CodeGen path, and shows the standardized decorator-based migration pattern. The other PRs are mostly separate model ports rather than duplicates of one another.", + "best_issue_reason": null, + "best_pr_reason": "PR 44066 is the best overall cluster representative because it most clearly captures the umbrella refactor pattern while still being concrete and issue-linked; most other PRs are narrower, model-specific variants.", + "soft_edge_verdicts": [ + { + "left": "pull_request:43996", + "right": "pull_request:44085", + "accept": false, + "reason": "Different model families and code paths (FNet/CVT vs GPT-J); same umbrella theme is not enough for duplicate PR triage." + }, + { + "left": "pull_request:43996", + "right": "pull_request:44044", + "accept": false, + "reason": "Different models (FNet/CVT vs DeBERTa-v2); both are output-tracing refactors, but not the same underlying change." + }, + { + "left": "pull_request:44066", + "right": "pull_request:44085", + "accept": true, + "reason": "Same GPT-J output-tracing refactor family on the same code path, with overlapping decorator migration and shared file-level changes." + }, + { + "left": "pull_request:44007", + "right": "pull_request:44072", + "accept": false, + "reason": "Different architectures and refactors (ResNet/RT-DETR/RegNet vs EfficientNet); broad standardized-output work is too generic to merge as duplicates." + }, + { + "left": "pull_request:44072", + "right": "pull_request:44722", + "accept": false, + "reason": "Different models (EfficientNet vs GPT-J/CodeGen); same issue tag does not make these the same concrete change." + }, + { + "left": "pull_request:44066", + "right": "pull_request:44072", + "accept": false, + "reason": "GPT-J/CodeGen output tracing vs EfficientNet output tracing are separate model-specific refactors." + }, + { + "left": "pull_request:44013", + "right": "pull_request:44044", + "accept": false, + "reason": "MobileNetV2 and DeBERTa-v2 are unrelated code paths; only the refactor pattern is shared." + }, + { + "left": "pull_request:44066", + "right": "pull_request:44086", + "accept": false, + "reason": "GPT-J/CodeGen refactor and MGP-STR refactor are different model implementations and not the same bug/change." + }, + { + "left": "pull_request:44066", + "right": "pull_request:44071", + "accept": false, + "reason": "GPT-J/CodeGen and MPT are distinct model refactors; same decorator transition is too broad to duplicate." + }, + { + "left": "pull_request:44018", + "right": "pull_request:44068", + "accept": true, + "reason": "Both target GPT-Neo output tracing on the same model/file and appear to be the same standardized-refactor change." + }, + { + "left": "pull_request:44066", + "right": "pull_request:44068", + "accept": false, + "reason": "GPT-J/CodeGen vs GPT-Neo are different model code paths; no duplicate concrete change." + }, + { + "left": "pull_request:44018", + "right": "pull_request:44066", + "accept": false, + "reason": "Different models (GPT-Neo vs GPT-J/CodeGen); shared issue target is not enough to merge as duplicates." + }, + { + "left": "pull_request:44019", + "right": "pull_request:44722", + "accept": false, + "reason": "ResNet output-tracing refactor is unrelated to GPT-J/CodeGen." + }, + { + "left": "pull_request:44019", + "right": "pull_request:44085", + "accept": false, + "reason": "ResNet and GPT-J are unrelated model implementations, so these are not the same underlying change." + }, + { + "left": "pull_request:44086", + "right": "pull_request:44722", + "accept": false, + "reason": "MGP-STR and GPT-J/CodeGen are different code paths and model families." + }, + { + "left": "pull_request:44018", + "right": "pull_request:44019", + "accept": false, + "reason": "Both are output-tracing refactors, but GPT-Neo and ResNet are different concrete changes." + }, + { + "left": "pull_request:44019", + "right": "pull_request:44071", + "accept": false, + "reason": "ResNet vs MPT; same general migration style, but not the same bug or patch." + }, + { + "left": "pull_request:44068", + "right": "pull_request:44116", + "accept": false, + "reason": "GPT-Neo is unrelated to Flaubert; shared refactor theme does not make them duplicates." + }, + { + "left": "pull_request:44068", + "right": "pull_request:44722", + "accept": false, + "reason": "GPT-Neo and GPT-J/CodeGen are separate model-specific refactors." + }, + { + "left": "pull_request:44018", + "right": "pull_request:44116", + "accept": false, + "reason": "GPT-Neo vs Flaubert; different concrete code paths." + }, + { + "left": "pull_request:44071", + "right": "pull_request:44722", + "accept": false, + "reason": "MPT and GPT-J/CodeGen are unrelated model implementations." + }, + { + "left": "pull_request:44018", + "right": "pull_request:44085", + "accept": false, + "reason": "GPT-Neo and GPT-J are different model refactors, despite both using the same output-tracing pattern." + }, + { + "left": "pull_request:44018", + "right": "pull_request:44086", + "accept": false, + "reason": "GPT-Neo vs MGP-STR; not the same underlying change." + }, + { + "left": "pull_request:44068", + "right": "pull_request:44086", + "accept": false, + "reason": "GPT-Neo and MGP-STR are distinct code paths." + }, + { + "left": "pull_request:44071", + "right": "pull_request:44086", + "accept": false, + "reason": "MPT and MGP-STR are unrelated models; no shared concrete fix." + }, + { + "left": "pull_request:44059", + "right": "pull_request:44086", + "accept": false, + "reason": "GPT-2 and MGP-STR are different models, so these are not duplicates." + }, + { + "left": "pull_request:44085", + "right": "pull_request:44086", + "accept": false, + "reason": "GPT-J and MGP-STR are unrelated code paths." + }, + { + "left": "pull_request:44098", + "right": "pull_request:44101", + "accept": false, + "reason": "ViLT vs Flaubert/XLM are different model families and changes." + }, + { + "left": "pull_request:44018", + "right": "pull_request:44071", + "accept": false, + "reason": "GPT-Neo and MPT are different model-specific output-tracing refactors." + }, + { + "left": "pull_request:44059", + "right": "pull_request:44071", + "accept": false, + "reason": "GPT-2 and MPT are separate model implementations." + }, + { + "left": "pull_request:44056", + "right": "pull_request:44101", + "accept": false, + "reason": "MPNet vs Flaubert/XLM are unrelated model refactors." + }, + { + "left": "pull_request:44018", + "right": "pull_request:44059", + "accept": false, + "reason": "GPT-Neo and GPT-2 are different code paths, so not the same change." + }, + { + "left": "pull_request:44059", + "right": "pull_request:44068", + "accept": false, + "reason": "GPT-2 and GPT-Neo are distinct model refactors." + }, + { + "left": "pull_request:44101", + "right": "pull_request:44161", + "accept": false, + "reason": "Flaubert/XLM vs LongT5 are unrelated model-specific refactors." + }, + { + "left": "pull_request:43997", + "right": "pull_request:44017", + "accept": false, + "reason": "RegNet and Segformer are different architectures; same standardized-output theme is too broad." + }, + { + "left": "pull_request:43997", + "right": "pull_request:44056", + "accept": false, + "reason": "RegNet and MPNet are unrelated model refactors." + }, + { + "left": "pull_request:44071", + "right": "pull_request:44076", + "accept": false, + "reason": "MPT vs ImageGPT are different model families and code paths." + }, + { + "left": "pull_request:44074", + "right": "pull_request:44101", + "accept": false, + "reason": "TextNet and Flaubert/XLM are different concrete changes." + }, + { + "left": "pull_request:44056", + "right": "pull_request:44161", + "accept": false, + "reason": "MPNet and LongT5 are unrelated model-specific patches." + }, + { + "left": "pull_request:43997", + "right": "pull_request:44018", + "accept": false, + "reason": "RegNet and GPT-Neo are different model implementations." + }, + { + "left": "pull_request:44010", + "right": "pull_request:44071", + "accept": false, + "reason": "SqueezeBert and MPT are unrelated code paths." + }, + { + "left": "pull_request:44010", + "right": "pull_request:44073", + "accept": false, + "reason": "SqueezeBert and VisualBert are different models; not the same concrete bug." + }, + { + "left": "pull_request:44018", + "right": "pull_request:44129", + "accept": false, + "reason": "GPT-Neo and SpeechT5 are unrelated refactors." + }, + { + "left": "pull_request:43997", + "right": "pull_request:44074", + "accept": false, + "reason": "RegNet and TextNet are different model families." + }, + { + "left": "pull_request:43997", + "right": "pull_request:44129", + "accept": false, + "reason": "RegNet and SpeechT5 are unrelated code paths." + }, + { + "left": "pull_request:44071", + "right": "pull_request:44129", + "accept": false, + "reason": "MPT and SpeechT5 are separate model-specific changes." + }, + { + "left": "pull_request:43996", + "right": "pull_request:44161", + "accept": false, + "reason": "FNet/CVT and LongT5 are unrelated models." + }, + { + "left": "pull_request:43996", + "right": "pull_request:44017", + "accept": false, + "reason": "FNet/CVT and Segformer are different architectures." + }, + { + "left": "pull_request:44010", + "right": "pull_request:44044", + "accept": false, + "reason": "SqueezeBert and DeBERTa-v2 are unrelated code paths." + }, + { + "left": "pull_request:44044", + "right": "pull_request:44074", + "accept": false, + "reason": "DeBERTa-v2 and TextNet are different model-specific refactors." + }, + { + "left": "pull_request:44066", + "right": "pull_request:44722", + "accept": true, + "reason": "Both are GPT-J/CodeGen output-tracing refactors on the same files and same underlying migration." + }, + { + "left": "pull_request:44085", + "right": "pull_request:44722", + "accept": true, + "reason": "Both are GPT-J output-tracing refactors on the same model/file, with the same standardized decorator migration." + }, + { + "left": "pull_request:44085", + "right": "pull_request:44129", + "accept": false, + "reason": "GPT-J and SpeechT5 are unrelated model implementations." + }, + { + "left": "pull_request:43996", + "right": "pull_request:44007", + "accept": false, + "reason": "FNet/CVT and ResNet/RT-DETR/RegNet are different model families." + }, + { + "left": "pull_request:43996", + "right": "pull_request:44129", + "accept": false, + "reason": "FNet/CVT and SpeechT5 are unrelated concrete changes." + }, + { + "left": "pull_request:44044", + "right": "pull_request:44129", + "accept": false, + "reason": "DeBERTa-v2 and SpeechT5 are different model-specific refactors." + }, + { + "left": "pull_request:44044", + "right": "pull_request:44154", + "accept": false, + "reason": "DeBERTa-v2 and ViTS are unrelated model code paths." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet: it treats the cluster as mostly model-specific output-tracing refactors under the shared umbrella issue, and it only calls out duplicate-like edges where the same model/file is involved (GPT-Neo, GPT-J/CodeGen). The false-vs-true soft-edge verdicts are generally conservative and do not overstate cross-model similarity." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 13, + "component_count": 36, + "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:44024", + "pull_request:44056", + "pull_request:44073", + "pull_request:44074", + "pull_request:44076", + "pull_request:44129" + ], + "soft_pairs": [ + "pull_request:44024|pull_request:44076", + "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": "83677dbdca63af4822d45b4020420db3ac0f94cf94d7ddc0bef5bbeaf7f3c53d", + "component_budget": { + "node_count": 28, + "item_count": 28, + "soft_pair_count": 78, + "serialized_chars": 84893, + "estimated_input_tokens": 21224, + "estimated_eval_tokens": 42704 + }, + "budget": { + "node_count": 8, + "item_count": 8, + "soft_pair_count": 6, + "serialized_chars": 17124, + "estimated_input_tokens": 4281, + "estimated_eval_tokens": 8818 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "All items are model-specific PRs in the same broad refactor wave toward standardized output tracing/capture_outputs, but they do not fix the same concrete bug or change the same code path. The cluster is best treated as related-by-theme, not duplicates.", + "confidence": 0.91, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 44129 is the most representative anchor: it has an explicit issue target and a concrete output-tracing refactor in SpeechT5, making it the clearest single PR to stand for this cluster theme.", + "best_issue_reason": null, + "best_pr_reason": "PR 44129 is the strongest standalone representative because it is tied to issue 43979 and has the most explicit problem framing; the others are similar refactors but on different models.", + "soft_edge_verdicts": [ + { + "left": "pull_request:44024", + "right": "pull_request:44076", + "accept": false, + "reason": "Both are standardized-output refactors, but they touch different models (FocalNet vs ImageGPT) and distinct forward paths, so they are not the same fix." + }, + { + "left": "pull_request:44073", + "right": "pull_request:44074", + "accept": false, + "reason": "VisualBert and TextNet are separate model implementations; this is the same refactor pattern, not one duplicate code-path change." + }, + { + "left": "pull_request:44056", + "right": "pull_request:44073", + "accept": false, + "reason": "MPNet and VisualBert share output-tracing cleanup theme, but they are unrelated model-specific changes and would not be merged as one PR." + }, + { + "left": "pull_request:44056", + "right": "pull_request:44074", + "accept": false, + "reason": "Same broad decorator/capture_outputs migration, but different models and code paths; not a single underlying bug or change." + }, + { + "left": "pull_request:43997", + "right": "pull_request:44010", + "accept": false, + "reason": "RegNet and SqueezeBert are separate model stacks; the PRs are similar refactors, not duplicates of the same fix." + }, + { + "left": "pull_request:44010", + "right": "pull_request:44129", + "accept": false, + "reason": "SqueezeBert and SpeechT5 both adopt standardized output capture, but they address different model code paths and cannot 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": 14, + "component_count": 36, + "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": "abcb4b2e2798401144f28fe2555c361dbe9291b4293758ad0b7031e813414cfb", + "component_budget": { + "node_count": 13, + "item_count": 13, + "soft_pair_count": 40, + "serialized_chars": 40168, + "estimated_input_tokens": 10042, + "estimated_eval_tokens": 20340 + }, + "budget": { + "node_count": 13, + "item_count": 13, + "soft_pair_count": 40, + "serialized_chars": 40168, + "estimated_input_tokens": 10042, + "estimated_eval_tokens": 20340 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are not duplicates; they are separate model-specific PRs all doing the same broad output-tracing refactor pattern in different code paths. I would not merge any pair as the same change.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 44025 is the best representative among the set because it is a broader, clearly scoped output-tracing refactor across two related depth-anything files; there is no single true canonical PR for the whole cluster.", + "best_issue_reason": null, + "best_pr_reason": "PR 44025 is the strongest representative candidate: it is self-contained, touches a cohesive model family, and follows the cluster\u2019s main decorator-based output-tracing pattern.", + "soft_edge_verdicts": [ + { + "left": "pull_request:43998", + "right": "pull_request:44002", + "accept": false, + "reason": "Same refactor theme, but different model code paths (`timm_backbone` vs `upernet`); not the same underlying change and not mergeable into one PR." + }, + { + "left": "pull_request:43998", + "right": "pull_request:44000", + "accept": false, + "reason": "Both are output-tracing refactors, but they target unrelated models (`timm_backbone` vs `vision_text_dual_encoder`) with different forward signatures." + }, + { + "left": "pull_request:43998", + "right": "pull_request:44001", + "accept": false, + "reason": "Different models and different implementation details; shared refactor style is not enough to treat them as duplicates." + }, + { + "left": "pull_request:43998", + "right": "pull_request:44003", + "accept": false, + "reason": "These touch separate model families (`timm_backbone` vs `mamba/falcon_mamba`) and are distinct code-path changes." + }, + { + "left": "pull_request:43998", + "right": "pull_request:44029", + "accept": false, + "reason": "Both are output-tracing cleanups, but they are in different architectures (`timm_backbone` vs `rwkv`) and not the same fix." + }, + { + "left": "pull_request:43999", + "right": "pull_request:44001", + "accept": false, + "reason": "`mobilenet_v1` and `univnet` are unrelated model implementations; same refactor pattern, but not the same underlying bug/change." + }, + { + "left": "pull_request:43999", + "right": "pull_request:44002", + "accept": false, + "reason": "Different model families and distinct forward/output handling changes; not plausibly one merged PR." + }, + { + "left": "pull_request:43999", + "right": "pull_request:44025", + "accept": false, + "reason": "These are separate model-specific output-tracing refactors in different files and subsystems." + }, + { + "left": "pull_request:43999", + "right": "pull_request:44026", + "accept": false, + "reason": "`mobilenet_v1` and `vision_encoder_decoder` are unrelated; no shared concrete code-path problem." + }, + { + "left": "pull_request:43999", + "right": "pull_request:44027", + "accept": false, + "reason": "Same general refactor idea, but different encoder-decoder implementation and different change scope." + }, + { + "left": "pull_request:44000", + "right": "pull_request:44001", + "accept": false, + "reason": "`vision_text_dual_encoder` and `univnet` are unrelated models; these are not duplicates." + }, + { + "left": "pull_request:44000", + "right": "pull_request:44002", + "accept": false, + "reason": "Different models and different forward signatures; too broad a similarity to merge as one PR." + }, + { + "left": "pull_request:44000", + "right": "pull_request:44003", + "accept": false, + "reason": "Separate model families (`vision_text_dual_encoder` vs `mamba/falcon_mamba`) with different output-capturing behavior." + }, + { + "left": "pull_request:44000", + "right": "pull_request:44025", + "accept": false, + "reason": "Both are output-tracing refactors, but they modify different model stacks and different concrete code paths." + }, + { + "left": "pull_request:44001", + "right": "pull_request:44002", + "accept": false, + "reason": "`univnet` and `upernet` are unrelated implementations; not the same bug or change." + }, + { + "left": "pull_request:44001", + "right": "pull_request:44003", + "accept": false, + "reason": "Different architectures and different output-capture integrations; not mergeable into a single PR." + }, + { + "left": "pull_request:44001", + "right": "pull_request:44025", + "accept": false, + "reason": "The refactor style matches, but the target models and call paths are different." + }, + { + "left": "pull_request:44002", + "right": "pull_request:44029", + "accept": false, + "reason": "`upernet` and `rwkv` are unrelated code paths; the shared output-tracing theme is too broad to be a duplicate." + }, + { + "left": "pull_request:44002", + "right": "pull_request:44003", + "accept": false, + "reason": "Different model families and different refactor mechanics; not the same concrete fix." + }, + { + "left": "pull_request:44003", + "right": "pull_request:44028", + "accept": false, + "reason": "`mamba/falcon_mamba` and `superpoint` are unrelated; these are separate model-specific changes." + }, + { + "left": "pull_request:44003", + "right": "pull_request:44004", + "accept": false, + "reason": "`mamba` and `codegen` are different code paths with different API changes, so they are not duplicates." + }, + { + "left": "pull_request:43998", + "right": "pull_request:44004", + "accept": false, + "reason": "`timm_backbone` and `codegen` change unrelated forward/output contracts; same refactor language, different underlying work." + }, + { + "left": "pull_request:44001", + "right": "pull_request:44004", + "accept": false, + "reason": "Separate models and separate output behavior; not the same underlying code-path problem." + }, + { + "left": "pull_request:44002", + "right": "pull_request:44004", + "accept": false, + "reason": "`upernet` and `codegen` are unrelated implementations, so they should not be merged as duplicates." + }, + { + "left": "pull_request:44000", + "right": "pull_request:44004", + "accept": false, + "reason": "Different model families and different signatures; this is only thematic similarity." + }, + { + "left": "pull_request:44025", + "right": "pull_request:44026", + "accept": false, + "reason": "`depth_anything` and `vision_encoder_decoder` are different subsystems; not the same concrete change." + }, + { + "left": "pull_request:44025", + "right": "pull_request:44027", + "accept": false, + "reason": "These are separate encoder-decoder/depth model refactors, not one shared code-path fix." + }, + { + "left": "pull_request:44025", + "right": "pull_request:44028", + "accept": false, + "reason": "Different model families and different output-capture implementations; not duplicates." + }, + { + "left": "pull_request:44025", + "right": "pull_request:44029", + "accept": false, + "reason": "`depth_anything` and `rwkv` have distinct internals and cannot be treated as one merged PR." + }, + { + "left": "pull_request:44025", + "right": "pull_request:44030", + "accept": false, + "reason": "Different models (`depth_anything` vs `dpr`) and different output-return behavior; not the same change." + }, + { + "left": "pull_request:44026", + "right": "pull_request:44029", + "accept": false, + "reason": "`vision_encoder_decoder` and `rwkv` are unrelated model implementations with no shared concrete bug." + }, + { + "left": "pull_request:44026", + "right": "pull_request:44027", + "accept": false, + "reason": "Different wrapper models and different API refactors; not a duplicate pair." + }, + { + "left": "pull_request:44026", + "right": "pull_request:44028", + "accept": false, + "reason": "These touch unrelated models and should not be collapsed together." + }, + { + "left": "pull_request:44026", + "right": "pull_request:44030", + "accept": false, + "reason": "`vision_encoder_decoder` and `dpr` are separate code paths; only the broad refactor theme is shared." + }, + { + "left": "pull_request:44027", + "right": "pull_request:44028", + "accept": false, + "reason": "`speech_encoder_decoder` and `superpoint` are unrelated and not the same underlying issue." + }, + { + "left": "pull_request:44027", + "right": "pull_request:44029", + "accept": false, + "reason": "Different model families and different output handling changes; not mergeable as one PR." + }, + { + "left": "pull_request:44027", + "right": "pull_request:44030", + "accept": false, + "reason": "`speech_encoder_decoder` and `dpr` are distinct code paths, so this is not a duplicate." + }, + { + "left": "pull_request:44028", + "right": "pull_request:44029", + "accept": false, + "reason": "`superpoint` and `rwkv` are unrelated models; same refactor style is not enough." + }, + { + "left": "pull_request:44028", + "right": "pull_request:44030", + "accept": false, + "reason": "Different implementations and different forward/output semantics; not the same change." + }, + { + "left": "pull_request:44029", + "right": "pull_request:44030", + "accept": false, + "reason": "`rwkv` and `dpr` are separate model-specific changes with no shared concrete code path." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 15, + "component_count": 36, + "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": "cf2c20a68a5b137faad39befa23698943f80a84fcf037d6da05c529902dfa11c", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 6551, + "estimated_input_tokens": 1638, + "estimated_eval_tokens": 3532 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 6551, + "estimated_input_tokens": 1638, + "estimated_eval_tokens": 3532 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both items are separate model-addition PRs for different architectures: PenguinVL and Voxtral TTS. They share Transformers boilerplate files, but the code paths, filenames, and targets are distinct, so they should not be clustered as duplicates.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #44662 is the more canonical representative because it is a mature, heavily reviewed, non-draft implementation PR for a complete new model integration.", + "best_issue_reason": null, + "best_pr_reason": "PR #44662 is the strongest standalone PR in the set: it has substantially more review activity and a fuller implementation footprint, so it is the better representative even though it is not a duplicate of #45401.", + "soft_edge_verdicts": [ + { + "left": "pull_request:44662", + "right": "pull_request:45401", + "accept": false, + "reason": "These are different model integrations (PenguinVL vs Voxtral TTS) with different model directories, configs, processors, and tests. Shared auto-registration and docs boilerplate is not enough to treat them as 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": 36, + "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": "aaa7499296e442d078007fdc909efa5e857a0c46bb8e6df8d7bea4c3a034f6c8", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4407, + "estimated_input_tokens": 1102, + "estimated_eval_tokens": 2460 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4407, + "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 pull requests are related to fusion, but they implement different layers of the feature: one introduces a generic Module Fusion API, while the other adds KernelConfig-based n-to-1 kernel fusion wiring and usage. They are conceptually adjacent, not duplicates.", + "confidence": 0.92, + "canonical_issue_reason": null, + "canonical_pr_reason": "No clear canonical PR: PR #44979 adds the underlying Module Fusion API, while PR #45363 adds a separate KernelConfig-driven fusion integration path. They overlap in theme but are different deliverables.", + "best_issue_reason": null, + "best_pr_reason": "PR #45363 is the better representative of the cluster because it is the more end-to-end, user-facing feature implementation, touching loading/integration code and including an example. PR #44979 is more of a lower-level API foundation.", + "soft_edge_verdicts": [ + { + "left": "pull_request:44979", + "right": "pull_request:45363", + "accept": false, + "reason": "Not the same concrete change: #44979 introduces a new generic module-fusion API and tests, while #45363 implements KernelConfig-based n-to-1 kernel fusion integration across model loading and utilities. Related, but not mergeable as one duplicate PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 17, + "component_count": 36, + "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": "eb3ba161859ec2dd84e2525c04c840573238e671bcaac2436e4ec1dac6826a37", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4414, + "estimated_input_tokens": 1104, + "estimated_eval_tokens": 2464 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4414, + "estimated_input_tokens": 1104, + "estimated_eval_tokens": 2464 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both pull requests address the same AttributeError in `_patch_mistral_regex` by switching from `backend_tokenizer.pre_tokenizer` to `tokenizer.pre_tokenizer` in the same file and targeting the same issue. They look like duplicate fixes for the same underlying bug; `45317` is the stronger PR because it includes tests.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": "`pull_request:45317` is the better canonical PR: it fixes the same code path as `45086` but also adds test coverage and appears to be the more complete treatment of the bug.", + "best_issue_reason": null, + "best_pr_reason": "`pull_request:45317` is the best overall PR choice because it addresses the same concrete AttributeError and includes validation via tests, making it the most suitable representative fix.", + "soft_edge_verdicts": [ + { + "left": "pull_request:45086", + "right": "pull_request:45317", + "accept": true, + "reason": "Same underlying bug and same fix in the same tokenizer code path; the second PR is essentially a more complete version of the first." + } + ] + }, + "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": 36, + "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-45115-4", + "nodes": [ + "pull_request:45115", + "pull_request:45569", + "pull_request:46332", + "pull_request:46565" + ], + "soft_pairs": [ + "pull_request:45115|pull_request:45569", + "pull_request:45569|pull_request:46565", + "pull_request:46332|pull_request:46565" + ], + "prepared_review_unit_hash": "1a42cdc71c27141ca0eed215963803f524aff935582de3c63d02dd6e74ebfed2", + "component_budget": { + "node_count": 4, + "item_count": 4, + "soft_pair_count": 3, + "serialized_chars": 13471, + "estimated_input_tokens": 3368, + "estimated_eval_tokens": 6992 + }, + "budget": { + "node_count": 4, + "item_count": 4, + "soft_pair_count": 3, + "serialized_chars": 13471, + "estimated_input_tokens": 3368, + "estimated_eval_tokens": 6992 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "The cluster is heterogeneous: one NemotronH refactor PR, one NemotronH dense/sparse architecture/docs PR, and two separate ASR streaming feature PRs for different Nemotron variants. None of the soft pairs look like the same concrete change, so all duplicate links should be rejected.", + "confidence": 0.94, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single canonical PR: the items cover distinct changes (NemotronH refactor, NemotronH dense/sparse split, Nemotron ASR streaming, and Nemotron 3.5 ASR streaming) rather than one duplicate fix or feature.", + "best_issue_reason": null, + "best_pr_reason": "PR 46332 is the most self-contained, broad feature PR in the cluster, but it is still distinct from the others; it is the closest thing to a representative standalone change.", + "soft_edge_verdicts": [ + { + "left": "pull_request:45115", + "right": "pull_request:45569", + "accept": false, + "reason": "Related NemotronH work, but not the same change: 45115 is a refactor/inheritance and conversion-mapping cleanup, while 45569 adds the dense/sparse split, docs, and broader model plumbing." + }, + { + "left": "pull_request:45569", + "right": "pull_request:46565", + "accept": false, + "reason": "Different model families and code paths: 45569 is NemotronH architecture/refactor work, while 46565 adds Nemotron 3.5 ASR streaming." + }, + { + "left": "pull_request:46332", + "right": "pull_request:46565", + "accept": false, + "reason": "Both are ASR streaming features, but for different concrete models (Nemotron ASR vs Nemotron 3.5 ASR) with overlapping infrastructure rather than the same underlying bug/change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 19, + "component_count": 36, + "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": "d9781ea5705a6e32105acf04c26c2c6f18292e9162c2081cda3a9b3cc295e5f3", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5466, + "estimated_input_tokens": 1367, + "estimated_eval_tokens": 2990 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5466, + "estimated_input_tokens": 1367, + "estimated_eval_tokens": 2990 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both PRs are part of the same heterogeneous-modeling effort, but they are not the same change: #45333 is config-only foundation work, while #45332 is a much broader modeling/cache/masking implementation. Shared files reflect common infrastructure, not a single duplicate fix.", + "confidence": 0.9, + "canonical_issue_reason": null, + "canonical_pr_reason": "#45333 is the more focused PR and best represents the shared per-layer configuration work; #45332 is a larger umbrella change that builds on that foundation with additional modeling behavior.", + "best_issue_reason": null, + "best_pr_reason": "#45333 is the best representative for the cluster because it is the narrower, cleaner config-support change, whereas #45332 spans extra code paths beyond the shared configuration layer.", + "soft_edge_verdicts": [ + { + "left": "pull_request:45332", + "right": "pull_request:45333", + "accept": false, + "reason": "Reject: they overlap on heterogeneous-config infrastructure, but #45332 adds broader model/cache/masking changes and is not the same concrete code-path fix as the config-only #45333. They would not be merged as one duplicate PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 20, + "component_count": 36, + "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": "02b0e53f0813d936c310d1e527f7d511f333f361248c9d6d115b99cd5b185ab4", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5514, + "estimated_input_tokens": 1379, + "estimated_eval_tokens": 3014 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5514, + "estimated_input_tokens": 1379, + "estimated_eval_tokens": 3014 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both pull requests are typing-related and overlap in a few files, but they address different changes: PR 45415 is a broad repository-wide type-checking sweep with config/ignore adjustments, while PR 45425 adds specific typing protocols and compatibility shims for model utilities. They do not look like the same underlying bug or change.", + "confidence": 0.91, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 45425 is the better representative artifact because it is a concrete, narrowly scoped typing fix with specific code-level compatibility changes, rather than a broad type-checking cleanup pass.", + "best_issue_reason": null, + "best_pr_reason": "PR 45425 is the most useful standalone PR in the cluster: it has a clear technical goal, focused implementation, and is easier to understand as a single change set.", + "soft_edge_verdicts": [ + { + "left": "pull_request:45415", + "right": "pull_request:45425", + "accept": false, + "reason": "Shared typing-related files are incidental; 45415 is a broad type-checking enablement/ignore update, while 45425 adds specific typing protocols and compatibility helpers. They are not the same concrete fix and would not naturally merge into one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 21, + "component_count": 36, + "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-45513-2", + "nodes": [ + "pull_request:45513", + "pull_request:46084" + ], + "soft_pairs": [ + "pull_request:45513|pull_request:46084" + ], + "prepared_review_unit_hash": "f77adfa46a4f3839b175af507a62f86460ade023c5bed60aa03ce5b9099271f4", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4949, + "estimated_input_tokens": 1238, + "estimated_eval_tokens": 2732 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4949, + "estimated_input_tokens": 1238, + "estimated_eval_tokens": 2732 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These PRs are only superficially similar: both mention cached multi-token forward paths, but they target different model families and different concrete bugs. #45513 fixes Qwen3.5/Olmo hybrid linear-attention caching behavior, while #46084 fixes Mamba2 use_cache behavior when seq_len > 1.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single canonical PR: the two PRs address different code paths in different model implementations, so they are not duplicates of the same fix.", + "best_issue_reason": null, + "best_pr_reason": "#45513 is the broader representative of the cached multi-token forward bug class because it spans several related model variants and includes direct test coverage, but it still should not be merged with #46084 as a duplicate.", + "soft_edge_verdicts": [ + { + "left": "pull_request:45513", + "right": "pull_request:46084", + "accept": false, + "reason": "Reject: they touch different architectures and fix different failure modes; the overlap is only at the high level of cached seq_len>1 inference." + } + ] + }, + "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": 36, + "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": "90dfe60dd2ec0085a9e7898353351a567d23e348be05be7db201a305b4b2b6a5", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4508, + "estimated_input_tokens": 1127, + "estimated_eval_tokens": 2510 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4508, + "estimated_input_tokens": 1127, + "estimated_eval_tokens": 2510 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are related to expert-parallel loading, but they fix different bugs in different code paths: 45548 addresses serialization/loading of `distributed_config` and the `has_ep` check for DeepSpeed ZeRO-3/accelerate, while 45662 fixes FSDP2/rank-0 broadcast overwriting expert parameters after sharding. They are not duplicates.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": "No canonical PR: the two PRs target different expert-parallel loading failures and should not be collapsed as the same change.", + "best_issue_reason": null, + "best_pr_reason": "45662 is the more concrete self-contained fix for a specific corruption bug (rank-0 broadcast overwriting experts), but it is still distinct from 45548's serialization/loading fix.", + "soft_edge_verdicts": [ + { + "left": "pull_request:45548", + "right": "pull_request:45662", + "accept": false, + "reason": "Different underlying bugs and code paths: 45548 is about `distributed_config` serialization and `has_ep` for ZeRO-3/accelerate, while 45662 prevents post-shard expert tensors from being overwritten during FSDP2 loading. They are related, not duplicates." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 23, + "component_count": 36, + "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": "2e5216b16e218040cf8dd58b78b50653a3ec9e971d035c167a69404252c866f6", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4589, + "estimated_input_tokens": 1148, + "estimated_eval_tokens": 2552 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4589, + "estimated_input_tokens": 1148, + "estimated_eval_tokens": 2552 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both PRs address the same xdist collision problem around captured_info debug artifacts. PR 45639 fixes the output-file write path in testing_utils; PR 45645 includes that fix plus CI/workflow changes to read multiple per-worker files, making it the more complete implementation of the same change.", + "confidence": 0.91, + "canonical_issue_reason": "Issue 45561 appears to be the shared underlying xdist-safe captured_info artifact problem referenced by both PRs.", + "canonical_pr_reason": "PR 45645 is the better canonical PR because it contains the core xdist-safe file-write fix and the necessary CI/workflow follow-up to consume multiple captured_info files.", + "best_issue_reason": "No standalone issue is present in the cluster; the shared target 45561 is the relevant tracking issue.", + "best_pr_reason": "PR 45645 is the best representative because it covers the full end-to-end fix, not just the writer-side change.", + "soft_edge_verdicts": [ + { + "left": "pull_request:45639", + "right": "pull_request:45645", + "accept": true, + "reason": "Both PRs fix the same concrete captured_info/xdist collision bug in testing_utils, and 45645 is an extension of the same fix with CI log aggregation." + } + ] + }, + "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": 36, + "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-45670-3", + "nodes": [ + "pull_request:45670", + "pull_request:46320", + "pull_request:46401" + ], + "soft_pairs": [ + "pull_request:45670|pull_request:46320", + "pull_request:45670|pull_request:46401", + "pull_request:46320|pull_request:46401" + ], + "prepared_review_unit_hash": "463fe5ab393f1cc3120a174248214125cff5e55fd390dcef745720e3a013b3ae", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 7525, + "estimated_input_tokens": 1882, + "estimated_eval_tokens": 4020 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 7525, + "estimated_input_tokens": 1882, + "estimated_eval_tokens": 4020 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are three separate PRs: one auto-model registration tweak and two model-specific attn_implementation dispatch additions. The latter two are related at a high level but target different model families and code paths, so they should not be merged as duplicates.", + "confidence": 0.95, + "canonical_issue_reason": null, + "canonical_pr_reason": "No clear duplicate canonical PR: 45670 is unrelated, and 46320/46401 are distinct model-specific dispatch implementations rather than the same concrete change.", + "best_issue_reason": null, + "best_pr_reason": "46401 is the broadest/most complete of the attn-dispatch PRs, but it is still not a duplicate of 46320; it best represents its own change set rather than a cluster canonical.", + "soft_edge_verdicts": [ + { + "left": "pull_request:45670", + "right": "pull_request:46320", + "accept": false, + "reason": "Unrelated changes: 45670 only updates AutoModel registration for glmasr/dia, while 46320 adds attn_implementation dispatch for RecurrentGemma in model code, docs, and tests." + }, + { + "left": "pull_request:45670", + "right": "pull_request:46401", + "accept": false, + "reason": "Unrelated changes: 45670 is an AutoModel mapping tweak, while 46401 adds attn_implementation dispatch for BLIP-2/InstructBLIP/InstructBLIPVideo plus docs/tests." + }, + { + "left": "pull_request:46320", + "right": "pull_request:46401", + "accept": false, + "reason": "Both are attn_implementation feature work, but they fix different model families with different code paths; they could not plausibly be merged into one duplicate PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 25, + "component_count": 36, + "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": "5900c32868f19603d2891623b9f0cb01b1e2c85f3fec422a621b8702a1a5f518", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 7248, + "estimated_input_tokens": 1812, + "estimated_eval_tokens": 3880 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 7248, + "estimated_input_tokens": 1812, + "estimated_eval_tokens": 3880 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These two PRs are both broad transformer model lint-rule changes, but they target different rules and different code paths: TRF003 tuple-return handling vs TRF018 model initialization cleanup. The overlap in model filenames looks incidental to the shared linter scope, not the same fix.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": "No canonical PR: the PRs implement different lint rules and fix different underlying behaviors, so they are not duplicates.", + "best_issue_reason": null, + "best_pr_reason": "Neither PR is a strong global representative for a duplicate cluster, but 45720 is slightly more representative of the shared-file overlap because it touches more of the same model forward methods; still, it should not be treated as the same change as 46259.", + "soft_edge_verdicts": [ + { + "left": "pull_request:45720", + "right": "pull_request:46259", + "accept": false, + "reason": "Different linter rules and different concrete code changes: one adjusts return-tuple handling in forwards, the other changes weight initialization logic. Shared model files alone are not enough to treat them as the same PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 26, + "component_count": 36, + "review_unit_index": 1, + "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": 27, + "component_count": 36, + "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-45932-2", + "nodes": [ + "pull_request:45932", + "pull_request:46100" + ], + "soft_pairs": [ + "pull_request:45932|pull_request:46100" + ], + "prepared_review_unit_hash": "e6a1df079b92efa81ef68892c2bca1504cc8cdd271d8fa6d29ee733fa318f303", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5083, + "estimated_input_tokens": 1271, + "estimated_eval_tokens": 2798 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5083, + "estimated_input_tokens": 1271, + "estimated_eval_tokens": 2798 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These two PRs are both test-related in a broad sense, but they target different changes: one is a noisy multi-file model/workflow update, the other is a focused ViT test migration. They do not look like the same underlying bug or change.", + "confidence": 0.94, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #46100 is the better representative because it is a single, coherent test-migration change with tightly scoped test-file edits, whereas PR #45932 is a broad mixed-change PR touching workflow plumbing and multiple model files.", + "best_issue_reason": null, + "best_pr_reason": "PR #46100 is the most self-contained and interpretable artifact in the set; #45932 appears to be a mixed-purpose or experimental PR and is not a good canonical anchor.", + "soft_edge_verdicts": [ + { + "left": "pull_request:45932", + "right": "pull_request:46100", + "accept": false, + "reason": "Both mention tests, but they are not the same concrete change: #45932 is a broad model/workflow update, while #46100 is a focused ViT test migration. They could not plausibly be merged as one PR for the same bug or feature." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 28, + "component_count": 36, + "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-46073-3", + "nodes": [ + "pull_request:46073", + "pull_request:46154", + "pull_request:46196" + ], + "soft_pairs": [ + "pull_request:46073|pull_request:46154", + "pull_request:46073|pull_request:46196", + "pull_request:46154|pull_request:46196" + ], + "prepared_review_unit_hash": "812777e820c55ba2c24a48095256e334cb4885dabf84e92d50ca19bfafe79d87", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 8400, + "estimated_input_tokens": 2100, + "estimated_eval_tokens": 4456 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 8400, + "estimated_input_tokens": 2100, + "estimated_eval_tokens": 4456 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Three model-specific PRs all add SDPA/attn_implementation dispatch support, but they target different concrete model code paths (CTRL, T5, and wav2vec2/audio family). They are thematically related, not true duplicates.", + "confidence": 0.9, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 46196 is the best representative of the cluster because it is the broadest and most substantial implementation, spanning multiple audio model files and the main code-path changes for SDPA dispatch.", + "best_issue_reason": null, + "best_pr_reason": "PR 46196 is the strongest cluster exemplar: it covers the largest set of affected models and contains the most complete attn_implementation dispatch work, while the others are narrower model-specific variants.", + "soft_edge_verdicts": [ + { + "left": "pull_request:46073", + "right": "pull_request:46154", + "accept": false, + "reason": "Both add attention-dispatch support, but for different model implementations (CTRL vs T5) with different code paths; this is a pattern match, not the same change." + }, + { + "left": "pull_request:46073", + "right": "pull_request:46196", + "accept": false, + "reason": "CTRL-specific dispatch support and wav2vec2/audio-family dispatch support are distinct model code paths and not plausibly one merged PR for the same bug/change." + }, + { + "left": "pull_request:46154", + "right": "pull_request:46196", + "accept": false, + "reason": "T5 and wav2vec2/audio-family SDPA dispatch patches target different architectures and different implementation details, so they are not duplicates." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 29, + "component_count": 36, + "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": "154028899b0b30f3ad9f821dd76ff7ebef7ac0a0a790eb65c74b04624304b270", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5523, + "estimated_input_tokens": 1381, + "estimated_eval_tokens": 3018 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5523, + "estimated_input_tokens": 1381, + "estimated_eval_tokens": 3018 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Two open PRs in the distributed area, but they are not duplicates: one is a focused EP/Trainer integration, the other is a broad distributed-branch base sweep. Shared files look like overlap from a common branch, not the same concrete change.", + "confidence": 0.88, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #46126 is the more focused, concrete change: EP + Trainer integration with targeted edits in the distributed/tensor-parallel path and trainer/config plumbing.", + "best_issue_reason": null, + "best_pr_reason": "PR #46126 is the best representative of a single actionable change. PR #46269 is too broad and spans many unrelated distributed subsystems, so it is not a good duplicate/merge target.", + "soft_edge_verdicts": [ + { + "left": "pull_request:46126", + "right": "pull_request:46269", + "accept": false, + "reason": "Reject: the overlap is only in shared distributed files; #46126 is a specific EP/Trainer integration, while #46269 is a large distributed base PR touching many unrelated areas. They do not appear to fix the same concrete code-path problem and would not plausibly merge into one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 30, + "component_count": 36, + "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-46138-9", + "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" + ], + "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:46504", + "pull_request:46138|pull_request:46432", + "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:46558|pull_request:46563", + "pull_request:46504|pull_request:46563", + "pull_request:46432|pull_request:46563" + ], + "prepared_review_unit_hash": "84bd9339541c09efceec87820aa81785499bb750413d45c8535023cecd1ea5cd", + "component_budget": { + "node_count": 9, + "item_count": 9, + "soft_pair_count": 26, + "serialized_chars": 23352, + "estimated_input_tokens": 5838, + "estimated_eval_tokens": 11932 + }, + "budget": { + "node_count": 9, + "item_count": 9, + "soft_pair_count": 26, + "serialized_chars": 23352, + "estimated_input_tokens": 5838, + "estimated_eval_tokens": 11932 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "This cluster is mostly a set of unrelated workflow-maintenance PRs touching different YAML files. The only near-duplicate subgroup is the pr-ci-security-gate hardening work, where one pair (46554/46558) looks mergeable into a single PR; the reusable-workflow rewrite (46563) is a different change path.", + "confidence": 0.64, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 46554 is the best representative anchor because it makes the most substantive, concrete change to the pr-ci-security-gate workflow by adding actual security checks and checkout behavior, whereas the rest of the cluster is either unrelated workflow churn or smaller follow-up hardening.", + "best_issue_reason": null, + "best_pr_reason": "PR 46554 is the strongest standalone representative of the only related subgroup in the cluster: it addresses the pr-ci-security-gate workflow with direct, code-path-level security checks and is broader than the nearby follow-up edits.", + "soft_edge_verdicts": [ + { + "left": "pull_request:46138", + "right": "pull_request:46554", + "accept": false, + "reason": "Different workflow files and different behavior: self-comment CI input handling vs PR security-gate hardening." + }, + { + "left": "pull_request:46138", + "right": "pull_request:46558", + "accept": false, + "reason": "Different workflow files and different behavior; only broad CI/workflow similarity, not the same underlying change." + }, + { + "left": "pull_request:46138", + "right": "pull_request:46563", + "accept": false, + "reason": "Self-comment CI stdin handling is unrelated to the pr-ci-security-gate workflow rewrite." + }, + { + "left": "pull_request:46138", + "right": "pull_request:46504", + "accept": false, + "reason": "These touch different workflows and different scripts/variables; no shared concrete bug or change." + }, + { + "left": "pull_request:46138", + "right": "pull_request:46432", + "accept": false, + "reason": "Different workflow files and unrelated fixes: comment parsing vs SSH runner security/runtime tweaks." + }, + { + "left": "pull_request:46432", + "right": "pull_request:46503", + "accept": false, + "reason": "SSH runner hardening and benchmark trigger logic are unrelated workflow changes." + }, + { + "left": "pull_request:46432", + "right": "pull_request:46501", + "accept": false, + "reason": "Different workflow files and different intent; no same underlying bug." + }, + { + "left": "pull_request:46432", + "right": "pull_request:46502", + "accept": false, + "reason": "SSH runner security/runtime changes do not match assign-reviewers checkout behavior." + }, + { + "left": "pull_request:46432", + "right": "pull_request:46554", + "accept": false, + "reason": "Different workflows and different problem domains; not the same concrete fix." + }, + { + "left": "pull_request:46432", + "right": "pull_request:46558", + "accept": false, + "reason": "Different workflow files and different behavior; only broad CI maintenance overlap." + }, + { + "left": "pull_request:46501", + "right": "pull_request:46503", + "accept": false, + "reason": "Anti-slop permissions cleanup and benchmark trigger support are unrelated changes." + }, + { + "left": "pull_request:46501", + "right": "pull_request:46502", + "accept": false, + "reason": "Different workflows and different fix targets; no common code-path issue." + }, + { + "left": "pull_request:46501", + "right": "pull_request:46504", + "accept": false, + "reason": "Unrelated workflow updates in different files." + }, + { + "left": "pull_request:46501", + "right": "pull_request:46554", + "accept": false, + "reason": "Anti-slop workflow permissions and pr-ci-security-gate hardening are different changes." + }, + { + "left": "pull_request:46502", + "right": "pull_request:46503", + "accept": false, + "reason": "Assign-reviewers checkout ref change and benchmark workflow_dispatch support are unrelated." + }, + { + "left": "pull_request:46502", + "right": "pull_request:46504", + "accept": false, + "reason": "Different workflow files and different behavior; no same underlying bug." + }, + { + "left": "pull_request:46502", + "right": "pull_request:46554", + "accept": false, + "reason": "Assign-reviewers and pr-ci-security-gate affect different workflows and code paths." + }, + { + "left": "pull_request:46503", + "right": "pull_request:46504", + "accept": false, + "reason": "Benchmark trigger logic and build-ci docker tag detection are unrelated changes." + }, + { + "left": "pull_request:46503", + "right": "pull_request:46554", + "accept": false, + "reason": "Different workflows and different functional changes." + }, + { + "left": "pull_request:46504", + "right": "pull_request:46554", + "accept": false, + "reason": "Build-ci docker image tagging and pr-ci-security-gate hardening are unrelated." + }, + { + "left": "pull_request:46504", + "right": "pull_request:46558", + "accept": false, + "reason": "Different workflows and different code paths; only generic YAML churn overlaps." + }, + { + "left": "pull_request:46554", + "right": "pull_request:46558", + "accept": true, + "reason": "Same pr-ci-security-gate workflow hardening path; both changes target the fork-PR security gate and could plausibly be merged into one PR." + }, + { + "left": "pull_request:46554", + "right": "pull_request:46563", + "accept": false, + "reason": "46554 adds local security checks, while 46563 switches to a reusable workflow and adds notes; different implementation paths." + }, + { + "left": "pull_request:46558", + "right": "pull_request:46563", + "accept": false, + "reason": "Permission/approval tightening in the local workflow is not the same change as replacing it with a reusable workflow." + }, + { + "left": "pull_request:46504", + "right": "pull_request:46563", + "accept": false, + "reason": "Build-ci docker image workflow changes are unrelated to the pr-ci-security-gate workflow." + }, + { + "left": "pull_request:46432", + "right": "pull_request:46563", + "accept": false, + "reason": "SSH runner workflow hardening is unrelated to the pr-ci-security-gate reusable-workflow migration." + } + ] + }, + "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": 36, + "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": "ee22f6e2763b52b5469bfd8adb2b41320a7f392143f99a8b050f9d4dbafc9e34", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4276, + "estimated_input_tokens": 1069, + "estimated_eval_tokens": 2394 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4276, + "estimated_input_tokens": 1069, + "estimated_eval_tokens": 2394 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are two separate pull requests that both replace source-file opening with source inspection, but they apply to different code paths and problems: one in import_utils for resolving package resources, the other in modeling_utils for attention-implementation detection.", + "confidence": 0.88, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 46152 is the cleaner canonical representative: it introduces a reusable source/file-resolution fix in import_utils, while 46207 is a separate function-specific change in modeling_utils.", + "best_issue_reason": null, + "best_pr_reason": "PR 46152 is the best single representative for the cluster because it is the broader, more self-contained source-reading fix and is not dependent on the unrelated modeling_utils change.", + "soft_edge_verdicts": [ + { + "left": "pull_request:46152", + "right": "pull_request:46207", + "accept": false, + "reason": "Both PRs modernize source access, but they fix different concrete paths in different modules with different behavior changes. They are not the same underlying bug and would not plausibly merge into one PR as duplicates." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 32, + "component_count": 36, + "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-46173-3", + "nodes": [ + "pull_request:46173", + "pull_request:46436", + "pull_request:46474" + ], + "soft_pairs": [ + "pull_request:46436|pull_request:46474", + "pull_request:46173|pull_request:46474" + ], + "prepared_review_unit_hash": "dabde6cd411e1011f9e3eb17a6894b97d8023dc5b980ff3585b2e01529c14b91", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 5920, + "estimated_input_tokens": 1480, + "estimated_eval_tokens": 3216 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 5920, + "estimated_input_tokens": 1480, + "estimated_eval_tokens": 3216 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "PRs #46173 and #46474 are duplicates: both make the same StaticLayer.get_seq_length type-hint change in src/transformers/cache_utils.py for issue #45987. PR #46436 is separate, changing static cache max length handling in generation/utils.py.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": "#46173 is the strongest canonical representative because it has the same cache_utils type-hint fix, an explicit link to issue #45987, and a clear, specific title.", + "best_issue_reason": null, + "best_pr_reason": "#46173 is the best representative of the duplicate pair: it directly targets the underlying StaticLayer.get_seq_length annotation fix and is explicitly tied to the tracking issue.", + "soft_edge_verdicts": [ + { + "left": "pull_request:46436", + "right": "pull_request:46474", + "accept": false, + "reason": "Different changes and code paths: one adjusts static cache max_len selection in generation/utils.py, the other only changes a return type hint in cache_utils.py." + }, + { + "left": "pull_request:46173", + "right": "pull_request:46474", + "accept": true, + "reason": "Same underlying bug/change: both update StaticLayer.get_seq_length return type annotation in cache_utils.py, with identical intent and nearly identical diffs." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet. The 46173/46474 pair is supported by the shared issue target, same filename, and matching type-hint diff, while 46436 is clearly a different change in a different code path. The soft-edge verdicts are conservative enough, and I don\u2019t see overstatement beyond what the diffs support." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 33, + "component_count": 36, + "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": "db1633bb030374040929897c0f2d071190f03fc5e089fd500803a2c5e4d621a1", + "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 Moonshine streaming sliding-window boundary bug (`-dist < right_window_size` vs `<=`), with 46546 appearing to be the fuller follow-up/superset change. The pair should be clustered together as duplicate fixes for the same underlying issue.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 46546 is the better canonical PR: it fixes the same boundary bug in Moonshine streaming, includes the related test update, and appears to be the more complete version of the change.", + "best_issue_reason": null, + "best_pr_reason": "PR 46546 is the strongest representative because it addresses the concrete bug in the target code path and adds tests; PR 46310 is the narrower earlier version of the same fix.", + "soft_edge_verdicts": [ + { + "left": "pull_request:46310", + "right": "pull_request:46546", + "accept": true, + "reason": "Same underlying Moonshine streaming right-window boundary bug in the same mask function; 46546 is effectively a broader version of the same fix and could plausibly subsume 46310." + } + ] + }, + "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": 36, + "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-46376-2", + "nodes": [ + "pull_request:46376", + "pull_request:46417" + ], + "soft_pairs": [ + "pull_request:46376|pull_request:46417" + ], + "prepared_review_unit_hash": "d30254547471604ad9340068dcd2954eacd2d40a42f72757687df0f7fb9f7fc8", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5939, + "estimated_input_tokens": 1485, + "estimated_eval_tokens": 3226 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5939, + "estimated_input_tokens": 1485, + "estimated_eval_tokens": 3226 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "The two pull requests are unrelated: one adds multimodal support to continuous batching infrastructure, while the other adds the HunYuan VL model and its docs/tests. They do not look like the same underlying change and should not be merged as duplicates.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #46417 is the better representative because it is a coherent, self-contained feature addition with extensive model, processor, docs, and tests coverage.", + "best_issue_reason": null, + "best_pr_reason": "PR #46417 is the strongest standalone PR in the set: it is a complete model integration rather than an in-progress infrastructure refactor.", + "soft_edge_verdicts": [ + { + "left": "pull_request:46376", + "right": "pull_request:46417", + "accept": false, + "reason": "Different change families and code paths: continuous batching/multimodal runtime support versus adding a new HunYuan VL model package, docs, and tests. Shared mention of multimodality is too broad to treat as a duplicate." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 35, + "component_count": 36, + "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": "e47f1ab560366d9e11fda24625d2c0e3e6b50b20841e84ad7cc4e5cb4d91c2a0", + "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": "These PRs are unrelated tokenizer fixes: one changes CodeLlama decoding to preserve real leading spaces, while the other changes AutoTokenizer checkpoint routing for specific DeepSeek models that load the wrong tokenizer class.", + "confidence": 0.99, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "pull_request:46574", + "right": "pull_request:46580", + "accept": false, + "reason": "Different bug classes and code paths: CodeLlama decoder stripping vs. DeepSeek model/tokenizer-class misrouting. They are not the same change and would not plausibly merge into one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 36, + "component_count": 36, + "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": "608abd69ddb2d867b72e1ac4feb9f089536abefbeafc9f17ede5e49729d99743", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5495, + "estimated_input_tokens": 1374, + "estimated_eval_tokens": 3004 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5495, + "estimated_input_tokens": 1374, + "estimated_eval_tokens": 3004 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These two Mistral PRs touch the same files but implement different changes: #46603 is a refactor that moves MistralConverter into the integrations package, while #46604 adds tekken tokenizer detection/conversion/save utilities. They are related, but not the same underlying change, so they should not be clustered as duplicates.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": "#46604 is the more substantive feature PR and best represents the cluster\u2019s Mistral tokenizer work; #46603 is primarily a code move/refactor.", + "best_issue_reason": null, + "best_pr_reason": "#46604 is the best representative PR because it adds the actual new tokenizer functionality, whereas #46603 is mostly an organizational extraction.", + "soft_edge_verdicts": [ + { + "left": "pull_request:46603", + "right": "pull_request:46604", + "accept": false, + "reason": "Different code-path changes: #46603 relocates MistralConverter, while #46604 implements tekken detection/conversion/save support. Shared files alone are not enough to treat them as the same PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + } + ] +} diff --git a/snapshots/20260616T120046Z/analysis-runs/hybrid-model-20260616t120046z/manifest.json b/snapshots/20260616T120046Z/analysis-runs/hybrid-model-20260616t120046z/manifest.json new file mode 100644 index 0000000000000000000000000000000000000000..25f6991f1db31c6e28c315ffd352f869c62abf3d --- /dev/null +++ b/snapshots/20260616T120046Z/analysis-runs/hybrid-model-20260616t120046z/manifest.json @@ -0,0 +1,14 @@ +{ + "analysis_id": "hybrid-model-20260616t120046z", + "artifacts": { + "hybrid": "snapshots/20260616T120046Z/analysis-runs/hybrid-model-20260616t120046z/analysis-report-hybrid.json", + "hybrid_reviews": "snapshots/20260616T120046Z/analysis-runs/hybrid-model-20260616t120046z/analysis-report-hybrid.llm-reviews.json" + }, + "channel": "canonical", + "model": null, + "published_at": "2026-06-16T12:10:43Z", + "repo": "huggingface/transformers", + "schema_version": 1, + "snapshot_id": "20260616T120046Z", + "variant": "hybrid" +} diff --git a/snapshots/20260616T120046Z/manifest.json b/snapshots/20260616T120046Z/manifest.json index 78a79712458e45c143327dddba3587a1cbff1243..fed600042a8426013b17c7a6648a4294447ca660 100644 --- a/snapshots/20260616T120046Z/manifest.json +++ b/snapshots/20260616T120046Z/manifest.json @@ -32,6 +32,24 @@ "timeline_events": 1203 }, "extracted_at": "2026-06-16T12:00:46Z", + "published_analysis": { + "canonical_analysis_id": "hybrid-model-20260616t120046z", + "runs": { + "hybrid-model-20260616t120046z": { + "analysis_id": "hybrid-model-20260616t120046z", + "artifacts": { + "hybrid": "snapshots/20260616T120046Z/analysis-runs/hybrid-model-20260616t120046z/analysis-report-hybrid.json", + "hybrid_reviews": "snapshots/20260616T120046Z/analysis-runs/hybrid-model-20260616t120046z/analysis-report-hybrid.llm-reviews.json" + }, + "channel": "canonical", + "manifest_path": "snapshots/20260616T120046Z/analysis-runs/hybrid-model-20260616t120046z/manifest.json", + "model": null, + "published_at": "2026-06-16T12:10:43Z", + "variant": "hybrid" + } + }, + "schema_version": 1 + }, "repo": "huggingface/transformers", "snapshot_id": "20260616T120046Z", "watermark": {